aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1703.07518 | 2949709872 | Social media expose millions of users every day to information campaigns --- some emerging organically from grassroots activity, others sustained by advertising or other coordinated efforts. These campaigns contribute to the shaping of collective opinions. While most information campaigns are benign, some may be deployed for nefarious purposes. It is therefore important to be able to detect whether a meme is being artificially promoted at the very moment it becomes wildly popular. This problem has important social implications and poses numerous technical challenges. As a first step, here we focus on discriminating between trending memes that are either organic or promoted by means of advertisement. The classification is not trivial: ads cause bursts of attention that can be easily mistaken for those of organic trends. We designed a machine learning framework to classify memes that have been labeled as trending on Twitter.After trending, we can rely on a large volume of activity data. Early detection, occurring immediately at trending time, is a more challenging problem due to the minimal volume of activity data that is available prior to trending.Our supervised learning framework exploits hundreds of time-varying features to capture changing network and diffusion patterns, content and sentiment information, timing signals, and user meta-data. We explore different methods for encoding feature time series. Using millions of tweets containing trending hashtags, we achieve 75 AUC score for early detection, increasing to above 95 after trending. We evaluate the robustness of the algorithms by introducing random temporal shifts on the trend time series. Feature selection analysis reveals that content cues provide consistently useful signals; user features are more informative for early detection, while network and timing features are more helpful once more data is available. | The present work, to the best of our knowledge, is the first to investigate the early detection of promoted content on social media. We focus our attention on advertisement, which can play an important role in information campaigns. Trending memes are considered an indicator of collective attention in social media @cite_3 @cite_64 , and as such they have been used to predict real-world events, like the winner of a popular reality TV show @cite_6 . Although emerging from collective attention, communication on social media can be manipulated, for example for political gain, as in the case of astroturf @cite_44 @cite_17 . | {
"cite_N": [
"@cite_64",
"@cite_6",
"@cite_3",
"@cite_44",
"@cite_17"
],
"mid": [
"2952051872",
"2113321055",
"2058465497",
"",
"202178741"
],
"abstract": [
"Micro-blogging systems such as Twitter expose digital traces of social discourse with an unprecedented degree of resolution of individual behaviors. They offer an opportunity to investigate how a large-scale social system responds to exogenous or endogenous stimuli, and to disentangle the temporal, spatial and topical aspects of users' activity. Here we focus on spikes of collective attention in Twitter, and specifically on peaks in the popularity of hashtags. Users employ hashtags as a form of social annotation, to define a shared context for a specific event, topic, or meme. We analyze a large-scale record of Twitter activity and find that the evolution of hastag popularity over time defines discrete classes of hashtags. We link these dynamical classes to the events the hashtags represent and use text mining techniques to provide a semantic characterization of the hastag classes. Moreover, we track the propagation of hashtags in the Twitter social network and find that epidemic spreading plays a minor role in hastag popularity, which is mostly driven by exogenous factors.",
"We present a contribution to the debate on the predictability of social events using big data analytics. We focus on the elimination of contestants in the American Idol TV shows as an example of a well defined electoral phenomenon that each week draws millions of votes in the USA. This event can be considered as basic test in a simplified environment to assess the predictive power of Twitter signals. We provide evidence that Twitter activity during the time span defined by the TV show airing and the voting period following it correlates with the contestants ranking and allows the anticipation of the voting outcome. Twitter data from the show and the voting period of the season finale have been analyzed to attempt the winner prediction ahead of the airing of the official result. We also show that the fraction of tweets that contain geolocation information allows us to map the fanbase of each contestant, both within the US and abroad, showing that strong regional polarizations occur. The geolocalized data are crucial for the correct prediction of the final outcome of the show, pointing out the importance of considering information beyond the aggregated Twitter signal. Although American Idol voting is just a minimal and simplified version of complex societ al phenomena such as political elections, this work shows that the volume of information available in online systems permits the real time gathering of quantitative indicators that may be able to anticipate the future unfolding of opinion formation events.",
"The subject of collective attention is central to an information age where millions of people are inundated with daily messages. It is thus of interest to understand how attention to novel items propagates and eventually fades among large populations. We have analyzed the dynamics of collective attention among 1 million users of an interactive web site, digg.com, devoted to thousands of novel news stories. The observations can be described by a dynamical model characterized by a single novelty factor. Our measurements indicate that novelty within groups decays with a stretched-exponential law, suggesting the existence of a natural time scale over which attention fades.",
"",
"We study astroturf political campaigns on microblogging platforms: politically-motivated individuals and organizations that use multiple centrally-controlled accounts to create the appearance of widespread support for a candidate or opinion. We describe a machine learning framework that combines topological, content-based and crowdsourced features of information diffusion networks on Twitter to detect the early stages of viral spreading of political misinformation. We present promising preliminary results with better than 96 accuracy in the detection of astroturf content in the run-up to the 2010 U.S. midterm elections."
]
} |
1703.07518 | 2949709872 | Social media expose millions of users every day to information campaigns --- some emerging organically from grassroots activity, others sustained by advertising or other coordinated efforts. These campaigns contribute to the shaping of collective opinions. While most information campaigns are benign, some may be deployed for nefarious purposes. It is therefore important to be able to detect whether a meme is being artificially promoted at the very moment it becomes wildly popular. This problem has important social implications and poses numerous technical challenges. As a first step, here we focus on discriminating between trending memes that are either organic or promoted by means of advertisement. The classification is not trivial: ads cause bursts of attention that can be easily mistaken for those of organic trends. We designed a machine learning framework to classify memes that have been labeled as trending on Twitter.After trending, we can rely on a large volume of activity data. Early detection, occurring immediately at trending time, is a more challenging problem due to the minimal volume of activity data that is available prior to trending.Our supervised learning framework exploits hundreds of time-varying features to capture changing network and diffusion patterns, content and sentiment information, timing signals, and user meta-data. We explore different methods for encoding feature time series. Using millions of tweets containing trending hashtags, we achieve 75 AUC score for early detection, increasing to above 95 after trending. We evaluate the robustness of the algorithms by introducing random temporal shifts on the trend time series. Feature selection analysis reveals that content cues provide consistently useful signals; user features are more informative for early detection, while network and timing features are more helpful once more data is available. | Recent work analyzes emerging topics, memes, and conversations triggered by real world events @cite_57 @cite_76 @cite_48 . Studies of information dissemination reveal mechanisms governing content production and consumption @cite_63 as well as prediction of future content popularity. Cheng study the prediction of photo-sharing cascade size @cite_31 and recurrence @cite_53 on Facebook. Machine learning models can predict future popularity of emerging hashtags and content on social media @cite_8 @cite_22 . Features extracted from content @cite_62 , sentiment @cite_67 @cite_74 , community structure @cite_38 @cite_54 , and temporal signatures @cite_13 @cite_86 @cite_71 are commonly used to train such models. In this paper we leverage similar features, but for the novel task of campaign detection. Furthermore, our task is more challenging because we deal with dynamic features whose changes over time are captured in high-dimensional time series. | {
"cite_N": [
"@cite_38",
"@cite_67",
"@cite_62",
"@cite_22",
"@cite_8",
"@cite_48",
"@cite_53",
"@cite_54",
"@cite_57",
"@cite_63",
"@cite_74",
"@cite_71",
"@cite_86",
"@cite_31",
"@cite_76",
"@cite_13"
],
"mid": [
"2156716308",
"",
"2164273822",
"2171410332",
"2114544578",
"1983012012",
"2257559988",
"",
"",
"1712130609",
"178030352",
"863049760",
"1987986834",
"1996263819",
"",
"2037970810"
],
"abstract": [
"How does network structure affect diffusion? Recent studies suggest that the answer depends on the type of contagion. Complex contagions, unlike infectious diseases (simple contagions), are affected by social reinforcement and homophily. Hence, the spread within highly clustered communities is enhanced, while diffusion across communities is hampered. A common hypothesis is that memes and behaviors are complex contagions. We show that, while most memes indeed spread like complex contagions, a few viral memes spread across many communities, like diseases. We demonstrate that the future popularity of a meme can be predicted by quantifying its early spreading pattern in terms of community concentration. The more communities a meme permeates, the more viral it is. We present a practical method to translate data about community structure into predictive knowledge about what information will spread widely. This connection contributes to our understanding in computational social science, social media analytics, and marketing applications.",
"",
"Using comment information available from Digg we define a co-participation network between users. We focus on the analysis of this implicit network, and study the behavioral characteristics of users. Using an entropy measure, we infer that users at Digg are not highly focused and participate across a wide range of topics. We also use the comment data and social network derived features to predict the popularity of online content linked at Digg using a classification and regression framework. We show promising results for predicting the popularity scores even after limiting our feature extraction to the first few hours of comment activity that follows a Digg submission.",
"Because of Twitter’s popularity and the viral nature of information dissemination on Twitter, predicting which Twitter topics will become popular in the near future becomes a task of considerable economic importance. Many Twitter topics are annotated by hashtags. In this article, we propose methods to predict the popularity of new hashtags on Twitter by formulating the problem as a classification task. We use five standard classification models (i.e., Naive bayes, k-nearest neighbors, decision trees, support vector machines, and logistic regression) for prediction. The main challenge is the identification of effective features for describing new hashtags. We extract 7 content features from a hashtag string and the collection of tweets containing the hashtag and 11 contextual features from the social graph formed by users who have adopted the hashtag. We conducted experiments on a Twitter data set consisting of 31 million tweets from 2 million Singapore-based users. The experimental results show that the standard classifiers using the extracted features significantly outperform the baseline methods that do not use these features. Among the five classifiers, the logistic regression model performs the best in terms of the Micro-F1 measure. We also observe that contextual features are more effective than content features.",
"Current social media research mainly focuses on temporal trends of the information flow and on the topology of the social graph that facilitates the propagation of information. In this paper we study the effect of the content of the idea on the information propagation. We present an efficient hybrid approach based on a linear regression for predicting the spread of an idea in a given time frame. We show that a combination of content features with temporal and topological features minimizes prediction error. Our algorithm is evaluated on Twitter hashtags extracted from a dataset of more than 400 million tweets. We analyze the contribution and the limitations of the various feature types to the spread of information, demonstrating that content aspects can be used as strong predictors thus should not be disregarded. We also study the dependencies between global features such as graph topology and content features.",
"Twitter is a user-generated content system that allows its users to share short text messages, called tweets, for a variety of purposes, including daily conversations, URLs sharing and information news. Considering its world-wide distributed network of users of any age and social condition, it represents a low level news flashes portal that, in its impressive short response time, has the principal advantage. In this paper we recognize this primary role of Twitter and we propose a novel topic detection technique that permits to retrieve in real-time the most emergent topics expressed by the community. First, we extract the contents (set of terms) of the tweets and model the term life cycle according to a novel aging theory intended to mine the emerging ones. A term can be defined as emerging if it frequently occurs in the specified time interval and it was relatively rare in the past. Moreover, considering that the importance of a content also depends on its source, we analyze the social relationships in the network with the well-known Page Rank algorithm in order to determine the authority of the users. Finally, we leverage a navigable topic graph which connects the emerging terms with other semantically related keywords, allowing the detection of the emerging topics, under user-specified time constraints. We provide different case studies which show the validity of the proposed approach.",
"Cascades of information-sharing are a primary mechanism by which content reaches its audience on social media, and an active line of research has studied how such cascades, which form as content is reshared from person to person, develop and subside. In this paper, we perform a large-scale analysis of cascades on Facebook over significantly longer time scales, and find that a more complex picture emerges, in which many large cascades recur, exhibiting multiple bursts of popularity with periods of quiescence in between. We characterize recurrence by measuring the time elapsed between bursts, their overlap and proximity in the social network, and the diversity in the demographics of individuals participating in each peak. We discover that content virality, as revealed by its initial popularity, is a main driver of recurrence, with the availability of multiple copies of that content helping to spark new bursts. Still, beyond a certain popularity of content, the rate of recurrence drops as cascades start exhausting the population of interested individuals. We reproduce these observed patterns in a simple model of content recurrence simulated on a real social network. Using only characteristics of a cascade's initial burst, we demonstrate strong performance in predicting whether it will recur in the future.",
"",
"",
"Online traces of human activity offer novel opportunities to study the dynamics of complex knowledge exchange networks, in particular how emergent patterns of collective attention determine what new information is generated and consumed. Can we measure the relationship between demand and supply for new information about a topic? We propose a normalization method to compare attention bursts statistics across topics with heterogeneous distribution of attention. Through analysis of a massive dataset on traffic to Wikipedia, we find that the production of new knowledge is associated to significant shifts of collective attention, which we take as proxy for its demand. This is consistent with a scenario in which allocation of attention toward a topic stimulates the demand for information about it, and in turn the supply of further novel information. However, attention spikes only for a limited time span, during which new content has higher chances of receiving traffic, compared to content created later or earlier on. Our attempt to quantify demand and supply of information, and our finding about their temporal ordering, may lead to the development of the fundamental laws of the attention economy, and to a better understanding of social exchange of knowledge information networks.",
"This paper introduces a new Web mining approach that combines social network analysis and automatic sentiment analysis. We show how weighting the forum posts of the contributors according to their network position allow us to predict trends and real world events in the movie business. To test our approach we conducted two experiments analyzing online forum discussions on the Internet movie database (IMDb) by examining the correlation of the social network structure with external metrics such as box office revenue and Oscar Awards. We find that discussion patterns on IMDb predict Academy Awards nominations and box office success. Two months before the Oscars were given we were able to correctly predict nine Oscar nominations. We also found that forum contributions correlated with box office success of 20 top grossing movies of 2006.",
"Studying the bursty nature of cascades in social media is practically important in many applications such as product sales prediction, disaster relief, and stock market prediction. Although the cascade volume prediction has been extensively studied, how to predict when a burst will come remains an open problem. It is challenging to predict the time of the burst due to the \"quick rise and fall\" pattern and the diverse time spans of the cascades. To this end, this paper proposes a classification based approach for burst time prediction by utilizing and modeling rich knowledge in information diffusion. Particularly, we first propose a time window based approach to predict in which time window the burst will appear. This paves the way to transform the time prediction task to a classification problem. To address the challenge that the original time series data of the cascade popularity only are not sufficient for predicting cascades with diverse magnitudes and time spans, we explore rich information diffusion related knowledge and model them in a scale-independent manner. Extensive experiments on a Sina Weibo reposting dataset demonstrate the superior performance of the proposed approach in accurately predicting the burst time of posts.",
"Understanding content popularity growth is of great importance to Internet service providers, content creators and online marketers. In this work, we characterize the growth patterns of video popularity on the currently most popular video sharing application, namely YouTube. Using newly provided data by the application, we analyze how the popularity of individual videos evolves since the video's upload time. Moreover, addressing a key aspect that has been mostly overlooked by previous work, we characterize the types of the referrers that most often attracted users to each video, aiming at shedding some light into the mechanisms (e.g., searching or external linking) that often drive users towards a video, and thus contribute to popularity growth. Our analyses are performed separately for three video datasets, namely, videos that appear in the YouTube top lists, videos removed from the system due to copyright violation, and videos selected according to random queries submitted to YouTube's search engine. Our results show that popularity growth patterns depend on the video dataset. In particular, copyright protected videos tend to get most of their views much earlier in their lifetimes, often exhibiting a popularity growth characterized by a viral epidemic-like propagation process. In contrast, videos in the top lists tend to experience sudden significant bursts of popularity. We also show that not only search but also other YouTube internal mechanisms play important roles to attract users to videos in all three datasets.",
"On many social networking web sites such as Facebook and Twitter, resharing or reposting functionality allows users to share others' content with their own friends or followers. As content is reshared from user to user, large cascades of reshares can form. While a growing body of research has focused on analyzing and characterizing such cascades, a recent, parallel line of work has argued that the future trajectory of a cascade may be inherently unpredictable. In this work, we develop a framework for addressing cascade prediction problems. On a large sample of photo reshare cascades on Facebook, we find strong performance in predicting whether a cascade will continue to grow in the future. We find that the relative growth of a cascade becomes more predictable as we observe more of its reshares, that temporal and structural features are key predictors of cascade size, and that initially, breadth, rather than depth in a cascade is a better indicator of larger cascades. This prediction performance is robust in the sense that multiple distinct classes of features all achieve similar performance. We also discover that temporal features are predictive of a cascade's eventual shape. Observing independent cascades of the same content, we find that while these cascades differ greatly in size, we are still able to predict which ends up the largest.",
"",
"Predicting Web content popularity is an important task for supporting the design and evaluation of a wide range of systems, from targeted advertising to effective search and recommendation services. We here present two simple models for predicting the future popularity of Web content based on historical information given by early popularity measures. Our approach is validated on datasets consisting of videos from the widely used YouTube video-sharing portal. Our experimental results show that, compared to a state-of-the-art baseline model, our proposed models lead to significant decreases in relative squared errors, reaching up to 20 reduction on average, and larger reductions (of up to 71 ) for videos that experience a high peak in popularity in their early days followed by a sharp decrease in popularity."
]
} |
1703.07518 | 2949709872 | Social media expose millions of users every day to information campaigns --- some emerging organically from grassroots activity, others sustained by advertising or other coordinated efforts. These campaigns contribute to the shaping of collective opinions. While most information campaigns are benign, some may be deployed for nefarious purposes. It is therefore important to be able to detect whether a meme is being artificially promoted at the very moment it becomes wildly popular. This problem has important social implications and poses numerous technical challenges. As a first step, here we focus on discriminating between trending memes that are either organic or promoted by means of advertisement. The classification is not trivial: ads cause bursts of attention that can be easily mistaken for those of organic trends. We designed a machine learning framework to classify memes that have been labeled as trending on Twitter.After trending, we can rely on a large volume of activity data. Early detection, occurring immediately at trending time, is a more challenging problem due to the minimal volume of activity data that is available prior to trending.Our supervised learning framework exploits hundreds of time-varying features to capture changing network and diffusion patterns, content and sentiment information, timing signals, and user meta-data. We explore different methods for encoding feature time series. Using millions of tweets containing trending hashtags, we achieve 75 AUC score for early detection, increasing to above 95 after trending. We evaluate the robustness of the algorithms by introducing random temporal shifts on the trend time series. Feature selection analysis reveals that content cues provide consistently useful signals; user features are more informative for early detection, while network and timing features are more helpful once more data is available. | The proposed framework is based on a mixture of features common in social media data, including emotional and sentiment information. The literature has reported extensively on the use of social media content to describe emotional and demographic characteristics of users @cite_11 @cite_67 @cite_1 . The use of language in online communities is the focus of two recent papers @cite_79 @cite_82 : the authors observe that the language of social media users evolves, and common patterns emerge over time. The language style of users adapts to achieve better fitness in the conversation @cite_35 . These findings suggest that language contains strong signals, in particular if studied in conjunction with other dimensions of the data. Our study confirms the importance of content for campaign detection. | {
"cite_N": [
"@cite_35",
"@cite_67",
"@cite_11",
"@cite_1",
"@cite_79",
"@cite_82"
],
"mid": [
"",
"",
"2167102709",
"2123329834",
"2127411301",
"2951727499"
],
"abstract": [
"",
"",
"Every second, the thoughts and feelings of millions of people across the world are recorded in the form of 140-character tweets using Twitter. However, despite the enormous potential presented by this remarkable data source, we still do not have an understanding of the Twitter population itself: Who are the Twitter users? How representative of the overall population are they? In this paper, we take the first steps towards answering these questions by analyzing data on a set of Twitter users representing over 1 of the U.S. population. We develop techniques that allow us to compare the Twitter population to the U.S. population along three axes (geography, gender, and race ethnicity), and find that the Twitter population is a highly non-uniform sample of the population.",
"We conduct a detailed investigation of correlations between real-time expressions of individuals made across the United States and a wide range of emotional, geographic, demographic, and health characteristics. We do so by combining (1) a massive, geo-tagged data set comprising over 80 million words generated in 2011 on the social network service Twitter and (2) annually-surveyed characteristics of all 50 states and close to 400 urban populations. Among many results, we generate taxonomies of states and cities based on their similarities in word use; estimate the happiness levels of states and cities; correlate highly-resolved demographic characteristics with happiness levels; and connect word choice and message length with urban characteristics such as education levels and obesity rates. Our results show how social media may potentially be used to estimate real-time levels and changes in population-scale measures such as obesity rates.",
"Vibrant online communities are in constant flux. As members join and depart, the interactional norms evolve, stimulating further changes to the membership and its social dynamics. Linguistic change --- in the sense of innovation that becomes accepted as the norm --- is essential to this dynamic process: it both facilitates individual expression and fosters the emergence of a collective identity. We propose a framework for tracking linguistic change as it happens and for understanding how specific users react to these evolving norms. By applying this framework to two large online communities we show that users follow a determined two-stage lifecycle with respect to their susceptibility to linguistic change: a linguistically innovative learning phase in which users adopt the language of the community followed by a conservative phase in which users stop changing and the evolving community norms pass them by. Building on this observation, we show how this framework can be used to detect, early in a user's career, how long she will stay active in the community. Thus, this work has practical significance for those who design and maintain online communities. It also yields new theoretical insights into the evolution of linguistic norms and the complex interplay between community-level and individual-level linguistic change.",
"Recommending products to consumers means not only understanding their tastes, but also understanding their level of experience. For example, it would be a mistake to recommend the iconic film Seven Samurai simply because a user enjoys other action movies; rather, we might conclude that they will eventually enjoy it -- once they are ready. The same is true for beers, wines, gourmet foods -- or any products where users have acquired tastes: the best' products may not be the most accessible'. Thus our goal in this paper is to recommend products that a user will enjoy now, while acknowledging that their tastes may have changed over time, and may change again in the future. We model how tastes change due to the very act of consuming more products -- in other words, as users become more experienced. We develop a latent factor recommendation system that explicitly accounts for each user's level of experience. We find that such a model not only leads to better recommendations, but also allows us to study the role of user experience and expertise on a novel dataset of fifteen million beer, wine, food, and movie reviews."
]
} |
1703.07518 | 2949709872 | Social media expose millions of users every day to information campaigns --- some emerging organically from grassroots activity, others sustained by advertising or other coordinated efforts. These campaigns contribute to the shaping of collective opinions. While most information campaigns are benign, some may be deployed for nefarious purposes. It is therefore important to be able to detect whether a meme is being artificially promoted at the very moment it becomes wildly popular. This problem has important social implications and poses numerous technical challenges. As a first step, here we focus on discriminating between trending memes that are either organic or promoted by means of advertisement. The classification is not trivial: ads cause bursts of attention that can be easily mistaken for those of organic trends. We designed a machine learning framework to classify memes that have been labeled as trending on Twitter.After trending, we can rely on a large volume of activity data. Early detection, occurring immediately at trending time, is a more challenging problem due to the minimal volume of activity data that is available prior to trending.Our supervised learning framework exploits hundreds of time-varying features to capture changing network and diffusion patterns, content and sentiment information, timing signals, and user meta-data. We explore different methods for encoding feature time series. Using millions of tweets containing trending hashtags, we achieve 75 AUC score for early detection, increasing to above 95 after trending. We evaluate the robustness of the algorithms by introducing random temporal shifts on the trend time series. Feature selection analysis reveals that content cues provide consistently useful signals; user features are more informative for early detection, while network and timing features are more helpful once more data is available. | Finally, our system builds on network features and diffusion patterns of social media messages. Network structure and information diffusion in social media have been studied extensively @cite_72 @cite_49 . Network features are highly predictive of certain types of social media abuse, like astroturf, that attempt to simulate grassroots online conversations @cite_17 @cite_36 @cite_19 @cite_81 @cite_60 . Such artificial campaigns produce peculiar patterns of information diffusion: the topology of retweet or mention networks is often a stronger signal than content or language. The present findings are consistent with this body of work, as network features are helpful in detecting promoted content after trending. | {
"cite_N": [
"@cite_60",
"@cite_36",
"@cite_19",
"@cite_72",
"@cite_81",
"@cite_49",
"@cite_17"
],
"mid": [
"2595521492",
"2100974526",
"2278635123",
"2432978112",
"1837843568",
"1752870744",
"202178741"
],
"abstract": [
"Increasing evidence suggests that a growing amount of social media content is generated by autonomous entities known as social bots. In this work we present a framework to detect such entities on Twitter. We leverage more than a thousand features extracted from public data and meta-data about users: friends, tweet content and sentiment, network patterns, and activity time series. We benchmark the classification framework by using a publicly available dataset of Twitter bots. This training data is enriched by a manually annotated collection of active Twitter users that include both humans and bots of varying sophistication. Our models yield high accuracy and agreement with each other and can detect bots of different nature. Our estimates suggest that between 9 and 15 of active Twitter accounts are bots. Characterizing ties among accounts, we observe that simple bots tend to interact with bots that exhibit more human-like behaviors. Analysis of content flows reveals retweet and mention strategies adopted by bots to interact with different target groups. Using clustering analysis, we characterize several subclasses of accounts, including spammers, self promoters, and accounts that post content from connected applications.",
"Online social media are complementing and in some cases replacing person-to-person social interaction and redefining the diffusion of information. In particular, microblogs have become crucial grounds on which public relations, marketing, and political battles are fought. We demonstrate a web service that tracks political memes in Twitter and helps detect astroturfing, smear campaigns, and other misinformation in the context of U.S. political elections. We also present some cases of abusive behaviors uncovered by our service. Our web service is based on an extensible framework that will enable the real-time analysis of meme diffusion in social media by mining, visualizing, mapping, classifying, and modeling massive streams of public microblogging events.",
"From politicians and nation states to terrorist groups, numerous organizations reportedly conduct explicit campaigns to influence opinions on social media, posing a risk to freedom of expression. Thus, there is a need to identify and eliminate \"influence bots\"--realistic, automated identities that illicitly shape discussions on sites like Twitter and Facebook--before they get too influential.",
"The processes by which communities come together, attract new members, and develop over time is a central research issue in the social sciences - political movements, professional organizations, and religious denominations all provide fundamental examples of such communities. In the digital domain, on-line groups are becoming increasingly prominent due to the growth of community and social networking sites such as MySpace and LiveJournal. However, the challenge of collecting and analyzing large-scale time-resolved data on social groups and communities has left most basic questions about the evolution of such groups largely unresolved: what are the structural features that influence whether individuals will join communities, which communities will grow rapidly, and how do the overlaps among pairs of communities change over time.Here we address these questions using two large sources of data: friendship links and community membership on LiveJournal, and co-authorship and conference publications in DBLP. Both of these datasets provide explicit user-defined communities, where conferences serve as proxies for communities in DBLP. We study how the evolution of these communities relates to properties such as the structure of the underlying social networks. We find that the propensity of individuals to join communities, and of communities to grow rapidly, depends in subtle ways on the underlying network structure. For example, the tendency of an individual to join a community is influenced not just by the number of friends he or she has within the community, but also crucially by how those friends are connected to one another. We use decision-tree techniques to identify the most significant structural determinants of these properties. We also develop a novel methodology for measuring movement of individuals between communities, and show how such movements are closely aligned with changes in the topics of interest within the communities.",
"Today's social bots are sophisticated and sometimes menacing. Indeed, their presence can endanger online ecosystems as well as our society.",
"Social networks have emerged as a critical factor in information dissemination, search, marketing, expertise and influence discovery, and potentially an important tool for mobilizing people. Social media has made social networks ubiquitous, and also given researchers access to massive quantities of data for empirical analysis. These data sets offer a rich source of evidence for studying dynamics of individual and group behavior, the structure of networks and global patterns of the flow of information on them. However, in most previous studies, the structure of the underlying networks was not directly visible but had to be inferred from the flow of information from one individual to another. As a result, we do not yet understand dynamics of information spread on networks or how the structure of the network affects it. We address this gap by analyzing data from two popular social news sites. Specifically, we extract social networks of active users on Digg and Twitter, and track how interest in news stories spreads among them. We show that social networks play a crucial role in the spread of information on these sites, and that network structure affects dynamics of information flow.",
"We study astroturf political campaigns on microblogging platforms: politically-motivated individuals and organizations that use multiple centrally-controlled accounts to create the appearance of widespread support for a candidate or opinion. We describe a machine learning framework that combines topological, content-based and crowdsourced features of information diffusion networks on Twitter to detect the early stages of viral spreading of political misinformation. We present promising preliminary results with better than 96 accuracy in the detection of astroturf content in the run-up to the 2010 U.S. midterm elections."
]
} |
1703.07309 | 2949681560 | Many interesting natural phenomena are sparsely distributed and discrete. Locating the hotspots of such sparsely distributed phenomena is often difficult because their density gradient is likely to be very noisy. We present a novel approach to this search problem, where we model the co-occurrence relations between a robot's observations with a Bayesian nonparametric topic model. This approach makes it possible to produce a robust estimate of the spatial distribution of the target, even in the absence of direct target observations. We apply the proposed approach to the problem of finding the spatial locations of the hotspots of a specific phytoplankton taxon in the ocean. We use classified image data from Imaging FlowCytobot (IFCB), which automatically measures individual microscopic cells and colonies of cells. Given these individual taxon-specific observations, we learn a phytoplankton community model that characterizes the co-occurrence relations between taxa. We present experiments with simulated robot missions drawn from real observation data collected during a research cruise traversing the US Atlantic coast. Our results show that the proposed approach outperforms nearest neighbor and k-means based methods for predicting the spatial distribution of hotspots from in-situ observations. | Chlorophyll fluorescence sensors provide a way to detect phytoplankton directly. For example, @cite_4 used fluorometers on AUVs and Lagrangian drifters to locate and track phytoplankton patches in the ocean. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2057235886"
],
"abstract": [
"We extend existing oceanographic sampling methodologies to sample an advecting feature of interest using autonomous robotic platforms. GPS-tracked Lagrangian drifters are used to tag and track a water patch of interest with position updates provided periodically to an autonomous underwater vehicle (AUV) for surveys around the drifter as it moves with ocean currents. Autonomous sampling methods currently rely on geographic waypoint track-line surveys that are suitable for static or slowly changing features. When studying dynamic, rapidly evolving oceanographic features, such methods at best introduce error through insufficient spatial and temporal resolution, and at worst, completely miss the spatial and temporal domain of interest. We demonstrate two approaches for tracking and sampling of advecting oceanographic features. The first relies on extending static-plan AUV surveys (the current state-of-the-art) to sample advecting features. The second approach involves planning of surveys in the drifter or patch frame of reference. We derive a quantitative envelope on patch speeds that can be tracked autonomously by AUVs and drifters and show results from a multi-day off-shore field trial. The results from the trial demonstrate the applicability of our approach to long-term tracking and sampling of advecting features. Additionally, we analyze the data from the trial to identify the sources of error that affect the quality of the surveys carried out. Our work presents the first set of experiments to autonomously observe advecting oceanographic features in the open ocean."
]
} |
1703.07309 | 2949681560 | Many interesting natural phenomena are sparsely distributed and discrete. Locating the hotspots of such sparsely distributed phenomena is often difficult because their density gradient is likely to be very noisy. We present a novel approach to this search problem, where we model the co-occurrence relations between a robot's observations with a Bayesian nonparametric topic model. This approach makes it possible to produce a robust estimate of the spatial distribution of the target, even in the absence of direct target observations. We apply the proposed approach to the problem of finding the spatial locations of the hotspots of a specific phytoplankton taxon in the ocean. We use classified image data from Imaging FlowCytobot (IFCB), which automatically measures individual microscopic cells and colonies of cells. Given these individual taxon-specific observations, we learn a phytoplankton community model that characterizes the co-occurrence relations between taxa. We present experiments with simulated robot missions drawn from real observation data collected during a research cruise traversing the US Atlantic coast. Our results show that the proposed approach outperforms nearest neighbor and k-means based methods for predicting the spatial distribution of hotspots from in-situ observations. | @cite_1 also developed an approach to predict the abundance of a particular species known to cause harmful algal blooms in the study region. Their objective was to optimize capture of the target species in a small, fixed number of physical samples taken by a robot. Their model is based on a Gaussian Process, with a set of environment variables including fluorescence, temperature, and other chemical properties as inputs, and the results of manual molecular analysis of historical data as training targets. Whereas their method focuses on predicting the abundance of the target species from environmental variables, our method predicts the abundance of a taxon from the distribution of other taxa. These two perspectives are complementary and both are useful for the problem of automatically choosing the best set of sample locations for extended ex-situ analysis. | {
"cite_N": [
"@cite_1"
],
"mid": [
"1938668859"
],
"abstract": [
"Robotic sampling is attractive in many field robotics applications that require persistent collection of physical samples for ex-situ analysis. Examples abound in the earth sciences in studies involving the collection of rock, soil, and water samples for laboratory analysis. In our test domain, marine ecosystem monitoring, detailed understanding of plankton ecology requires laboratory analysis of water samples, but predictions using physical and chemical properties measured in real-time by sensors aboard an autonomous underwater vehicle AUV can guide sample collection decisions. In this paper, we present a data-driven and opportunistic sampling strategy to minimize cumulative regret for batches of plankton samples acquired by an AUV over multiple surveys. Samples are labeled at the end of each survey, and used to update a probabilistic model that guides sampling during subsequent surveys. During a survey, the AUV makes irrevocable sample collection decisions online for a sequential stream of candidates, with no knowledge of the quality of future samples. In addition to extensive simulations using historical field data, we present results from a one-day field trial where beginning with a prior model learned from data collected and labeled in an earlier campaign, the AUV collected water samples with a high abundance of a pre-specified planktonic target. This is the first time such a field experiment has been carried out in its entirety in a data-driven fashion, in effect ?closing the loop? on a significant and relevant ecosystem monitoring problem while allowing domain experts marine ecologists to specify the mission at a relatively high level."
]
} |
1703.07309 | 2949681560 | Many interesting natural phenomena are sparsely distributed and discrete. Locating the hotspots of such sparsely distributed phenomena is often difficult because their density gradient is likely to be very noisy. We present a novel approach to this search problem, where we model the co-occurrence relations between a robot's observations with a Bayesian nonparametric topic model. This approach makes it possible to produce a robust estimate of the spatial distribution of the target, even in the absence of direct target observations. We apply the proposed approach to the problem of finding the spatial locations of the hotspots of a specific phytoplankton taxon in the ocean. We use classified image data from Imaging FlowCytobot (IFCB), which automatically measures individual microscopic cells and colonies of cells. Given these individual taxon-specific observations, we learn a phytoplankton community model that characterizes the co-occurrence relations between taxa. We present experiments with simulated robot missions drawn from real observation data collected during a research cruise traversing the US Atlantic coast. Our results show that the proposed approach outperforms nearest neighbor and k-means based methods for predicting the spatial distribution of hotspots from in-situ observations. | @cite_10 proposed the use of a neural network to learn a shared representation over multiple sensor modalities for underwater vehicles (imagery and bathymetry). The learned model is then used to identify information-rich locations given exclusively the bathymetric data. For a small number of classes, this type of multimodal learning framework might capture more of the spatial or temporal complexities of plankton taxon associations. However, as the number of modalities increases this approach is not scalable and therefore it is not suitable for modelling the numerous plankton taxa we consider from this dataset. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2413870294"
],
"abstract": [
"Autonomous underwater vehicles (AUVs) are widely used to perform information gathering missions in unseen environments. Given the sheer size of the ocean environment, and the time and energy constraints of an AUV, it is important to consider the potential utility of candidate missions when performing survey planning. In this paper, we utilise a multimodal learning approach to capture the relationship between in-situ visual observations, and shipborne bathymetry (ocean depth) data that are freely available a priori. We then derive information-theoretic measures under this model that predict the amount of visual information gain at an unobserved location based on the bathymetric features. Unlike previous approaches, these measures consider the value of additional visual features, rather than just the habitat labels obtained. Experimental results with a toy dataset and real marine data demonstrate that the approach can be used to predict the true utility of unexplored areas."
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | Direct policy search (PS) methods have been successful in robotics as they can easily be applied in high-dimensional continuous state-action RL problems @cite_3 . REINFORCE @cite_40 is an early policy gradient method which performs exploration of the action space using probabilistic policies. It suffers, however, from slow convergence due to the high variance in its gradient estimates. Policy gradients with parameter-based exploration (PGPE) @cite_36 address this problem by transferring exploration to parameter space. In particular, PGPE samples deterministic policies at the start of each episode by maintaining a separate Gaussian distribution for each parameter of the policy, whose mean and variance are adapted during training. The PoWER (Policy learning by Weighting Exploration with the Returns) algorithm @cite_32 uses probability-weighted averaging, which has the property of following the natural gradient without computing it @cite_32 . PoWER, however, assumes that the immediate rewards sum to a constant number and are always positive, which complicates the design of reward functions. The Policy Improvements with Path Integrals (PI @math ) @cite_14 algorithm does not make such an assumption. When the reward function is compatible with both PoWER and PI @math , the algorithms have identical performance @cite_14 . | {
"cite_N": [
"@cite_14",
"@cite_36",
"@cite_32",
"@cite_3",
"@cite_40"
],
"mid": [
"1925816294",
"2116468180",
"2167117957",
"2012587148",
"2119717200"
],
"abstract": [
"With the goal to generate more scalable algorithms with higher efficiency and fewer open parameters, reinforcement learning (RL) has recently moved towards combining classical techniques from optimal control and dynamic programming with modern learning techniques from statistical estimation theory. In this vein, this paper suggests to use the framework of stochastic optimal control with path integrals to derive a novel approach to RL with parameterized policies. While solidly grounded in value function estimation and optimal control based on the stochastic Hamilton-Jacobi-Bellman (HJB) equations, policy improvements can be transformed into an approximation problem of a path integral which has no open algorithmic parameters other than the exploration noise. The resulting algorithm can be conceived of as model-based, semi-model-based, or even model free, depending on how the learning problem is structured. The update equations have no danger of numerical instabilities as neither matrix inversions nor gradient learning rates are required. Our new algorithm demonstrates interesting similarities with previous RL research in the framework of probability matching and provides intuition why the slightly heuristically motivated probability matching approach can actually perform well. Empirical evaluations demonstrate significant performance improvements over gradient-based policy learning and scalability to high-dimensional control problems. Finally, a learning experiment on a simulated 12 degree-of-freedom robot dog illustrates the functionality of our algorithm in a complex robot learning scenario. We believe that Policy Improvement with Path Integrals (PI2) offers currently one of the most efficient, numerically robust, and easy to implement algorithms for RL based on trajectory roll-outs.",
"We present a model-free reinforcement learning method for partially observable Markov decision problems. Our method estimates a likelihood gradient by sampling directly in parameter space, which leads to lower variance gradient estimates than those obtained by policy gradient methods such as REINFORCE. For several complex control tasks, including robust standing with a humanoid robot, we show that our method outperforms well-known algorithms from the fields of policy gradients, finite difference methods and population based heuristics. We also provide a detailed analysis of the differences between our method and the other algorithms.",
"Many motor skills in humanoid robotics can be learned using parametrized motor primitives as done in imitation learning. However, most interesting motor learning problems are high-dimensional reinforcement learning problems often beyond the reach of current methods. In this paper, we extend previous work on policy learning from the immediate reward case to episodic reinforcement learning. We show that this results in a general, common framework also connected to policy gradient methods and yielding a novel algorithm for policy learning that is particularly well-suited for dynamic motor primitives. The resulting algorithm is an EM-inspired algorithm applicable to complex motor learning tasks. We compare this algorithm to several well-known parametrized policy search methods and show that it outperforms them. We apply it in the context of motor learning and show that it can learn a complex Ball-in-a-Cup task using a real Barrett WAM™ robot arm.",
"Policy search is a subfield in reinforcement learning which focuses on finding good parameters for a given policy parametrization. It is well suited for robotics as it can cope with high-dimensional state and action spaces, one of the main challenges in robot learning. We review recent successes of both model-free and model-based policy search in robot learning.Model-free policy search is a general approach to learn policies based on sampled trajectories. We classify model-free methods based on their policy evaluation strategy, policy update strategy, and exploration strategy and present a unified view on existing algorithms. Learning a policy is often easier than learning an accurate forward model, and, hence, model-free methods are more frequently used in practice. However, for each sampled trajectory, it is necessary to interact with the robot, which can be time consuming and challenging in practice. Model-based policy search addresses this problem by first learning a simulator of the robot's dynamics from data. Subsequently, the simulator generates trajectories that are used for policy learning. For both model-free and model-based policy search methods, we review their respective properties and their applicability to robotic systems.",
"This article presents a general class of associative reinforcement learning algorithms for connectionist networks containing stochastic units. These algorithms, called REINFORCE algorithms, are shown to make weight adjustments in a direction that lies along the gradient of expected reinforcement in both immediate-reinforcement tasks and certain limited forms of delayed-reinforcement tasks, and they do this without explicitly computing gradient estimates or even storing information from which such estimates could be computed. Specific examples of such algorithms are presented, some of which bear a close relationship to certain existing algorithms while others are novel but potentially interesting in their own right. Also given are results that show how such algorithms can be naturally integrated with backpropagation. We close with a brief discussion of a number of additional issues surrounding the use of such algorithms, including what is known about their limiting behaviors as well as further considerations that might be used to help develop similar but potentially more powerful reinforcement learning algorithms."
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | A limitation of PGPE is that it does not consider any correlations between dimensions in parameter space. This can be addressed by the Natural Evolution Strategies (NES) @cite_28 and Covariance Matrix Adaptation ES (CMA-ES) @cite_29 families of algorithms, which are population-based Black-Box Optimizers (BBO). Both NES and CMA-ES iteratively update a search distribution by calculating an estimated gradient on the distribution parameters (mean and covariance matrix). At each generation, they sample a set of solutions (i.e., policy parameters) and rank them based on their fitness (i.e., expected return). NES performs gradient ascent along the natural gradient, which normalizes the update with respect to uncertainty. CMA-ES updates the distribution by exploiting the technique of evolution paths to average-out random effects over the generations. NES and CMA-ES are closely related, as the latter performs an approximate natural gradient ascent @cite_31 . Interestingly, a variant of PI @math with a simplified parameter perturbation and update method outperforms PI @math and was shown to be a special case of CMA-ES @cite_6 . | {
"cite_N": [
"@cite_28",
"@cite_29",
"@cite_31",
"@cite_6"
],
"mid": [
"2151965738",
"",
"1579744901",
"2085627234"
],
"abstract": [
"This paper presents Natural Evolution Strategies (NES), a recent family of black-box optimization algorithms that use the natural gradient to update a parameterized search distribution in the direction of higher expected fitness. We introduce a collection of techniques that address issues of convergence, robustness, sample complexity, computational complexity and sensitivity to hyperparameters. This paper explores a number of implementations of the NES family, such as general-purpose multi-variate normal distributions and separable distributions tailored towards search in high dimensional spaces. Experimental results show best published performance on various standard benchmarks, as well as competitive performance on others.",
"",
"This paper investigates the relation between the covariance matrix adaptation evolution strategy and the natural evolution strategy, the latter of which is recently proposed and is formulated as a natural gradient based method on the expected fitness under the mutation distribution. To enable to compare these algorithms, we derive the explicit form of the natural gradient of the expected fitness and transform it into the forms corresponding to the mean vector and the covariance matrix of the mutation distribution. We show that the natural evolution strategy can be viewed as a variant of covariance matrix adaptation evolution strategies using Cholesky update and also that the covariance matrix adaptation evolution strategy can be formulated as a variant of natural evolution strategies.",
"Policy improvement methods seek to optimize the parameters of a policy with respect to a utility function. Owing to current trends involving searching in parameter space (rather than action space) and using reward-weighted averaging (rather than gradient estimation), reinforcement learning algorithms for policy improvement, e.g. PoWER and PI2, are now able to learn sophisticated high-dimensional robot skills. A side-effect of these trends has been that, over the last 15 years, reinforcement learning (RL) algorithms have become more and more similar to evolution strategies such as (μW , λ)-ES and CMA-ES. Evolution strategies treat policy improvement as a black-box optimization problem, and thus do not leverage the problem structure, whereas RL algorithms do. In this paper, we demonstrate how two straightforward simplifications to the state-of-the-art RL algorithm PI2 suffice to convert it into the black-box optimization algorithm (μW, λ)-ES. Furthermore, we show that (μW , λ)-ES empirically outperforms PI2 on the tasks in [36]. It is striking that PI2 and (μW , λ)-ES share a common core, and that the simpler algorithm converges faster and leads to similar or lower final costs. We argue that this difference is due to a third trend in robot skill learning: the predominant use of dynamic movement primitives (DMPs). We show how DMPs dramatically simplify the learning problem, and discuss the implications of this for past and future work on policy improvement for robot skill learning"
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | The data-efficiency of direct PS can be further increased by learning the model (i.e., transition and reward function) of the system from data and inferring the optimal policy from the model @cite_3 . Probabilistic models have been more successful than deterministic ones, as they provide an estimate about the uncertainty of their approximation which can be incorporated into long-term planning @cite_10 . For example, local linear models have been used in @cite_17 @cite_19 @cite_24 , Gaussian processes (GPs) in @cite_18 @cite_10 @cite_13 and least-squares conditional density estimation in @cite_7 . | {
"cite_N": [
"@cite_13",
"@cite_18",
"@cite_7",
"@cite_3",
"@cite_24",
"@cite_19",
"@cite_10",
"@cite_17"
],
"mid": [
"1972063518",
"2162717641",
"2041851440",
"2012587148",
"2121103318",
"",
"2018705428",
"2130105540"
],
"abstract": [
"In robotics, lower-level controllers are typically used to make the robot solve a specific task in a fixed context. For example, the lower-level controller can encode a hitting movement while the context defines the target coordinates to hit. However, in many learning problems the context may change between task executions. To adapt the policy to a new context, we utilize a hierarchical approach by learning an upper-level policy that generalizes the lower-level controllers to new contexts. A common approach to learn such upper-level policies is to use policy search. However, the majority of current contextual policy search approaches are model-free and require a high number of interactions with the robot and its environment. Model-based approaches are known to significantly reduce the amount of robot experiments, however, current model-based techniques cannot be applied straightforwardly to the problem of learning contextual upper-level policies. They rely on specific parametrizations of the policy and the reward function, which are often unrealistic in the contextual policy search formulation. In this paper, we propose a novel model-based contextual policy search algorithm that is able to generalize lower-level controllers, and is data-efficient. Our approach is based on learned probabilistic forward models and information theoretic policy search. Unlike current algorithms, our method does not require any assumption on the parametrization of the policy or the reward function. We show on complex simulated robotic tasks and in a real robot experiment that the proposed learning framework speeds up the learning process by up to two orders of magnitude in comparison to existing methods, while learning high quality policies.",
"Blimps are a promising platform for aerial robotics and have been studied extensively for this purpose. Unlike other aerial vehicles, blimps are relatively safe and also possess the ability to loiter for long periods. These advantages, however, have been difficult to exploit because blimp dynamics are complex and inherently non-linear. The classical approach to system modeling represents the system as an ordinary differential equation (ODE) based on Newtonian principles. A more recent modeling approach is based on representing state transitions as a Gaussian process (GP). In this paper, we present a general technique for system identification that combines these two modeling approaches into a single formulation. This is done by training a Gaussian process on the residual between the non-linear model and ground truth training data. The result is a GP-enhanced model that provides an estimate of uncertainty in addition to giving better state predictions than either ODE or GP alone. We show how the GP-enhanced model can be used in conjunction with reinforcement learning to generate a blimp controller that is superior to those learned with ODE or GP models alone.",
"The goal of reinforcement learning (RL) is to let an agent learn an optimal control policy in an unknown environment so that future expected rewards are maximized. The model-free RL approach directly learns the policy based on data samples. Although using many samples tends to improve the accuracy of policy learning, collecting a large number of samples is often expensive in practice. On the other hand, the model-based RL approach first estimates the transition model of the environment and then learns the policy based on the estimated transition model. Thus, if the transition model is accurately learned from a small amount of data, the model-based approach is a promising alternative to the model-free approach. In this paper, we propose a novel model-based RL method by combining a recently proposed model-free policy search method called policy gradients with parameter-based exploration and the state-of-the-art transition model estimator called least-squares conditional density estimation. Through experiments, we demonstrate the practical usefulness of the proposed method.",
"Policy search is a subfield in reinforcement learning which focuses on finding good parameters for a given policy parametrization. It is well suited for robotics as it can cope with high-dimensional state and action spaces, one of the main challenges in robot learning. We review recent successes of both model-free and model-based policy search in robot learning.Model-free policy search is a general approach to learn policies based on sampled trajectories. We classify model-free methods based on their policy evaluation strategy, policy update strategy, and exploration strategy and present a unified view on existing algorithms. Learning a policy is often easier than learning an accurate forward model, and, hence, model-free methods are more frequently used in practice. However, for each sampled trajectory, it is necessary to interact with the robot, which can be time consuming and challenging in practice. Model-based policy search addresses this problem by first learning a simulator of the robot's dynamics from data. Subsequently, the simulator generates trajectories that are used for policy learning. For both model-free and model-based policy search methods, we review their respective properties and their applicability to robotic systems.",
"We present a policy search method that uses iteratively refitted local linear models to optimize trajectory distributions for large, continuous problems. These trajectory distributions can be used within the framework of guided policy search to learn policies with an arbitrary parameterization. Our method fits time-varying linear dynamics models to speed up learning, but does not rely on learning a global model, which can be difficult when the dynamics are complex and discontinuous. We show that this hybrid approach requires many fewer samples than model-free methods, and can handle complex, nonsmooth dynamics that can pose a challenge for model-based techniques. We present experiments showing that our method can be used to learn complex neural network policies that successfully execute simulated robotic manipulation tasks in partially observed environments with numerous contact discontinuities and underactuation.",
"",
"Autonomous learning has been a promising direction in control and robotics for more than a decade since data-driven learning allows to reduce the amount of engineering knowledge, which is otherwise required. However, autonomous reinforcement learning (RL) approaches typically require many interactions with the system to learn controllers, which is a practical limitation in real systems, such as robots, where many interactions can be impractical and time consuming. To address this problem, current learning approaches typically require task-specific knowledge in form of expert demonstrations, realistic simulators, pre-shaped policies, or specific knowledge about the underlying dynamics. In this paper, we follow a different approach and speed up learning by extracting more information from data. In particular, we learn a probabilistic, non-parametric Gaussian process transition model of the system. By explicitly incorporating model uncertainty into long-term planning and controller learning our approach reduces the effects of model errors, a key problem in model-based learning. Compared to state-of-the art RL our model-based policy search method achieves an unprecedented speed of learning. We demonstrate its applicability to autonomous learning in real robot and control tasks.",
"Many control problems in the robotics field can be cast as partially observed Markovian decision problems (POMDPs), an optimal control formalism. Finding optimal solutions to such problems in general, however is known to be intractable. It has often been observed that in practice, simple structured controllers suffice for good sub-optimal control, and recent research in the artificial intelligence community has focused on policy search methods as techniques for finding sub-optimal controllers when such structured controllers do exist. Traditional model-based reinforcement learning algorithms make a certainty equivalence assumption on their learned models and calculate optimal policies for a maximum-likelihood Markovian model. We consider algorithms that evaluate and synthesize controllers under distributions of Markovian models. Previous work has demonstrated that algorithms that maximize mean reward with respect to model uncertainty leads to safer and more robust controllers. We consider briefly other performance criterion that emphasize robustness and exploration in the search for controllers, and note the relation with experiment design and active learning. To validate the power of the approach on a robotic application we demonstrate the presented learning control algorithm by flying an autonomous helicopter. We show that the controller learned is robust and delivers good performance in this real-world domain."
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | Early examples of such model-based PS include applications on helicopter hovering @cite_17 @cite_19 and blimp control @cite_18 . These works employ the PEGASUS algorithm which can transform a stochastic Markov Decision Process (MDP) or partially-observable MDP (POMDP) into a deterministic POMDP @cite_23 . It does so by fixing in advance the sequence of random numbers associated with the state transitions. This simple modification significantly reduces the time needed to optimize the policy, as it removes the noise from the evaluation of an initially noisy objective function. | {
"cite_N": [
"@cite_19",
"@cite_18",
"@cite_23",
"@cite_17"
],
"mid": [
"",
"2162717641",
"2950462606",
"2130105540"
],
"abstract": [
"",
"Blimps are a promising platform for aerial robotics and have been studied extensively for this purpose. Unlike other aerial vehicles, blimps are relatively safe and also possess the ability to loiter for long periods. These advantages, however, have been difficult to exploit because blimp dynamics are complex and inherently non-linear. The classical approach to system modeling represents the system as an ordinary differential equation (ODE) based on Newtonian principles. A more recent modeling approach is based on representing state transitions as a Gaussian process (GP). In this paper, we present a general technique for system identification that combines these two modeling approaches into a single formulation. This is done by training a Gaussian process on the residual between the non-linear model and ground truth training data. The result is a GP-enhanced model that provides an estimate of uncertainty in addition to giving better state predictions than either ODE or GP alone. We show how the GP-enhanced model can be used in conjunction with reinforcement learning to generate a blimp controller that is superior to those learned with ODE or GP models alone.",
"We propose a new approach to the problem of searching a space of policies for a Markov decision process (MDP) or a partially observable Markov decision process (POMDP), given a model. Our approach is based on the following observation: Any (PO)MDP can be transformed into an \"equivalent\" POMDP in which all state transitions (given the current state and action) are deterministic. This reduces the general problem of policy search to one in which we need only consider POMDPs with deterministic transitions. We give a natural way of estimating the value of all policies in these transformed POMDPs. Policy search is then simply performed by searching for a policy with high estimated value. We also establish conditions under which our value estimates will be good, recovering theoretical results similar to those of Kearns, Mansour and Ng (1999), but with \"sample complexity\" bounds that have only a polynomial rather than exponential dependence on the horizon time. Our method applies to arbitrary POMDPs, including ones with infinite state and action spaces. We also present empirical results for our approach on a small discrete problem, and on a complex continuous state continuous action problem involving learning to ride a bicycle.",
"Many control problems in the robotics field can be cast as partially observed Markovian decision problems (POMDPs), an optimal control formalism. Finding optimal solutions to such problems in general, however is known to be intractable. It has often been observed that in practice, simple structured controllers suffice for good sub-optimal control, and recent research in the artificial intelligence community has focused on policy search methods as techniques for finding sub-optimal controllers when such structured controllers do exist. Traditional model-based reinforcement learning algorithms make a certainty equivalence assumption on their learned models and calculate optimal policies for a maximum-likelihood Markovian model. We consider algorithms that evaluate and synthesize controllers under distributions of Markovian models. Previous work has demonstrated that algorithms that maximize mean reward with respect to model uncertainty leads to safer and more robust controllers. We consider briefly other performance criterion that emphasize robustness and exploration in the search for controllers, and note the relation with experiment design and active learning. To validate the power of the approach on a robotic application we demonstrate the presented learning control algorithm by flying an autonomous helicopter. We show that the controller learned is robust and delivers good performance in this real-world domain."
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | Both the model-based PGPE @cite_7 and the PILCO @cite_10 algorithm use gradient-based policy updates. Rather than using Monte Carlo sampling, as in model-based PGPE, PILCO performs deterministic approximate inference by explicitly incorporating the model uncertainty into long-term predictions. This procedure is done by approximating the probability distribution over trajectories with a Gaussian that has the same mean and covariance (moment matching). The gradient of the expected return is then computed analytically with respect to the policy parameters. This makes PILCO dependent on differentiable reward and policy functions. | {
"cite_N": [
"@cite_10",
"@cite_7"
],
"mid": [
"2018705428",
"2041851440"
],
"abstract": [
"Autonomous learning has been a promising direction in control and robotics for more than a decade since data-driven learning allows to reduce the amount of engineering knowledge, which is otherwise required. However, autonomous reinforcement learning (RL) approaches typically require many interactions with the system to learn controllers, which is a practical limitation in real systems, such as robots, where many interactions can be impractical and time consuming. To address this problem, current learning approaches typically require task-specific knowledge in form of expert demonstrations, realistic simulators, pre-shaped policies, or specific knowledge about the underlying dynamics. In this paper, we follow a different approach and speed up learning by extracting more information from data. In particular, we learn a probabilistic, non-parametric Gaussian process transition model of the system. By explicitly incorporating model uncertainty into long-term planning and controller learning our approach reduces the effects of model errors, a key problem in model-based learning. Compared to state-of-the art RL our model-based policy search method achieves an unprecedented speed of learning. We demonstrate its applicability to autonomous learning in real robot and control tasks.",
"The goal of reinforcement learning (RL) is to let an agent learn an optimal control policy in an unknown environment so that future expected rewards are maximized. The model-free RL approach directly learns the policy based on data samples. Although using many samples tends to improve the accuracy of policy learning, collecting a large number of samples is often expensive in practice. On the other hand, the model-based RL approach first estimates the transition model of the environment and then learns the policy based on the estimated transition model. Thus, if the transition model is accurately learned from a small amount of data, the model-based approach is a promising alternative to the model-free approach. In this paper, we propose a novel model-based RL method by combining a recently proposed model-free policy search method called policy gradients with parameter-based exploration and the state-of-the-art transition model estimator called least-squares conditional density estimation. Through experiments, we demonstrate the practical usefulness of the proposed method."
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | Gradient-free methods, such as the Model-Based Relative Entropy PS (M-REPS) @cite_13 and the Model-Based Guided PS (M-GPS) @cite_24 , do not have these requirements. Both algorithms place a KL-divergence constraint on the cost function to bound the distance between the old trajectory distribution and the newly estimated one at each policy improvement step. This constraint limits the information loss of the updates @cite_2 . M-GPS turns the policy optimization problem into a supervised learning one, allowing the use of high-dimensional policy representations such as deep neural networks. However, M-GPS makes strong assumptions about the task at hand, by assuming that time-varying Gaussians can approximate the local dynamics. In contrast, M-REPS uses GPs for model learning, and the REPS algorithm (which can be seen as a BBO) for policy search. | {
"cite_N": [
"@cite_24",
"@cite_13",
"@cite_2"
],
"mid": [
"2121103318",
"1972063518",
"2125612430"
],
"abstract": [
"We present a policy search method that uses iteratively refitted local linear models to optimize trajectory distributions for large, continuous problems. These trajectory distributions can be used within the framework of guided policy search to learn policies with an arbitrary parameterization. Our method fits time-varying linear dynamics models to speed up learning, but does not rely on learning a global model, which can be difficult when the dynamics are complex and discontinuous. We show that this hybrid approach requires many fewer samples than model-free methods, and can handle complex, nonsmooth dynamics that can pose a challenge for model-based techniques. We present experiments showing that our method can be used to learn complex neural network policies that successfully execute simulated robotic manipulation tasks in partially observed environments with numerous contact discontinuities and underactuation.",
"In robotics, lower-level controllers are typically used to make the robot solve a specific task in a fixed context. For example, the lower-level controller can encode a hitting movement while the context defines the target coordinates to hit. However, in many learning problems the context may change between task executions. To adapt the policy to a new context, we utilize a hierarchical approach by learning an upper-level policy that generalizes the lower-level controllers to new contexts. A common approach to learn such upper-level policies is to use policy search. However, the majority of current contextual policy search approaches are model-free and require a high number of interactions with the robot and its environment. Model-based approaches are known to significantly reduce the amount of robot experiments, however, current model-based techniques cannot be applied straightforwardly to the problem of learning contextual upper-level policies. They rely on specific parametrizations of the policy and the reward function, which are often unrealistic in the contextual policy search formulation. In this paper, we propose a novel model-based contextual policy search algorithm that is able to generalize lower-level controllers, and is data-efficient. Our approach is based on learned probabilistic forward models and information theoretic policy search. Unlike current algorithms, our method does not require any assumption on the parametrization of the policy or the reward function. We show on complex simulated robotic tasks and in a real robot experiment that the proposed learning framework speeds up the learning process by up to two orders of magnitude in comparison to existing methods, while learning high quality policies.",
"Autonomous learning is one of the hallmarks of human and animal behavior, and understanding the principles of learning will be crucial in order to achieve true autonomy in advanced machines like humanoid robots. In this paper, we examine learning of complex motor skills with human-like limbs. While supervised learning can offer useful tools for bootstrapping behavior, e.g., by learning from demonstration, it is only reinforcement learning that offers a general approach to the final trial-and-error improvement that is needed by each individual acquiring a skill. Neither neurobiological nor machine learning studies have, so far, offered compelling results on how reinforcement learning can be scaled to the high-dimensional continuous state and action spaces of humans or humanoids. Here, we combine two recent research developments on learning motor control in order to achieve this scaling. First, we interpret the idea of modular motor control by means of motor primitives as a suitable way to generate parameterized control policies for reinforcement learning. Second, we combine motor primitives with the theory of stochastic policy gradient learning, which currently seems to be the only feasible framework for reinforcement learning for humanoids. We evaluate different policy gradient methods with a focus on their applicability to parameterized motor primitives. We compare these algorithms in the context of motor primitive learning, and show that our most modern algorithm, the Episodic Natural Actor-Critic outperforms previous algorithms by at least an order of magnitude. We demonstrate the efficiency of this reinforcement learning method in the application of learning to hit a baseball with an anthropomorphic robot arm."
]
} |
1703.07261 | 2949535390 | The most data-efficient algorithms for reinforcement learning (RL) in robotics are based on uncertain dynamical models: after each episode, they first learn a dynamical model of the robot, then they use an optimization algorithm to find a policy that maximizes the expected return given the model and its uncertainties. It is often believed that this optimization can be tractable only if analytical, gradient-based algorithms are used; however, these algorithms require using specific families of reward functions and policies, which greatly limits the flexibility of the overall approach. In this paper, we introduce a novel model-based RL algorithm, called Black-DROPS (Black-box Data-efficient RObot Policy Search) that: (1) does not impose any constraint on the reward function or the policy (they are treated as black-boxes), (2) is as data-efficient as the state-of-the-art algorithm for data-efficient RL in robotics, and (3) is as fast (or faster) than analytical approaches when several cores are available. The key idea is to replace the gradient-based optimization algorithm with a parallel, black-box algorithm that takes into account the model uncertainties. We demonstrate the performance of our new algorithm on two standard control benchmark problems (in simulation) and a low-cost robotic manipulator (with a real robot). | Overall, the current consensus @cite_3 is that (1) model-based algorithms are more data-efficient than direct PS, (2) in model-based PS, it is crucial to account for potential model errors during policy learning, and (3) deterministic approximate inference and analytic computation of policy gradients is required to make model-based PS computationally tractable. In this paper, we focus on the latter and explore a parallel BBO algorithm for policy optimization. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2012587148"
],
"abstract": [
"Policy search is a subfield in reinforcement learning which focuses on finding good parameters for a given policy parametrization. It is well suited for robotics as it can cope with high-dimensional state and action spaces, one of the main challenges in robot learning. We review recent successes of both model-free and model-based policy search in robot learning.Model-free policy search is a general approach to learn policies based on sampled trajectories. We classify model-free methods based on their policy evaluation strategy, policy update strategy, and exploration strategy and present a unified view on existing algorithms. Learning a policy is often easier than learning an accurate forward model, and, hence, model-free methods are more frequently used in practice. However, for each sampled trajectory, it is necessary to interact with the robot, which can be time consuming and challenging in practice. Model-based policy search addresses this problem by first learning a simulator of the robot's dynamics from data. Subsequently, the simulator generates trajectories that are used for policy learning. For both model-free and model-based policy search methods, we review their respective properties and their applicability to robotic systems."
]
} |
1703.07417 | 2950374614 | Solving linear programs is often a challenging task in distributed settings. While there are good algorithms for solving packing and covering linear programs in a distributed manner ( 2006), this is essentially the only class of linear programs for which such an algorithm is known. In this work we provide a distributed algorithm for solving a different class of convex programs which we call "distance-bounded network design convex programs". These can be thought of as relaxations of network design problems in which the connectivity requirement includes a distance constraint (most notably, graph spanners). Our algorithm runs in @math rounds in the @math model and finds a @math -approximation to the optimal LP solution for any @math , where @math is the largest distance constraint. While solving linear programs in a distributed setting is interesting in its own right, this class of convex programs is particularly important because solving them is often a crucial step when designing approximation algorithms. Hence we almost immediately obtain new and improved distributed approximation algorithms for a variety of network design problems, including Basic @math - and @math -Spanner, Directed @math -Spanner, Lowest Degree @math -Spanner, and Shallow-Light Steiner Network Design with a spanning demand graph. Our algorithms do not require any "heavy" computation and essentially match the best-known centralized approximation algorithms, while previous approaches which do not use heavy computation give approximations which are worse than the best-known centralized bounds. | A special case of our result was proved earlier in @cite_1 , who showed how to solve the LP relaxation of [2] in the model in @math rounds (they actually show more than this, by giving a distributed algorithm for the version of [2], but that is not germane to our results). Our techniques are heavily based on @cite_1 , which is itself based on the ideas from @cite_19 . In particular, @cite_19 uses a Linial-Saks decomposition @cite_5 to solve local'' versions of the linear program in different parts of the graph, and then combines these appropriately. To make this work for the [2] LP relaxation, @cite_1 had to use , which can be thought of as a variant of Linial-Saks with slightly different guarantees which, for technical reasons, are more useful for network design LPs. In this paper we extend these techniques further by giving a more general definition of padded decomposition which works for larger distance requirements, showing how to construct them in the model, and then showing that the basic combining'' idea from @cite_1 can be extended to handle these more general decompositions and far more general constraints and objective functions. | {
"cite_N": [
"@cite_19",
"@cite_5",
"@cite_1"
],
"mid": [
"1998137177",
"1998544343",
"2952422587"
],
"abstract": [
"Achieving a global goal based on local information is challenging, especially in complex and large-scale networks such as the Internet or even the human brain. In this paper, we provide an almost tight classification of the possible trade-off between the amount of local information and the quality of the global solution for general covering and packing problems. Specifically, we give a distributed algorithm using only small messages which obtains an (ρΔ)1 k-approximation for general covering and packing problems in time O(k2), where ρ depends on the LP's coefficients. If message size is unbounded, we present a second algorithm that achieves an O(n1 k) approximation in O(k) rounds. Finally, we prove that these algorithms are close to optimal by giving a lower bound on the approximability of packing problems given that each node has to base its decision on information from its k-neighborhood.",
"Adecomposition of a graphG=(V,E) is a partition of the vertex set into subsets (calledblocks). Thediameter of a decomposition is the leastd such that any two vertices belonging to the same connected component of a block are at distance ≤d. In this paper we prove (nearly best possible) statements, of the form: Anyn-vertex graph has a decomposition into a small number of blocks each having small diameter. Such decompositions provide a tool for efficiently decentralizing distributed computations. In [4] it was shown that every graph has a decomposition into at mosts(n) blocks of diameter at mosts(n) for (s(n) = n^ O( n n) ). Using a technique of Awerbuch [3] and Awerbuch and Peleg [5], we improve this result by showing that every graph has a decomposition of diameterO (logn) intoO(logn) blocks. In addition, we give a randomized distributed algorithm that produces such a decomposition and runs in timeO(log2n). The construction can be parameterized to provide decompositions that trade-off between the number of blocks and the diameter. We show that this trade-off is nearly best possible, for two families of graphs: the first consists of skeletons of certain triangulations of a simplex and the second consists of grid graphs with added diagonals. The proofs in both cases rely on basic results in combinatorial topology, Sperner's lemma for the first class and Tucker's lemma for the second.",
"A natural requirement of many distributed structures is fault-tolerance: after some failures, whatever remains from the structure should still be effective for whatever remains from the network. In this paper we examine spanners of general graphs that are tolerant to vertex failures, and significantly improve their dependence on the number of faults @math , for all stretch bounds. For stretch @math we design a simple transformation that converts every @math -spanner construction with at most @math edges into an @math -fault-tolerant @math -spanner construction with at most @math edges. Applying this to standard greedy spanner constructions gives @math -fault tolerant @math -spanners with @math edges. The previous construction by Chechik, Langberg, Peleg, and Roddity [STOC 2009] depends similarly on @math but exponentially on @math (approximately like @math ). For the case @math and unit-length edges, an @math -approximation algorithm is known from recent work of Dinitz and Krauthgamer [arXiv 2010], where several spanner results are obtained using a common approach of rounding a natural flow-based linear programming relaxation. Here we use a different (stronger) LP relaxation and improve the approximation ratio to @math , which is, notably, independent of the number of faults @math . We further strengthen this bound in terms of the maximum degree by using the Local Lemma. Finally, we show that most of our constructions are inherently local by designing equivalent distributed algorithms in the LOCAL model of distributed computation."
]
} |
1703.07144 | 2950001648 | Finding image correspondences remains a challenging problem in the presence of intra-class variations and large changes in scene layout. Semantic flow methods are designed to handle images depicting different instances of the same object or scene category. We introduce a novel approach to semantic flow, dubbed proposal flow, that establishes reliable correspondences using object proposals. Unlike prevailing semantic flow approaches that operate on pixels or regularly sampled local regions, proposal flow benefits from the characteristics of modern object proposals, that exhibit high repeatability at multiple scales, and can take advantage of both local and geometric consistency constraints among proposals. We also show that the corresponding sparse proposal flow can effectively be transformed into a conventional dense flow field. We introduce two new challenging datasets that can be used to evaluate both general semantic flow techniques and region-based approaches such as proposal flow. We use these benchmarks to compare different matching algorithms, object proposals, and region features within proposal flow, to the state of the art in semantic flow. This comparison, along with experiments on standard datasets, demonstrates that proposal flow significantly outperforms existing semantic flow methods in various settings. | Despite differences in graph construction, optimization, and similarity computation, existing semantic flow approaches share grid-based regular sampling and spatial regularization: The appearance similarity is defined at each region or pixel on (a pyramid of) regular grids, and spatial regularization is imposed between neighboring regions in the pyramid models @cite_69 @cite_39 @cite_64 @cite_0 . In contrast, our work builds on generic object proposals with diverse spatial supports @cite_5 @cite_47 @cite_76 @cite_74 @cite_79 , and uses an irregular form of spatial regularization based on co-occurrence and overlap of the proposals. We show that the use of local regularization with object proposals yields substantial gains in generic region matching and semantic flow, in particular when handling images with significant clutter, intra-class variations and scaling changes, establishing a new state of the art on the task. | {
"cite_N": [
"@cite_69",
"@cite_64",
"@cite_39",
"@cite_0",
"@cite_79",
"@cite_74",
"@cite_5",
"@cite_47",
"@cite_76"
],
"mid": [
"2124861766",
"",
"",
"2464606141",
"7746136",
"2088049833",
"",
"",
"2121660792"
],
"abstract": [
"We introduce a fast deformable spatial pyramid (DSP) matching algorithm for computing dense pixel correspondences. Dense matching methods typically enforce both appearance agreement between matched pixels as well as geometric smoothness between neighboring pixels. Whereas the prevailing approaches operate at the pixel level, we propose a pyramid graph model that simultaneously regularizes match consistency at multiple spatial extents-ranging from an entire image, to coarse grid cells, to every single pixel. This novel regularization substantially improves pixel-level matching in the face of challenging image variations, while the \"deformable\" aspect of our model overcomes the strict rigidity of traditional spatial pyramids. Results on Label Me and Caltech show our approach outperforms state-of-the-art methods (SIFT Flow [15] and Patch-Match [2]), both in terms of accuracy and run time.",
"",
"",
"We propose a new technique to jointly recover cosegmentation and dense per-pixel correspondence in two images. Our method parameterizes the correspondence field using piecewise similarity transformations and recovers a mapping between the estimated common \"foreground\" regions in the two images allowing them to be precisely aligned. Our formulation is based on a hierarchical Markov random field model with segmentation and transformation labels. The hierarchical structure uses nested image regions to constrain inference across multiple scales. Unlike prior hierarchical methods which assume that the structure is given, our proposed iterative technique dynamically recovers the structure along with the labeling. This joint inference is performed in an energy minimization framework using iterated graph cuts. We evaluate our method on a new dataset of 400 image pairs with manually obtained ground truth, where it outperforms state-of-the-art methods designed specifically for either cosegmentation or correspondence estimation.",
"The use of object proposals is an effective recent approach for increasing the computational efficiency of object detection. We propose a novel method for generating object bounding box proposals using edges. Edges provide a sparse yet informative representation of an image. Our main observation is that the number of contours that are wholly contained in a bounding box is indicative of the likelihood of the box containing an object. We propose a simple box objectness score that measures the number of edges that exist in the box minus those that are members of contours that overlap the box’s boundary. Using efficient data structures, millions of candidate boxes can be evaluated in a fraction of a second, returning a ranked set of a few thousand top-scoring proposals. Using standard metrics, we show results that are significantly more accurate than the current state-of-the-art while being faster to compute. In particular, given just 1000 proposals we achieve over 96 object recall at overlap threshold of 0.5 and over 75 recall at the more challenging overlap of 0.7. Our approach runs in 0.25 seconds and we additionally demonstrate a near real-time variant with only minor loss in accuracy.",
"This paper addresses the problem of generating possible object locations for use in object recognition. We introduce selective search which combines the strength of both an exhaustive search and segmentation. Like segmentation, we use the image structure to guide our sampling process. Like exhaustive search, we aim to capture all possible object locations. Instead of a single technique to generate possible object locations, we diversify our search and use a variety of complementary image partitionings to deal with as many image conditions as possible. Our selective search results in a small set of data-driven, class-independent, high quality locations, yielding 99 recall and a Mean Average Best Overlap of 0.879 at 10,097 locations. The reduced number of locations compared to an exhaustive search enables the use of stronger machine learning techniques and stronger appearance models for object recognition. In this paper we show that our selective search enables the use of the powerful Bag-of-Words model for recognition. The selective search software is made publicly available (Software: http: disi.unitn.it uijlings SelectiveSearch.html ).",
"",
"",
"Generic object detection is the challenging task of proposing windows that localize all the objects in an image, regardless of their classes. Such detectors have recently been shown to benefit many applications such as speeding-up class-specific object detection, weakly supervised learning of object detectors and object discovery. In this paper, we introduce a novel and very efficient method for generic object detection based on a randomized version of Prim's algorithm. Using the connectivity graph of an image's super pixels, with weights modelling the probability that neighbouring super pixels belong to the same object, the algorithm generates random partial spanning trees with large expected sum of edge weights. Object localizations are proposed as bounding-boxes of those partial trees. Our method has several benefits compared to the state-of-the-art. Thanks to the efficiency of Prim's algorithm, it samples proposals very quickly: 1000 proposals are obtained in about 0.7s. With proposals bound to super pixel boundaries yet diversified by randomization, it yields very high detection rates and windows that tightly fit objects. In extensive experiments on the challenging PASCAL VOC 2007 and 2012 and SUN2012 benchmark datasets, we show that our method improves over state-of-the-art competitors for a wide range of evaluation scenarios."
]
} |
1703.07144 | 2950001648 | Finding image correspondences remains a challenging problem in the presence of intra-class variations and large changes in scene layout. Semantic flow methods are designed to handle images depicting different instances of the same object or scene category. We introduce a novel approach to semantic flow, dubbed proposal flow, that establishes reliable correspondences using object proposals. Unlike prevailing semantic flow approaches that operate on pixels or regularly sampled local regions, proposal flow benefits from the characteristics of modern object proposals, that exhibit high repeatability at multiple scales, and can take advantage of both local and geometric consistency constraints among proposals. We also show that the corresponding sparse proposal flow can effectively be transformed into a conventional dense flow field. We introduce two new challenging datasets that can be used to evaluate both general semantic flow techniques and region-based approaches such as proposal flow. We use these benchmarks to compare different matching algorithms, object proposals, and region features within proposal flow, to the state of the art in semantic flow. This comparison, along with experiments on standard datasets, demonstrates that proposal flow significantly outperforms existing semantic flow methods in various settings. | While improving over pairwise correspondence results at the expense of runtime, these multi-image methods all use a pairwise method to find initial matches before refining them, (e.g., with cycle consistency @cite_28 ). Our correspondence method outperforms current pairwise methods, and its output could be used as a good initialization for multi-image methods. | {
"cite_N": [
"@cite_28"
],
"mid": [
"2949254743"
],
"abstract": [
"In this paper we propose a global optimization-based approach to jointly matching a set of images. The estimated correspondences simultaneously maximize pairwise feature affinities and cycle consistency across multiple images. Unlike previous convex methods relying on semidefinite programming, we formulate the problem as a low-rank matrix recovery problem and show that the desired semidefiniteness of a solution can be spontaneously fulfilled. The low-rank formulation enables us to derive a fast alternating minimization algorithm in order to handle practical problems with thousands of features. Both simulation and real experiments demonstrate that the proposed algorithm can achieve a competitive performance with an order of magnitude speedup compared to the state-of-the-art algorithm. In the end, we demonstrate the applicability of the proposed method to match the images of different object instances and as a result the potential to reconstruct category-specific object models from those images."
]
} |
1703.07144 | 2950001648 | Finding image correspondences remains a challenging problem in the presence of intra-class variations and large changes in scene layout. Semantic flow methods are designed to handle images depicting different instances of the same object or scene category. We introduce a novel approach to semantic flow, dubbed proposal flow, that establishes reliable correspondences using object proposals. Unlike prevailing semantic flow approaches that operate on pixels or regularly sampled local regions, proposal flow benefits from the characteristics of modern object proposals, that exhibit high repeatability at multiple scales, and can take advantage of both local and geometric consistency constraints among proposals. We also show that the corresponding sparse proposal flow can effectively be transformed into a conventional dense flow field. We introduce two new challenging datasets that can be used to evaluate both general semantic flow techniques and region-based approaches such as proposal flow. We use these benchmarks to compare different matching algorithms, object proposals, and region features within proposal flow, to the state of the art in semantic flow. This comparison, along with experiments on standard datasets, demonstrates that proposal flow significantly outperforms existing semantic flow methods in various settings. | Object proposals @cite_5 @cite_47 @cite_76 @cite_74 @cite_79 have originally been developed for object detection, where they are used to reduce the search space as well as false alarms. They are now an important component in many state-of-the-art detection pipelines @cite_41 @cite_65 and other computer vision applications, including object tracking @cite_15 , action recognition @cite_36 , weakly supervised localization @cite_72 , and semantic segmentation @cite_3 . Despite their success for object detection and segmentation, object proposals have seldom been used in matching tasks @cite_71 @cite_49 . In particular, while Cho . @cite_71 have shown that object proposals are useful for region matching due to their high repeatability on salient part regions, the use of object proposals has never been thoroughly investigated in semantic flow computation. The approach proposed in this paper is a first step in this direction, and we explore how the choice of object proposals, matching algorithms, and features affects matching robustness and accuracy. | {
"cite_N": [
"@cite_15",
"@cite_41",
"@cite_36",
"@cite_65",
"@cite_3",
"@cite_79",
"@cite_72",
"@cite_74",
"@cite_71",
"@cite_49",
"@cite_5",
"@cite_47",
"@cite_76"
],
"mid": [
"",
"",
"2950209802",
"2179352600",
"2949086864",
"7746136",
"",
"2088049833",
"",
"",
"",
"",
"2121660792"
],
"abstract": [
"",
"",
"There are multiple cues in an image which reveal what action a person is performing. For example, a jogger has a pose that is characteristic for jogging, but the scene (e.g. road, trail) and the presence of other joggers can be an additional source of information. In this work, we exploit the simple observation that actions are accompanied by contextual cues to build a strong action recognition system. We adapt RCNN to use more than one region for classification while still maintaining the ability to localize the action. We call our system R*CNN. The action-specific models and the feature maps are trained jointly, allowing for action specific representations to emerge. R*CNN achieves 90.2 mean AP on the PASAL VOC Action dataset, outperforming all other approaches in the field by a significant margin. Last, we show that R*CNN is not limited to action recognition. In particular, R*CNN can also be used to tackle fine-grained tasks such as attribute classification. We validate this claim by reporting state-of-the-art performance on the Berkeley Attributes of People dataset.",
"Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g. 224×224) input image. This requirement is “artificial” and may hurt the recognition accuracy for the images or sub-images of an arbitrary size scale. In this work, we equip the networks with a more principled pooling strategy, “spatial pyramid pooling”, to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size scale. By removing the fixed-size limitation, we can improve all CNN-based image classification methods in general. Our SPP-net achieves state-of-the-art accuracy on the datasets of ImageNet 2012, Pascal VOC 2007, and Caltech101.",
"Recent leading approaches to semantic segmentation rely on deep convolutional networks trained with human-annotated, pixel-level segmentation masks. Such pixel-accurate supervision demands expensive labeling effort and limits the performance of deep networks that usually benefit from more training data. In this paper, we propose a method that achieves competitive accuracy but only requires easily obtained bounding box annotations. The basic idea is to iterate between automatically generating region proposals and training convolutional networks. These two steps gradually recover segmentation masks for improving the networks, and vise versa. Our method, called BoxSup, produces competitive results supervised by boxes only, on par with strong baselines fully supervised by masks under the same setting. By leveraging a large amount of bounding boxes, BoxSup further unleashes the power of deep convolutional networks and yields state-of-the-art results on PASCAL VOC 2012 and PASCAL-CONTEXT.",
"The use of object proposals is an effective recent approach for increasing the computational efficiency of object detection. We propose a novel method for generating object bounding box proposals using edges. Edges provide a sparse yet informative representation of an image. Our main observation is that the number of contours that are wholly contained in a bounding box is indicative of the likelihood of the box containing an object. We propose a simple box objectness score that measures the number of edges that exist in the box minus those that are members of contours that overlap the box’s boundary. Using efficient data structures, millions of candidate boxes can be evaluated in a fraction of a second, returning a ranked set of a few thousand top-scoring proposals. Using standard metrics, we show results that are significantly more accurate than the current state-of-the-art while being faster to compute. In particular, given just 1000 proposals we achieve over 96 object recall at overlap threshold of 0.5 and over 75 recall at the more challenging overlap of 0.7. Our approach runs in 0.25 seconds and we additionally demonstrate a near real-time variant with only minor loss in accuracy.",
"",
"This paper addresses the problem of generating possible object locations for use in object recognition. We introduce selective search which combines the strength of both an exhaustive search and segmentation. Like segmentation, we use the image structure to guide our sampling process. Like exhaustive search, we aim to capture all possible object locations. Instead of a single technique to generate possible object locations, we diversify our search and use a variety of complementary image partitionings to deal with as many image conditions as possible. Our selective search results in a small set of data-driven, class-independent, high quality locations, yielding 99 recall and a Mean Average Best Overlap of 0.879 at 10,097 locations. The reduced number of locations compared to an exhaustive search enables the use of stronger machine learning techniques and stronger appearance models for object recognition. In this paper we show that our selective search enables the use of the powerful Bag-of-Words model for recognition. The selective search software is made publicly available (Software: http: disi.unitn.it uijlings SelectiveSearch.html ).",
"",
"",
"",
"",
"Generic object detection is the challenging task of proposing windows that localize all the objects in an image, regardless of their classes. Such detectors have recently been shown to benefit many applications such as speeding-up class-specific object detection, weakly supervised learning of object detectors and object discovery. In this paper, we introduce a novel and very efficient method for generic object detection based on a randomized version of Prim's algorithm. Using the connectivity graph of an image's super pixels, with weights modelling the probability that neighbouring super pixels belong to the same object, the algorithm generates random partial spanning trees with large expected sum of edge weights. Object localizations are proposed as bounding-boxes of those partial trees. Our method has several benefits compared to the state-of-the-art. Thanks to the efficiency of Prim's algorithm, it samples proposals very quickly: 1000 proposals are obtained in about 0.7s. With proposals bound to super pixel boundaries yet diversified by randomization, it yields very high detection rates and windows that tightly fit objects. In extensive experiments on the challenging PASCAL VOC 2007 and 2012 and SUN2012 benchmark datasets, we show that our method improves over state-of-the-art competitors for a wide range of evaluation scenarios."
]
} |
1703.07131 | 2604342492 | Current approaches for Knowledge Distillation (KD) either directly use training data or sample from the training data distribution. In this paper, we demonstrate effectiveness of 'mismatched' unlabeled stimulus to perform KD for image classification networks. For illustration, we consider scenarios where this is a complete absence of training data, or mismatched stimulus has to be used for augmenting a small amount of training data. We demonstrate that stimulus complexity is a key factor for distillation's good performance. Our examples include use of various datasets for stimulating MNIST and CIFAR teachers. | Knowledge Distillation (KD) is the process of transferring the generalization ability of a teacher model (usually large neural network) to a student model (usually a small neural network). @cite_3 demonstrated that the student network can be trained using an input data with combination of hard labels as well as soft labels. The hard labels are the ground truth labels (one-hot vectors) available for the training data while the soft labels are the output of the teacher network on the input data. Most of the distillation approaches either directly use training data or device a strategy to learn the training data distribution and then sample from it. However, the assumption of availability of labeled training data may not always hold true, due to various reasons. | {
"cite_N": [
"@cite_3"
],
"mid": [
"1821462560"
],
"abstract": [
"A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel."
]
} |
1703.07355 | 2604283646 | In online discussion communities, users can interact and share information and opinions on a wide variety of topics. However, some users may create multiple identities, or sockpuppets, and engage in undesired behavior by deceiving others or manipulating discussions. In this work, we study sockpuppetry across nine discussion communities, and show that sockpuppets differ from ordinary users in terms of their posting behavior, linguistic traits, as well as social network structure. Sockpuppets tend to start fewer discussions, write shorter posts, use more personal pronouns such as I'', and have more clustered ego-networks. Further, pairs of sockpuppets controlled by the same individual are more likely to interact on the same discussion at the same time than pairs of ordinary users. Our analysis suggests a taxonomy of deceptive behavior in discussion communities. Pairs of sockpuppets can vary in their deceptiveness, i.e., whether they pretend to be different users, or their supportiveness, i.e., if they support arguments of other sockpuppets controlled by the same user. We apply these findings to a series of prediction tasks, notably, to identify whether a pair of accounts belongs to the same underlying user or not. Altogether, this work presents a data-driven view of deception in online discussion communities and paves the way towards the automatic detection of sockpuppets. | Sockpuppetry is situated in the broader field of deception. Deception online is aided by the virtue of anonymity @cite_3 . It can occur as deceptive content as well as a deceptive agent @cite_38 . The behavior of people changes when they deceive, for example, they reduce communication @cite_43 and change the focus of their presentation @cite_44 . When writing deceptively to hide their identity, authors tend to increase use of particles and personal pronouns, write shorter sentences, and show nervousness @cite_23 @cite_27 @cite_36 @cite_15 . Our work adds to this line of research by finding evidence of deceptive writing styles and presentation by sockpuppets -- pretender sockpuppets may pretend to be different people by using different display names and they tend to write deceptively as well. | {
"cite_N": [
"@cite_38",
"@cite_36",
"@cite_3",
"@cite_44",
"@cite_43",
"@cite_27",
"@cite_23",
"@cite_15"
],
"mid": [
"1993637905",
"2136000463",
"1535174579",
"2136978983",
"2116625462",
"",
"1996665163",
""
],
"abstract": [
"The unknown and the invisible exploit the unwary and the uninformed for illicit financial gain and reputation damage.",
"Whistleblowers and activists need the ability to communicate without disclosing their identity, as of course do kidnappers and terrorists. Recent advances in the technology of stylometry (the study of authorial style) or \"authorship attribution\" have made it possible to identify the author with high reliability in a non-confrontational setting. In a confrontational setting, where the author is deliberately masking their identity (i.e. attempting to deceive), the results are much less promising. In this paper, we show that although the specific author may not be identifiable, the intent to deceive and to hide his identity can be. We show this by a reanalysis of the Brennan and Greenstadt (2009) deception corpus and discuss some of the implications of this surprising finding.",
"Identity plays a key role in virtual communities. In communication, which is the primary activity, knowing the identity of those with whom you communicate is essential for understanding and evaluating an interaction. Yet in the disembodied world of the virtual community, identity is also ambiguous. Many of the basic cues about personality and social role we are accustomed to in the physical world are absent. The goal of this paper is to understand how identity is established in an online community and to examine the effects of identity deception and the conditions that give rise to it.",
"This article investigates whether deceptions in online dating profiles correlate with changes in the way daters write about themselves in the free-text portion of the profile, and whether these changes are detectable by both computerized linguistic analyses and human judges. Computerized analyses (Study 1) found that deceptions manifested themselves through linguistic cues pertaining to (a) liars’ emotions and cognitions and (b) liars’ strategic efforts to manage their self-presentations. Technological affordances (i.e., asynchronicity and editability) affected the production of cognitive cues more than that of emotional cues. Human judges (Study 2) relied on different and nonpredictive linguistic cues to assess daters’ trustworthiness. The findings inform theories concerned with deception, media, and self-presentation, and also expound on how writing style influences perceived trustworthiness.",
"Advancing our knowledge about cues to deception is crucial to successful deception detection. A lengthy list of cues to deception has been identified via a myriad of deception studies. Nonetheless, we identified two major limitations of existing cues to deception: the lack of cues in computer-mediated communication and in non-Western group communication. In this research, we aim to make some contributions to addressing this line of inquiry. We conducted an empirical study on cues to deception using a large real-world online Chinese community. Through hypotheses testing, we observed a number of interesting findings. For example, we found that deceivers tended to communicate less and showed low complexity and high diversity in their messages. These findings provide significant implications to deception research and the broad online communication community.",
"",
"In digital forensics, questions often arise about the authors of documents: their identity, demographic background, and whether they can be linked to other documents. The field of stylometry uses linguistic features and machine learning techniques to answer these questions. While stylometry techniques can identify authors with high accuracy in non-adversarial scenarios, their accuracy is reduced to random guessing when faced with authors who intentionally obfuscate their writing style or attempt to imitate that of another author. While these results are good for privacy, they raise concerns about fraud. We argue that some linguistic features change when people hide their writing style and by identifying those features, stylistic deception can be recognized. The major contribution of this work is a method for detecting stylistic deception in written documents. We show that using a large feature set, it is possible to distinguish regular documents from deceptive documents with 96.6 accuracy (F-measure). We also present an analysis of linguistic features that can be modified to hide writing style.",
""
]
} |
1703.07355 | 2604283646 | In online discussion communities, users can interact and share information and opinions on a wide variety of topics. However, some users may create multiple identities, or sockpuppets, and engage in undesired behavior by deceiving others or manipulating discussions. In this work, we study sockpuppetry across nine discussion communities, and show that sockpuppets differ from ordinary users in terms of their posting behavior, linguistic traits, as well as social network structure. Sockpuppets tend to start fewer discussions, write shorter posts, use more personal pronouns such as I'', and have more clustered ego-networks. Further, pairs of sockpuppets controlled by the same individual are more likely to interact on the same discussion at the same time than pairs of ordinary users. Our analysis suggests a taxonomy of deceptive behavior in discussion communities. Pairs of sockpuppets can vary in their deceptiveness, i.e., whether they pretend to be different users, or their supportiveness, i.e., if they support arguments of other sockpuppets controlled by the same user. We apply these findings to a series of prediction tasks, notably, to identify whether a pair of accounts belongs to the same underlying user or not. Altogether, this work presents a data-driven view of deception in online discussion communities and paves the way towards the automatic detection of sockpuppets. | Turning to research that studies sockpuppetry specifically, one line of work has studied their motivations. Sockpuppetry is often used to avoid being banned, to create false consensus @cite_7 @cite_23 and support a person or a position @cite_29 , or vandalize content ( on Wikipedia @cite_41 ). Relatedly, motivations for multiple account creation in online multiplayer games can either be benign ( experimentation with different identities) or malicious ( increasing in-game profit, cheating) @cite_45 @cite_40 @cite_56 @cite_8 . In our work, we find evidence for these motivations -- sockpuppets in discussion communities sometimes support each other, and beyond malicious uses, some uses of sockpuppetry may be benign ( a user may simply use different accounts to post in different topics). | {
"cite_N": [
"@cite_7",
"@cite_8",
"@cite_41",
"@cite_29",
"@cite_56",
"@cite_40",
"@cite_45",
"@cite_23"
],
"mid": [
"21058638",
"",
"2127310906",
"2057310979",
"",
"",
"2073296652",
"1996665163"
],
"abstract": [
"A hydraulic control system for use in an automatic transmission including a fluid coupling type torque converter, and a gear system provided with two or more frictional engaging devices adapted to establish two or more speed conditions. The hydraulic system further includes a hydraulic pressure source, a line pressure control valve for supplying a regulated line pressure, by receiving a hydraulic pressure from the hydraulic pressure source, a throttle pressure control valve for generating a throttle pressure commensurate with an intake throttle opening, a governor pressure control valve for generating a governor pressure commensurate with vehicle speed, a manual change-over valve for enabling manual change-over of the transmission, two or more shift valves for shifting the path of the line pressure to be supplied to the frictional engaging devices in response to abalancing of the relationship between the throttle pressure and the governor pressure, the shift valves including one shift valve operative to establish the highest speed ratio for the transmission, and a highest speed control device for controlling to a low-speed-side shift position the one shift valve of the aforementioned shift valves which is operative to establish the highest speed ratio irrespective of the governor pressure resulting when the intake throttle opening remains over a given valve.",
"",
"This paper presents preliminary results of using authorship attribution methods for the detection of sockpuppeteering in Wikipedia. Sockpuppets are fake accounts created by malicious users to bypass Wikipedia’s regulations. Our dataset is composed of the comments made by the editors on the talk pages. To overcome the limitations of the short lengths of these comments, we use an voting scheme to combine predictions made on individual user entries. We show that this approach is promising and that it can be a viable alternative to the current human process that Wikipedia uses to resolve suspected sockpuppet cases.",
"This research has three goals: first, to find out how prevalent online deception is within a sample of Israeli users, second, to explore the underlying motivations to deceive online, and third, to discover the emotions that accompany online deception. A web-based survey was distributed in 14 discussion groups, and the answers of 257 respondents were analyzed. It was found that, while most of the respondents believe that online deception is very widespread, only about one-third of them reported engaging in online deception. Frequent users deceive online more than infrequent users, young users more than old, and competent users more than non-competent. The most common motivations to deceive online were \"play\" on the one hand and privacy concerns on the other. Most people felt a sense of enjoyment while engaging in online deception. The results are discussed in light of a possible mechanism for changing personal moral standards.",
"",
"",
"Multiple identity systems involve a physical self and a set of virtual identities.An 8-item model of avatar functions within a multiple identity system was identified.The vast majority of participants noted all or most functions in their interviews.The combined physical and virtual selves form a wider system of personality. Prior research has shown that approximately 50 of active participants in the 3D virtual world of Second Life have one or more secondary avatars or \"alts\" in addition to their primary avatar. Thus, these individuals are operating a \"multiple or poly-identity system\" composed of a physical self, a primary avatar, and one or more alts. However, little is known about the functions these virtual identities serve for the virtual-world user. The current study involved qualitative analysis of semistructured interviews with Second Life participants (N=24) who had a primary avatar and at least one alt. Interviews were coded to examine the functions that primary avatars and alts served. Eight functions-seven suggested by previous research on virtual world identity and one that emerged from analyses-were reflected in a large majority of the transcribed interviews and are described in the article. The current findings add to our understanding of how multifaceted identity systems operate, as more individuals augment their physical self with a set of virtual identities.",
"In digital forensics, questions often arise about the authors of documents: their identity, demographic background, and whether they can be linked to other documents. The field of stylometry uses linguistic features and machine learning techniques to answer these questions. While stylometry techniques can identify authors with high accuracy in non-adversarial scenarios, their accuracy is reduced to random guessing when faced with authors who intentionally obfuscate their writing style or attempt to imitate that of another author. While these results are good for privacy, they raise concerns about fraud. We argue that some linguistic features change when people hide their writing style and by identifying those features, stylistic deception can be recognized. The major contribution of this work is a method for detecting stylistic deception in written documents. We show that using a large feature set, it is possible to distinguish regular documents from deceptive documents with 96.6 accuracy (F-measure). We also present an analysis of linguistic features that can be modified to hide writing style."
]
} |
1703.07355 | 2604283646 | In online discussion communities, users can interact and share information and opinions on a wide variety of topics. However, some users may create multiple identities, or sockpuppets, and engage in undesired behavior by deceiving others or manipulating discussions. In this work, we study sockpuppetry across nine discussion communities, and show that sockpuppets differ from ordinary users in terms of their posting behavior, linguistic traits, as well as social network structure. Sockpuppets tend to start fewer discussions, write shorter posts, use more personal pronouns such as I'', and have more clustered ego-networks. Further, pairs of sockpuppets controlled by the same individual are more likely to interact on the same discussion at the same time than pairs of ordinary users. Our analysis suggests a taxonomy of deceptive behavior in discussion communities. Pairs of sockpuppets can vary in their deceptiveness, i.e., whether they pretend to be different users, or their supportiveness, i.e., if they support arguments of other sockpuppets controlled by the same user. We apply these findings to a series of prediction tasks, notably, to identify whether a pair of accounts belongs to the same underlying user or not. Altogether, this work presents a data-driven view of deception in online discussion communities and paves the way towards the automatic detection of sockpuppets. | Another line of work has also identified sockpuppets using textual information, link analysis and temporal information, both in online discussion forums @cite_34 @cite_30 and social networks @cite_25 @cite_39 @cite_30 @cite_32 . However, definitions of sockpuppets from previous research have tended to make assumptions about the usernames that sockpuppets use ( , that they are similar @cite_22 ), their opinion towards topics ( , they have the same opinion @cite_34 ), and their interactions (e.g., that they reply in support of each other's posts @cite_20 ). As such, these definitions tend to miss several types of sockpuppetry. In this work, we developed a robust methodology for identifying sockpuppets that makes fewer assumptions, and showed that a significant fraction of sockpuppets do use different names (i.e., the non-pretenders), and tend not to support each other in discussions (i.e., the non-supporters). | {
"cite_N": [
"@cite_30",
"@cite_22",
"@cite_32",
"@cite_39",
"@cite_34",
"@cite_25",
"@cite_20"
],
"mid": [
"",
"2398048567",
"2153644028",
"",
"2061364406",
"2009014039",
"2004661748"
],
"abstract": [
"",
"Users of social media sites can use more than one account. These identities have pseudo anonymous properties, and as such some users abuse multiple accounts to perform undesirable actions, such as posting false or misleading remarks comments that praise or defame the work of others. The detection of multiple user accounts that are controlled by an individual or organization is important. Herein, we define the problem as sockpuppet gang (SPG) detection. First, we analyze user sentiment orientation to topics based on emotional phrases extracted from their posted comments. Then we evaluate the similarity between sentiment orientations of user account pairs, and build a similar-orientation network (SON) where each vertex represents a user account on a social media site. In an SON, an edge exists only if the two user accounts have similar sentiment orientations to most topics. The boundary between detected SPGs may be indistinct, thus by analyzing account posting behavior features we propose a multiple random walk method to iteratively remeasure the weight of each edge. Finally, we adopt multiple community detection algorithms to detect SPGs in the network. User accounts in the same SPG are considered to be controlled by the same individual or organization. In our experiments on real world datasets, our method shows better performance than other contemporary methods.",
"Recently, there has been much excitement in the research community over using social networks to mitigate multiple identity, or Sybil, attacks. A number of schemes have been proposed, but they differ greatly in the algorithms they use and in the networks upon which they are evaluated. As a result, the research community lacks a clear understanding of how these schemes compare against each other, how well they would work on real-world social networks with different structural properties, or whether there exist other (potentially better) ways of Sybil defense. In this paper, we show that, despite their considerable differences, existing Sybil defense schemes work by detecting local communities (i.e., clusters of nodes more tightly knit than the rest of the graph) around a trusted node. Our finding has important implications for both existing and future designs of Sybil defense schemes. First, we show that there is an opportunity to leverage the substantial amount of prior work on general community detection algorithms in order to defend against Sybils. Second, our analysis reveals the fundamental limits of current social network-based Sybil defenses: We demonstrate that networks with well-defined community structure are inherently more vulnerable to Sybil attacks, and that, in such networks, Sybils can carefully target their links in order make their attacks more effective.",
"",
"On virtual spaces, some individuals use multiple usernames or copycat forge other users (usually called ''sock puppet'') to communicate with others. Those sock puppets are fake identities through which members of Internet community praise or create the illusion of support for the product or one's work, pretending to be a different person. A fundamental problem is how to identify these sock puppets. In this paper, we propose a sock puppet detection algorithm which combines authorship-identification techniques and link analysis. Firstly, we propose an interesting social network model in which links between two IDs are built if they have similar attitudes to most topics that both of them participate in; then, the edges are pruned according a hypothesis test, which consider the impact of their writing styles; finally, the link-based community detection for pruned network is performed. Compared to traditional methods, our approach has three advantages: (1) it conforms to the practical meanings of sock puppet community; (2) it can be applied in online situation; (3) it increases the efficiency of link analysis. In the experimental work, we evaluate our method using real datasets and compared our approach with several previous methods; the results have proved above advantages.",
"In this paper we discuss a piece of work which intends to provide some insights regarding the resolution of the hard problem of multiple identities detection. Based on hypothesis that each person is unique and identifiable whether in its writing style or social behavior, we propose a Framework relying on machine learning models and a deep analysis of social interactions, towards such detection.",
"One common trick to cheat people to believe fake products or a high-return low-risk investment scheme in the online discussion groups is t0 make use of sock puppets (i.e., use different fake identities pretending to be different persons to praise or create the illusion of support for the product). A fundamental problem is how to identify these sock puppets. In this paper, we propose two approaches to identify these puppet pairs that occur in the same forum as well as cross forum. The evaluation based on millions of real posts in two popular discussion forums in Hong Kong shows that the methods are effective."
]
} |
1703.07255 | 2605028456 | Many problems in image processing and computer vision (e.g. colorization, style transfer) can be posed as 'manipulating' an input image into a corresponding output image given a user-specified guiding signal. A holy-grail solution towards generic image manipulation should be able to efficiently alter an input image with any personalized signals (even signals unseen during training), such as diverse paintings and arbitrary descriptive attributes. However, existing methods are either inefficient to simultaneously process multiple signals (let alone generalize to unseen signals), or unable to handle signals from other modalities. In this paper, we make the first attempt to address the zero-shot image manipulation task. We cast this problem as manipulating an input image according to a parametric model whose key parameters can be conditionally generated from any guiding signal (even unseen ones). To this end, we propose the Zero-shot Manipulation Net (ZM-Net), a fully-differentiable architecture that jointly optimizes an image-transformation network (TNet) and a parameter network (PNet). The PNet learns to generate key transformation parameters for the TNet given any guiding signal while the TNet performs fast zero-shot image manipulation according to both signal-dependent parameters from the PNet and signal-invariant parameters from the TNet itself. Extensive experiments show that our ZM-Net can perform high-quality image manipulation conditioned on different forms of guiding signals (e.g. style images and attributes) in real-time (tens of milliseconds per image) even for unseen signals. Moreover, a large-scale style dataset with over 20,000 style images is also constructed to promote further research. | A lot of research efforts have been devoted to the image manipulation task, among which the most common and efficient approach is to train a convolutional neural network (CNN) which directly outputs a transformed image for the input content image @cite_15 @cite_2 @cite_5 @cite_16 @cite_17 @cite_24 @cite_6 . For example, in @cite_8 @cite_0 , a CNN is trained to perform colorization on input images, and in @cite_1 @cite_21 @cite_12 @cite_4 to transform content images according to specific styles. Although the most recent method by @cite_1 can process images (nearly) in real-time, it has to train a single network for each specific type of manipulation (e.g. a specific style image in style transfer) and cannot generalize to other types of manipulation (new style images or other forms of guiding signals) unless retraining the model for every type, which usually takes several hours and prevents them from being scaled to real-world applications. One of the most relevant works with ours in @cite_4 tries to encode multiple styles within a single network; however, their model focuses on increasing the diversity of output images and are still unable to handle diverse and unseen guiding signals from distinct modalities (e.g. text attributes). | {
"cite_N": [
"@cite_4",
"@cite_8",
"@cite_21",
"@cite_1",
"@cite_6",
"@cite_24",
"@cite_0",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_12",
"@cite_17"
],
"mid": [
"2952849295",
"",
"2952226636",
"2950689937",
"",
"",
"2326925005",
"2951277909",
"2952632681",
"2022508996",
"2951713345",
"2953054324",
"2952637581"
],
"abstract": [
"Recent progresses on deep discriminative and generative modeling have shown promising results on texture synthesis. However, existing feed-forward based methods trade off generality for efficiency, which suffer from many issues, such as shortage of generality (i.e., build one network per texture), lack of diversity (i.e., always produce visually identical output) and suboptimality (i.e., generate less satisfying visual effects). In this work, we focus on solving these issues for improved texture synthesis. We propose a deep generative feed-forward network which enables efficient synthesis of multiple textures within one single network and meaningful interpolation between them. Meanwhile, a suite of important techniques are introduced to achieve better convergence and diversity. With extensive experiments, we demonstrate the effectiveness of the proposed model and techniques for synthesizing a large number of textures and show its applications with the stylization.",
"",
"recently demonstrated that deep networks can generate beautiful textures and stylized images from a single texture example. However, their methods requires a slow and memory-consuming optimization process. We propose here an alternative approach that moves the computational burden to a learning stage. Given a single example of a texture, our approach trains compact feed-forward convolutional networks to generate multiple samples of the same texture of arbitrary size and to transfer artistic style from a given image to any other image. The resulting networks are remarkably light-weight and can generate textures of quality comparable to Gatys et al., but hundreds of times faster. More generally, our approach highlights the power and flexibility of generative feed-forward models trained with complex and expressive loss functions.",
"We consider image transformation problems, where an input image is transformed into an output image. Recent methods for such problems typically train feed-forward convolutional neural networks using a loss between the output and ground-truth images. Parallel work has shown that high-quality images can be generated by defining and optimizing loss functions based on high-level features extracted from pretrained networks. We combine the benefits of both approaches, and propose the use of perceptual loss functions for training feed-forward networks for image transformation tasks. We show results on image style transfer, where a feed-forward network is trained to solve the optimization problem proposed by in real-time. Compared to the optimization-based method, our network gives similar qualitative results but is three orders of magnitude faster. We also experiment with single-image super-resolution, where replacing a per-pixel loss with a perceptual loss gives visually pleasing results.",
"",
"",
"Given a grayscale photograph as input, this paper attacks the problem of hallucinating a plausible color version of the photograph. This problem is clearly underconstrained, so previous approaches have either relied on significant user interaction or resulted in desaturated colorizations. We propose a fully automatic approach that produces vibrant and realistic colorizations. We embrace the underlying uncertainty of the problem by posing it as a classification task and use class-rebalancing at training time to increase the diversity of colors in the result. The system is implemented as a feed-forward pass in a CNN at test time and is trained on over a million color images. We evaluate our algorithm using a “colorization Turing test,” asking human participants to choose between a generated and ground truth color image. Our method successfully fools humans on 32 of the trials, significantly higher than previous methods. Moreover, we show that colorization can be a powerful pretext task for self-supervised feature learning, acting as a cross-channel encoder. This approach results in state-of-the-art performance on several feature learning benchmarks.",
"Scene parsing is a technique that consist on giving a label to all pixels in an image according to the class they belong to. To ensure a good visual coherence and a high class accuracy, it is essential for a scene parser to capture image long range dependencies. In a feed-forward architecture, this can be simply achieved by considering a sufficiently large input context patch, around each pixel to be labeled. We propose an approach consisting of a recurrent convolutional neural network which allows us to consider a large input context, while limiting the capacity of the model. Contrary to most standard approaches, our method does not rely on any segmentation methods, nor any task-specific features. The system is trained in an end-to-end manner over raw pixels, and models complex spatial dependencies with low inference cost. As the context size increases with the built-in recurrence, the system identifies and corrects its own errors. Our approach yields state-of-the-art performance on both the Stanford Background Dataset and the SIFT Flow Dataset, while remaining very fast at test time.",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build \"fully convolutional\" networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet, the VGG net, and GoogLeNet) into fully convolutional networks and transfer their learned representations by fine-tuning to the segmentation task. We then define a novel architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes one third of a second for a typical image.",
"Scene labeling consists of labeling each pixel in an image with the category of the object it belongs to. We propose a method that uses a multiscale convolutional network trained from raw pixels to extract dense feature vectors that encode regions of multiple sizes centered on each pixel. The method alleviates the need for engineered features, and produces a powerful representation that captures texture, shape, and contextual information. We report results using multiple postprocessing methods to produce the final labeling. Among those, we propose a technique to automatically retrieve, from a pool of segmentation components, an optimal set of components that best explain the scene; these components are arbitrary, for example, they can be taken from a segmentation tree or from any family of oversegmentations. The system yields record accuracies on the SIFT Flow dataset (33 classes) and the Barcelona dataset (170 classes) and near-record accuracy on Stanford background dataset (eight classes), while being an order of magnitude faster than competing approaches, producing a 320×240 image labeling in less than a second, including feature extraction.",
"In this paper we address three different computer vision tasks using a single basic architecture: depth prediction, surface normal estimation, and semantic labeling. We use a multiscale convolutional network that is able to adapt easily to each task using only small modifications, regressing from the input image to the output map directly. Our method progressively refines predictions using a sequence of scales, and captures many image details without any superpixels or low-level segmentation. We achieve state-of-the-art performance on benchmarks for all three tasks.",
"The diversity of painting styles represents a rich visual vocabulary for the construction of an image. The degree to which one may learn and parsimoniously capture this visual vocabulary measures our understanding of the higher level features of paintings, if not images in general. In this work we investigate the construction of a single, scalable deep network that can parsimoniously capture the artistic style of a diversity of paintings. We demonstrate that such a network generalizes across a diversity of artistic styles by reducing a painting to a point in an embedding space. Importantly, this model permits a user to explore new painting styles by arbitrarily combining the styles learned from individual paintings. We hope that this work provides a useful step towards building rich models of paintings and offers a window on to the structure of the learned representation of artistic style.",
"We propose a novel semantic segmentation algorithm by learning a deconvolution network. We learn the network on top of the convolutional layers adopted from VGG 16-layer net. The deconvolution network is composed of deconvolution and unpooling layers, which identify pixel-wise class labels and predict segmentation masks. We apply the trained network to each proposal in an input image, and construct the final semantic segmentation map by combining the results from all proposals in a simple manner. The proposed algorithm mitigates the limitations of the existing methods based on fully convolutional networks by integrating deep deconvolution network and proposal-wise prediction; our segmentation method typically identifies detailed structures and handles objects in multiple scales naturally. Our network demonstrates outstanding performance in PASCAL VOC 2012 dataset, and we achieve the best accuracy (72.5 ) among the methods trained with no external data through ensemble with the fully convolutional network."
]
} |
1703.07255 | 2605028456 | Many problems in image processing and computer vision (e.g. colorization, style transfer) can be posed as 'manipulating' an input image into a corresponding output image given a user-specified guiding signal. A holy-grail solution towards generic image manipulation should be able to efficiently alter an input image with any personalized signals (even signals unseen during training), such as diverse paintings and arbitrary descriptive attributes. However, existing methods are either inefficient to simultaneously process multiple signals (let alone generalize to unseen signals), or unable to handle signals from other modalities. In this paper, we make the first attempt to address the zero-shot image manipulation task. We cast this problem as manipulating an input image according to a parametric model whose key parameters can be conditionally generated from any guiding signal (even unseen ones). To this end, we propose the Zero-shot Manipulation Net (ZM-Net), a fully-differentiable architecture that jointly optimizes an image-transformation network (TNet) and a parameter network (PNet). The PNet learns to generate key transformation parameters for the TNet given any guiding signal while the TNet performs fast zero-shot image manipulation according to both signal-dependent parameters from the PNet and signal-invariant parameters from the TNet itself. Extensive experiments show that our ZM-Net can perform high-quality image manipulation conditioned on different forms of guiding signals (e.g. style images and attributes) in real-time (tens of milliseconds per image) even for unseen signals. Moreover, a large-scale style dataset with over 20,000 style images is also constructed to promote further research. | On the other hand, some iterative approaches @cite_13 @cite_7 @cite_10 @cite_19 have been proposed to manipulate image, either patch by patch @cite_13 @cite_10 , or by iteratively updating the input image with hundreds of refinement @cite_7 to obtain the transformed image. Although these methods require no additional training for each new guiding signal, the iterative evaluation process usually takes tens of seconds even with GPU acceleration @cite_1 , which might be impractical especially for online users. | {
"cite_N": [
"@cite_13",
"@cite_7",
"@cite_1",
"@cite_19",
"@cite_10"
],
"mid": [
"",
"2475287302",
"2950689937",
"2952008036",
"2471440592"
],
"abstract": [
"",
"Rendering the semantic content of an image in different styles is a difficult image processing task. Arguably, a major limiting factor for previous approaches has been the lack of image representations that explicitly represent semantic information and, thus, allow to separate image content from style. Here we use image representations derived from Convolutional Neural Networks optimised for object recognition, which make high level image information explicit. We introduce A Neural Algorithm of Artistic Style that can separate and recombine the image content and style of natural images. The algorithm allows us to produce new images of high perceptual quality that combine the content of an arbitrary photograph with the appearance of numerous wellknown artworks. Our results provide new insights into the deep image representations learned by Convolutional Neural Networks and demonstrate their potential for high level image synthesis and manipulation.",
"We consider image transformation problems, where an input image is transformed into an output image. Recent methods for such problems typically train feed-forward convolutional neural networks using a loss between the output and ground-truth images. Parallel work has shown that high-quality images can be generated by defining and optimizing loss functions based on high-level features extracted from pretrained networks. We combine the benefits of both approaches, and propose the use of perceptual loss functions for training feed-forward networks for image transformation tasks. We show results on image style transfer, where a feed-forward network is trained to solve the optimization problem proposed by in real-time. Compared to the optimization-based method, our network gives similar qualitative results but is three orders of magnitude faster. We also experiment with single-image super-resolution, where replacing a per-pixel loss with a perceptual loss gives visually pleasing results.",
"Neural Style Transfer has recently demonstrated very exciting results which catches eyes in both academia and industry. Despite the amazing results, the principle of neural style transfer, especially why the Gram matrices could represent style remains unclear. In this paper, we propose a novel interpretation of neural style transfer by treating it as a domain adaptation problem. Specifically, we theoretically show that matching the Gram matrices of feature maps is equivalent to minimize the Maximum Mean Discrepancy (MMD) with the second order polynomial kernel. Thus, we argue that the essence of neural style transfer is to match the feature distributions between the style images and the generated images. To further support our standpoint, we experiment with several other distribution alignment methods, and achieve appealing results. We believe this novel interpretation connects these two important research fields, and could enlighten future researches.",
"This paper presents a novel unsupervised method to transfer the style of an example image to a source image. The complex notion of image style is here considered as a local texture transfer, eventually coupled with a global color transfer. For the local texture transfer, we propose a new method based on an adaptive patch partition that captures the style of the example image and preserves the structure of the source image. More precisely, this example-based partition predicts how well a source patch matches an example patch. Results on various images show that our method outperforms the most recent techniques."
]
} |
1703.06680 | 2601039388 | In this paper we consider the problem of identifying intersections between two sets of d-dimensional axis-parallel rectangles. This is a common operation that arises in many agent-based simulation studies, and is of central importance in the context of High Level Architecture (HLA), where it is at the core of the Data Distribution Management (DDM) service. Several realizations of the DDM service have been proposed; however, many of them are either inefficient or inherently sequential. We propose a parallel version of the Sort-Based Matching algorithm for shared-memory multiprocessors. Sort-Based Matching is one of the most efficient serial algorithms for the DDM problem, but is quite difficult to parallelize because of data dependencies. We describe the algorithm and compute its asymptotic running time; we complete the analysis by assessing its performance and scalability through extensive experiments on two commodity multicore systems based on a dual socket Intel Xeon processor, and a single socket Intel Core i7 processor. | Data structures such as @math - @math trees @cite_25 and R-trees @cite_8 are able to efficiently store volumetric objects and identify intersections. Such data structures are quite complex to implement and, in many real-world situations, slower than less efficient but simpler solutions @cite_18 . For example, in @cite_24 the authors introduced a rectangle-intersection algorithm that is implemented using only simple data structures (i.e., arrays) and that can enumerate all @math intersections among @math rectangles with complexity @math time and @math space. | {
"cite_N": [
"@cite_24",
"@cite_18",
"@cite_25",
"@cite_8"
],
"mid": [
"2072647561",
"117292834",
"2122630376",
"2118269922"
],
"abstract": [
"It is demonstrated that power-efficient software also requires simplicity and the use of elementary data structures in addition to asymptotically optimal CPU and memory requirements. Though in the past few decades much effort has been devoted to reporting all k intersecting pairs in a planar set of n iso-oriented rectangles, all the known algorithms using elementary data structures, such as linked lists, are either not optimal, report some intersections repeatedly or fail to report some altogether. A simpler algorithm is proposed that uses only linear arrays and that takes O(n log n + k) time and O(n) space, which are the best possible under the algebraic RAM model of computation. The algorithm is designed for systems with limited resources, such as mobile 3D graphics, and can be implemented in less than 100 lines of Java code.",
"The High Level Architecture (HLA) is a standard for constructing distributed simulations. The Data Distribution Management services of HLA reduce the amount of data delivered to an HLA federate by allowing communications connections to be based on federates’ expressed data production and requirements. At the core of determining which connections to make is a geometric problem: finding the dynamic intersection of d-dimensional rectilinear hyperrectangles in d-space. Four different algorithms for solving that problem are described, including a new one developed through application of a data structure from computational geometry. Those algorithms are then compared in an experiment designed to reveal how well they perform in the specific context of the data distribution application. Both intersection performance and connectivity efficiency results are reported.",
"This paper compares three data structures that support area operations on 2-space: linked lists, quad trees, and multidimensional binary trees (k-d trees). Region searching is the most important operation these data structures must support in many applications. Insertion and deletion must also be reasonably fast. The three data structures are described and implementation considerations are discussed. Extensive experimentation was done using an experimental program that executed algorithms for all three data structures on the same problem. Results of this experimentation is presented to show the superior performance of k-d trees. The conclusion of this paper is that in applications where region search (or point search) on large problems is crucial (as in computer-aided design), k-d trees provide superior performance. Quad trees have become quite popular recently to solve the same problem--this may be due to unfamiliarity with k-d trees or a feeling that they are hard to implement. Actual working code for k-d trees (and quad trees) is included in this paper in an attempt to address both these problems.",
"In order to handle spatial data efficiently, as required in computer aided design and geo-data applications, a database system needs an index mechanism that will help it retrieve data items quickly according to their spatial locations However, traditional indexing methods are not well suited to data objects of non-zero size located m multi-dimensional spaces In this paper we describe a dynamic index structure called an R-tree which meets this need, and give algorithms for searching and updating it. We present the results of a series of tests which indicate that the structure performs well, and conclude that it is useful for current database systems in spatial applications"
]
} |
1703.06680 | 2601039388 | In this paper we consider the problem of identifying intersections between two sets of d-dimensional axis-parallel rectangles. This is a common operation that arises in many agent-based simulation studies, and is of central importance in the context of High Level Architecture (HLA), where it is at the core of the Data Distribution Management (DDM) service. Several realizations of the DDM service have been proposed; however, many of them are either inefficient or inherently sequential. We propose a parallel version of the Sort-Based Matching algorithm for shared-memory multiprocessors. Sort-Based Matching is one of the most efficient serial algorithms for the DDM problem, but is quite difficult to parallelize because of data dependencies. We describe the algorithm and compute its asymptotic running time; we complete the analysis by assessing its performance and scalability through extensive experiments on two commodity multicore systems based on a dual socket Intel Xeon processor, and a single socket Intel Core i7 processor. | Among the many matching algorithms that have been proposed for enumerating all intersections among subscription and update extents, the SBM @cite_16 proved to be very efficient. SBM solves the region matching problem in one dimension; SBM first sorts the endpoints, and then scans the sorted set. @cite_13 , SBM has been extended to deal with dynamic environments in which extents are dynamic (both in terms of placement and size). On the other hand, SBM has the drawback that it can not be trivially parallelized due to the presence of a sequential scan phase that is intrinsically serial. This is a serious limitation since the most of modern processing architectures are multi or many-cores. | {
"cite_N": [
"@cite_16",
"@cite_13"
],
"mid": [
"2054316261",
"2013573646"
],
"abstract": [
"The High Level Architecture (HLA) is an architecture for reuse and interoperation of simulations. It provides several Data Distribution Management (DDM) services to reduce the transmission and reception of irrelevant data. These services rely on the computation of the intersection between \"update\" and \"subscription\" regions. Currently, there are several main DDM filtering algorithms. Since each approach still has some shortcomings, we have focused our research on the design and the evaluation of intersection algorithms for the DDM. In this article, we introduce a new algorithm in which extents are sorted before computing the intersections. Our experiments show that usually the sort-based algorithm has the best performance among all approaches. The improvement of its performance ranges between 30 and 99 over the brute force and hybrid approaches.",
"Simulation is a low-cost and safe alternative to solve complex problems in various areas. To promote reuse and interoperability of simulation applications and link geographically dispersed simulation components, distributed simulation was introduced. The High-Level Architecture (HLA) is the IEEE standard for distributed simulation. To optimize communication efficiency between simulation components, HLA defines a Data Distribution Management (DDM) service group for filtering out unnecessary data exchange. It relies on the computation of overlap between update and subscription regions, which is called “matching”. There are many existing matching algorithms, among which a sort-based approach improves efficiency by sorting region bounds before the actual matching process, and is found to outperform other existing matching algorithms in many situations. However, the existing algorithm performs matching for all regions in one round and cannot dynamically deal with a selective region modification without processing all the regions once again. Realizing that in many spatial applications, only a small subset of all regions are actually modified in each time step, this article proposes a dynamic sort-based matching algorithm to deal with this efficiently. Theoretical analysis has been carried out for the proposed algorithm and experimental results show that the proposed algorithm has significantly better performance than major existing matching algorithms at dynamic matching."
]
} |
1703.06680 | 2601039388 | In this paper we consider the problem of identifying intersections between two sets of d-dimensional axis-parallel rectangles. This is a common operation that arises in many agent-based simulation studies, and is of central importance in the context of High Level Architecture (HLA), where it is at the core of the Data Distribution Management (DDM) service. Several realizations of the DDM service have been proposed; however, many of them are either inefficient or inherently sequential. We propose a parallel version of the Sort-Based Matching algorithm for shared-memory multiprocessors. Sort-Based Matching is one of the most efficient serial algorithms for the DDM problem, but is quite difficult to parallelize because of data dependencies. We describe the algorithm and compute its asymptotic running time; we complete the analysis by assessing its performance and scalability through extensive experiments on two commodity multicore systems based on a dual socket Intel Xeon processor, and a single socket Intel Core i7 processor. | Only few parallel solutions for DDM and interest matching @cite_15 have been proposed. Among them, the authors of this paper have proposed the ITM algorithm for computing intersections among @math -rectangles @cite_10 . ITM is based on an interval tree data structure, and after the tree is built, exhibits an embarrassingly parallel structure. The performance evaluation reported in @cite_10 shows that the sequential implementation of ITM is competitive with the sequential implementation of SBM . | {
"cite_N": [
"@cite_15",
"@cite_10"
],
"mid": [
"2099783956",
"1975191193"
],
"abstract": [
"Interest management is a filtering technique which is designed to reduce bandwidth consumption in Distributed Virtual Environments. This technique usually involves a process called \"interest matching\", which determines what data should be filtered. Existing interest matching algorithms, however, are mainly designed for serial processing which is supposed to be run on a single processor. As the problem size grows, these algorithms may not be scalable since the single processor may eventually become a bottleneck. In this paper, a parallel approach for interest matching is presented which is suitable to deploy on both shared-memory and distributed-memory multiprocessors. We also provide an analysis of speed-up and efficiency for the simulation results of the parallel algorithms.",
"Identifying intersections among a set of d-dimensional rectangular regions (d-rectangles) is a common problem in many simulation and modeling applications. Since algorithms for computing intersections over a large number of regions can be computationally demanding, an obvious solution is to take advantage of the multiprocessing capabilities of modern multicore processors. Unfortunately, many solutions employed for the Data Distribution Management service of the High Level Architecture are either inefficient, or can only partially be parallelized. In this paper we propose the Interval Tree Matching(ITM) algorithm for computing intersections among d-rectangles. ITMis based on a simple Interval Tree data structure, and exhibits an embarrassingly parallel structure. We implement the ITM algorithm, and compare its sequential performance with two widely used solutions(brute force and sort-based matching). We also analyze the scalability of ITM on shared-memory multicore processors. The results show that the sequential implementation of ITM is competitive with sort-based matching, moreover, the parallel implementation provides good speed upon multicore processors."
]
} |
1703.06680 | 2601039388 | In this paper we consider the problem of identifying intersections between two sets of d-dimensional axis-parallel rectangles. This is a common operation that arises in many agent-based simulation studies, and is of central importance in the context of High Level Architecture (HLA), where it is at the core of the Data Distribution Management (DDM) service. Several realizations of the DDM service have been proposed; however, many of them are either inefficient or inherently sequential. We propose a parallel version of the Sort-Based Matching algorithm for shared-memory multiprocessors. Sort-Based Matching is one of the most efficient serial algorithms for the DDM problem, but is quite difficult to parallelize because of data dependencies. We describe the algorithm and compute its asymptotic running time; we complete the analysis by assessing its performance and scalability through extensive experiments on two commodity multicore systems based on a dual socket Intel Xeon processor, and a single socket Intel Core i7 processor. | @cite_14 , a parallel ordered-relation-based matching algorithm is proposed. The algorithm is composed of five phases: projection, sorting, task decomposition, internal matching and external matching. In the experimental evaluation, a MATLAB implementation is compared with the sequential SBM . The results show that, with a high number of extents the proposed algorithm is faster than SBM . | {
"cite_N": [
"@cite_14"
],
"mid": [
"2001104644"
],
"abstract": [
"In distribution simulation based on High-level architecture (HLA), data distribution management (DDM) is one of HLA services for the purpose of filtering the unnecessary data transferring over the network. DDM admits the sending federates and the receiving federates to express their interest using update regions and subscription regions in a multidimensional routing space. There are several matching algorithms to obtain overlap information between the update regions and subscription regions. When the number of regions increase sharply, the matching process is time consuming. However, the existing algorithms is hard to be parallelized to take advantage of the computing capabilities of multi-core processors. To reduce the computational overhead of region matching, we propose a parallel algorithm based on order relation to accelerate the matching process. The new matching algorithm adopts divide-and-conquer approach to divide the regions into multiple region bound sublists, each of which comprises parts of region bounds. To calculate the intersection inside and amongst the region bound sublists, two matching rules are presented. This approach has good performance since it performs region matching on the sublists parallel and does not require unnecessary comparisons within regions in different sublists. Theoretical analysis has been carried out for the proposed algorithm and experimental result shows that the proposed algorithm has better performance than major existing DDM matching algorithms."
]
} |
1703.07023 | 2949285404 | In contrast to the widely studied problem of recognizing an action given a complete sequence, action anticipation aims to identify the action from only partially available videos. As such, it is therefore key to the success of computer vision applications requiring to react as early as possible, such as autonomous navigation. In this paper, we propose a new action anticipation method that achieves high prediction accuracy even in the presence of a very small percentage of a video sequence. To this end, we develop a multi-stage LSTM architecture that leverages context-aware and action-aware features, and introduce a novel loss function that encourages the model to predict the correct class as early as possible. Our experiments on standard benchmark datasets evidence the benefits of our approach; We outperform the state-of-the-art action anticipation methods for early prediction by a relative increase in accuracy of 22.0 on JHMDB-21, 14.0 on UT-Interaction and 49.9 on UCF-101. | The idea of action anticipation was introduced by @cite_9 , which models causal relationships to predict human activities. This was followed by several attempts to model the dynamics of the observed actions, such as by introducing integral and dynamic bag-of-words @cite_44 , using spatial-temporal implicit shape models @cite_16 , extracting human body movements via skeleton information @cite_24 , and accounting for the complete and partial history of observed features @cite_54 . | {
"cite_N": [
"@cite_54",
"@cite_9",
"@cite_44",
"@cite_24",
"@cite_16"
],
"mid": [
"66452226",
"2533503513",
"2147615062",
"2168328261",
"2150354123"
],
"abstract": [
"The speed with which intelligent systems can react to an action depends on how soon it can be recognized. The ability to recognize ongoing actions is critical in many applications, for example, spotting criminal activity. It is challenging, since decisions have to be made based on partial videos of temporally incomplete action executions. In this paper, we propose a novel discriminative multi-scale model for predicting the action class from a partially observed video. The proposed model captures temporal dynamics of human actions by explicitly considering all the history of observed features as well as features in smaller temporal segments. We develop a new learning formulation, which elegantly captures the temporal evolution over time, and enforces the label consistency between segments and corresponding partial videos. Experimental results on two public datasets show that the proposed approach outperforms state-of-the-art action prediction methods.",
"Human activity recognition is a challenging task, especially when its background is unknown or changing, and when scale or illumination differs in each video. Approaches utilizing spatio-temporal local features have proved that they are able to cope with such difficulties, but they mainly focused on classifying short videos of simple periodic actions. In this paper, we present a new activity recognition methodology that overcomes the limitations of the previous approaches using local features. We introduce a novel matching, spatio-temporal relationship match, which is designed to measure structural similarity between sets of features extracted from two videos. Our match hierarchically considers spatio-temporal relationships among feature points, thereby enabling detection and localization of complex non-periodic activities. In contrast to previous approaches to ‘classify’ videos, our approach is designed to ‘detect and localize’ all occurring activities from continuous videos where multiple actors and pedestrians are present. We implement and test our methodology on a newly-introduced dataset containing videos of multiple interacting persons and individual pedestrians. The results confirm that our system is able to recognize complex non-periodic activities (e.g. ‘push’ and ‘hug’) from sets of spatio-temporal features even when multiple activities are present in the scene",
"In this paper, we present a novel approach of human activity prediction. Human activity prediction is a probabilistic process of inferring ongoing activities from videos only containing onsets (i.e. the beginning part) of the activities. The goal is to enable early recognition of unfinished activities as opposed to the after-the-fact classification of completed activities. Activity prediction methodologies are particularly necessary for surveillance systems which are required to prevent crimes and dangerous activities from occurring. We probabilistically formulate the activity prediction problem, and introduce new methodologies designed for the prediction. We represent an activity as an integral histogram of spatio-temporal features, efficiently modeling how feature distributions change over time. The new recognition methodology named dynamic bag-of-words is developed, which considers sequential nature of human activities while maintaining advantages of the bag-of-words to handle noisy observations. Our experiments confirm that our approach reliably recognizes ongoing activities from streaming videos with a high accuracy.",
"Online human gesture recognition has a wide range of applications in computer vision, especially in human-computer interaction applications. Recent introduction of cost-effective depth cameras brings on a new trend of research on body-movement gesture recognition. However, there are two major challenges: i) how to continuously recognize gestures from unsegmented streams, and ii) how to differentiate different styles of a same gesture from other types of gestures. In this paper, we solve these two problems with a new effective and efficient feature extraction method that uses a dynamic matching approach to construct a feature vector for each frame and improves sensitivity to the features of different gestures and decreases sensitivity to the features of gestures within the same class. Our comprehensive experiments on MSRC-12 Kinect Gesture and MSR-Action3D datasets have demonstrated a superior performance than the stat-of-the-art approaches.",
"Early recognition and prediction of human activities are of great importance in video surveillance, e.g., by recognizing a criminal activity at its beginning stage, it is possible to avoid unfortunate outcomes. We address early activity recognition by developing a Spatial-Temporal Implicit Shape Model (STISM), which characterizes the space-time structure of the sparse local features extracted from a video. The early recognition of human activities is accomplished by pattern matching through STISM. To enable efficient and robust matching, we propose a new random forest structure, called multi-class balanced random forest, which makes a good trade-off between the balance of the trees and the discriminative abilities. The prediction is done simultaneously for multiple classes, which saves both the memory and computational cost. The experiments show that our algorithm significantly outperforms the state of the arts for the human activity prediction problem."
]
} |
1703.07023 | 2949285404 | In contrast to the widely studied problem of recognizing an action given a complete sequence, action anticipation aims to identify the action from only partially available videos. As such, it is therefore key to the success of computer vision applications requiring to react as early as possible, such as autonomous navigation. In this paper, we propose a new action anticipation method that achieves high prediction accuracy even in the presence of a very small percentage of a video sequence. To this end, we develop a multi-stage LSTM architecture that leverages context-aware and action-aware features, and introduce a novel loss function that encourages the model to predict the correct class as early as possible. Our experiments on standard benchmark datasets evidence the benefits of our approach; We outperform the state-of-the-art action anticipation methods for early prediction by a relative increase in accuracy of 22.0 on JHMDB-21, 14.0 on UT-Interaction and 49.9 on UCF-101. | More recently, @cite_37 proposed to make use of binary SVMs to classify video snippets into sub-action categories and obtain the final class label in an online manner using dynamic programming. To overcome the need to train one classifier per sub-action, @cite_47 extended this approach to using a structural SVM. Importantly, this work further introduced a new objective function to encourage the score of the correct action to increase as time progresses. | {
"cite_N": [
"@cite_37",
"@cite_47"
],
"mid": [
"2460134573",
"2560009937"
],
"abstract": [
"This paper proposes a novel approach to tackle the challenging problem of 'online action localization' which entails predicting actions and their locations as they happen in a video. Typically, action localization or recognition is performed in an offline manner where all the frames in the video are processed together and action labels are not predicted for the future. This disallows timely localization of actions - an important consideration for surveillance tasks. In our approach, given a batch of frames from the immediate past in a video, we estimate pose and oversegment the current frame into superpixels. Next, we discriminatively train an actor foreground model on the superpixels using the pose bounding boxes. A Conditional Random Field with superpixels as nodes, and edges connecting spatio-temporal neighbors is used to obtain action segments. The action confidence is predicted using dynamic programming on SVM scores obtained on short segments of the video, thereby capturing sequential information of the actions. The issue of visual drift is handled by updating the appearance model and pose refinement in an online manner. Lastly, we introduce a new measure to quantify the performance of action prediction (i.e. online action localization), which analyzes how the prediction accuracy varies as a function of observed portion of the video. Our experiments suggest that despite using only a few frames to localize actions at each time instant, we are able to predict the action and obtain competitive results to state-of-the-art offline methods.",
"This paper proposes a person-centric and online approach to the challenging problem of localization and prediction of actions and interactions in videos. Typically, localization or recognition is performed in an offline manner where all the frames in the video are processed together. This prevents timely localization and prediction of actions and interactions - an important consideration for many tasks including surveillance and human-machine interaction. In our approach, we estimate human poses at each frame and train discriminative appearance models using the superpixels inside the pose bounding boxes. Since the pose estimation per frame is inherently noisy, the conditional probability of pose hypotheses at current time-step (frame) is computed using pose estimations in the current frame and their consistency with poses in the previous frames. Next, both the superpixel and pose-based foreground likelihoods are used to infer the location of actors at each time through a Conditional Random. The issue of visual drift is handled by updating the appearance models, and refining poses using motion smoothness on joint locations, in an online manner. For online prediction of action (interaction) confidences, we propose an approach based on Structural SVM that operates on short video segments, and is trained with the objective that confidence of an action or interaction increases as time progresses. Lastly, we quantify the performance of both detection and prediction together, and analyze how the prediction accuracy varies as a time function of observed action (interaction) at different levels of detection performance. Our experiments on several datasets suggest that despite using only a few frames to localize actions (interactions) at each time instant, we are able to obtain competitive results to state-of-the-art offline methods."
]
} |
1703.07023 | 2949285404 | In contrast to the widely studied problem of recognizing an action given a complete sequence, action anticipation aims to identify the action from only partially available videos. As such, it is therefore key to the success of computer vision applications requiring to react as early as possible, such as autonomous navigation. In this paper, we propose a new action anticipation method that achieves high prediction accuracy even in the presence of a very small percentage of a video sequence. To this end, we develop a multi-stage LSTM architecture that leverages context-aware and action-aware features, and introduce a novel loss function that encourages the model to predict the correct class as early as possible. Our experiments on standard benchmark datasets evidence the benefits of our approach; We outperform the state-of-the-art action anticipation methods for early prediction by a relative increase in accuracy of 22.0 on JHMDB-21, 14.0 on UT-Interaction and 49.9 on UCF-101. | While the above-mentioned work made use of handcrafted features, recent advances have naturally led to the development of deep learning approaches to action anticipation. In this context, @cite_28 proposed to combine a Convolutional Neural Network (CNN) with an LSTM to model both spatial and temporal information. The authors further introduced new ranking losses whose goal is to enforce either the score of the correct class or the margin between the score of the correct class and that of the best score to be non-decreasing over time. Similarly, in @cite_12 , a new loss that penalizes classification mistakes increasingly strongly over time was introduced in an LSTM-based framework that used multiple modalities. While the two above-mentioned methods indeed aim at improving classification accuracy over time, they do not explicitly encourage making correct predictions as early as possible. By contrast, while accounting for ambiguities in early stages, our new loss still aims to prevent false negatives from the beginning of the sequence. | {
"cite_N": [
"@cite_28",
"@cite_12"
],
"mid": [
"2472970127",
"2174887554"
],
"abstract": [
"In this work we improve training of temporal deep models to better learn activity progression for activity detection and early detection tasks. Conventionally, when training a Recurrent Neural Network, specifically a Long Short Term Memory (LSTM) model, the training loss only considers classification error. However, we argue that the detection score of the correct activity category, or the detection score margin between the correct and incorrect categories, should be monotonically non-decreasing as the model observes more of the activity. We design novel ranking losses that directly penalize the model on violation of such monotonicities, which are used together with classification loss in training of LSTM models. Evaluation on ActivityNet shows significant benefits of the proposed ranking losses in both activity detection and early detection tasks.",
"Anticipating the future actions of a human is a widely studied problem in robotics that requires spatio-temporal reasoning. In this work we propose a deep learning approach for anticipation in sensory-rich robotics applications. We introduce a sensory-fusion architecture which jointly learns to anticipate and fuse information from multiple sensory streams. Our architecture consists of Recurrent Neural Networks (RNNs) that use Long Short-Term Memory (LSTM) units to capture long temporal dependencies. We train our architecture in a sequence-to-sequence prediction manner, and it explicitly learns to predict the future given only a partial temporal context. We further introduce a novel loss layer for anticipation which prevents over-fitting and encourages early anticipation. We use our architecture to anticipate driving maneuvers several seconds before they happen on a natural driving data set of 1180 miles. The context for maneuver anticipation comes from multiple sensors installed on the vehicle. Our approach shows significant improvement over the state-of-the-art in maneuver anticipation by increasing the precision from 77.4 to 90.5 and recall from 71.2 to 87.4 ."
]
} |
1703.07023 | 2949285404 | In contrast to the widely studied problem of recognizing an action given a complete sequence, action anticipation aims to identify the action from only partially available videos. As such, it is therefore key to the success of computer vision applications requiring to react as early as possible, such as autonomous navigation. In this paper, we propose a new action anticipation method that achieves high prediction accuracy even in the presence of a very small percentage of a video sequence. To this end, we develop a multi-stage LSTM architecture that leverages context-aware and action-aware features, and introduce a novel loss function that encourages the model to predict the correct class as early as possible. Our experiments on standard benchmark datasets evidence the benefits of our approach; We outperform the state-of-the-art action anticipation methods for early prediction by a relative increase in accuracy of 22.0 on JHMDB-21, 14.0 on UT-Interaction and 49.9 on UCF-101. | Instead of predicting the future class label, in @cite_30 , the authors proposed to predict the future visual representation. However, the main motivation for this was to work with unlabeled videos, and the learned representation is therefore not always related to the action itself. | {
"cite_N": [
"@cite_30"
],
"mid": [
"2951242004"
],
"abstract": [
"Anticipating actions and objects before they start or appear is a difficult problem in computer vision with several real-world applications. This task is challenging partly because it requires leveraging extensive knowledge of the world that is difficult to write down. We believe that a promising resource for efficiently learning this knowledge is through readily available unlabeled video. We present a framework that capitalizes on temporal structure in unlabeled video to learn to anticipate human actions and objects. The key idea behind our approach is that we can train deep networks to predict the visual representation of images in the future. Visual representations are a promising prediction target because they encode images at a higher semantic level than pixels yet are automatic to compute. We then apply recognition algorithms on our predicted representation to anticipate objects and actions. We experimentally validate this idea on two datasets, anticipating actions one second in the future and objects five seconds in the future."
]
} |
1703.07023 | 2949285404 | In contrast to the widely studied problem of recognizing an action given a complete sequence, action anticipation aims to identify the action from only partially available videos. As such, it is therefore key to the success of computer vision applications requiring to react as early as possible, such as autonomous navigation. In this paper, we propose a new action anticipation method that achieves high prediction accuracy even in the presence of a very small percentage of a video sequence. To this end, we develop a multi-stage LSTM architecture that leverages context-aware and action-aware features, and introduce a novel loss function that encourages the model to predict the correct class as early as possible. Our experiments on standard benchmark datasets evidence the benefits of our approach; We outperform the state-of-the-art action anticipation methods for early prediction by a relative increase in accuracy of 22.0 on JHMDB-21, 14.0 on UT-Interaction and 49.9 on UCF-101. | Most recent action approaches extract global representations for the entire image @cite_10 @cite_50 @cite_39 or video sequence @cite_40 @cite_38 . As such, these methods do not truly focus on the actions of interest, but rather compute a representation. Unfortunately, context does not always bring reliable information about the action. For example, one can play guitar in a bedroom, a concert hall or a yard. To overcome this, some methods localize the feature extraction process by exploiting dense trajectories @cite_33 @cite_4 @cite_35 or optical flow @cite_3 . Inspired by objectness, the notion of actionness @cite_29 @cite_7 @cite_14 @cite_52 @cite_1 @cite_32 has recently also been proposed to localize the regions where a generic action occurs. The resulting methods can then be thought of as extracting representations. In other words, these methods go to the other extreme and completely discard the notion of context which can be useful for some actions, such as playing football on a grass field. | {
"cite_N": [
"@cite_38",
"@cite_35",
"@cite_14",
"@cite_4",
"@cite_33",
"@cite_7",
"@cite_29",
"@cite_1",
"@cite_52",
"@cite_3",
"@cite_39",
"@cite_32",
"@cite_40",
"@cite_50",
"@cite_10"
],
"mid": [
"2308045930",
"2471775118",
"2018068650",
"2105101328",
"1944615693",
"",
"2410118306",
"",
"1945129080",
"2465313502",
"2951183276",
"",
"2952633803",
"",
"2465503160"
],
"abstract": [
"",
"We present hierarchical rank pooling, a video sequence encoding method for activity recognition. It consists of a network of rank pooling functions which captures the dynamics of rich convolutional neural network features within a video sequence. By stacking non-linear feature functions and rank pooling over one another, we obtain a high capacity dynamic encoding mechanism, which is used for action recognition. We present a method for jointly learning the video representation and activity classifier parameters. Our method obtains state-of-the art results on three important activity recognition benchmarks: 76.7 on Hollywood2, 66.9 on HMDB51 and, 91.4 on UCF101.",
"This paper considers the problem of action localization, where the objective is to determine when and where certain actions appear. We introduce a sampling strategy to produce 2D+t sequences of bounding boxes, called tubelets. Compared to state-of-the-art alternatives, this drastically reduces the number of hypotheses that are likely to include the action of interest. Our method is inspired by a recent technique introduced in the context of image localization. Beyond considering this technique for the first time for videos, we revisit this strategy for 2D+t sequences obtained from super-voxels. Our sampling strategy advantageously exploits a criterion that reflects how action related motion deviates from background motion. We demonstrate the interest of our approach by extensive experiments on two public datasets: UCF Sports and MSR-II. Our approach significantly outperforms the state-of-the-art on both datasets, while restricting the search of actions to a fraction of possible bounding box sequences.",
"Recently dense trajectories were shown to be an efficient video representation for action recognition and achieved state-of-the-art results on a variety of datasets. This paper improves their performance by taking into account camera motion to correct them. To estimate camera motion, we match feature points between frames using SURF descriptors and dense optical flow, which are shown to be complementary. These matches are, then, used to robustly estimate a homography with RANSAC. Human motion is in general different from camera motion and generates inconsistent matches. To improve the estimation, a human detector is employed to remove these matches. Given the estimated camera motion, we remove trajectories consistent with it. We also use this estimation to cancel out camera motion from the optical flow. This significantly improves motion-based descriptors, such as HOF and MBH. Experimental results on four challenging action datasets (i.e., Hollywood2, HMDB51, Olympic Sports and UCF50) significantly outperform the current state of the art.",
"Visual features are of vital importance for human action understanding in videos. This paper presents a new video representation, called trajectory-pooled deep-convolutional descriptor (TDD), which shares the merits of both hand-crafted features [31] and deep-learned features [24]. Specifically, we utilize deep architectures to learn discriminative convolutional feature maps, and conduct trajectory-constrained pooling to aggregate these convolutional features into effective descriptors. To enhance the robustness of TDDs, we design two normalization methods to transform convolutional feature maps, namely spatiotemporal normalization and channel normalization. The advantages of our features come from (i) TDDs are automatically learned and contain high discriminative capacity compared with those hand-crafted features; (ii) TDDs take account of the intrinsic characteristics of temporal dimension and introduce the strategies of trajectory-constrained sampling and pooling for aggregating deep-learned features. We conduct experiments on two challenging datasets: HMD-B51 and UCF101. Experimental results show that TDDs outperform previous hand-crafted features [31] and deep-learned features [24]. Our method also achieves superior performance to the state of the art on these datasets.",
"",
"Actionness was introduced to quantify the likelihood of containing a generic action instance at a specific location. Accurate and efficient estimation of actionness is important in video analysis and may benefit other relevant tasks such as action recognition and action detection. This paper presents a new deep architecture for actionness estimation, called hybrid fully convolutional network (H-FCN), which is composed of appearance FCN (A-FCN) and motion FCN (M-FCN). These two FCNs leverage the strong capacity of deep models to estimate actionness maps from the perspectives of static appearance and dynamic motion, respectively. In addition, the fully convolutional nature of H-FCN allows it to efficiently process videos with arbitrary sizes. Experiments are conducted on the challenging datasets of Stanford40, UCF Sports, and JHMDB to verify the effectiveness of H-FCN on actionness estimation, which demonstrate that our method achieves superior performance to previous ones. Moreover, we apply the estimated actionness maps on action proposal generation and action detection. Our actionness maps advance the current state-of-the-art performance of these tasks substantially.",
"",
"In this paper we target at generating generic action proposals in unconstrained videos. Each action proposal corresponds to a temporal series of spatial bounding boxes, i.e., a spatio-temporal video tube, which has a good potential to locate one human action. Assuming each action is performed by a human with meaningful motion, both appearance and motion cues are utilized to measure the actionness of the video tubes. After picking those spatiotemporal paths of high actionness scores, our action proposal generation is formulated as a maximum set coverage problem, where greedy search is performed to select a set of action proposals that can maximize the overall actionness score. Compared with existing action proposal approaches, our action proposals do not rely on video segmentation and can be generated in nearly real-time. Experimental results on two challenging datasets, MSRII and UCF 101, validate the superior performance of our action proposals as well as competitive results on action detection and search.",
"Previous approaches to action recognition with deep features tend to process video frames only within a small temporal region, and do not model long-range dynamic information explicitly. However, such information is important for the accurate recognition of actions, especially for the discrimination of complex activities that share sub-actions, and when dealing with untrimmed videos. Here, we propose a representation, VLAD for Deep Dynamics (VLAD3), that accounts for different levels of video dynamics. It captures short-term dynamics with deep convolutional neural network features, relying on linear dynamic systems (LDS) to model medium-range dynamics. To account for long-range inhomogeneous dynamics, a VLAD descriptor is derived for the LDS and pooled over the whole video, to arrive at the final VLAD3 representation. An extensive evaluation was performed on Olympic Sports, UCF101 and THUMOS15, where the use of the VLAD3 representation leads to stateof-the-art results.",
"Models based on deep convolutional networks have dominated recent image interpretation tasks; we investigate whether models which are also recurrent, or \"temporally deep\", are effective for tasks involving sequences, visual and otherwise. We develop a novel recurrent convolutional architecture suitable for large-scale visual learning which is end-to-end trainable, and demonstrate the value of these models on benchmark video recognition tasks, image description and retrieval problems, and video narration challenges. In contrast to current models which assume a fixed spatio-temporal receptive field or simple temporal averaging for sequential processing, recurrent convolutional models are \"doubly deep\"' in that they can be compositional in spatial and temporal \"layers\". Such models may have advantages when target concepts are complex and or training data are limited. Learning long-term dependencies is possible when nonlinearities are incorporated into the network state updates. Long-term RNN models are appealing in that they directly can map variable-length inputs (e.g., video frames) to variable length outputs (e.g., natural language text) and can model complex temporal dynamics; yet they can be optimized with backpropagation. Our recurrent long-term models are directly connected to modern visual convnet models and can be jointly trained to simultaneously learn temporal dynamics and convolutional perceptual representations. Our results show such models have distinct advantages over state-of-the-art models for recognition or generation which are separately defined and or optimized.",
"",
"We propose a simple, yet effective approach for spatiotemporal feature learning using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset. Our findings are three-fold: 1) 3D ConvNets are more suitable for spatiotemporal feature learning compared to 2D ConvNets; 2) A homogeneous architecture with small 3x3x3 convolution kernels in all layers is among the best performing architectures for 3D ConvNets; and 3) Our learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks. In addition, the features are compact: achieving 52.8 accuracy on UCF101 dataset with only 10 dimensions and also very efficient to compute due to the fast inference of ConvNets. Finally, they are conceptually very simple and easy to train and use.",
"",
"The recognition of human actions and the determination of human attributes are two tasks that call for fine-grained classification. Indeed, often rather small and inconspicuous objects and features have to be detected to tell their classes apart. In order to deal with this challenge, we propose a novel convolutional neural network that mines mid-level image patches that are sufficiently dedicated to resolve the corresponding subtleties. In particular, we train a newly designed CNN (DeepPattern) that learns discriminative patch groups. There are two innovative aspects to this. On the one hand we pay attention to contextual information in an original fashion. On the other hand, we let an iteration of feature learning and patch clustering purify the set of dedicated patches that we use. We validate our method for action classification on two challenging datasets: PASCAL VOC 2012 Action and Stanford 40 Actions, and for attribute recognition we use the Berkeley Attributes of People dataset. Our discriminative mid-level mining CNN obtains state-of-theart results on these datasets, without a need for annotations about parts and poses."
]
} |
1703.06749 | 2949677752 | In this work we perform outlier detection using ensembles of neural networks obtained by variational approximation of the posterior in a Bayesian neural network setting. The variational parameters are obtained by sampling from the true posterior by gradient descent. We show our outlier detection results are comparable to those obtained using other efficient ensembling methods. | In @cite_15 , the authors train @math independent networks from scratch (with respective uncertainty estimates for regression) and use the ensemble to obtain good uncertainty estimates and classification performance. We view this model as an upper bound on how good our model can be, since the goal is to achieve similarly good results training only a single network. | {
"cite_N": [
"@cite_15"
],
"mid": [
"2560321925"
],
"abstract": [
"Deep neural networks (NNs) are powerful black box predictors that have recently achieved impressive performance on a wide spectrum of tasks. Quantifying predictive uncertainty in NNs is a challenging and yet unsolved problem. Bayesian NNs, which learn a distribution over weights, are currently the state-of-the-art for estimating predictive uncertainty; however these require significant modifications to the training procedure and are computationally expensive compared to standard (non-Bayesian) NNs. We propose an alternative to Bayesian NNs that is simple to implement, readily parallelizable, requires very little hyperparameter tuning, and yields high quality predictive uncertainty estimates. Through a series of experiments on classification and regression benchmarks, we demonstrate that our method produces well-calibrated uncertainty estimates which are as good or better than approximate Bayesian NNs. To assess robustness to dataset shift, we evaluate the predictive uncertainty on test examples from known and unknown distributions, and show that our method is able to express higher uncertainty on out-of-distribution examples. We demonstrate the scalability of our method by evaluating predictive uncertainty estimates on ImageNet."
]
} |
1703.06749 | 2949677752 | In this work we perform outlier detection using ensembles of neural networks obtained by variational approximation of the posterior in a Bayesian neural network setting. The variational parameters are obtained by sampling from the true posterior by gradient descent. We show our outlier detection results are comparable to those obtained using other efficient ensembling methods. | This work is most comparable to ( @cite_11 ) and other Dropout-related methods ( @cite_16 ). This is because only a single network is trained and the ensembles are only created for testing. | {
"cite_N": [
"@cite_16",
"@cite_11"
],
"mid": [
"2907176385",
"582134693"
],
"abstract": [
"As deep neural networks (DNNs) are applied to increasingly challenging problems, they will need to be able to represent their own uncertainty. Modeling uncertainty is one of the key features of Bayesian methods. Using Bernoulli dropout with sampling at prediction time has recently been proposed as an efficient and well performing variational inference method for DNNs. However, sampling from other multiplicative noise based variational distributions has not been investigated in depth. We evaluated Bayesian DNNs trained with Bernoulli or Gaussian multiplicative masking of either the units (dropout) or the weights (dropconnect). We tested the calibration of the probabilistic predictions of Bayesian convolutional neural networks (CNNs) on MNIST and CIFAR-10. Sampling at prediction time increased the calibration of the DNNs' probabalistic predictions. Sampling weights, whether Gaussian or Bernoulli, led to more robust representation of uncertainty compared to sampling of units. However, using either Gaussian or Bernoulli dropout led to increased test set classification accuracy. Based on these findings we used both Bernoulli dropout and Gaussian dropconnect concurrently, which we show approximates the use of a spike-and-slab variational distribution without increasing the number of learned parameters. We found that spike-and-slab sampling had higher test set performance than Gaussian dropconnect and more robustly represented its uncertainty compared to Bernoulli dropout.",
"Deep learning tools have gained tremendous attention in applied machine learning. However such tools for regression and classification do not capture model uncertainty. In comparison, Bayesian models offer a mathematically grounded framework to reason about model uncertainty, but usually come with a prohibitive computational cost. In this paper we develop a new theoretical framework casting dropout training in deep neural networks (NNs) as approximate Bayesian inference in deep Gaussian processes. A direct result of this theory gives us tools to model uncertainty with dropout NNs -- extracting information from existing models that has been thrown away so far. This mitigates the problem of representing uncertainty in deep learning without sacrificing either computational complexity or test accuracy. We perform an extensive study of the properties of dropout's uncertainty. Various network architectures and non-linearities are assessed on tasks of regression and classification, using MNIST as an example. We show a considerable improvement in predictive log-likelihood and RMSE compared to existing state-of-the-art methods, and finish by using dropout's uncertainty in deep reinforcement learning."
]
} |
1703.06995 | 2952478944 | Automated Facial Expression Recognition (FER) has been a challenging task for decades. Many of the existing works use hand-crafted features such as LBP, HOG, LPQ, and Histogram of Optical Flow (HOF) combined with classifiers such as Support Vector Machines for expression recognition. These methods often require rigorous hyperparameter tuning to achieve good results. Recently Deep Neural Networks (DNN) have shown to outperform traditional methods in visual object recognition. In this paper, we propose a two-part network consisting of a DNN-based architecture followed by a Conditional Random Field (CRF) module for facial expression recognition in videos. The first part captures the spatial relation within facial images using convolutional layers followed by three Inception-ResNet modules and two fully-connected layers. To capture the temporal relation between the image frames, we use linear chain CRF in the second part of our network. We evaluate our proposed network on three publicly available databases, viz. CK+, MMI, and FERA. Experiments are performed in subject-independent and cross-database manners. Our experimental results show that cascading the deep network architecture with the CRF module considerably increases the recognition of facial expressions in videos and in particular it outperforms the state-of-the-art methods in the cross-database experiments and yields comparable results in the subject-independent experiments. | Szegedy @cite_16 introduced GoogLeNet architecture which uses a novel multi-scale approach by using multiple classifier structures, combined with multiple sources for back propagation. Their architecture increases both width and depth of the network without causing significant penalties. The architecture is composed of multiple Inception" layers, which applies convolution on the input feature map in different scales, allowing the architecture to make more complex decisions. Different variations of the Inception layer have been proposed @cite_13 @cite_14 . These architectures have shown remarkable recognition results in object recognition tasks. Mollahosseini @cite_17 have used the traditional Inception layer for the task of facial expression recognition and achieved state-of-the-art results. | {
"cite_N": [
"@cite_14",
"@cite_16",
"@cite_13",
"@cite_17"
],
"mid": [
"2949117887",
"2950179405",
"2949605076",
"2244142460"
],
"abstract": [
"Training Deep Neural Networks is complicated by the fact that the distribution of each layer's inputs changes during training, as the parameters of the previous layers change. This slows down the training by requiring lower learning rates and careful parameter initialization, and makes it notoriously hard to train models with saturating nonlinearities. We refer to this phenomenon as internal covariate shift, and address the problem by normalizing layer inputs. Our method draws its strength from making normalization a part of the model architecture and performing the normalization for each training mini-batch. Batch Normalization allows us to use much higher learning rates and be less careful about initialization. It also acts as a regularizer, in some cases eliminating the need for Dropout. Applied to a state-of-the-art image classification model, Batch Normalization achieves the same accuracy with 14 times fewer training steps, and beats the original model by a significant margin. Using an ensemble of batch-normalized networks, we improve upon the best published result on ImageNet classification: reaching 4.9 top-5 validation error (and 4.8 test error), exceeding the accuracy of human raters.",
"We propose a deep convolutional neural network architecture codenamed \"Inception\", which was responsible for setting the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC 2014). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. This was achieved by a carefully crafted design that allows for increasing the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC 2014 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.",
"Convolutional networks are at the core of most state-of-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we explore ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21.2 top-1 and 5.6 top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3.5 top-5 error on the validation set (3.6 error on the test set) and 17.3 top-1 error on the validation set.",
"Automated Facial Expression Recognition (FER) has remained a challenging and interesting problem in computer vision. Despite efforts made in developing various methods for FER, existing approaches lack generalizability when applied to unseen images or those that are captured in wild setting (i.e. the results are not significant). Most of the existing approaches are based on engineered features (e.g. HOG, LBPH, and Gabor) where the classifier's hyper-parameters are tuned to give best recognition accuracies across a single database, or a small collection of similar databases. This paper proposes a deep neural network architecture to address the FER problem across multiple well-known standard face datasets. Specifically, our network consists of two convolutional layers each followed by max pooling and then four Inception layers. The network is a single component architecture that takes registered facial images as the input and classifies them into either of the six basic or the neutral expressions. We conducted comprehensive experiments on seven publicly available facial expression databases, viz. MultiPIE, MMI, CK+, DISFA, FERA, SFEW, and FER2013. The results of our proposed architecture are comparable to or better than the state-of-the-art methods and better than traditional convolutional neural networks in both accuracy and training time."
]
} |
1703.06995 | 2952478944 | Automated Facial Expression Recognition (FER) has been a challenging task for decades. Many of the existing works use hand-crafted features such as LBP, HOG, LPQ, and Histogram of Optical Flow (HOF) combined with classifiers such as Support Vector Machines for expression recognition. These methods often require rigorous hyperparameter tuning to achieve good results. Recently Deep Neural Networks (DNN) have shown to outperform traditional methods in visual object recognition. In this paper, we propose a two-part network consisting of a DNN-based architecture followed by a Conditional Random Field (CRF) module for facial expression recognition in videos. The first part captures the spatial relation within facial images using convolutional layers followed by three Inception-ResNet modules and two fully-connected layers. To capture the temporal relation between the image frames, we use linear chain CRF in the second part of our network. We evaluate our proposed network on three publicly available databases, viz. CK+, MMI, and FERA. Experiments are performed in subject-independent and cross-database manners. Our experimental results show that cascading the deep network architecture with the CRF module considerably increases the recognition of facial expressions in videos and in particular it outperforms the state-of-the-art methods in the cross-database experiments and yields comparable results in the subject-independent experiments. | Residual connections were introduced by He @cite_15 . ResNets consist of many stacked Residual Units" and each of these units can be formulated as follows: | {
"cite_N": [
"@cite_15"
],
"mid": [
"2949650786"
],
"abstract": [
"Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation."
]
} |
1703.06995 | 2952478944 | Automated Facial Expression Recognition (FER) has been a challenging task for decades. Many of the existing works use hand-crafted features such as LBP, HOG, LPQ, and Histogram of Optical Flow (HOF) combined with classifiers such as Support Vector Machines for expression recognition. These methods often require rigorous hyperparameter tuning to achieve good results. Recently Deep Neural Networks (DNN) have shown to outperform traditional methods in visual object recognition. In this paper, we propose a two-part network consisting of a DNN-based architecture followed by a Conditional Random Field (CRF) module for facial expression recognition in videos. The first part captures the spatial relation within facial images using convolutional layers followed by three Inception-ResNet modules and two fully-connected layers. To capture the temporal relation between the image frames, we use linear chain CRF in the second part of our network. We evaluate our proposed network on three publicly available databases, viz. CK+, MMI, and FERA. Experiments are performed in subject-independent and cross-database manners. Our experimental results show that cascading the deep network architecture with the CRF module considerably increases the recognition of facial expressions in videos and in particular it outperforms the state-of-the-art methods in the cross-database experiments and yields comparable results in the subject-independent experiments. | Where @math and @math are input and output of the @math -th unit and @math is the residual function. The main idea in ResNets is to learn the additive residual function @math with respect to @math with a choice of using identity mapping @math @cite_23 . Moreover, Inception layer is combined with residual unit and it shows that the resulting architecture accelerates the training of Inception networks significantly @cite_4 . | {
"cite_N": [
"@cite_4",
"@cite_23"
],
"mid": [
"2274287116",
"2949427019"
],
"abstract": [
"Very deep convolutional networks have been central to the largest advances in image recognition performance in recent years. One example is the Inception architecture that has been shown to achieve very good performance at relatively low computational cost. Recently, the introduction of residual connections in conjunction with a more traditional architecture has yielded state-of-the-art performance in the 2015 ILSVRC challenge; its performance was similar to the latest generation Inception-v3 network. This raises the question of whether there are any benefit in combining the Inception architecture with residual connections. Here we give clear empirical evidence that training with residual connections accelerates the training of Inception networks significantly. There is also some evidence of residual Inception networks outperforming similarly expensive Inception networks without residual connections by a thin margin. We also present several new streamlined architectures for both residual and non-residual Inception networks. These variations improve the single-frame recognition performance on the ILSVRC 2012 classification task significantly. We further demonstrate how proper activation scaling stabilizes the training of very wide residual Inception networks. With an ensemble of three residual and one Inception-v4, we achieve 3.08 percent top-5 error on the test set of the ImageNet classification (CLS) challenge",
"Deep residual networks have emerged as a family of extremely deep architectures showing compelling accuracy and nice convergence behaviors. In this paper, we analyze the propagation formulations behind the residual building blocks, which suggest that the forward and backward signals can be directly propagated from one block to any other block, when using identity mappings as the skip connections and after-addition activation. A series of ablation experiments support the importance of these identity mappings. This motivates us to propose a new residual unit, which makes training easier and improves generalization. We report improved results using a 1001-layer ResNet on CIFAR-10 (4.62 error) and CIFAR-100, and a 200-layer ResNet on ImageNet. Code is available at: this https URL"
]
} |
1703.06995 | 2952478944 | Automated Facial Expression Recognition (FER) has been a challenging task for decades. Many of the existing works use hand-crafted features such as LBP, HOG, LPQ, and Histogram of Optical Flow (HOF) combined with classifiers such as Support Vector Machines for expression recognition. These methods often require rigorous hyperparameter tuning to achieve good results. Recently Deep Neural Networks (DNN) have shown to outperform traditional methods in visual object recognition. In this paper, we propose a two-part network consisting of a DNN-based architecture followed by a Conditional Random Field (CRF) module for facial expression recognition in videos. The first part captures the spatial relation within facial images using convolutional layers followed by three Inception-ResNet modules and two fully-connected layers. To capture the temporal relation between the image frames, we use linear chain CRF in the second part of our network. We evaluate our proposed network on three publicly available databases, viz. CK+, MMI, and FERA. Experiments are performed in subject-independent and cross-database manners. Our experimental results show that cascading the deep network architecture with the CRF module considerably increases the recognition of facial expressions in videos and in particular it outperforms the state-of-the-art methods in the cross-database experiments and yields comparable results in the subject-independent experiments. | In @cite_25 it is shown that CRFs are more effective in recognizing some of human activities like walking, running, etc. Their results show that CRFs outperform HMMs and even provide good results for distinguishing between subtle motion patterns like normal walk vs. wander walk. There are several extensions of CRFs like Latent-Dynamic Conditional Random Fields (LD-CRFs) and Hidden Conditional Random Fields (HCRFs) @cite_8 which incorporate hidden states in the CRF model. In @cite_27 these models have been used for facial expression recognition task. | {
"cite_N": [
"@cite_27",
"@cite_25",
"@cite_8"
],
"mid": [
"2141990597",
"2151214862",
""
],
"abstract": [
"Conditional Random Fields (CRFs) can be used as a discriminative approach for simultaneous sequence segmentation and frame labeling. Latent-Dynamic Conditional Random Fields (LDCRFs) incorporates hidden state variables within CRFs which model sub-structure motion patterns and dynamics between labels. Motivated by the success of LDCRFs in gesture recognition, we propose a framework for automatic facial expression recognition from continuous video sequence by modeling temporal variations within shapes using LDCRFs. We show that the proposed approach outperforms CRFs for recognizing facial expressions. Using Principal Component Analysis (PCA) we study the separability of various expression classes in lower dimension projected spaces. By comparing the performance of CRFs and LDCRFs against that of Support Vector Machines (SVMs), we demonstrate that temporal variations within shapes are crucial in classifying expressions especially for those with a small range of facial motion like anger and sadness. We also show empirically that only using changes in facial appearance over time, without using shape variations, is not sufficient to obtain high performance for facial expression recognition.",
"Abstract We describe algorithms for recognizing human motion in monocular video sequences, based on discriminative conditional random fields (CRFs) and maximum entropy Markov models (MEMMs). Existing approaches to this problem typically use generative structures like the hidden Markov model (HMM). Therefore, they have to make simplifying, often unrealistic assumptions on the conditional independence of observations given the motion class labels and cannot accommodate rich overlapping features of the observation or long-term contextual dependencies among observations at multiple timesteps. This makes them prone to myopic failures in recognizing many human motions, because even the transition between simple human activities naturally has temporal segments of ambiguity and overlap. The correct interpretation of these sequences requires more holistic, contextual decisions, where the estimate of an activity at a particular timestep could be constrained by longer windows of observations, prior and even posterior to that timestep. This would not be computationally feasible with a HMM which requires the enumeration of a number of observation sequences exponential in the size of the context window. In this work we follow a different philosophy: instead of restrictively modeling the complex image generation process – the observation, we work with models that can unrestrictedly take it as an input, hence condition on it. Conditional models like the proposed CRFs seamlessly represent contextual dependencies and have computationally attractive properties: they support efficient, exact recognition using dynamic programming, and their parameters can be learned using convex optimization. We introduce conditional graphical models as complementary tools for human motion recognition and present an extensive set of experiments that show not only how these can successfully classify diverse human activities like walking, jumping, running, picking or dancing, but also how they can discriminate among subtle motion styles like normal walks and wander walks.",
""
]
} |
1703.06995 | 2952478944 | Automated Facial Expression Recognition (FER) has been a challenging task for decades. Many of the existing works use hand-crafted features such as LBP, HOG, LPQ, and Histogram of Optical Flow (HOF) combined with classifiers such as Support Vector Machines for expression recognition. These methods often require rigorous hyperparameter tuning to achieve good results. Recently Deep Neural Networks (DNN) have shown to outperform traditional methods in visual object recognition. In this paper, we propose a two-part network consisting of a DNN-based architecture followed by a Conditional Random Field (CRF) module for facial expression recognition in videos. The first part captures the spatial relation within facial images using convolutional layers followed by three Inception-ResNet modules and two fully-connected layers. To capture the temporal relation between the image frames, we use linear chain CRF in the second part of our network. We evaluate our proposed network on three publicly available databases, viz. CK+, MMI, and FERA. Experiments are performed in subject-independent and cross-database manners. Our experimental results show that cascading the deep network architecture with the CRF module considerably increases the recognition of facial expressions in videos and in particular it outperforms the state-of-the-art methods in the cross-database experiments and yields comparable results in the subject-independent experiments. | Discriminative models are reliable tools for the task of sequence labeling @cite_2 @cite_25 @cite_27 . These models have shown better results in various modeling problems. Conditional Random Fields are commonly used in Natural Language Processing tasks like part-of-speech tagging and word recognition @cite_31 . Researchers have used these models in the field of computer vision as well @cite_27 @cite_25 . | {
"cite_N": [
"@cite_27",
"@cite_31",
"@cite_25",
"@cite_2"
],
"mid": [
"2141990597",
"2147880316",
"2151214862",
"2117497855"
],
"abstract": [
"Conditional Random Fields (CRFs) can be used as a discriminative approach for simultaneous sequence segmentation and frame labeling. Latent-Dynamic Conditional Random Fields (LDCRFs) incorporates hidden state variables within CRFs which model sub-structure motion patterns and dynamics between labels. Motivated by the success of LDCRFs in gesture recognition, we propose a framework for automatic facial expression recognition from continuous video sequence by modeling temporal variations within shapes using LDCRFs. We show that the proposed approach outperforms CRFs for recognizing facial expressions. Using Principal Component Analysis (PCA) we study the separability of various expression classes in lower dimension projected spaces. By comparing the performance of CRFs and LDCRFs against that of Support Vector Machines (SVMs), we demonstrate that temporal variations within shapes are crucial in classifying expressions especially for those with a small range of facial motion like anger and sadness. We also show empirically that only using changes in facial appearance over time, without using shape variations, is not sufficient to obtain high performance for facial expression recognition.",
"We present conditional random fields , a framework for building probabilistic models to segment and label sequence data. Conditional random fields offer several advantages over hidden Markov models and stochastic grammars for such tasks, including the ability to relax strong independence assumptions made in those models. Conditional random fields also avoid a fundamental limitation of maximum entropy Markov models (MEMMs) and other discriminative Markov models based on directed graphical models, which can be biased towards states with few successor states. We present iterative parameter estimation algorithms for conditional random fields and compare the performance of the resulting models to HMMs and MEMMs on synthetic and natural-language data.",
"Abstract We describe algorithms for recognizing human motion in monocular video sequences, based on discriminative conditional random fields (CRFs) and maximum entropy Markov models (MEMMs). Existing approaches to this problem typically use generative structures like the hidden Markov model (HMM). Therefore, they have to make simplifying, often unrealistic assumptions on the conditional independence of observations given the motion class labels and cannot accommodate rich overlapping features of the observation or long-term contextual dependencies among observations at multiple timesteps. This makes them prone to myopic failures in recognizing many human motions, because even the transition between simple human activities naturally has temporal segments of ambiguity and overlap. The correct interpretation of these sequences requires more holistic, contextual decisions, where the estimate of an activity at a particular timestep could be constrained by longer windows of observations, prior and even posterior to that timestep. This would not be computationally feasible with a HMM which requires the enumeration of a number of observation sequences exponential in the size of the context window. In this work we follow a different philosophy: instead of restrictively modeling the complex image generation process – the observation, we work with models that can unrestrictedly take it as an input, hence condition on it. Conditional models like the proposed CRFs seamlessly represent contextual dependencies and have computationally attractive properties: they support efficient, exact recognition using dynamic programming, and their parameters can be learned using convex optimization. We introduce conditional graphical models as complementary tools for human motion recognition and present an extensive set of experiments that show not only how these can successfully classify diverse human activities like walking, jumping, running, picking or dancing, but also how they can discriminate among subtle motion styles like normal walks and wander walks.",
"Many problems in vision involve the prediction of a class label for each frame in an unsegmented sequence. In this paper, we develop a discriminative framework for simultaneous sequence segmentation and labeling which can capture both intrinsic and extrinsic class dynamics. Our approach incorporates hidden state variables which model the sub-structure of a class sequence and learn dynamics between class labels. Each class label has a disjoint set of associated hidden states, which enables efficient training and inference in our model. We evaluated our method on the task of recognizing human gestures from unsegmented video streams and performed experiments on three different datasets of head and eye gestures. Our results demonstrate that our model compares favorably to Support Vector Machines, Hidden Markov Models, and Conditional Random Fields on visual gesture recognition tasks."
]
} |
1703.06870 | 2599765304 | We present a conceptually simple, flexible, and general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. The method, called Mask R-CNN, extends Faster R-CNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition. Mask R-CNN is simple to train and adds only a small overhead to Faster R-CNN, running at 5 fps. Moreover, Mask R-CNN is easy to generalize to other tasks, e.g., allowing us to estimate human poses in the same framework. We show top results in all three tracks of the COCO suite of challenges, including instance segmentation, bounding-box object detection, and person keypoint detection. Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners. We hope our simple and effective approach will serve as a solid baseline and help ease future research in instance-level recognition. Code has been made available at: this https URL | Another family of solutions @cite_13 @cite_17 @cite_26 @cite_23 to instance segmentation are driven by the success of semantic segmentation. Starting from per-pixel classification results ( , FCN outputs), these methods attempt to cut the pixels of the same category into different instances. In contrast to the strategy of these methods, Mask R-CNN is based on an strategy. We expect a deeper incorporation of both strategies will be studied in the future. | {
"cite_N": [
"@cite_26",
"@cite_13",
"@cite_23",
"@cite_17"
],
"mid": [
"2949242507",
"2949556967",
"2777795072",
"2557889580"
],
"abstract": [
"Semantic segmentation and object detection research have recently achieved rapid progress. However, the former task has no notion of different instances of the same object, and the latter operates at a coarse, bounding-box level. We propose an Instance Segmentation system that produces a segmentation map where each pixel is assigned an object class and instance identity label. Most approaches adapt object detectors to produce segments instead of boxes. In contrast, our method is based on an initial semantic segmentation module, which feeds into an instance subnetwork. This subnetwork uses the initial category-level segmentation, along with cues from the output of an object detector, within an end-to-end CRF to predict instances. This part of our model is dynamically instantiated to produce a variable number of instances per image. Our end-to-end approach requires no post-processing and considers the image holistically, instead of processing independent proposals. Therefore, unlike some related work, a pixel cannot belong to multiple instances. Furthermore, far more precise segmentations are achieved, as shown by our state-of-the-art results (particularly at high IoU thresholds) on the Pascal VOC and Cityscapes datasets.",
"This work addresses the task of instance-aware semantic segmentation. Our key motivation is to design a simple method with a new modelling-paradigm, which therefore has a different trade-off between advantages and disadvantages compared to known approaches. Our approach, we term InstanceCut, represents the problem by two output modalities: (i) an instance-agnostic semantic segmentation and (ii) all instance-boundaries. The former is computed from a standard convolutional neural network for semantic segmentation, and the latter is derived from a new instance-aware edge detection model. To reason globally about the optimal partitioning of an image into instances, we combine these two modalities into a novel MultiCut formulation. We evaluate our approach on the challenging CityScapes dataset. Despite the conceptual simplicity of our approach, we achieve the best result among all published methods, and perform particularly well for rare object classes.",
"In this paper, we propose Sequential Grouping Networks (SGN) to tackle the problem of object instance segmentation. SGNs employ a sequence of neural networks, each solving a sub-grouping problem of increasing semantic complexity in order to gradually compose objects out of pixels. In particular, the first network aims to group pixels along each image row and column by predicting horizontal and vertical object breakpoints. These breakpoints are then used to create line segments. By exploiting two-directional information, the second network groups horizontal and vertical lines into connected components. Finally, the third network groups the connected components into object instances. Our experiments show that our SGN significantly outperforms state-of-the-art approaches in both, the Cityscapes dataset as well as PASCAL VOC.",
"Most contemporary approaches to instance segmentation use complex pipelines involving conditional random fields, recurrent neural networks, object proposals, or template matching schemes. In this paper, we present a simple yet powerful end-to-end convolutional neural network to tackle this task. Our approach combines intuitions from the classical watershed transform and modern deep learning to produce an energy map of the image where object instances are unambiguously represented as energy basins. We then perform a cut at a single energy level to directly yield connected components corresponding to object instances. Our model achieves more than double the performance over the state-of-the-art on the challenging Cityscapes Instance Level Segmentation task."
]
} |
1703.07022 | 2951684117 | A natural image usually conveys rich semantic content and can be viewed from different angles. Existing image description methods are largely restricted by small sets of biased visual paragraph annotations, and fail to cover rich underlying semantics. In this paper, we investigate a semi-supervised paragraph generative framework that is able to synthesize diverse and semantically coherent paragraph descriptions by reasoning over local semantic regions and exploiting linguistic knowledge. The proposed Recurrent Topic-Transition Generative Adversarial Network (RTT-GAN) builds an adversarial framework between a structured paragraph generator and multi-level paragraph discriminators. The paragraph generator generates sentences recurrently by incorporating region-based visual and language attention mechanisms at each step. The quality of generated paragraph sentences is assessed by multi-level adversarial discriminators from two aspects, namely, plausibility at sentence level and topic-transition coherence at paragraph level. The joint adversarial training of RTT-GAN drives the model to generate realistic paragraphs with smooth logical transition between sentence topics. Extensive quantitative experiments on image and video paragraph datasets demonstrate the effectiveness of our RTT-GAN in both supervised and semi-supervised settings. Qualitative results on telling diverse stories for an image also verify the interpretability of RTT-GAN. | Paragraph generation overcomes shortcomings of standard captioning and dense captioning by producing a coherent and fine-grained natural language description. To reason about long-term linguistic structures with multiple sentences, hierarchical recurrent network @cite_32 @cite_15 @cite_20 @cite_35 has been widely used to directly simulate the hierarchy of language. For example, @cite_20 generate multi-sentence video descriptions for cooking videos to capture strong temporal dependencies. @cite_35 combine semantics of all regions of interest to produce a generic paragraph for an image. However, all these methods suffer from the overfitting problem due to the lack of sufficient paragraph descriptions. In contrast, we propose a generative model to automatically synthesize a large amount of diverse and reasonable paragraph descriptions by learning the implicit linguistic interplay between sentences. Our RTT-GAN has better interpretability by imposing the sentence plausibility and topic-transition coherence on the generator with two adversarial discriminators. The generator selectively incorporates visual and language cues of semantic regions to produce each sentence. | {
"cite_N": [
"@cite_35",
"@cite_15",
"@cite_32",
"@cite_20"
],
"mid": [
"2549599535",
"2251849926",
"2115613106",
"1957740064"
],
"abstract": [
"Recent progress on image captioning has made it possible to generate novel sentences describing images in natural language, but compressing an image into a single sentence can describe visual content in only coarse detail. While one new captioning approach, dense captioning, can potentially describe images in finer levels of detail by captioning many regions within an image, it in turn is unable to produce a coherent story for an image. In this paper we overcome these limitations by generating entire paragraphs for describing images, which can tell detailed, unified stories. We develop a model that decomposes both images and paragraphs into their constituent parts, detecting semantic regions in images and using a hierarchical recurrent neural network to reason about language. Linguistic analysis confirms the complexity of the paragraph generation task, and thorough experiments on a new dataset of image and paragraph pairs demonstrate the effectiveness of our approach.",
"This paper proposes a novel hierarchical recurrent neural network language model (HRNNLM) for document modeling. After establishing a RNN to capture the coherence between sentences in a document, HRNNLM integrates it as the sentence history information into the word level RNN to predict the word sequence with cross-sentence contextual information. A two-step training approach is designed, in which sentence-level and word-level language models are approximated for the convergence in a pipeline style. Examined by the standard sentence reordering scenario, HRNNLM is proved for its better accuracy in modeling the sentence coherence. And at the word level, experimental results also indicate a significant lower model perplexity, followed by a practical better translation result when applied to a Chinese-English document translation reranking task.",
"Natural language generation of coherent long texts like paragraphs or longer documents is a challenging problem for recurrent networks models. In this paper, we explore an important step toward this generation task: training an LSTM (Longshort term memory) auto-encoder to preserve and reconstruct multi-sentence paragraphs. We introduce an LSTM model that hierarchically builds an embedding for a paragraph from embeddings for sentences and words, then decodes this embedding to reconstruct the original paragraph. We evaluate the reconstructed paragraph using standard metrics like ROUGE and Entity Grid, showing that neural models are able to encode texts in a way that preserve syntactic, semantic, and discourse coherence. While only a first step toward generating coherent text units from neural models, our work has the potential to significantly impact natural language generation and summarization1.",
"We present an approach that exploits hierarchical Recurrent Neural Networks (RNNs) to tackle the video captioning problem, i.e., generating one or multiple sentences to describe a realistic video. Our hierarchical framework contains a sentence generator and a paragraph generator. The sentence generator produces one simple short sentence that describes a specific short video interval. It exploits both temporal- and spatial-attention mechanisms to selectively focus on visual elements during generation. The paragraph generator captures the inter-sentence dependency by taking as input the sentential embedding produced by the sentence generator, combining it with the paragraph history, and outputting the new initial state for the sentence generator. We evaluate our approach on two large-scale benchmark datasets: YouTubeClips and TACoS-MultiLevel. The experiments demonstrate that our approach significantly outperforms the current state-of-the-art methods with BLEU@4 scores 0.499 and 0.305 respectively."
]
} |
1703.06361 | 2953189960 | The friendship paradox states that in a social network, egos tend to have lower degree than their alters, or, "your friends have more friends than you do". Most research has focused on the friendship paradox and its implications for information transmission, but treating the network as static and unweighted. Yet, people can dedicate only a finite fraction of their attention budget to each social interaction: a high-degree individual may have less time to dedicate to individual social links, forcing them to modulate the quantities of contact made to their different social ties. Here we study the friendship paradox in the context of differing contact volumes between egos and alters, finding a connection between contact volume and the strength of the friendship paradox. The most frequently contacted alters exhibit a less pronounced friendship paradox compared with the ego, whereas less-frequently contacted alters are more likely to be high degree and give rise to the paradox. We argue therefore for a more nuanced version of the friendship paradox: "your closest friends have slightly more friends than you do", and in certain networks even: "your best friend has no more friends than you do". We demonstrate that this relationship is robust, holding in both a social media and a mobile phone dataset. These results have implications for information transfer and influence in social networks, which we explore using a simple dynamical model. | The key observation of the friendship paradox is that, because highly popular individuals are over-represented across the sets of alters of egos, the average degree of alters is right-skewed, leading to the colloquial your friends have more friends than you'' @cite_14 . Perceptions can be biased because of the paradox @cite_25 and and personality traits which correlate with network degree, such as extroversion @cite_23 will be influenced by the paradox. | {
"cite_N": [
"@cite_14",
"@cite_25",
"@cite_23"
],
"mid": [
"2084862036",
"1971930416",
"2111761833"
],
"abstract": [
"It is reasonable to suppose that individuals use the number of friends that their friends have as one basis for determining whether they, themselves, have an adequate number of friends. This article shows that, if individuals compare themselves with their friends, it is likely that most of them will feel relatively inadequate. Data on friendship drawn from James Coleman's (1961) classic study The Adolescent Society are used to illustrate the phenomenon that most people have fewer friends than their friends have. The logic underlying the phenomenon is mathematically explored, showing that the mean number of friends of friends is always greater than the mean number of friends of individuals. Further analysis shows that the proportion of individuals who have fewer friends than the mean number of friends their own friends have is affected by the exact arrangement of friendships in a social network. This disproportionate experiencing of friends with many friends is related to a set of",
"We report on a survey of undergraduates at the University of Chicago in which respondents were asked to assess their popularity relative to others. Popularity estimates were related to actual popularity, but we also found strong evidence of self-enhancement in self-other comparisons of popularity. In particular, self-enhancement was stronger for self versus friend comparisons than for self versus typical other comparisons; this is contrary to the reality demonstrated in Feld's friendship paradox and suggests that people are more threatened by the success of friends than of strangers. At the same time, people with relatively popular friends tended to make more self-serving estimates of their own popularity than did people with less popular friends. These results clarify how objective patterns of interpersonal contact work together with cognitive and motivational tendencies to shape perceptions of one's location in the social world.",
"Previous studies showed that extraversion influences social network size. However, it is unclear how extraversion affects the size of different layers of the network, and how extraversion relates to the emotional intensity of social relationships. We examined the relationships between extraversion, network size, and emotional closeness for 117 individuals. The results demonstrated that extraverts had larger networks at every layer (support clique, sympathy group, outer layer). The results were robust and were not attributable to potential confounds such as sex, though they were modest in size (raw correlations between extraversion and size of network layer, .20 < r < .23). However, extraverts were not emotionally closer to individuals in their network, even after controlling for network size. These results highlight the importance of considering not just social network size in relation to personality, but also the quality of relationships with network members."
]
} |
1703.06361 | 2953189960 | The friendship paradox states that in a social network, egos tend to have lower degree than their alters, or, "your friends have more friends than you do". Most research has focused on the friendship paradox and its implications for information transmission, but treating the network as static and unweighted. Yet, people can dedicate only a finite fraction of their attention budget to each social interaction: a high-degree individual may have less time to dedicate to individual social links, forcing them to modulate the quantities of contact made to their different social ties. Here we study the friendship paradox in the context of differing contact volumes between egos and alters, finding a connection between contact volume and the strength of the friendship paradox. The most frequently contacted alters exhibit a less pronounced friendship paradox compared with the ego, whereas less-frequently contacted alters are more likely to be high degree and give rise to the paradox. We argue therefore for a more nuanced version of the friendship paradox: "your closest friends have slightly more friends than you do", and in certain networks even: "your best friend has no more friends than you do". We demonstrate that this relationship is robust, holding in both a social media and a mobile phone dataset. These results have implications for information transfer and influence in social networks, which we explore using a simple dynamical model. | The work which is perhaps most similar to our own is the work of @cite_21 . They showed that, not only does the paradox hold in Twitter, it is very strong, affecting $ 99 work was limited to Twitter only and considered numbers of tweets and retweets, but unlike the work here did not investigate contact volume (at-mentions) between users. | {
"cite_N": [
"@cite_21"
],
"mid": [
"1959141096"
],
"abstract": [
"Feld's friendship paradox states that \"your friends have more friends than you, on average.\" This paradox arises because extremely popular people, despite being rare, are overrepresented when averaging over friends. Using a sample of the Twitter firehose, we confirm that the friendship paradox holds for >98 of Twitter users. Because of the directed nature of the follower graph on Twitter, we are further able to confirm more detailed forms of the friendship paradox: everyone you follow or who follows you has more friends and followers than you. This is likely caused by a correlation we demonstrate between Twitter activity, number of friends, and number of followers. In addition, we discover two new paradoxes: the virality paradox that states \"your friends receive more viral content than you, on average,\" and the activity paradox, which states \"your friends are more active than you, on average.\" The latter paradox is important in regulating online communication. It may result in users having difficulty maintaining optimal incoming information rates, because following additional users causes the volume of incoming tweets to increase super-linearly. While users may compensate for increased information flow by increasing their own activity, users become information overloaded when they receive more information than they are able or willing to process. We compare the average size of cascades that are sent and received by overloaded and underloaded users. And we show that overloaded users post and receive larger cascades and they are poor detector of small cascades."
]
} |
1703.06527 | 2953295199 | The paper focuses on the problem of vision-based obstacle detection and tracking for unmanned aerial vehicle navigation. A real-time object localization and tracking strategy from monocular image sequences is developed by effectively integrating the object detection and tracking into a dynamic Kalman model. At the detection stage, the object of interest is automatically detected and localized from a saliency map computed via the image background connectivity cue at each frame; at the tracking stage, a Kalman filter is employed to provide a coarse prediction of the object state, which is further refined via a local detector incorporating the saliency map and the temporal information between two consecutive frames. Compared to existing methods, the proposed approach does not require any manual initialization for tracking, runs much faster than the state-of-the-art trackers of its kind, and achieves competitive tracking performance on a large number of image sequences. Extensive experiments demonstrate the effectiveness and superior performance of the proposed approach. | The goal of visual tracking is to estimate the boundary and trajectory of the object in every frame of an image sequence. Designing an efficient and robust tracker is a critical issue in visual tracking, especially in challenging situations, such as illumination variation, in-plane rotation, out-of-plane rotation, scale variation, occlusion, background clutter and so on @cite_31 . Over the past decades, various tracking algorithms have been proposed to cope with the challenges in visual tracking. According to the models adopted, these approaches can be generally classified into generative models @cite_24 @cite_14 @cite_23 @cite_28 , and discriminative models @cite_41 @cite_6 @cite_38 . Ross et al @cite_24 exploited an incremental subspace learning to visual tracking, which assumes that the obtained temporal targets reside in a low-dimensional subspace. Sui et al @cite_14 proposed a sparsity-induced subspace learning which selects effective features to construct the target subspace. Yin et al @cite_23 proposed a hierarchical tracking method based on the subspace representation and Kalman filter. Yu et al @cite_28 introduced a large-scale fiber tracking approach based on Kalman filter and group-wise thin-plate spline point matching. | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_28",
"@cite_41",
"@cite_6",
"@cite_24",
"@cite_23",
"@cite_31"
],
"mid": [
"2124211486",
"1534421354",
"2460993045",
"",
"2157381954",
"2139047213",
"2084758167",
"2089961441"
],
"abstract": [
"This paper investigates long-term tracking of unknown objects in a video stream. The object is defined by its location and extent in a single frame. In every frame that follows, the task is to determine the object's location and extent or indicate that the object is not present. We propose a novel tracking framework (TLD) that explicitly decomposes the long-term tracking task into tracking, learning, and detection. The tracker follows the object from frame to frame. The detector localizes all appearances that have been observed so far and corrects the tracker if necessary. The learning estimates the detector's errors and updates it to avoid these errors in the future. We study how to identify the detector's errors and learn from them. We develop a novel learning method (P-N learning) which estimates the errors by a pair of “experts”: (1) P-expert estimates missed detections, and (2) N-expert estimates false alarms. The learning process is modeled as a discrete dynamical system and the conditions under which the learning guarantees improvement are found. We describe our real-time implementation of the TLD framework and the P-N learning. We carry out an extensive quantitative evaluation which shows a significant improvement over state-of-the-art approaches.",
"Target representation is a necessary component for a robust tracker. However, during tracking, many complicated factors may make the accumulated errors in the representation significantly large, leading to tracking drift. This paper aims to improve the robustness of target representation to avoid the influence of the accumulated errors, such that the tracker only acquires the information that facilitates tracking and ignores the distractions. We observe that the locally mutual relations between the feature observations of temporally obtained targets are beneficial to the subspace representation in visual tracking. Thus, we propose a novel subspace learning algorithm for visual tracking, which imposes joint row-wise sparsity structure on the target subspace to adaptively exclude distractive information. The sparsity is induced by exploiting the locally mutual relations between the feature observations during learning. To this end, we formulate tracking as a subspace sparsity inducing problem. A large number of experiments on various challenging video sequences demonstrate that our tracker outperforms many other state-of-the-art trackers.",
"Automatic tracking of large-scale crowded targets are of particular importance in many applications, such as crowded people vehicle tracking in video surveillance, fiber tracking in materials science, and cell tracking in biomedical imaging. This problem becomes very challenging when the targets show similar appearance and the interslice inter-frame continuity is low due to sparse sampling, camera motion and target occlusion. The main challenge comes from the step of association which aims at matching the predictions and the observations of the multiple targets. In this paper we propose a new groupwise method to explore the target group information and employ the within-group correlations for association and tracking. In particular, the within-group association is modeled by a nonrigid 2D Thin-Plate transform and a sequence of group shrinking, group growing and group merging operations are then developed to refine the composition of each group. We apply the proposed method to track large-scale fibers from microscopy material images and compare its performance against several other multi-target tracking methods. We also apply the proposed method to track crowded people from videos with poor inter-frame continuity.",
"",
"We propose a biologically inspired framework for visual tracking based on discriminant center surround saliency. At each frame, discrimination of the target from the background is posed as a binary classification problem. From a pool of feature descriptors for the target and background, a subset that is most informative for classification between the two is selected using the principle of maximum marginal diversity. Using these features, the location of the target in the next frame is identified using top-down saliency, completing one iteration of the tracking algorithm. We also show that a simple extension of the framework to include motion features in a bottom-up saliency mode can robustly identify salient moving objects and automatically initialize the tracker. The connections of the proposed method to existing works on discriminant tracking are discussed. Experimental results comparing the proposed method to the state of the art in tracking are presented, showing improved performance.",
"Visual tracking, in essence, deals with non-stationary image streams that change over time. While most existing algorithms are able to track objects well in controlled environments, they usually fail in the presence of significant variation of the object's appearance or surrounding illumination. One reason for such failures is that many algorithms employ fixed appearance models of the target. Such models are trained using only appearance data available before tracking begins, which in practice limits the range of appearances that are modeled, and ignores the large volume of information (such as shape changes or specific lighting conditions) that becomes available during tracking. In this paper, we present a tracking method that incrementally learns a low-dimensional subspace representation, efficiently adapting online to changes in the appearance of the target. The model update, based on incremental algorithms for principal component analysis, includes two important features: a method for correctly updating the sample mean, and a forgetting factor to ensure less modeling power is expended fitting older observations. Both of these features contribute measurably to improving overall tracking performance. Numerous experiments demonstrate the effectiveness of the proposed tracking algorithm in indoor and outdoor environments where the target objects undergo large changes in pose, scale, and illumination.",
"We present a new tracking method with improved efficiency and accuracy based on the subspace representation and particle filter. The subspace representation has been successfully adopted in tracking, e.g., the Eigen-tracking algorithm, and it has shown considerable robustness for tracking an object with changing appearance. Particle filters are widely used for a wide range of tracking problems since they can efficiently handle non-Gaussian and nonlinearity. Their combination has shown superior performance in terms of accuracy and robustness, but it suffers from the heavy computational load. Our tracking algorithm requires a significantly small number of particles while maintaining robustness and accuracy. We propose two methods in our tracking algorithm: first, we analyze object motion in a coarse-to-fine way and use hierarchical strategy to estimate it, in which the Kalman filter estimates global linear motion and the particle filter handles the local nonlinear motion, second, we give a more physically meaningful proposal distribution of the particle filter with consideration of the nature of motion. Experiments demonstrate the effectiveness of our tracking algorithm in real video sequences in which the target objects undergo rapid and abrupt motion. Furthermore, we provide quantitative comparisons between the existing tracking algorithm and the proposed tracking algorithm.",
"Object tracking is one of the most important components in numerous applications of computer vision. While much progress has been made in recent years with efforts on sharing code and datasets, it is of great importance to develop a library and benchmark to gauge the state of the art. After briefly reviewing recent advances of online object tracking, we carry out large scale experiments with various evaluation criteria to understand how these algorithms perform. The test image sequences are annotated with different attributes for performance evaluation and analysis. By analyzing quantitative results, we identify effective approaches for robust tracking and provide potential future research directions in this field."
]
} |
1703.06527 | 2953295199 | The paper focuses on the problem of vision-based obstacle detection and tracking for unmanned aerial vehicle navigation. A real-time object localization and tracking strategy from monocular image sequences is developed by effectively integrating the object detection and tracking into a dynamic Kalman model. At the detection stage, the object of interest is automatically detected and localized from a saliency map computed via the image background connectivity cue at each frame; at the tracking stage, a Kalman filter is employed to provide a coarse prediction of the object state, which is further refined via a local detector incorporating the saliency map and the temporal information between two consecutive frames. Compared to existing methods, the proposed approach does not require any manual initialization for tracking, runs much faster than the state-of-the-art trackers of its kind, and achieves competitive tracking performance on a large number of image sequences. Extensive experiments demonstrate the effectiveness and superior performance of the proposed approach. | The discriminative tracking-by-detection approaches have become increasingly popular in recent years. Zhang et al @cite_41 proposed a real-time tracker based on compressive sensing. Mahadevan et al @cite_6 proposed a saliency-based discriminative tracker, which learns the salient features based on Bayesian framework. Kalal et al @cite_38 introduced a long-term tracker which enables a re-initialization in case of tracking failures. | {
"cite_N": [
"@cite_41",
"@cite_38",
"@cite_6"
],
"mid": [
"",
"2124211486",
"2157381954"
],
"abstract": [
"",
"This paper investigates long-term tracking of unknown objects in a video stream. The object is defined by its location and extent in a single frame. In every frame that follows, the task is to determine the object's location and extent or indicate that the object is not present. We propose a novel tracking framework (TLD) that explicitly decomposes the long-term tracking task into tracking, learning, and detection. The tracker follows the object from frame to frame. The detector localizes all appearances that have been observed so far and corrects the tracker if necessary. The learning estimates the detector's errors and updates it to avoid these errors in the future. We study how to identify the detector's errors and learn from them. We develop a novel learning method (P-N learning) which estimates the errors by a pair of “experts”: (1) P-expert estimates missed detections, and (2) N-expert estimates false alarms. The learning process is modeled as a discrete dynamical system and the conditions under which the learning guarantees improvement are found. We describe our real-time implementation of the TLD framework and the P-N learning. We carry out an extensive quantitative evaluation which shows a significant improvement over state-of-the-art approaches.",
"We propose a biologically inspired framework for visual tracking based on discriminant center surround saliency. At each frame, discrimination of the target from the background is posed as a binary classification problem. From a pool of feature descriptors for the target and background, a subset that is most informative for classification between the two is selected using the principle of maximum marginal diversity. Using these features, the location of the target in the next frame is identified using top-down saliency, completing one iteration of the tracking algorithm. We also show that a simple extension of the framework to include motion features in a bottom-up saliency mode can robustly identify salient moving objects and automatically initialize the tracker. The connections of the proposed method to existing works on discriminant tracking are discussed. Experimental results comparing the proposed method to the state of the art in tracking are presented, showing improved performance."
]
} |
1703.06527 | 2953295199 | The paper focuses on the problem of vision-based obstacle detection and tracking for unmanned aerial vehicle navigation. A real-time object localization and tracking strategy from monocular image sequences is developed by effectively integrating the object detection and tracking into a dynamic Kalman model. At the detection stage, the object of interest is automatically detected and localized from a saliency map computed via the image background connectivity cue at each frame; at the tracking stage, a Kalman filter is employed to provide a coarse prediction of the object state, which is further refined via a local detector incorporating the saliency map and the temporal information between two consecutive frames. Compared to existing methods, the proposed approach does not require any manual initialization for tracking, runs much faster than the state-of-the-art trackers of its kind, and achieves competitive tracking performance on a large number of image sequences. Extensive experiments demonstrate the effectiveness and superior performance of the proposed approach. | In particular, the correlation filter-based discriminative tracking methods have attracted much attention and achieved significant progress @cite_10 . Henriques et al @cite_40 proposed a tracker using kernelized correlation filters (KCF). Zhu et al @cite_30 extended the KCF to a multi-scale kernelized tracker in order to deal with the scale variation. Zhang et al @cite_17 proposed a tracker via dense spatio-temporal context learning. Danelljan et al @cite_5 introduced a discriminative tracker using a scale pyramid representation. Li et al @cite_11 proposed to tackle the scale variation by integrating different low-level features. Danelljan et al @cite_26 designed a tracker by adaptive extension of color attributes. Readers can refer to @cite_35 @cite_0 and the references therein for details about visual tracking. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_11",
"@cite_26",
"@cite_0",
"@cite_40",
"@cite_5",
"@cite_10",
"@cite_17"
],
"mid": [
"2315667193",
"1995903777",
"",
"",
"2126302311",
"",
"1997121481",
"",
"29474918"
],
"abstract": [
"Correlation filter based tracking has attracted many researchers’ attention in recent years for high efficiency and robustness. Most existing works focus on exploiting different characteristics with correlation filters for visual tracking, e.g. circulant structure, kernel trick, effective feature representation and context information. However, how to handle the scale variation and the model drift is still an open problem. In this paper, we propose a collaborative correlation tracker to deal with the above problems. Firstly, we extend the correlation tracking filter by embedding the scale factor into the kernelized matrix to handle the scale variation. Then a novel long-term CUR filter for detection is learnt efficiently with random sampling to alleviate model drift by detecting effective object candidates in the collaborative tracker. In this way, the proposed approach could estimate the object state accurately and handle the model drift problem effectively. Extensive experiments show the superiority of the proposed method.",
"The goal of this article is to review the state-of-the-art tracking methods, classify them into different categories, and identify new trends. Object tracking, in general, is a challenging problem. Difficulties in tracking objects can arise due to abrupt object motion, changing appearance patterns of both the object and the scene, nonrigid object structures, object-to-object and object-to-scene occlusions, and camera motion. Tracking is usually performed in the context of higher-level applications that require the location and or shape of the object in every frame. Typically, assumptions are made to constrain the tracking problem in the context of a particular application. In this survey, we categorize the tracking methods on the basis of the object and motion representations used, provide detailed descriptions of representative methods in each category, and examine their pros and cons. Moreover, we discuss the important issues related to tracking including the use of appropriate image features, selection of motion models, and detection of objects.",
"",
"",
"There is a large variety of trackers, which have been proposed in the literature during the last two decades with some mixed success. Object tracking in realistic scenarios is a difficult problem, therefore, it remains a most active area of research in computer vision. A good tracker should perform well in a large number of videos involving illumination changes, occlusion, clutter, camera motion, low contrast, specularities, and at least six more aspects. However, the performance of proposed trackers have been evaluated typically on less than ten videos, or on the special purpose datasets. In this paper, we aim to evaluate trackers systematically and experimentally on 315 video fragments covering above aspects. We selected a set of nineteen trackers to include a wide variety of algorithms often cited in literature, supplemented with trackers appearing in 2010 and 2011 for which the code was publicly available. We demonstrate that trackers can be evaluated objectively by survival curves, Kaplan Meier statistics, and Grubs testing. We find that in the evaluation practice the F-score is as effective as the object tracking accuracy (OTA) score. The analysis under a large variety of circumstances provides objective insight into the strengths and weaknesses of trackers.",
"",
"Robust scale estimation is a challenging problem in visual object tracking. Most existing methods fail to handle large scale variations in complex image sequences. This paper presents a novel approach for robust scale estimation in a tracking-by-detection framework. The proposed approach works by learning discriminative correlation filters based on a scale pyramid representation. We learn separate filters for translation and scale estimation, and show that this improves the performance compared to an exhaustive scale search. Our scale estimation approach is generic as it can be incorporated into any tracking method with no inherent scale estimation.Experiments are performed on 28 benchmark sequences with significant scale variations. Our results show that the proposed approach significantly improves the performance by 18.8 in median distance precision compared to our baseline. Finally, we provide both quantitative and qualitative comparison of our approach with state-of-the-art trackers in literature. The proposed method is shown to outperform the best existing tracker by 16.6 in median distance precision, while operating at real-time.",
"",
"In this paper, we present a simple yet fast and robust algorithm which exploits the dense spatio-temporal context for visual tracking. Our approach formulates the spatio-temporal relationships between the object of interest and its locally dense contexts in a Bayesian framework, which models the statistical correlation between the simple low-level features (i.e., image intensity and position) from the target and its surrounding regions. The tracking problem is then posed by computing a confidence map which takes into account the prior information of the target location and thereby alleviates target location ambiguity effectively. We further propose a novel explicit scale adaptation scheme, which is able to deal with target scale variations efficiently and effectively. The Fast Fourier Transform (FFT) is adopted for fast learning and detection in this work, which only needs 4 FFT operations. Implemented in MATLAB without code optimization, the proposed tracker runs at 350 frames per second on an i7 machine. Extensive experimental results show that the proposed algorithm performs favorably against state-of-the-art methods in terms of efficiency, accuracy and robustness."
]
} |
1703.06408 | 2604913478 | In this work, we propose the combined usage of low- and high-level blocks of convolutional neural networks (CNNs) for improving object recognition. While recent research focused on either propagating the context from all layers, e.g. ResNet, (including the very low-level layers) or having multiple loss layers (e.g. GoogLeNet), the importance of the features close to the higher layers is ignored. This paper postulates that the use of context closer to the high-level layers provides the scale and translation invariance and works better than using the top layer only. In particular, we extend AlexNet and GoogLeNet by additional connections in the top n layers. In order to demonstrate the effectiveness of the proposed approach, we evaluated it on the standard ImageNet task. The relative reduction of the classification error is around 1-2 without affecting the computational cost. Furthermore, we show that this approach is orthogonal to typical test data augmentation techniques, as recently introduced by (leading to a runtime reduction of 144 during test time). | One well-known benchmark for several visual tasks, including image classification, is the annual ImageNet Large Scale Visual Recognition Challenge (ILSRVC) @cite_13 . It allows the participating teams to compare the performance of their developed models. The data set used for the competition consists of 1.000 image classes containing 1.2 million labeled images for training, 50.000 labeled validation images and 100.000 unlabeled images which are used to compare the performance of the submitted entries. | {
"cite_N": [
"@cite_13"
],
"mid": [
"2117539524"
],
"abstract": [
"The ImageNet Large Scale Visual Recognition Challenge is a benchmark in object category classification and detection on hundreds of object categories and millions of images. The challenge has been run annually from 2010 to present, attracting participation from more than fifty institutions. This paper describes the creation of this benchmark dataset and the advances in object recognition that have been possible as a result. We discuss the challenges of collecting large-scale ground truth annotation, highlight key breakthroughs in categorical object recognition, provide a detailed analysis of the current state of the field of large-scale image classification and object detection, and compare the state-of-the-art computer vision accuracy with human accuracy. We conclude with lessons learned in the 5 years of the challenge, and propose future directions and improvements."
]
} |
1703.06408 | 2604913478 | In this work, we propose the combined usage of low- and high-level blocks of convolutional neural networks (CNNs) for improving object recognition. While recent research focused on either propagating the context from all layers, e.g. ResNet, (including the very low-level layers) or having multiple loss layers (e.g. GoogLeNet), the importance of the features close to the higher layers is ignored. This paper postulates that the use of context closer to the high-level layers provides the scale and translation invariance and works better than using the top layer only. In particular, we extend AlexNet and GoogLeNet by additional connections in the top n layers. In order to demonstrate the effectiveness of the proposed approach, we evaluated it on the standard ImageNet task. The relative reduction of the classification error is around 1-2 without affecting the computational cost. Furthermore, we show that this approach is orthogonal to typical test data augmentation techniques, as recently introduced by (leading to a runtime reduction of 144 during test time). | In 2012, have, for the first time, used a deep CNN in the ILSVRC and won the image classification with a significant margin, outperforming all traditional methods @cite_6 . The winning architecture consists of five convolutional layers which are followed by three fully connected layers. The first and second layer use local response normalizations and max-pooling before passing the activations to the next layer. The fifth convolutional layer is again followed by a max-pooling layer which provides the input to the fully-connected layers. To prevent overfitting, the fully-connected layers use dropout @cite_11 . Rectified linear units (ReLU) are used as activation function in all layers @cite_15 . AlexNet achieves a top-5 error rate of @math 6.656 -0.2in | {
"cite_N": [
"@cite_15",
"@cite_6",
"@cite_11"
],
"mid": [
"1665214252",
"2163605009",
"1904365287"
],
"abstract": [
"Restricted Boltzmann machines were developed using binary stochastic hidden units. These can be generalized by replacing each binary unit by an infinite number of copies that all have the same weights but have progressively more negative biases. The learning and inference rules for these \"Stepped Sigmoid Units\" are unchanged. They can be approximated efficiently by noisy, rectified linear units. Compared with binary units, these units learn features that are better for object recognition on the NORB dataset and face verification on the Labeled Faces in the Wild dataset. Unlike binary units, rectified linear units preserve information about relative intensities as information travels through multiple layers of feature detectors.",
"We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5 and 17.0 which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called \"dropout\" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3 , compared to 26.2 achieved by the second-best entry.",
"When a large feedforward neural network is trained on a small training set, it typically performs poorly on held-out test data. This \"overfitting\" is greatly reduced by randomly omitting half of the feature detectors on each training case. This prevents complex co-adaptations in which a feature detector is only helpful in the context of several other specific feature detectors. Instead, each neuron learns to detect a feature that is generally helpful for producing the correct answer given the combinatorially large variety of internal contexts in which it must operate. Random \"dropout\" gives big improvements on many benchmark tasks and sets new records for speech and object recognition."
]
} |
1703.06408 | 2604913478 | In this work, we propose the combined usage of low- and high-level blocks of convolutional neural networks (CNNs) for improving object recognition. While recent research focused on either propagating the context from all layers, e.g. ResNet, (including the very low-level layers) or having multiple loss layers (e.g. GoogLeNet), the importance of the features close to the higher layers is ignored. This paper postulates that the use of context closer to the high-level layers provides the scale and translation invariance and works better than using the top layer only. In particular, we extend AlexNet and GoogLeNet by additional connections in the top n layers. In order to demonstrate the effectiveness of the proposed approach, we evaluated it on the standard ImageNet task. The relative reduction of the classification error is around 1-2 without affecting the computational cost. Furthermore, we show that this approach is orthogonal to typical test data augmentation techniques, as recently introduced by (leading to a runtime reduction of 144 during test time). | Note that the general idea of adding skip-connections (so-called residual connections) has been recently introduced by in @cite_35 @cite_0 . However, the motivation and effects are fundamentally different. In ResNet, the skip-connections go through all the layers, letting the layers mainly pass the information through the network with minor additive terms. This mainly overcomes the vanishing gradient problem and allows learning architectures with hundreds of layers. Contrary, the main idea in this paper is to use the extracted features of multiple layers directly for classification. The only work which comes close to our idea is the proposal of Center-Multilayer Features (CMF) as proposed by @cite_21 . They used stacked convolutional auto-encoders (SCAE) for classifying an image or the center pixel of a patch. This idea is similar to our pre-study (section ) where we use multiple feature representations from fine-tuned networks. However, their network architecture is shallow and they use features from all the layers. | {
"cite_N": [
"@cite_0",
"@cite_35",
"@cite_21"
],
"mid": [
"",
"2194775991",
"2576489129"
],
"abstract": [
"",
"Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.",
"This paper presents a novel, highly-adaptable Java framework N-light-N, for the work with deep neural networks, especially with CAEs. While the most popular deep learning libraries focus on fast processing and high performance, they only implement the main-stream network architectures and network units. In recent research in the document domain, however, we have shown that modified networks, units, and training processes significantly improve the performance in various tasks. To enable the document research community with such capabilities, in this paper we introduce a novel, publicly available Deep Learning framework which is easy to use, adapt, and extend. Furthermore, we present successful applications for three tasks, including two in the domain of handwritten historical documents, and show how the framework can be used for adaptation, optimization, and deeper analysis."
]
} |
1703.06408 | 2604913478 | In this work, we propose the combined usage of low- and high-level blocks of convolutional neural networks (CNNs) for improving object recognition. While recent research focused on either propagating the context from all layers, e.g. ResNet, (including the very low-level layers) or having multiple loss layers (e.g. GoogLeNet), the importance of the features close to the higher layers is ignored. This paper postulates that the use of context closer to the high-level layers provides the scale and translation invariance and works better than using the top layer only. In particular, we extend AlexNet and GoogLeNet by additional connections in the top n layers. In order to demonstrate the effectiveness of the proposed approach, we evaluated it on the standard ImageNet task. The relative reduction of the classification error is around 1-2 without affecting the computational cost. Furthermore, we show that this approach is orthogonal to typical test data augmentation techniques, as recently introduced by (leading to a runtime reduction of 144 during test time). | Apart from novel network architectures, a lot of work has been done to understand how the networks are learning and how they can be improved @cite_9 @cite_51 . Especially visualization techniques have helped to get an understanding of the convolutional layers @cite_30 . Despite all the research that has been done to develop new CNN architectures, novel architectures typically employ more layers or make use ensembling techniques @cite_34 . However, both approaches typically require more computational resources at training and inference time. | {
"cite_N": [
"@cite_30",
"@cite_9",
"@cite_34",
"@cite_51"
],
"mid": [
"1849277567",
"1932847118",
"2100128988",
""
],
"abstract": [
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark [18]. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we explore both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. Used in a diagnostic role, these visualizations allow us to find model architectures that outperform on the ImageNet classification benchmark. We also perform an ablation study to discover the performance contribution from different model layers. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.",
"Though recent advanced convolutional neural networks (CNNs) have been improving the image recognition accuracy, the models are getting more complex and time-consuming. For real-world applications in industrial and commercial scenarios, engineers and developers are often faced with the requirement of constrained time budget. In this paper, we investigate the accuracy of CNNs under constrained time cost. Under this constraint, the designs of the network architectures should exhibit as trade-offs among the factors like depth, numbers of filters, filter sizes, etc. With a series of controlled comparisons, we progressively modify a baseline model while preserving its time complexity. This is also helpful for understanding the importance of the factors in network designs. We present an architecture that achieves very competitive accuracy in the ImageNet dataset (11.8 top-5 error, 10-view test), yet is 20 faster than “AlexNet” [14] (16.0 top-5 error, 10-view test).",
"Neural network ensemble is a learning paradigm where many neural networks are jointly used to solve a problem. In this paper, the relationship between the ensemble and its component neural networks is analyzed from the context of both regression and classification, which reveals that it may be better to ensemble many instead of all of the neural networks at hand. This result is interesting because at present, most approaches ensemble all the available neural networks for prediction. Then, in order to show that the appropriate neural networks for composing an ensemble can be effectively selected from a set of available neural networks, an approach named GASEN is presented. GASEN trains a number of neural networks at first. Then it assigns random weights to those networks and employs genetic algorithm to evolve the weights so that they can characterize to some extent the fitness of the neural networks in constituting an ensemble. Finally it selects some neural networks based on the evolved weights to make up the ensemble. A large empirical study shows that, compared with some popular ensemble approaches such as Bagging and Boosting, GASEN can generate neural network ensembles with far smaller sizes but stronger generalization ability. Furthermore, in order to understand the working mechanism of GASEN, the bias-variance decomposition of the error is provided in this paper, which shows that the success of GASEN may lie in that it can significantly reduce the bias as well as the variance.",
""
]
} |
1703.06217 | 2949427550 | We propose and systematically evaluate three strategies for training dynamically-routed artificial neural networks: graphs of learned transformations through which different input signals may take different paths. Though some approaches have advantages over others, the resulting networks are often qualitatively similar. We find that, in dynamically-routed networks trained to classify images, layers and branches become specialized to process distinct categories of images. Additionally, given a fixed computational budget, dynamically-routed networks tend to perform better than comparable statically-routed networks. | Since the late 1980s, researchers have combined artificial neural networks with decision trees in various ways @cite_18 @cite_25 . More recently, performed joint optimization of ANN and decision tree parameters, and used randomized multi-layer networks to compute decision tree split functions. To our knowledge, the family of inference systems we discuss was first described by . Additionally, explored dynamically skipping layers in neural networks, and explored dynamic routing in networks with equal-length paths. Some recently-developed visual detection systems perform cascaded evaluation of convolutional neural network layers @cite_16 @cite_11 @cite_15 @cite_2 ; though highly specialized for the task of visual detection, these modifications can radically improve efficiency. While these approaches lend evidence that dynamic routing can be effective, they either ignore the cost of computation, or do not represent it explicitly, and instead use opaque heuristics to trade accuracy for efficiency. We build on this foundation by deriving training procedures from arbitrary application-provided costs of error and computation, comparing one actor-style and two critic-style strategies, and considering regularization and optimization in the context of dynamically-routed networks. | {
"cite_N": [
"@cite_18",
"@cite_2",
"@cite_15",
"@cite_16",
"@cite_25",
"@cite_11"
],
"mid": [
"2163142059",
"639708223",
"",
"1934410531",
"2179423374",
"2950167387"
],
"abstract": [
"Abstract This article presents a case study in examining the bias of two particular formalisms: decision trees and linear threshold units. The immediate result is a new hybrid representation, called a ‘perceptron tree’, and an associated learning algorithm called the ‘percepton tree error correction procedure’. The longer term result is a model for exploring issues related to understanding representational bias and constructing other useful hybrid representations.",
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet [1] and Fast R-CNN [2] have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features—using the recently popular terminology of neural networks with ’attention’ mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model [3] , our detection system has a frame rate of 5 fps ( including all steps ) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.",
"",
"In real-world face detection, large visual variations, such as those due to pose, expression, and lighting, demand an advanced discriminative model to accurately differentiate faces from the backgrounds. Consequently, effective models for the problem tend to be computationally prohibitive. To address these two conflicting challenges, we propose a cascade architecture built on convolutional neural networks (CNNs) with very powerful discriminative capability, while maintaining high performance. The proposed CNN cascade operates at multiple resolutions, quickly rejects the background regions in the fast low resolution stages, and carefully evaluates a small number of challenging candidates in the last high resolution stage. To improve localization effectiveness, and reduce the number of candidates at later stages, we introduce a CNN-based calibration stage after each of the detection stages in the cascade. The output of each calibration stage is used to adjust the detection window position for input to the subsequent stage. The proposed method runs at 14 FPS on a single CPU core for VGA-resolution images and 100 FPS using a GPU, and achieves state-of-the-art detection performance on two public face detection benchmarks.",
"Deep learning has become the state-of-art tool in many applications, but the evaluation and training of deep models can be time-consuming and computationally expensive. The conditional computation approach has been proposed to tackle this problem (, 2013; Davis & Arel, 2013). It operates by selectively activating only parts of the network at a time. In this paper, we use reinforcement learning as a tool to optimize conditional computation policies. More specifically, we cast the problem of learning activation-dependent policies for dropping out blocks of units as a reinforcement learning problem. We propose a learning scheme motivated by computation speed, capturing the idea of wanting to have parsimonious activations while maintaining prediction accuracy. We apply a policy gradient algorithm for learning policies that optimize this loss function and propose a regularization mechanism that encourages diversification of the dropout policy. We present encouraging empirical results showing that this approach improves the speed of computation without impacting the quality of the approximation.",
"The design of complexity-aware cascaded detectors, combining features of very different complexities, is considered. A new cascade design procedure is introduced, by formulating cascade learning as the Lagrangian optimization of a risk that accounts for both accuracy and complexity. A boosting algorithm, denoted as complexity aware cascade training (CompACT), is then derived to solve this optimization. CompACT cascades are shown to seek an optimal trade-off between accuracy and complexity by pushing features of higher complexity to the later cascade stages, where only a few difficult candidate patches remain to be classified. This enables the use of features of vastly different complexities in a single detector. In result, the feature pool can be expanded to features previously impractical for cascade design, such as the responses of a deep convolutional neural network (CNN). This is demonstrated through the design of a pedestrian detector with a pool of features whose complexities span orders of magnitude. The resulting cascade generalizes the combination of a CNN with an object proposal mechanism: rather than a pre-processing stage, CompACT cascades seamlessly integrate CNNs in their stages. This enables state of the art performance on the Caltech and KITTI datasets, at fairly fast speeds."
]
} |
1703.06189 | 2953229046 | Temporal Action Proposal (TAP) generation is an important problem, as fast and accurate extraction of semantically important (e.g. human actions) segments from untrimmed videos is an important step for large-scale video analysis. We propose a novel Temporal Unit Regression Network (TURN) model. There are two salient aspects of TURN: (1) TURN jointly predicts action proposals and refines the temporal boundaries by temporal coordinate regression; (2) Fast computation is enabled by unit feature reuse: a long untrimmed video is decomposed into video units, which are reused as basic building blocks of temporal proposals. TURN outperforms the state-of-the-art methods under average recall (AR) by a large margin on THUMOS-14 and ActivityNet datasets, and runs at over 880 frames per second (FPS) on a TITAN X GPU. We further apply TURN as a proposal generation stage for existing temporal action localization pipelines, it outperforms state-of-the-art performance on THUMOS-14 and ActivityNet. | Sparse-prop @cite_36 proposes the use of STIPs @cite_29 and dictionary learning for class-independent proposal generation. S-CNN @cite_1 presents a two-stage action localization system, in which the first stage is temporal proposal generation, and shows the effectiveness of temporal proposals for action localization. S-CNN's proposal network is based on fine-tuning 3D convolutional networks (C3D) @cite_14 to binary classification task. DAPs @cite_21 adopts LSTM networks to encode a video stream and produce proposals inside the video stream. | {
"cite_N": [
"@cite_14",
"@cite_36",
"@cite_29",
"@cite_21",
"@cite_1"
],
"mid": [
"2952633803",
"2471143248",
"2020163092",
"2519328139",
"2394849137"
],
"abstract": [
"We propose a simple, yet effective approach for spatiotemporal feature learning using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset. Our findings are three-fold: 1) 3D ConvNets are more suitable for spatiotemporal feature learning compared to 2D ConvNets; 2) A homogeneous architecture with small 3x3x3 convolution kernels in all layers is among the best performing architectures for 3D ConvNets; and 3) Our learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks. In addition, the features are compact: achieving 52.8 accuracy on UCF101 dataset with only 10 dimensions and also very efficient to compute due to the fast inference of ConvNets. Finally, they are conceptually very simple and easy to train and use.",
"In many large-scale video analysis scenarios, one is interested in localizing and recognizing human activities that occur in short temporal intervals within long untrimmed videos. Current approaches for activity detection still struggle to handle large-scale video collections and the task remains relatively unexplored. This is in part due to the computational complexity of current action recognition approaches and the lack of a method that proposes fewer intervals in the video, where activity processing can be focused. In this paper, we introduce a proposal method that aims to recover temporal segments containing actions in untrimmed videos. Building on techniques for learning sparse dictionaries, we introduce a learning framework to represent and retrieve activity proposals. We demonstrate the capabilities of our method in not only producing high quality proposals but also in its efficiency. Finally, we show the positive impact our method has on recognition performance when it is used for action detection, while running at 10FPS.",
"Local image features or interest points provide compact and abstract representations of patterns in an image. In this paper, we extend the notion of spatial interest points into the spatio-temporal domain and show how the resulting features often reflect interesting events that can be used for a compact representation of video data as well as for interpretation of spatio-temporal events. To detect spatio-temporal events, we build on the idea of the Harris and Forstner interest point operators and detect local structures in space-time where the image values have significant local variations in both space and time. We estimate the spatio-temporal extents of the detected events by maximizing a normalized spatio-temporal Laplacian operator over spatial and temporal scales. To represent the detected events, we then compute local, spatio-temporal, scale-invariant N-jets and classify each event with respect to its jet descriptor. For the problem of human motion analysis, we illustrate how a video representation in terms of local space-time features allows for detection of walking people in scenes with occlusions and dynamic cluttered backgrounds.",
"Object proposals have contributed significantly to recent advances in object understanding in images. Inspired by the success of this approach, we introduce Deep Action Proposals (DAPs), an effective and efficient algorithm for generating temporal action proposals from long videos. We show how to take advantage of the vast capacity of deep learning models and memory cells to retrieve from untrimmed videos temporal segments, which are likely to contain actions. A comprehensive evaluation indicates that our approach outperforms previous work on a large scale action benchmark, runs at 134 FPS making it practical for large-scale scenarios, and exhibits an appealing ability to generalize, i.e. to retrieve good quality temporal proposals of actions unseen in training.",
"We address temporal action localization in untrimmed long videos. This is important because videos in real applications are usually unconstrained and contain multiple action instances plus video content of background scenes or other activities. To address this challenging issue, we exploit the effectiveness of deep networks in temporal action localization via three segment-based 3D ConvNets: (1) a proposal network identifies candidate segments in a long video that may contain actions; (2) a classification network learns one-vs-all action classification model to serve as initialization for the localization network; and (3) a localization network fine-tunes on the learned classification network to localize each action instance. We propose a novel loss function for the localization network to explicitly consider temporal overlap and therefore achieve high temporal localization accuracy. Only the proposal network and the localization network are used during prediction. On two large-scale benchmarks, our approach achieves significantly superior performances compared with other state-of-the-art systems: mAP increases from 1.7 to 7.4 on MEXaction2 and increases from 15.0 to 19.0 on THUMOS 2014, when the overlap threshold for evaluation is set to 0.5."
]
} |
1703.06189 | 2953229046 | Temporal Action Proposal (TAP) generation is an important problem, as fast and accurate extraction of semantically important (e.g. human actions) segments from untrimmed videos is an important step for large-scale video analysis. We propose a novel Temporal Unit Regression Network (TURN) model. There are two salient aspects of TURN: (1) TURN jointly predicts action proposals and refines the temporal boundaries by temporal coordinate regression; (2) Fast computation is enabled by unit feature reuse: a long untrimmed video is decomposed into video units, which are reused as basic building blocks of temporal proposals. TURN outperforms the state-of-the-art methods under average recall (AR) by a large margin on THUMOS-14 and ActivityNet datasets, and runs at over 880 frames per second (FPS) on a TITAN X GPU. We further apply TURN as a proposal generation stage for existing temporal action localization pipelines, it outperforms state-of-the-art performance on THUMOS-14 and ActivityNet. | A handful of efforts have been seen in spatio-temporal action localization. Gkioxari @cite_33 extract proposals from RGB images with SelectiveSearch @cite_17 and then apply R-CNN @cite_30 on both RGB and optical flow images for action detection. Weinzaepfel @cite_26 replace SelectiveSearch @cite_17 with EdgeBoxes @cite_19 . Mettes @cite_22 propose to use sparse points as supervision for action detection to save tedious annotation work. | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_33",
"@cite_22",
"@cite_19",
"@cite_17"
],
"mid": [
"2102605133",
"2950966695",
"1923332106",
"2950281962",
"7746136",
""
],
"abstract": [
"Object detection performance, as measured on the canonical PASCAL VOC dataset, has plateaued in the last few years. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30 relative to the previous best result on VOC 2012 -- achieving a mAP of 53.3 . Our approach combines two key insights: (1) one can apply high-capacity convolutional neural networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (2) when labeled training data is scarce, supervised pre-training for an auxiliary task, followed by domain-specific fine-tuning, yields a significant performance boost. Since we combine region proposals with CNNs, we call our method R-CNN: Regions with CNN features. We also present experiments that provide insight into what the network learns, revealing a rich hierarchy of image features. Source code for the complete system is available at http: www.cs.berkeley.edu rbg rcnn.",
"We propose an effective approach for spatio-temporal action localization in realistic videos. The approach first detects proposals at the frame-level and scores them with a combination of static and motion CNN features. It then tracks high-scoring proposals throughout the video using a tracking-by-detection approach. Our tracker relies simultaneously on instance-level and class-level detectors. The tracks are scored using a spatio-temporal motion histogram, a descriptor at the track level, in combination with the CNN features. Finally, we perform temporal localization of the action using a sliding-window approach at the track level. We present experimental results for spatio-temporal localization on the UCF-Sports, J-HMDB and UCF-101 action localization datasets, where our approach outperforms the state of the art with a margin of 15 , 7 and 12 respectively in mAP.",
"We address the problem of action detection in videos. Driven by the latest progress in object detection from 2D images, we build action models using rich feature hierarchies derived from shape and kinematic cues. We incorporate appearance and motion in two ways. First, starting from image region proposals we select those that are motion salient and thus are more likely to contain the action. This leads to a significant reduction in the number of regions being processed and allows for faster computations. Second, we extract spatio-temporal feature representations to build strong classifiers using Convolutional Neural Networks. We link our predictions to produce detections consistent in time, which we call action tubes. We show that our approach outperforms other techniques in the task of action detection.",
"We strive for spatio-temporal localization of actions in videos. The state-of-the-art relies on action proposals at test time and selects the best one with a classifier trained on carefully annotated box annotations. Annotating action boxes in video is cumbersome, tedious, and error prone. Rather than annotating boxes, we propose to annotate actions in video with points on a sparse subset of frames only. We introduce an overlap measure between action proposals and points and incorporate them all into the objective of a non-convex Multiple Instance Learning optimization. Experimental evaluation on the UCF Sports and UCF 101 datasets shows that (i) spatio-temporal proposals can be used to train classifiers while retaining the localization performance, (ii) point annotations yield results comparable to box annotations while being significantly faster to annotate, (iii) with a minimum amount of supervision our approach is competitive to the state-of-the-art. Finally, we introduce spatio-temporal action annotations on the train and test videos of Hollywood2, resulting in Hollywood2Tubes, available at this http URL",
"The use of object proposals is an effective recent approach for increasing the computational efficiency of object detection. We propose a novel method for generating object bounding box proposals using edges. Edges provide a sparse yet informative representation of an image. Our main observation is that the number of contours that are wholly contained in a bounding box is indicative of the likelihood of the box containing an object. We propose a simple box objectness score that measures the number of edges that exist in the box minus those that are members of contours that overlap the box’s boundary. Using efficient data structures, millions of candidate boxes can be evaluated in a fraction of a second, returning a ranked set of a few thousand top-scoring proposals. Using standard metrics, we show results that are significantly more accurate than the current state-of-the-art while being faster to compute. In particular, given just 1000 proposals we achieve over 96 object recall at overlap threshold of 0.5 and over 75 recall at the more challenging overlap of 0.7. Our approach runs in 0.25 seconds and we additionally demonstrate a near real-time variant with only minor loss in accuracy.",
""
]
} |
1703.06630 | 2604637576 | The number of documents available into Internet moves each day up. For this reason, processing this amount of information effectively and expressibly becomes a major concern for companies and scientists. Methods that represent a textual document by a topic representation are widely used in Information Retrieval (IR) to process big data such as Wikipedia articles. One of the main difficulty in using topic model on huge data collection is related to the material resources (CPU time and memory) required for model estimate. To deal with this issue, we propose to build topic spaces from summarized documents. In this paper, we present a study of topic space representation in the context of big data. The topic space representation behavior is analyzed on different languages. Experiments show that topic spaces estimated from text summaries are as relevant as those estimated from the complete documents. The real advantage of such an approach is the processing time gain: we showed that the processing time can be drastically reduced using summarized documents (more than 60 in general). This study finally points out the differences between thematic representations of documents depending on the targeted languages such as English or latin languages. | This method was improved by @cite_15 which proposed a Probabilistic LSA (PLSA). PLSA models each word in a document as a sample from a mixture model, where the mixture components are multinomial random variables that can be viewed as representations of topics. This method demonstrated its performance on various tasks, such as sentence @cite_6 or keyword @cite_8 extraction. In spite of the effectiveness of the PLSA approach, this method has two main drawbacks. The distribution of topics in PLSA is indexed by training documents. Thus, the number of its parameters grows with the training document set size and then, the model is prone to overfitting which is a main issue in an IR task such as documents clustering. However, to address this shortcoming, a tempering heuristic is used to smooth the parameter of PLSA models for acceptable predictive performance: the authors in @cite_1 showed that overfitting can occur even if tempering process is used. | {
"cite_N": [
"@cite_15",
"@cite_1",
"@cite_6",
"@cite_8"
],
"mid": [
"2953062473",
"2097129520",
"2118714763",
""
],
"abstract": [
"Probabilistic Latent Semantic Analysis is a novel statistical technique for the analysis of two-mode and co-occurrence data, which has applications in information retrieval and filtering, natural language processing, machine learning from text, and in related areas. Compared to standard Latent Semantic Analysis which stems from linear algebra and performs a Singular Value Decomposition of co-occurrence tables, the proposed method is based on a mixture decomposition derived from a latent class model. This results in a more principled approach which has a solid foundation in statistics. In order to avoid overfitting, we propose a widely applicable generalization of maximum likelihood model fitting by tempered EM. Our approach yields substantial and consistent improvements over Latent Semantic Analysis in a number of experiments.",
"Recommender systems leverage product and community information to target products to consumers. Researchers have developed collaborative recommenders, content-based recommenders, and a few hybrid systems. We propose a unified probabilistic framework for merging collaborative and content-based recommendations. We extend Hofmarm's (1999) aspect model to incorporate three-way co-occurrence data among users, items, and item content. The relative influence of collaboration data versus content data is not imposed as an exogenous parameter, but rather emerges naturally from the given data sources. However, global probabilistic models coupled with standard EM learning algorithms tend to drastically overfit in the sparsedata situations typical of recommendation applications. We show that secondary content information can often be used to overcome sparsity. Experiments on data from the Researchlndex library of Computer Science publications show that appropriate mixture models incorporating secondary data produce significantly better quality recommenders than k-nearest neighbors (k-NN). Global probabilistic models also allow more general inferences than local methods like k-NN.",
"Statistical language models used in large-vocabulary speech recognition must properly encapsulate the various constraints, both local and global, present in the language. While local constraints are readily captured through n-gram modeling, global constraints, such as long-term semantic dependencies, have been more difficult to handle within a data-driven formalism. This paper focuses on the use of latent semantic analysis, a paradigm that automatically uncovers the salient semantic relationships between words and documents in a given corpus. In this approach, (discrete) words and documents are mapped onto a (continuous) semantic vector space, in which familiar clustering techniques can be applied. This leads to the specification of a powerful framework for automatic semantic classification, as well as the derivation of several language model families with various smoothing properties. Because of their large-span nature, these language models are well suited to complement conventional n-grams. An integrative formulation is proposed for harnessing this synergy, in which the latent semantic information is used to adjust the standard n-gram probability. Such hybrid language modeling compares favorably with the corresponding n-gram baseline: experiments conducted on the Wall Street Journal domain show a reduction in average word error rate of over 20 . This paper concludes with a discussion of intrinsic tradeoffs, such as the influence of training data selection on the resulting performance.",
""
]
} |
1703.06630 | 2604637576 | The number of documents available into Internet moves each day up. For this reason, processing this amount of information effectively and expressibly becomes a major concern for companies and scientists. Methods that represent a textual document by a topic representation are widely used in Information Retrieval (IR) to process big data such as Wikipedia articles. One of the main difficulty in using topic model on huge data collection is related to the material resources (CPU time and memory) required for model estimate. To deal with this issue, we propose to build topic spaces from summarized documents. In this paper, we present a study of topic space representation in the context of big data. The topic space representation behavior is analyzed on different languages. Experiments show that topic spaces estimated from text summaries are as relevant as those estimated from the complete documents. The real advantage of such an approach is the processing time gain: we showed that the processing time can be drastically reduced using summarized documents (more than 60 in general). This study finally points out the differences between thematic representations of documents depending on the targeted languages such as English or latin languages. | To overcome these two issues, the latent Dirichlet allocation (LDA) @cite_3 method was proposed. Thus, the number of LDA parameters does not grow with the size of the training corpus and LDA is not candidate for overfitting. Next section describes more precisely the LDA approach that will be used in our experimental study. | {
"cite_N": [
"@cite_3"
],
"mid": [
"1880262756"
],
"abstract": [
"We describe latent Dirichlet allocation (LDA), a generative probabilistic model for collections of discrete data such as text corpora. LDA is a three-level hierarchical Bayesian model, in which each item of a collection is modeled as a finite mixture over an underlying set of topics. Each topic is, in turn, modeled as an infinite mixture over an underlying set of topic probabilities. In the context of text modeling, the topic probabilities provide an explicit representation of a document. We present efficient approximate inference techniques based on variational methods and an EM algorithm for empirical Bayes parameter estimation. We report results in document modeling, text classification, and collaborative filtering, comparing to a mixture of unigrams model and the probabilistic LSI model."
]
} |
1703.06630 | 2604637576 | The number of documents available into Internet moves each day up. For this reason, processing this amount of information effectively and expressibly becomes a major concern for companies and scientists. Methods that represent a textual document by a topic representation are widely used in Information Retrieval (IR) to process big data such as Wikipedia articles. One of the main difficulty in using topic model on huge data collection is related to the material resources (CPU time and memory) required for model estimate. To deal with this issue, we propose to build topic spaces from summarized documents. In this paper, we present a study of topic space representation in the context of big data. The topic space representation behavior is analyzed on different languages. Experiments show that topic spaces estimated from text summaries are as relevant as those estimated from the complete documents. The real advantage of such an approach is the processing time gain: we showed that the processing time can be drastically reduced using summarized documents (more than 60 in general). This study finally points out the differences between thematic representations of documents depending on the targeted languages such as English or latin languages. | The authors in @cite_4 evaluated the effectiveness of the Jensen-Shannon ( @math ) theoretic measure @cite_12 in predicting systems ranks in two summarization tasks: query-focused and update summarization . They have shown that ranks produced by Pyramids and those produced by @math measure correlate. However, they did not investigate the effect of the measure in summarization tasks such as generic multi-document summarization (DUC 2004 Task 2), biographical summarization (DUC 2004 Task 5), opinion summarization (TAC 2008 OS), and summarization in languages other than English. | {
"cite_N": [
"@cite_4",
"@cite_12"
],
"mid": [
"2141927646",
"2146950091"
],
"abstract": [
"We present a fully automatic method for content selection evaluation in summarization that does not require the creation of human model summaries. Our work capitalizes on the assumption that the distribution of words in the input and an informative summary of that input should be similar to each other. Results on a large scale evaluation from the Text Analysis Conference show that input-summary comparisons are very effective for the evaluation of content selection. Our automatic methods rank participating systems similarly to manual model-based pyramid evaluation and to manual human judgments of responsiveness. The best feature, Jensen-Shannon divergence, leads to a correlation as high as 0.88 with manual pyramid and 0.73 with responsiveness evaluations.",
"A novel class of information-theoretic divergence measures based on the Shannon entropy is introduced. Unlike the well-known Kullback divergences, the new measures do not require the condition of absolute continuity to be satisfied by the probability distributions involved. More importantly, their close relationship with the variational distance and the probability of misclassification error are established in terms of bounds. These bounds are crucial in many applications of divergence measures. The measures are also well characterized by the properties of nonnegativity, finiteness, semiboundedness, and boundedness. >"
]
} |
1703.06169 | 2604725943 | Massive Open Online Courses (MOOCs) use peer assessment to grade open ended questions at scale, allowing students to provide feedback. Relative to teacher based grading, peer assessment on MOOCs traditionally delivers lower quality feedback and fewer learner interactions. We present the identified peer review (IPR) framework, which provides non-blind peer assessment and incentives driving high quality feedback. We show that, compared to traditional peer assessment methods, IPR leads to significantly longer and more useful feedback as well as more discussion between peers. | Assessment is important to the pedagogy of MOOCs @cite_6 . MOOCs utilize peer assessment to assess student's work in a scalable way, not dependent on a one to many teacher to student relationship, but, optimal methods of choosing graders and assignments to grade remains an open question @cite_3 . The benefits of peer assessment include improvement of higher order thinking skills, consolidation of topical knowledge, and individualized feedback for each participant @cite_0 . Giving and getting feedback has been identified as an effective way to learn in online @cite_2 and in the classroom. | {
"cite_N": [
"@cite_0",
"@cite_3",
"@cite_6",
"@cite_2"
],
"mid": [
"2106189547",
"2337971266",
"2077286707",
""
],
"abstract": [
"We have implemented a peer grading system for review of student assignments over the World Wide Web and used it in approximately eight courses in computing and electrical engineering. Students prepare their assignments and submit them to our Peer Grader (PG) system. Other students are then assigned to review and grade the assignments. The system allows authors and reviewers to communicate, with authors being able to update their submissions. Unique features of our approach include the ability to submit arbitrary sets of Web pages for review, and mechanisms for encouraging careful review of submissions. Electronic peer review facilitates collaborative learning in several ways. First, there is the obvious fact that students can learn from their reviewers' comments. Second, students help each other to improve their communication skills. Third, in team projects, peer review allows team members to be assessed by each other. Finally, peer review makes it possible to break up a large project into small chunks. In fact, new releases of PG are being developed in exactly this way.",
"Joshua Brown, Adam Mikeal, and Alysha Clark provided substantial feedback that greatly enhanced the value and clarity of the information in this article. Abstract Two of the largest Massive Open Online Course (MOOC) organizations have chosen different methods for the way they will score and provide feedback on essays students submit. EdX, MIT and Harvard’s non-profit MOOC federation, recently announced that they will use a machine-based Automated Essay Scoring (AES) application to assess written work in their MOOCs. Coursera, a Stanford startup for MOOCs, has been skeptical of AES applications and therefore has held that it will use some form of human-based “calibrated peer review” to score and provide feedback on student writing. This essay reviews the relevant literature on AES and UCLA’s Calibrated Peer ReviewTM (CPR) product at a high level, outlines the capabilities and limitations of both AES and CPR, and provides a table and framework for comparing these forms of assessment of student writing in MOOCs. Stephen Balfour is an instructional associate professor of psychology and the Director of Information Technology for the College of Liberal Arts at Texas A&M University.",
"Complex, interactive, and far-reaching change ushered in by globalisation processes compels educators, scholars, and policymakers to consider a pedagogy of the future. No other institution than public education is in the position to take on the \"reeducation of humankind\" that Paul Kennedy professed is needed to face the continuous change taking place worldwide. A pedagogy of the future—one that addresses a social reality in flux and an ongoing redefinition of political, moral, and social foundations—is one fundamental way to respond to such an immense undertaking. The pedagogy I propose in this article aims at nothing less than creating a new consciousness, beginning with preparing socially conscious, multidimensional citizens of the world. Using a sociocultural perspective on learning and development, I lay out a few key pedagogical goals that show great promise for facilitating the necessary shift from an industrial model of preparing individuals for a hierarchical, routinised, and predicable workplace ...",
""
]
} |
1703.06169 | 2604725943 | Massive Open Online Courses (MOOCs) use peer assessment to grade open ended questions at scale, allowing students to provide feedback. Relative to teacher based grading, peer assessment on MOOCs traditionally delivers lower quality feedback and fewer learner interactions. We present the identified peer review (IPR) framework, which provides non-blind peer assessment and incentives driving high quality feedback. We show that, compared to traditional peer assessment methods, IPR leads to significantly longer and more useful feedback as well as more discussion between peers. | However, with the potential benefits, peer grading still faces challenges. Providing accurate grading where the performance of a novice is being judged by other novices is problematic. Students mistrust peer grades and anonymity lead students to be unscrupulous. Reviews are often short and not well considered, which is especially problematic when students work hard on an assignment and receive comments they can't learn from @cite_19 . | {
"cite_N": [
"@cite_19"
],
"mid": [
"1520074575"
],
"abstract": [
"The teach-learn-assess cycle in education is broken in a typical massive open online course (MOOC). Without formative assessment and feedback, MOOCs amount to information dump or broadcasting shows, not educational experiences. A number of remedies have been attempted to bring formative assessment back into MOOCs, each with its own limits and problems. The most widely applicable approach for all MOOCs to date is to use peer assessment to provide the necessary feedback. However, unmoderated peer assessment results suffer from a lack of credibility. Several methods are available today to improve on the accuracy of peer assessment results. Some combination of these methods may be necessary to make peer assessment results sufficiently accurate to be useful for formative assessment. Such results can also help to facilitate peer learning, online discussion forums, and may possibly augment summative evaluation for credentialing."
]
} |
1703.06169 | 2604725943 | Massive Open Online Courses (MOOCs) use peer assessment to grade open ended questions at scale, allowing students to provide feedback. Relative to teacher based grading, peer assessment on MOOCs traditionally delivers lower quality feedback and fewer learner interactions. We present the identified peer review (IPR) framework, which provides non-blind peer assessment and incentives driving high quality feedback. We show that, compared to traditional peer assessment methods, IPR leads to significantly longer and more useful feedback as well as more discussion between peers. | Feedback improves performance by changing students' locus of attention, focusing them on productive aspects of their work @cite_20 @cite_11 . However, not all peers provide great feedback and some leave limited comments with no coherent message for improvement, or they are rogue reviews @cite_1 . Rogue reviews are insufficient reviews caused by laziness, collusion, dishonesty, ret aliation, competition, or malevolence @cite_23 . To improve on this, PeerStudio @cite_4 peer assessment system is designed to encourage more feedback comments by showing short tips for writing comments just below the comment box. For example, if a response has no constructive feedback, it may remind students with phrases like: ''Quick check: Is your feedback actionable?'' by triggering heuristics word count on feedback @cite_4 . Students see such comments as more useful than rubrics in reviewing @cite_22 . Similar techniques are used to improve the quality of product reviews online @cite_24 . Our framework uses a simple interface design reflecting these lessons by integrating four pointed questions with separate response areas. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_1",
"@cite_24",
"@cite_23",
"@cite_20",
"@cite_11"
],
"mid": [
"2038924341",
"2168105487",
"2116119142",
"2087294982",
"1607629406",
"2030441548",
""
],
"abstract": [
"Rapid feedback is a core component of mastery learning, but feedback on open-ended work requires days or weeks in most classes today. This paper introduces PeerStudio, an assessment platform that leverages the large number of students' peers in online classes to enable rapid feedback on in-progress work. Students submit their draft, give rubric-based feedback on two peers' drafts, and then receive peer feedback. Students can integrate the feedback and repeat this process as often as they desire. In MOOC deployments, the median student received feedback in just twenty minutes. Rapid feedback on in-progress work improves course outcomes: in a controlled experiment, students' final grades improved when feedback was delivered quickly, but not if delayed by 24 hours. More than 3,600 students have used PeerStudio in eight classes, both massive and in-person. This research demonstrates how large classes can leverage their scale to encourage mastery through rapid feedback and revision.",
"When students work with peers, they learn more actively, build richer knowledge structures, and connect material to their lives. However, not every peer learning experience online sees successful adoption. This paper articulates and addresses three adoption challenges for global-scale peer learning. First, peer interactions struggle to bootstrap critical mass. However, class incentives can signal importance and spur initial usage. Second, online classes have limited peer visibility and awareness, so students often feel alone even when surrounded by peers. We find that highlighting interdependence and strengthening norms can mitigate this issue. Third, teachers can readily access \"big\" aggregate data but not \"thick\" contextual data that helps build intuitions, so software should guide teachers' scaffolding of peer interactions. We illustrate these challenges through studying 8,500 students' usage of two peer learning platforms, Talkabout and PeerStudio. This paper measures efficacy through sign-up and participation rates and the structure and duration of student interactions.",
"Scientific peer review, open source software development, wikis, and other domains use distributed review to improve quality of created content by providing feedback to the work's creator. Distributed review is used to assess or improve the quality of a work (e.g., an article). However, it can also provide learning benefits to the participants in the review process. We developed an online review system for beginning computer programming students; it gathers multiple anonymous peer reviews to give students feedback on their programming work. We deployed the system in an introductory programming class and evaluated it in a controlled study. We find that: peer reviews are accurate compared to an accepted evaluation standard, that students prefer reviews from other students with less experience than themselves, and that participating in a peer review process results in better learning outcomes.",
"User-supplied reviews are widely and increasingly used to enhance e-commerce and other websites. Because reviews can be numerous and varying in quality, it is important to assess how helpful each review is. While review helpfulness is currently assessed manually, in this paper we consider the task of automatically assessing it. Experiments using SVM regression on a variety of features over Amazon.com product reviews show promising results, with rank correlations of up to 0.66. We found that the most useful features include the length of the review, its unigrams, and its product rating.",
"With demonstrated benefits to higher level learning, peer review in the classroom has been well researched and popular since at least the 1990s. However, little or no prior studies exist into the peer review process for online courses. Further, we found no prior research specifically addressing the operational aspects of online peer review. This research addresses that gap by comparing the issues involved in managing peer review for an online course with those for a traditional classroom course. In an exploratory case study, two sections of the same introductory level course were taught by the same professor in the same academic term, one section in the traditional classroom and one as an online section. Both sections covered the same material in the same order. Online students had access to narrated PowerPoint recordings that tracked in-class lectures. The same assignments and exams were used. The two sections used a joint discussion board for posting questions and answers about the course material. In short, the two courses were almost identical, except for the steps necessary to make peer review operate in an online environment.",
"",
""
]
} |
1703.05693 | 2605212061 | This paper proposes the SVDNet for retrieval problems, with focus on the application of person re-identification (re-ID). We view each weight vector within a fully connected (FC) layer in a convolutional neuron network (CNN) as a projection basis. It is observed that the weight vectors are usually highly correlated. This problem leads to correlations among entries of the FC descriptor, and compromises the retrieval performance based on the Euclidean distance. To address the problem, this paper proposes to optimize the deep representation learning process with Singular Vector Decomposition (SVD). Specifically, with the restraint and relaxation iteration (RRI) training scheme, we are able to iteratively integrate the orthogonality constraint in CNN training, yielding the so-called SVDNet. We conduct experiments on the Market-1501, CUHK03, and Duke datasets, and show that RRI effectively reduces the correlation among the projection vectors, produces more discriminative FC descriptors, and significantly improves the re-ID accuracy. On the Market-1501 dataset, for instance, rank-1 accuracy is improved from 55.3 to 80.5 for CaffeNet, and from 73.8 to 82.3 for ResNet-50. | The second type of CNN-based re-ID methods focuses on feature learning, which categorizes the training samples into pre-defined classes and the FC descriptor is used for retrieval @cite_4 @cite_34 @cite_15 . In @cite_4 @cite_35 , the classification CNN model is fine-tuned using either the video frames or image bounding boxes to learn a discriminative embedding for pedestrian retrieval. Xiao al @cite_15 propose learning generic feature representations from multiple re-ID datasets jointly. To deal with spatial misalignment, Zheng al @cite_27 propose the PoseBox structure similar to the pictorial structure @cite_5 to learn pose invariant embeddings. To take advantage of both the feature learning and similarity learning, Zheng al @cite_29 and Geng al @cite_33 combine the contrastive loss and the identification loss to improve the discriminative ability of the learned feature embedding, following the success in face verification @cite_8 . This paper adopts the classification mode, which is shown to produce competitive accuracy without losing efficiency potentials. | {
"cite_N": [
"@cite_35",
"@cite_4",
"@cite_33",
"@cite_8",
"@cite_29",
"@cite_27",
"@cite_5",
"@cite_15",
"@cite_34"
],
"mid": [
"",
"2531440880",
"2953350812",
"",
"2549957142",
"2583528645",
"",
"2342611082",
""
],
"abstract": [
"",
"Person re-identification (re-ID) has become increasingly popular in the community due to its application and research significance. It aims at spotting a person of interest in other cameras. In the early days, hand-crafted algorithms and small-scale evaluation were predominantly reported. Recent years have witnessed the emergence of large-scale datasets and deep learning systems which make use of large data volumes. Considering different tasks, we classify most current re-ID methods into two classes, i.e., image-based and video-based; in both tasks, hand-crafted and deep learning systems will be reviewed. Moreover, two new re-ID tasks which are much closer to real-world applications are described and discussed, i.e., end-to-end re-ID and fast re-ID in very large galleries. This paper: 1) introduces the history of person re-ID and its relationship with image classification and instance retrieval; 2) surveys a broad selection of the hand-crafted systems and the large-scale methods in both image- and video-based re-ID; 3) describes critical future directions in end-to-end re-ID and fast retrieval in large galleries; and 4) finally briefs some important yet under-developed issues.",
"Person re-identification (Re-ID) poses a unique challenge to deep learning: how to learn a deep model with millions of parameters on a small training set of few or no labels. In this paper, a number of deep transfer learning models are proposed to address the data sparsity problem. First, a deep network architecture is designed which differs from existing deep Re-ID models in that (a) it is more suitable for transferring representations learned from large image classification datasets, and (b) classification loss and verification loss are combined, each of which adopts a different dropout strategy. Second, a two-stepped fine-tuning strategy is developed to transfer knowledge from auxiliary datasets. Third, given an unlabelled Re-ID dataset, a novel unsupervised deep transfer learning model is developed based on co-training. The proposed models outperform the state-of-the-art deep Re-ID models by large margins: we achieve Rank-1 accuracy of 85.4 , 83.7 and 56.3 on CUHK03, Market1501, and VIPeR respectively, whilst on VIPeR, our unsupervised model (45.1 ) beats most supervised models.",
"",
"In this article, we revisit two popular convolutional neural networks in person re-identification (re-ID): verification and identification models. The two models have their respective advantages and limitations due to different loss functions. Here, we shed light on how to combine the two models to learn more discriminative pedestrian descriptors. Specifically, we propose a Siamese network that simultaneously computes the identification loss and verification loss. Given a pair of training images, the network predicts the identities of the two input images and whether they belong to the same identity. Our network learns a discriminative embedding and a similarity measurement at the same time, thus taking full usage of the re-ID annotations. Our method can be easily applied on different pretrained networks. Albeit simple, the learned embedding improves the state-of-the-art performance on two public person re-ID benchmarks. Further, we show that our architecture can also be applied to image retrieval. The code is available at https: github.com layumi 2016_person_re-ID.",
"Pedestrian misalignment, which mainly arises from detector errors and pose variations, is a critical problem for a robust person re-identification (re-ID) system. With bad alignment, the background noise will significantly compromise the feature learning and matching process. To address this problem, this paper introduces the pose invariant embedding (PIE) as a pedestrian descriptor. First, in order to align pedestrians to a standard pose, the PoseBox structure is introduced, which is generated through pose estimation followed by affine transformations. Second, to reduce the impact of pose estimation errors and information loss during PoseBox construction, we design a PoseBox fusion (PBF) CNN architecture that takes the original image, the PoseBox, and the pose estimation confidence as input. The proposed PIE descriptor is thus defined as the fully connected layer of the PBF network for the retrieval task. Experiments are conducted on the Market-1501, CUHK03, and VIPeR datasets. We show that PoseBox alone yields decent re-ID accuracy and that when integrated in the PBF network, the learned PIE descriptor produces competitive performance compared with the state-of-the-art approaches.",
"",
"Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.",
""
]
} |
1703.05693 | 2605212061 | This paper proposes the SVDNet for retrieval problems, with focus on the application of person re-identification (re-ID). We view each weight vector within a fully connected (FC) layer in a convolutional neuron network (CNN) as a projection basis. It is observed that the weight vectors are usually highly correlated. This problem leads to correlations among entries of the FC descriptor, and compromises the retrieval performance based on the Euclidean distance. To address the problem, this paper proposes to optimize the deep representation learning process with Singular Vector Decomposition (SVD). Specifically, with the restraint and relaxation iteration (RRI) training scheme, we are able to iteratively integrate the orthogonality constraint in CNN training, yielding the so-called SVDNet. We conduct experiments on the Market-1501, CUHK03, and Duke datasets, and show that RRI effectively reduces the correlation among the projection vectors, produces more discriminative FC descriptors, and significantly improves the re-ID accuracy. On the Market-1501 dataset, for instance, rank-1 accuracy is improved from 55.3 to 80.5 for CaffeNet, and from 73.8 to 82.3 for ResNet-50. | Truncated SVD @cite_18 @cite_36 is widely used for CNN model compression. SVDNet departs from it in two aspects. First, truncated SVD decomposes the weight matrix in FC layers and reconstructs it with several dominant singular vectors and values. SVDNet does not reconstruct the weight matrix but replaces it with an orthogonal matrix, which is the product of the left unitary matrix and the singular value matrix. Second, Truncated SVD reduces the model size and testing time at the cost of acceptable precision loss, while SVDNet significantly improves the retrieval accuracy without impact on the model size. | {
"cite_N": [
"@cite_36",
"@cite_18"
],
"mid": [
"2294543795",
"2167215970"
],
"abstract": [
"Recently proposed deep neural network (DNN) obtains significant accuracy improvements in many large vocabulary continuous speech recognition (LVCSR) tasks. However, DNN requires much more parameters than traditional systems, which brings huge cost during online evaluation, and also limits the application of DNN in a lot of scenarios. In this paper we present our new effort on DNN aiming at reducing the model size while keeping the accuracy improvements. We apply singular value decomposition (SVD) on the weight matrices in DNN, and then restructure the model based on the inherent sparseness of the original matrices. After restructuring we can reduce the DNN model size significantly with negligible accuracy loss. We also fine-tune the restructured model using the regular back-propagation method to get the accuracy back when reducing the DNN model size heavily. The proposed method has been evaluated on two LVCSR tasks, with context-dependent DNN hidden Markov model (CD-DNN-HMM). Experimental results show that the proposed approach dramatically reduces the DNN model size by more than 80 without losing any accuracy. Index Terms: deep neural network, singular value decomposition, model restructuring",
"We present techniques for speeding up the test-time evaluation of large convolutional networks, designed for object recognition tasks. These models deliver impressive accuracy, but each image evaluation requires millions of floating point operations, making their deployment on smartphones and Internet-scale clusters problematic. The computation is dominated by the convolution operations in the lower layers of the model. We exploit the redundancy present within the convolutional filters to derive approximations that significantly reduce the required computation. Using large state-of-the-art models, we demonstrate speedups of convolutional layers on both CPU and GPU by a factor of 2 x, while keeping the accuracy within 1 of the original model."
]
} |
1703.05693 | 2605212061 | This paper proposes the SVDNet for retrieval problems, with focus on the application of person re-identification (re-ID). We view each weight vector within a fully connected (FC) layer in a convolutional neuron network (CNN) as a projection basis. It is observed that the weight vectors are usually highly correlated. This problem leads to correlations among entries of the FC descriptor, and compromises the retrieval performance based on the Euclidean distance. To address the problem, this paper proposes to optimize the deep representation learning process with Singular Vector Decomposition (SVD). Specifically, with the restraint and relaxation iteration (RRI) training scheme, we are able to iteratively integrate the orthogonality constraint in CNN training, yielding the so-called SVDNet. We conduct experiments on the Market-1501, CUHK03, and Duke datasets, and show that RRI effectively reduces the correlation among the projection vectors, produces more discriminative FC descriptors, and significantly improves the re-ID accuracy. On the Market-1501 dataset, for instance, rank-1 accuracy is improved from 55.3 to 80.5 for CaffeNet, and from 73.8 to 82.3 for ResNet-50. | We note a concurrent work @cite_10 which also aims to orthogonalize the CNN filters, yet our work is different from @cite_10 . In @cite_10 , the regularization effect of orthogonalization benefits the back-propagation of very deep networks, thus improving the classification accuracy. The regularization proposed in @cite_10 may not directly benefit the embedding learning process. But in this paper, orthogonalization is used to generate decorrelated descriptors suitable for retrieval. Our network may not be suitable for improving classification. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2593110912"
],
"abstract": [
"Deep neural network is difficult to train and this predicament becomes worse as the depth increases. The essence of this problem exists in the magnitude of backpropagated errors that will result in gradient vanishing or exploding phenomenon. We show that a variant of regularizer which utilizes orthonormality among different filter banks can alleviate this problem. Moreover, we design a backward error modulation mechanism based on the quasi-isometry assumption between two consecutive parametric layers. Equipped with these two ingredients, we propose several novel optimization solutions that can be utilized for training a specific-structured (repetitively triple modules of Conv-BNReLU) extremely deep convolutional neural network (CNN) WITHOUT any shortcuts identity mappings from scratch. Experiments show that our proposed solutions can achieve distinct improvements for a 44-layer and a 110-layer plain networks on both the CIFAR-10 and ImageNet datasets. Moreover, we can successfully train plain CNNs to match the performance of the residual counterparts. Besides, we propose new principles for designing network structure from the insights evoked by orthonormality. Combined with residual structure, we achieve comparative performance on the ImageNet dataset."
]
} |
1703.05465 | 2949737549 | This paper describes a neural-network model which performed competitively (top 6) at the SemEval 2017 cross-lingual Semantic Textual Similarity (STS) task. Our system employs an attention-based recurrent neural network model that optimizes the sentence similarity. In this paper, we describe our participation in the multilingual STS task which measures similarity across English, Spanish, and Arabic. | Most proposed approaches in the past adopted a hybrid of varying text unit sizes ranging from character-based, token-based, to knowledge-based similarity measure @cite_26 . The linguistic depths of these measures often vary between lexical, syntactic, and semantic levels. | {
"cite_N": [
"@cite_26"
],
"mid": [
"2171313960"
],
"abstract": [
"ABSTRACT Measuring the similarity between words, sentences, paragraphs and documents is an important component in various tasks such as information retrieval, document clustering, word-sense disambiguation, automatic essay scoring, short answer grading, machine translation and text summarization. This survey discusses the existing works on text similarity through partitioning them into three approaches; String-based, Corpus-based and Knowledge-based similarities. Furthermore, samples of combination between these similarities are presented. General Terms Text Mining, Natural Language Processing. Keywords BasedText Similarity, Semantic Similarity, String-Based Similarity, Corpus-Based Similarity, Knowledge-Based Similarity. NeedlemanWunsch 1. INTRODUCTION Text similarity measures play an increasingly important role in text related research and applications in tasks Nsuch as information retrieval, text classification, document clustering, topic detection, topic tracking, questions generation, question answering, essay scoring, short answer scoring, machine translation, text summarization and others. Finding similarity between words is a fundamental part of text similarity which is then used as a primary stage for sentence, paragraph and document similarities. Words can be similar in two ways lexically and semantically. Words are similar lexically if they have a similar character sequence. Words are similar semantically if they have the same thing, are opposite of each other, used in the same way, used in the same context and one is a type of another. DistanceLexical similarity is introduced in this survey though different String-Based algorithms, Semantic similarity is introduced through Corpus-Based and Knowledge-Based algorithms. String-Based measures operate on string sequences and character composition. A string metric is a metric that measures similarity or dissimilarity (distance) between two text strings for approximate string matching or comparison. Corpus-Based similarity is a semantic similarity measure that determines the similarity between words according to information gained from large corpora. Knowledge-Based similarity is a semantic similarity measure that determines the degree of similarity between words using information derived from semantic networks. The most popular for each type will be presented briefly. This paper is organized as follows: Section two presents String-Based algorithms by partitioning them into two types character-based and term-based measures. Sections three and four introduce Corpus-Based and knowledge-Based algorithms respectively. Samples of combinations between similarity algorithms are introduced in section five and finally section six presents conclusion of the survey."
]
} |
1703.05446 | 2950107101 | Human parsing has recently attracted a lot of research interests due to its huge application potentials. However existing datasets have limited number of images and annotations, and lack the variety of human appearances and the coverage of challenging cases in unconstrained environment. In this paper, we introduce a new benchmark "Look into Person (LIP)" that makes a significant advance in terms of scalability, diversity and difficulty, a contribution that we feel is crucial for future developments in human-centric analysis. This comprehensive dataset contains over 50,000 elaborately annotated images with 19 semantic part labels, which are captured from a wider range of viewpoints, occlusions and background complexity. Given these rich annotations we perform detailed analyses of the leading human parsing approaches, gaining insights into the success and failures of these methods. Furthermore, in contrast to the existing efforts on improving the feature discriminative capability, we solve human parsing by exploring a novel self-supervised structure-sensitive learning approach, which imposes human pose structures into parsing results without resorting to extra supervision (i.e., no need for specifically labeling human joints in model training). Our self-supervised learning framework can be injected into any advanced neural networks to help incorporate rich high-level knowledge regarding human joints from a global perspective and improve the parsing results. Extensive evaluations on our LIP and the public PASCAL-Person-Part dataset demonstrate the superiority of our method. | The commonly used publicly available datasets for human parsing are summarized in Table. . The previous datasets were labeled with limited number of images or categories. The largest one @cite_21 so far only contains 17,000 fashion images with mostly upright fashion models. Containing 50,462 images annotated with 20 categories, our LIP dataset is the largest and most comprehensive human parsing dataset to date. Some other datasets in the vision community were dedicated to the tasks of clothes recognition, retrieval @cite_4 @cite_27 and human pose estimation @cite_13 @cite_22 , while our LIP dataset only focuses on human parsing. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_21",
"@cite_27",
"@cite_13"
],
"mid": [
"2471768434",
"2101032778",
"2204578866",
"2094311777",
"2080873731"
],
"abstract": [
"Recent advances in clothes recognition have been driven by the construction of clothes datasets. Existing datasets are limited in the amount of annotations and are difficult to cope with the various challenges in real-world applications. In this work, we introduce DeepFashion1, a large-scale clothes dataset with comprehensive annotations. It contains over 800,000 images, which are richly annotated with massive attributes, clothing landmarks, and correspondence of images taken under different scenarios including store, street snapshot, and consumer. Such rich annotations enable the development of powerful algorithms in clothes recognition and facilitating future researches. To demonstrate the advantages of DeepFashion, we propose a new deep model, namely FashionNet, which learns clothing features by jointly predicting clothing attributes and landmarks. The estimated landmarks are then employed to pool or gate the learned features. It is optimized in an iterative manner. Extensive experiments demonstrate the effectiveness of FashionNet and the usefulness of DeepFashion.",
"We introduce a new dataset, Human3.6M, of 3.6 Million accurate 3D Human poses, acquired by recording the performance of 5 female and 6 male subjects, under 4 different viewpoints, for training realistic human sensing systems and for evaluating the next generation of human pose estimation models and algorithms. Besides increasing the size of the datasets in the current state-of-the-art by several orders of magnitude, we also aim to complement such datasets with a diverse set of motions and poses encountered as part of typical human activities (taking photos, talking on the phone, posing, greeting, eating, etc.), with additional synchronized image, human motion capture, and time of flight (depth) data, and with accurate 3D body scans of all the subject actors involved. We also provide controlled mixed reality evaluation scenarios where 3D human models are animated using motion capture and inserted using correct 3D geometry, in complex real environments, viewed with moving cameras, and under occlusion. Finally, we provide a set of large-scale statistical models and detailed evaluation baselines for the dataset illustrating its diversity and the scope for improvement by future work in the research community. Our experiments show that our best large-scale model can leverage our full training set to obtain a 20 improvement in performance compared to a training set of the scale of the largest existing public dataset for this problem. Yet the potential for improvement by leveraging higher capacity, more complex models with our large dataset, is substantially vaster and should stimulate future research. The dataset together with code for the associated large-scale learning models, features, visualization tools, as well as the evaluation server, is available online at http: vision.imar.ro human3.6m .",
"In this work, we address the human parsing task with a novel Contextualized Convolutional Neural Network (Co-CNN) architecture, which well integrates the cross-layer context, global image-level context, within-super-pixel context and cross-super-pixel neighborhood context into a unified network. Given an input human image, Co-CNN produces the pixel-wise categorization in an end-to-end way. First, the cross-layer context is captured by our basic local-to-global-to-local structure, which hierarchically combines the global semantic structure and the local fine details within the cross-layers. Second, the global image-level label prediction is used as an auxiliary objective in the intermediate layer of the Co-CNN, and its outputs are further used for guiding the feature learning in subsequent convolutional layers to leverage the global image-level context. Finally, to further utilize the local super-pixel contexts, the within-super-pixel smoothing and cross-super-pixel neighbourhood voting are formulated as natural sub-components of the Co-CNN to achieve the local label consistency in both training and testing process. Comprehensive evaluations on two public datasets well demonstrate the significant superiority of our Co-CNN architecture over other state-of-the-arts for human parsing. In particular, the F-1 score on the large dataset [15] reaches 76.95 by Co-CNN, significantly higher than 62.81 and 64.38 by the state-of-the-art algorithms, M-CNN [21] and ATR [15], respectively.",
"Clothing is one of the most informative cues of human appearance. In this paper, we propose a novel multi-person clothing segmentation algorithm for highly occluded images. The key idea is combining blocking models to address the person-wise occlusions. In contrary to the traditional layered model that tries to solve the full layer ranking problem, the proposed blocking model partitions the problem into a series of pair-wise ones and then determines the local blocking relationship based on individual and contextual information. Thus, it is capable of dealing with cases with a large number of people. Additionally, we propose a layout model formulated as Markov Network which incorporates the blocking relationship to pursue an approximately optimal clothing layout for group people. Experiments demonstrated on a group images dataset show the effectiveness of our algorithm.",
"Human pose estimation has made significant progress during the last years. However current datasets are limited in their coverage of the overall pose estimation challenges. Still these serve as the common sources to evaluate, train and compare different models on. In this paper we introduce a novel benchmark \"MPII Human Pose\" that makes a significant advance in terms of diversity and difficulty, a contribution that we feel is required for future developments in human body models. This comprehensive dataset was collected using an established taxonomy of over 800 human activities [1]. The collected images cover a wider variety of human activities than previous datasets including various recreational, occupational and householding activities, and capture people from a wider range of viewpoints. We provide a rich set of labels including positions of body joints, full 3D torso and head orientation, occlusion labels for joints and body parts, and activity labels. For each image we provide adjacent video frames to facilitate the use of motion information. Given these rich annotations we perform a detailed analysis of leading human pose estimation approaches and gaining insights for the success and failures of these methods."
]
} |
1703.05446 | 2950107101 | Human parsing has recently attracted a lot of research interests due to its huge application potentials. However existing datasets have limited number of images and annotations, and lack the variety of human appearances and the coverage of challenging cases in unconstrained environment. In this paper, we introduce a new benchmark "Look into Person (LIP)" that makes a significant advance in terms of scalability, diversity and difficulty, a contribution that we feel is crucial for future developments in human-centric analysis. This comprehensive dataset contains over 50,000 elaborately annotated images with 19 semantic part labels, which are captured from a wider range of viewpoints, occlusions and background complexity. Given these rich annotations we perform detailed analyses of the leading human parsing approaches, gaining insights into the success and failures of these methods. Furthermore, in contrast to the existing efforts on improving the feature discriminative capability, we solve human parsing by exploring a novel self-supervised structure-sensitive learning approach, which imposes human pose structures into parsing results without resorting to extra supervision (i.e., no need for specifically labeling human joints in model training). Our self-supervised learning framework can be injected into any advanced neural networks to help incorporate rich high-level knowledge regarding human joints from a global perspective and improve the parsing results. Extensive evaluations on our LIP and the public PASCAL-Person-Part dataset demonstrate the superiority of our method. | Recently many research efforts have been devoted to human parsing @cite_21 @cite_20 @cite_0 @cite_5 @cite_15 @cite_32 @cite_14 . For example, Liang al @cite_21 proposed a novel Co-CNN architecture which integrates multiple levels of image contexts into a unified nerwork. Besides human parsing, there has also been increasing research interest on the part segmentation of other objects such as animals or cars @cite_2 @cite_25 @cite_3 . To capture the rich structure information based on the advanced CNN architecture, common solutions inlcude the combination of CNNs and CRFs @cite_1 @cite_26 and the adoptions of multi-scale feature representations @cite_1 @cite_14 @cite_32 . Chen al @cite_14 proposed an attention mechanism that learns to weight the multi-scale features at each pixel location. Some previous works @cite_33 @cite_6 explored human pose information to guide human parsing by generating pose-guided'' part segment proposals. To leverage human joint structure more effortlessly and efficiently, the focus in our approach is nevertheless a new self-supervised structure-sensitive learning approach, which actually can be embedded in any networks. | {
"cite_N": [
"@cite_14",
"@cite_26",
"@cite_33",
"@cite_21",
"@cite_1",
"@cite_32",
"@cite_3",
"@cite_6",
"@cite_0",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_25",
"@cite_20"
],
"mid": [
"2158865742",
"",
"",
"2204578866",
"1923697677",
"2236679687",
"2020206277",
"2514518917",
"2121339428",
"1903370114",
"1195044660",
"",
"792160549",
"2074621908"
],
"abstract": [
"Incorporating multi-scale features in fully convolutional neural networks (FCNs) has been a key element to achieving state-of-the-art performance on semantic image segmentation. One common way to extract multi-scale features is to feed multiple resized input images to a shared deep network and then merge the resulting features for pixelwise classification. In this work, we propose an attention mechanism that learns to softly weight the multi-scale features at each pixel location. We adapt a state-of-the-art semantic image segmentation model, which we jointly train with multi-scale input images and the attention model. The proposed attention model not only outperforms average- and max-pooling, but allows us to diagnostically visualize the importance of features at different positions and scales. Moreover, we show that adding extra supervision to the output at each scale is essential to achieving excellent performance when merging multi-scale features. We demonstrate the effectiveness of our model with extensive experiments on three challenging datasets, including PASCAL-Person-Part, PASCAL VOC 2012 and a subset of MS-COCO 2014.",
"",
"",
"In this work, we address the human parsing task with a novel Contextualized Convolutional Neural Network (Co-CNN) architecture, which well integrates the cross-layer context, global image-level context, within-super-pixel context and cross-super-pixel neighborhood context into a unified network. Given an input human image, Co-CNN produces the pixel-wise categorization in an end-to-end way. First, the cross-layer context is captured by our basic local-to-global-to-local structure, which hierarchically combines the global semantic structure and the local fine details within the cross-layers. Second, the global image-level label prediction is used as an auxiliary objective in the intermediate layer of the Co-CNN, and its outputs are further used for guiding the feature learning in subsequent convolutional layers to leverage the global image-level context. Finally, to further utilize the local super-pixel contexts, the within-super-pixel smoothing and cross-super-pixel neighbourhood voting are formulated as natural sub-components of the Co-CNN to achieve the local label consistency in both training and testing process. Comprehensive evaluations on two public datasets well demonstrate the significant superiority of our Co-CNN architecture over other state-of-the-arts for human parsing. In particular, the F-1 score on the large dataset [15] reaches 76.95 by Co-CNN, significantly higher than 62.81 and 64.38 by the state-of-the-art algorithms, M-CNN [21] and ATR [15], respectively.",
"Deep Convolutional Neural Networks (DCNNs) have recently shown state of the art performance in high level vision tasks, such as image classification and object detection. This work brings together methods from DCNNs and probabilistic graphical models for addressing the task of pixel-level classification (also called \"semantic image segmentation\"). We show that responses at the final layer of DCNNs are not sufficiently localized for accurate object segmentation. This is due to the very invariance properties that make DCNNs good for high level tasks. We overcome this poor localization property of deep networks by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF). Qualitatively, our \"DeepLab\" system is able to localize segment boundaries at a level of accuracy which is beyond previous methods. Quantitatively, our method sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 71.6 IOU accuracy in the test set. We show how these results can be obtained efficiently: Careful network re-purposing and a novel application of the 'hole' algorithm from the wavelet community allow dense computation of neural net responses at 8 frames per second on a modern GPU.",
"Parsing human regions into semantic parts, e.g., body, head and arms etc., from a random natural image is challenging while fundamental for computer vision and widely applicable in industry. One major difficulty to handle such a problem is the high flexibility of scale and location of a human instance and its corresponding parts, making the parsing task either lack of boundary details or suffer from local confusions. To tackle such problems, in this work, we propose the \"Auto-Zoom Net\" (AZN) for human part parsing, which is a unified fully convolutional neural network structure that: (1) parses each human instance into detailed parts. (2) predicts the locations and scales of human instances and their corresponding parts. In our unified network, the two tasks are mutually beneficial. The score maps obtained for parsing help estimate the locations and scales for human instances and their parts. With the predicted locations and scales, our model \"zooms\" the region into a right scale to further refine the parsing. In practice, we perform the two tasks iteratively so that detailed human parts are gradually recovered. We conduct extensive experiments over the challenging PASCAL-Person-Part segmentation, and show our approach significantly outperforms the state-of-art parsing techniques especially for instances and parts at small scale. In addition, we perform experiments for horse and cow segmentation and also obtain results which are considerably better than state-of-the-art methods (by over 5 )., which is contribued by the proposed iterative zooming process.",
"This paper addresses the problem of semantic part parsing (segmentation) of cars, i.e.assigning every pixel within the car to one of the parts (e.g.body, window, lights, license plates and wheels). We formulate this as a landmark identification problem, where a set of landmarks specifies the boundaries of the parts. A novel mixture of graphical models is proposed, which dynamically couples the landmarks to a hierarchy of segments. When modeling pairwise relation between landmarks, this coupling enables our model to exploit the local image contents in addition to spatial deformation, an aspect that most existing graphical models ignore. In particular, our model enforces appearance consistency between segments within the same part. Parsing the car, including finding the optimal coupling between landmarks and segments in the hierarchy, is performed by dynamic programming. We evaluate our method on a subset of PASCAL VOC 2010 car images and on the car subset of 3D Object Category dataset (CAR3D). We show good results and, in particular, quantify the effectiveness of using the segment appearance consistency in terms of accuracy of part localization and segmentation.",
"Parsing human into semantic parts is crucial to human-centric analysis. In this paper, we propose a human parsing pipeline that uses pose cues, i.e., estimates of human joint locations, to provide pose-guided segment proposals for semantic parts. These segment proposals are ranked using standard appearance cues, deep-learned semantic feature, and a novel pose feature called pose-context. Then these proposals are selected and assembled using an And-Or graph to output a parse of the person. The And-Or graph is able to deal with large human appearance variability due to pose, choice of clothes, etc. We evaluate our approach on the popular Penn-Fudan pedestrian parsing dataset, showing that it significantly outperforms the state-of-the-arts, and perform diagnostics to demonstrate the effectiveness of different stages of our pipeline.",
"Clothing recognition is an extremely challenging problem due to wide variation in clothing item appearance, layering, and style. In this paper, we tackle the clothing parsing problem using a retrieval based approach. For a query image, we find similar styles from a large database of tagged fashion images and use these examples to parse the query. Our approach combines parsing from: pre-trained global clothing models, local clothing models learned on the fly from retrieved examples, and transferred parse masks (paper doll item transfer) from retrieved examples. Experimental evaluation shows that our approach significantly outperforms state of the art in parsing accuracy.",
"In this paper, we study the problem of semantic part segmentation for animals. This is more challenging than standard object detection, object segmentation and pose estimation tasks because semantic parts of animals often have similar appearance and highly varying shapes. To tackle these challenges, we build a mixture of compositional models to represent the object boundary and the boundaries of semantic parts. And we incorporate edge, appearance, and semantic part cues into the compositional model. Given part-level segmentation annotation, we develop a novel algorithm to learn a mixture of compositional models under various poses and viewpoints for certain animal classes. Furthermore, a linear complexity algorithm is offered for efficient inference of the compositional model using dynamic programming. We evaluate our method for horse and cow using a newly annotated dataset on Pascal VOC 2010 which has pixelwise part labels. Experimental results demonstrate the effectiveness of our method.",
"In this paper we tackle the problem of clothing parsing: Our goal is to segment and classify different garments a person is wearing. We frame the problem as the one of inference in a pose-aware Conditional Random Field (CRF) which exploits appearance, figure ground segmentation, shape and location priors for each garment as well as similarities between segments, and symmetries between different human body parts. We demonstrate the effectiveness of our approach on the Fashionista dataset [1] and show that we can obtain a significant improvement over the state-of-the-art.",
"",
"Segmenting semantic objects from images and parsing them into their respective semantic parts are fundamental steps towards detailed object understanding in computer vision. In this paper, we propose a joint solution that tackles semantic object and part segmentation simultaneously, in which higher object-level context is provided to guide part segmentation, and more detailed part-level localization is utilized to refine object segmentation. Specifically, we first introduce the concept of semantic compositional parts (SCP) in which similar semantic parts are grouped and shared among different objects. A two-channel fully convolutional network (FCN) is then trained to provide the SCP and object potentials at each pixel. At the same time, a compact set of segments can also be obtained from the SCP predictions of the network. Given the potentials and the generated segments, in order to explore long-range context, we finally construct an efficient fully connected conditional random field (FCRF) to jointly predict the final object and part labels. Extensive evaluation on three different datasets shows that our approach can mutually enhance the performance of object and part segmentation, and outperforms the current state-of-the-art on both tasks.",
"In this paper we demonstrate an effective method for parsing clothing in fashion photographs, an extremely challenging problem due to the large number of possible garment items, variations in configuration, garment appearance, layering, and occlusion. In addition, we provide a large novel dataset and tools for labeling garment items, to enable future research on clothing estimation. Finally, we present intriguing initial results on using clothing estimates to improve pose identification, and demonstrate a prototype application for pose-independent visual garment retrieval."
]
} |
1703.05423 | 2951357606 | End-to-end design of dialogue systems has recently become a popular research topic thanks to powerful tools such as encoder-decoder architectures for sequence-to-sequence learning. Yet, most current approaches cast human-machine dialogue management as a supervised learning problem, aiming at predicting the next utterance of a participant given the full history of the dialogue. This vision is too simplistic to render the intrinsic planning problem inherent to dialogue as well as its grounded nature, making the context of a dialogue larger than the sole history. This is why only chit-chat and question answering tasks have been addressed so far using end-to-end architectures. In this paper, we introduce a Deep Reinforcement Learning method to optimize visually grounded task-oriented dialogues, based on the policy gradient algorithm. This approach is tested on a dataset of 120k dialogues collected through Mechanical Turk and provides encouraging results at solving both the problem of generating natural dialogues and the task of discovering a specific object in a complex picture. | Outside of the dialogue literature, RL methods have been applied to encoder-decoder architectures in machine translation @cite_11 @cite_26 and image captioning @cite_2 . In those scenarios, the BLEU score is used as a reward signal to fine-tune a network trained with a cross-entropy loss. However, the BLEU score is a surrogate for human evaluation of naturalness, so directly optimizing this measure does not guarantee improvement in the translation captioning quality. In contrast, our reward function encodes task completion, and optimizing this metric is exactly what we aim for. Finally, the BLEU score can only be used in a batch setting because it requires the ground-truth labels from the dataset. In , the computed reward is independent from the generated dialogue. | {
"cite_N": [
"@cite_26",
"@cite_2",
"@cite_11"
],
"mid": [
"2487501366",
"",
"2176263492"
],
"abstract": [
"We present an approach to training neural networks to generate sequences using actor-critic methods from reinforcement learning (RL). Current log-likelihood training methods are limited by the discrepancy between their training and testing modes, as models must generate tokens conditioned on their previous guesses rather than the ground-truth tokens. We address this problem by introducing a network that is trained to predict the value of an output token, given the policy of an network. This results in a training procedure that is much closer to the test phase, and allows us to directly optimize for a task-specific score such as BLEU. Crucially, since we leverage these techniques in the supervised learning setting rather than the traditional RL setting, we condition the critic network on the ground-truth output. We show that our method leads to improved performance on both a synthetic task, and for German-English machine translation. Our analysis paves the way for such methods to be applied in natural language generation tasks, such as machine translation, caption generation, and dialogue modelling.",
"",
"Many natural language processing applications use language models to generate text. These models are typically trained to predict the next word in a sequence, given the previous words and some context such as an image. However, at test time the model is expected to generate the entire sequence from scratch. This discrepancy makes generation brittle, as errors may accumulate along the way. We address this issue by proposing a novel sequence level training algorithm that directly optimizes the metric used at test time, such as BLEU or ROUGE. On three different tasks, our approach outperforms several strong baselines for greedy generation. The method is also competitive when these baselines employ beam search, while being several times faster."
]
} |
1703.04826 | 2951545716 | Semantic role labeling (SRL) is the task of identifying the predicate-argument structure of a sentence. It is typically regarded as an important step in the standard NLP pipeline. As the semantic representations are closely related to syntactic ones, we exploit syntactic information in our model. We propose a version of graph convolutional networks (GCNs), a recent class of neural networks operating on graphs, suited to model syntactic dependency graphs. GCNs over syntactic dependency trees are used as sentence encoders, producing latent feature representations of words in a sentence. We observe that GCN layers are complementary to LSTM ones: when we stack both GCN and LSTM layers, we obtain a substantial improvement over an already state-of-the-art LSTM SRL model, resulting in the best reported scores on the standard benchmark (CoNLL-2009) both for Chinese and English. | Perhaps the earliest methods modeling syntax-semantics interface with RNNs are due to @cite_18 @cite_34 @cite_4 , they used shift-reduce parsers for joint SRL and syntactic parsing, and relied on RNNs to model statistical dependencies across syntactic and semantic parsing actions. A more modern (e.g., based on LSTMs) and effective reincarnation of this line of research has been proposed in . Other recent work which considered incorporation of syntactic information in neural SRL models include: who use standard syntactic features within an MLP calculating potentials of a CRF model; who enriched standard features for SRL with LSTM representations of syntactic paths between arguments and predicates; who relied on low-rank tensor factorizations for modeling syntax. Also used (non-graph) convolutional networks and provided syntactic features as input. A very different line of research, but with similar goals to ours (i.e. integrating syntax with minimal feature engineering), used tree kernels @cite_28 . | {
"cite_N": [
"@cite_28",
"@cite_18",
"@cite_34",
"@cite_4"
],
"mid": [
"1987486061",
"2144784110",
"2099162170",
"2100439229"
],
"abstract": [
"The availability of large scale data sets of manually annotated predicate-argument structures has recently favored the use of machine learning approaches to the design of automated semantic role labeling (SRL) systems. The main research in this area relates to the design choices for feature representation and for effective decompositions of the task in different learning models. Regarding the former choice, structural properties of full syntactic parses are largely employed as they represent ways to encode different principles suggested by the linking theory between syntax and semantics. The latter choice relates to several learning schemes over global views of the parses. For example, re-ranking stages operating over alternative predicate-argument sequences of the same sentence have shown to be very effective. In this article, we propose several kernel functions to model parse tree properties in kernel-based machines, for example, perceptrons or support vector machines. In particular, we define different kinds of tree kernels as general approaches to feature engineering in SRL. Moreover, we extensively experiment with such kernels to investigate their contribution to individual stages of an SRL architecture both in isolation and in combination with other traditional manually coded features. The results for boundary recognition, classification, and re-ranking stages provide systematic evidence about the significant impact of tree kernels on the overall accuracy, especially when the amount of training data is small. As a conclusive result, tree kernels allow for a general and easily portable feature engineering method which is applicable to a large family of natural language processing tasks.",
"We propose a solution to the challenge of the CoNLL 2008 shared task that uses a generative history-based latent variable model to predict the most likely derivation of a synchronous dependency parser for both syntactic and semantic dependencies. The submitted model yields 79.1 macro-average F1 performance, for the joint task, 86.9 syntactic dependencies LAS and 71.0 semantic dependencies F1. A larger model trained after the deadline achieves 80.5 macro-average F1, 87.6 syntactic dependencies LAS, and 73.1 semantic dependencies F1.",
"This demonstration presents a high-performance syntactic and semantic dependency parser. The system consists of a pipeline of modules that carry out the to-kenization, lemmatization, part-of-speech tagging, dependency parsing, and semantic role labeling of a sentence. The system's two main components draw on improved versions of a state-of-the-art dependency parser (Bohnet, 2009) and semantic role labeler (, 2009) developed independently by the authors. The system takes a sentence as input and produces a syntactic and semantic annotation using the CoNLL 2009 format. The processing time needed for a sentence typically ranges from 10 to 1000 milliseconds. The predicate--argument structures in the final output are visualized in the form of segments, which are more intuitive for a user.",
"Motivated by the large number of languages (seven) and the short development time (two months) of the 2009 CoNLL shared task, we exploited latent variables to avoid the costly process of hand-crafted feature engineering, allowing the latent variables to induce features from the data. We took a pre-existing generative latent variable model of joint syntactic-semantic dependency parsing, developed for English, and applied it to six new languages with minimal adjustments. The parser's robustness across languages indicates that this parser has a very general feature set. The parser's high performance indicates that its latent variables succeeded in inducing effective features. This system was ranked third overall with a macro averaged F1 score of 82.14 , only 0.5 worse than the best system."
]
} |
1703.04826 | 2951545716 | Semantic role labeling (SRL) is the task of identifying the predicate-argument structure of a sentence. It is typically regarded as an important step in the standard NLP pipeline. As the semantic representations are closely related to syntactic ones, we exploit syntactic information in our model. We propose a version of graph convolutional networks (GCNs), a recent class of neural networks operating on graphs, suited to model syntactic dependency graphs. GCNs over syntactic dependency trees are used as sentence encoders, producing latent feature representations of words in a sentence. We observe that GCN layers are complementary to LSTM ones: when we stack both GCN and LSTM layers, we obtain a substantial improvement over an already state-of-the-art LSTM SRL model, resulting in the best reported scores on the standard benchmark (CoNLL-2009) both for Chinese and English. | Beyond SRL, there have been many proposals on how to incorporate syntactic information in RNN models, for example, in the context of neural machine translation @cite_30 @cite_29 . One of the most popular and attractive approaches is to use tree-structured recursive neural networks @cite_19 @cite_5 @cite_35 , including stacking them on top of a sequential BiLSTM @cite_26 . An approach of to sentiment analysis and question classification, introduced even before GCNs became popular in the machine learning community, is related to graph convolution. However, it is inherently single-layer and tree-specific, uses bottom-up computations, does not share parameters across syntactic functions and does not use gates. Gates have been previously used in GCNs @cite_27 but between GCN layers rather than for individual edges. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_26",
"@cite_29",
"@cite_19",
"@cite_27",
"@cite_5"
],
"mid": [
"2951883919",
"2949952998",
"2229639163",
"2410082850",
"2251939518",
"2244807774",
"2251628756"
],
"abstract": [
"There has been relatively little attention to incorporating linguistic prior to neural machine translation. Much of the previous work was further constrained to considering linguistic prior on the source side. In this paper, we propose a hybrid model, called NMT+RNNG, that learns to parse and translate by combining the recurrent neural network grammar into the attention-based neural machine translation. Our approach encourages the neural machine translation model to incorporate linguistic prior during training, and lets it translate on its own afterward. Extensive experiments with four language pairs show the effectiveness of the proposed NMT+RNNG.",
"We propose a technique for learning representations of parser states in transition-based dependency parsers. Our primary innovation is a new control structure for sequence-to-sequence neural networks---the stack LSTM. Like the conventional stack data structures used in transition-based parsing, elements can be pushed to or popped from the top of the stack in constant time, but, in addition, an LSTM maintains a continuous space embedding of the stack contents. This lets us formulate an efficient parsing model that captures three facets of a parser's state: (i) unbounded look-ahead into the buffer of incoming words, (ii) the complete history of actions taken by the parser, and (iii) the complete contents of the stack of partially built tree fragments, including their internal structures. Standard backpropagation techniques are used for training and yield state-of-the-art parsing performance.",
"We present a novel end-to-end neural model to extract entities and relations between them. Our recurrent neural network based model captures both word sequence and dependency tree substructure information by stacking bidirectional tree-structured LSTM-RNNs on bidirectional sequential LSTM-RNNs. This allows our model to jointly represent both entities and relations with shared parameters in a single model. We further encourage detection of entities during training and use of entity information in relation extraction via entity pretraining and scheduled sampling. Our model improves over the state-of-the-art feature-based model on end-to-end relation extraction, achieving 12.1 and 5.7 relative error reductions in F1-score on ACE2005 and ACE2004, respectively. We also show that our LSTM-RNN based model compares favorably to the state-of-the-art CNN based model (in F1-score) on nominal relation classification (SemEval-2010 Task 8). Finally, we present an extensive ablation analysis of several model components.",
"Neural machine translation has recently achieved impressive results, while using little in the way of external linguistic information. In this paper we show that the strong learning capability of neural MT models does not make linguistic features redundant; they can be easily incorporated to provide further improvements in performance. We generalize the embedding layer of the encoder in the attentional encoder--decoder architecture to support the inclusion of arbitrary features, in addition to the baseline word feature. We add morphological features, part-of-speech tags, and syntactic dependency labels as input features to English German, and English->Romanian neural machine translation systems. In experiments on WMT16 training and test sets, we find that linguistic input features improve model quality according to three metrics: perplexity, BLEU and CHRF3. An open-source implementation of our neural MT system is available, as are sample files and configurations.",
"Semantic word spaces have been very useful but cannot express the meaning of longer phrases in a principled way. Further progress towards understanding compositionality in tasks such as sentiment detection requires richer supervised training and evaluation resources and more powerful models of composition. To remedy this, we introduce a Sentiment Treebank. It includes fine grained sentiment labels for 215,154 phrases in the parse trees of 11,855 sentences and presents new challenges for sentiment compositionality. To address them, we introduce the Recursive Neural Tensor Network. When trained on the new treebank, this model outperforms all previous methods on several metrics. It pushes the state of the art in single sentence positive negative classification from 80 up to 85.4 . The accuracy of predicting fine-grained sentiment labels for all phrases reaches 80.7 , an improvement of 9.7 over bag of features baselines. Lastly, it is the only model that can accurately capture the effects of negation and its scope at various tree levels for both positive and negative phrases.",
"Graph-structured data appears frequently in domains including chemistry, natural language semantics, social networks, and knowledge bases. In this work, we study feature learning techniques for graph-structured inputs. Our starting point is previous work on Graph Neural Networks (, 2009), which we modify to use gated recurrent units and modern optimization techniques and then extend to output sequences. The result is a flexible and broadly useful class of neural network models that has favorable inductive biases relative to purely sequence-based models (e.g., LSTMs) when the problem is graph-structured. We demonstrate the capabilities on some simple AI (bAbI) and graph algorithm learning tasks. We then show it achieves state-of-the-art performance on a problem from program verification, in which subgraphs need to be matched to abstract data structures.",
"We propose the first implementation of an infinite-order generative dependency model. The model is based on a new recursive neural network architecture, the Inside-Outside Recursive Neural Network. This architecture allows information to flow not only bottom-up, as in traditional recursive neural networks, but also topdown. This is achieved by computing content as well as context representations for any constituent, and letting these representations interact. Experimental results on the English section of the Universal Dependency Treebank show that the infinite-order model achieves a perplexity seven times lower than the traditional third-order model using counting, and tends to choose more accurate parses in k-best lists. In addition, reranking with this model achieves state-of-the-art unlabelled attachment scores and unlabelled exact match scores."
]
} |
1703.05129 | 2601774179 | Vertex Descent is a local search algorithm which forms the basis of a wide spectrum of tabu search, simulated annealing and hybrid evolutionary algorithms for graph colouring. These algorithms are usually treated as experimental and provide strong results on established benchmarks. As a step towards studying these heuristics analytically, an analysis of the behaviour of Vertex Descent is provided. It is shown that Vertex Descent is able to find feasible colourings for several types of instances in expected polynomial time. This includes 2-colouring of paths and 3-colouring of graphs with maximum degree 3. The same also holds for 3-colouring of a subset of 3-colourable graphs with maximum degree 4. As a consequence, Vertex Descent finds a 3-colouring in expected polynomial time for the smallest graph for which Br 'elaz's heuristic DSATUR needs 4 colours. On the other hand, Vertex Descent may fail for forests with maximum degree 3 with high probability. | Most of the modern graph colouring heuristics are . Currently, the most successful algorithms include a QACol @cite_20 @cite_12 , an algorithm IE @math Col, based on of independent sets @cite_35 and a HEAD with a very small population of two colourings @cite_5 . All three of these algorithms use Vertex Descent combined with several different ideas. These include solution populations, tabu lists, thermal and quantum fluctuations, partition crossovers and preprocessing. However, Vertex Descent remains at the core of all these approaches. | {
"cite_N": [
"@cite_35",
"@cite_5",
"@cite_12",
"@cite_20"
],
"mid": [
"2064850189",
"2121391146",
"2104695995",
"2003291739"
],
"abstract": [
"Graph coloring is one of the most studied combinatorial optimization problems. This paper presents an improved extraction and expansion method (IE^2COL), initially introduced in Wu and Hao (2012) [44]. IE^2COL employs a forward independent set extraction strategy to reduce the initial graph G. From the reduced graph, IE^2COL triggers a backward coloring process which uses extracted independent sets as new color classes for intermediate subgraph coloring. The proposed method is assessed on 20 large benchmark graphs with 900 to 4000 vertices. Computational results show that it provides new upper bounds for 6 graphs and consistently matches the current best-known results for 12 other graphs.",
"This paper presents an effective memetic approach (HEAD ) designed for coloring difficult graphs. In this algorithm a powerful tabu search is used inside a very specific population of individuals. Indeed, the main characteristic of (HEAD ) is to work with a population of only two individuals. This provides a very simple algorithm with neither selection operator nor replacement strategy. Because of its simplicity, (HEAD ) allows an easy way for managing the diversity. We focus this work on the impact of this diversity management on well-studied graphs of the DIMACS challenge benchmarks, known to be very difficult to solve. A detailed analysis is provided for three graphs on which (HEAD ) finds a legal coloring with less colors than reference algorithms: DSJC500.5 with 47 colors, DSJC1000.5 with 82 colors and flat1000_76_0 with 81 colors. The analysis performed in this work will allow to improve (HEAD ) efficiency in terms of computation time and maybe to decrease the number of needed colors for other graphs.",
"Quantum annealing is a combinatorial optimization technique inspired by quantum mechanics. Here we show that a spin model for the k-coloring of large dense random graphs can be field tuned so that its acceptance ratio diverges during Monte Carlo quantum annealing, until a ground state is reached. We also find that simulations exhibiting such a diverging acceptance ratio are generally more effective than those tuned to the more conventional pattern of a declining and or stagnating acceptance ratio. This observation facilitates the discovery of solutions to several well-known benchmark k-coloring instances, some of which have been open for almost two decades.",
"Quantum annealing extends simulated annealing by introducing artificial quantum fluctuations. The path-integral Monte Carlo version chosen is population-based and designed to be implemented on a classical computer. Its first application to the graph coloring problem is presented in this paper. It is shown by experiments that quantum annealing can outperform classical thermal simulated annealing for this particular problem. Moreover, quantum annealing proved competitive when compared with the best algorithms on most of the difficult instances from the DIMACS benchmarks. The quantum annealing algorithm has even found that the well-known benchmark graph dsjc1000.9 has a chromatic number of at most 222. This is an improvement on its best upper-bound from a large body of literature."
]
} |
1703.05129 | 2601774179 | Vertex Descent is a local search algorithm which forms the basis of a wide spectrum of tabu search, simulated annealing and hybrid evolutionary algorithms for graph colouring. These algorithms are usually treated as experimental and provide strong results on established benchmarks. As a step towards studying these heuristics analytically, an analysis of the behaviour of Vertex Descent is provided. It is shown that Vertex Descent is able to find feasible colourings for several types of instances in expected polynomial time. This includes 2-colouring of paths and 3-colouring of graphs with maximum degree 3. The same also holds for 3-colouring of a subset of 3-colourable graphs with maximum degree 4. As a consequence, Vertex Descent finds a 3-colouring in expected polynomial time for the smallest graph for which Br 'elaz's heuristic DSATUR needs 4 colours. On the other hand, Vertex Descent may fail for forests with maximum degree 3 with high probability. | One of the most well-known hybrid evolutionary algorithms for GCP was introduced by Galinier and Hao @cite_25 . It used the tabu search algorithm TabuCol @cite_41 as an intensification subroutine. TabuCol practically represents the Vertex Descent algorithm enhanced by a tabu list to prevent it from cycling. Glass and Pr " u gel-Bennett investigated an adaptation of the hybrid algorithm by Galinier and Hao obtained by substituting TabuCol with Vertex Descent @cite_27 . They concluded that this version can perform comparably to the original variant but requires a larger population. Vertex Descent has also been used as a technique of search space sampling in an evaluation of different objective functions for the problem @cite_17 . | {
"cite_N": [
"@cite_41",
"@cite_27",
"@cite_25",
"@cite_17"
],
"mid": [
"2168000780",
"1606171447",
"1582334898",
"1598002709"
],
"abstract": [
"Tabu search techniques are used for moving step by step towards the minimum value of a function. A tabu list of forbidden movements is updated during the iterations to avoid cycling and being trapped in local minima. Such techniques are adapted to graph coloring problems. We show that they provide almost optimal colorings of graphs having up to 1000 nodes and their efficiency is shown to be significantly superior to the famous simulated annealing.",
"This paper examines the best current algorithm for solving the Chromatic Number Problem, due to Galinier and Hao (Journal of Combinatorial Optimization,1999, 3(4), pp 379-397). The algorithm combines a Genetic Algorithm with Tabu Search. We show that the algorithm remains powerful even if the Tabu Search component is eliminated, and explore the reasons for its success where other Genetic Algorithms have failed. In addition we propose a generalized algorithm for the Frequency Assignment Problem.",
"A recent and very promising approach for combinatorial optimization is to embed local search into the framework of evolutionary algorithms. In this paper, we present such hybrid algorithms for the graph coloring problem. These algorithms combine a new class of highly specialized crossover operators and a well-known tabu search algorithm. Experiments of such a hybrid algorithm are carried out on large DIMACS Challenge benchmark graphs. Results prove very competitive with and even better than those of state-of-the-art algorithms. Analysis of the behavior of the algorithm sheds light on ways to further improvement.",
"The evaluation or fitness function is a key component ofany heuristic search algorithm. This paper introduces a new evaluationfunction for the well-known graph K-coloring problem. This functiontakes into account not only the number of conflicting vertices, but alsoinherent information related to the structure of the graph. To assessthe effectiveness of this new evaluation function, we carry out a numberof experiments using a set of DIMACS benchmark graphs. Based onstatistic data obtained with a parameter free steepest descent, we showan improvement of the new evaluation function over the classical one."
]
} |
1703.05129 | 2601774179 | Vertex Descent is a local search algorithm which forms the basis of a wide spectrum of tabu search, simulated annealing and hybrid evolutionary algorithms for graph colouring. These algorithms are usually treated as experimental and provide strong results on established benchmarks. As a step towards studying these heuristics analytically, an analysis of the behaviour of Vertex Descent is provided. It is shown that Vertex Descent is able to find feasible colourings for several types of instances in expected polynomial time. This includes 2-colouring of paths and 3-colouring of graphs with maximum degree 3. The same also holds for 3-colouring of a subset of 3-colourable graphs with maximum degree 4. As a consequence, Vertex Descent finds a 3-colouring in expected polynomial time for the smallest graph for which Br 'elaz's heuristic DSATUR needs 4 colours. On the other hand, Vertex Descent may fail for forests with maximum degree 3 with high probability. | Numerous analytical results have also been obtained for other combinatorial optimisation problems. For the maximum matching problem, behaviours of local search and evolutionary algorithms for paths and worst-case approximations were investigated @cite_15 . For the vertex cover problem, behaviour of hybrid algorithms was analysed for specific types of graphs @cite_37 and an iterated local search algorithm was investigated in the context of sparse random graphs @cite_51 . Fixed-parameter evolutionary algorithms were also analysed @cite_14 . Other studied problems include makespan scheduling @cite_32 , the Euclidean travelling salesperson problem @cite_2 , the Eulerian cycle problem @cite_22 , the minimum spanning tree problem @cite_45 or the minimum cut problem @cite_43 . | {
"cite_N": [
"@cite_37",
"@cite_14",
"@cite_22",
"@cite_32",
"@cite_43",
"@cite_45",
"@cite_2",
"@cite_15",
"@cite_51"
],
"mid": [
"2157816158",
"2088415500",
"1781730958",
"1510977876",
"1978984733",
"2153401898",
"2086885650",
"2565562138",
"2052468331"
],
"abstract": [
"Hybrid methods are very popular for solving problems from combinatorial optimization. In contrast, the theoretical understanding of the interplay of different optimization methods is rare. In this paper, we make a first step into the rigorous analysis of such combinations for combinatorial optimization problems. The subject of our analyses is the vertex cover problem for which several approximation algorithms have been proposed. We point out specific instances where solutions can (or cannot) be improved by the search process of a simple evolutionary algorithm in expected polynomial time.",
"In this paper, we consider multi-objective evolutionary algorithms for the Vertex Cover problem in the context of parameterized complexity. We consider two different measures for the problem. The first measure is a very natural multi-objective one for the use of evolutionary algorithms and takes into account the number of chosen vertices and the number of edges that remain uncovered. The second fitness function is based on a linear programming formulation and proves to give better results. We point out that both approaches lead to a kernelization for the Vertex Cover problem. Based on this, we show that evolutionary algorithms solve the vertex cover problem efficiently if the size of a minimum vertex cover is not too large, i.e., the expected runtime is bounded by O(f(OPT)źnc), where c is a constant and f a function that only depends on OPT. This shows that evolutionary algorithms are randomized fixed-parameter tractable algorithms for the vertex cover problem.",
"Evolutionary algorithms are randomized search heuristics, which are applied to problems whose structure is not well understood, as well as to problems in combinatorial optimization. They have successfully been applied to different kinds of arc routing problems. To start the analysis of evolutionary algorithms with respect to the expected optimization time on these problems, we consider the Eulerian cycle problem. We show that a variant of the well-known (1+1) EA working on the important encoding of permutations is able to find an Eulerian tour of an Eulerian graph in expected polynomial time. Altering the operator used for mutation in the considered algorithms, the expected optimization time changes from polynomial to exponential.",
"In recent years, probabilistic analyses of algorithms have received increasing attention. Despite results on the average-case complexity and smoothed complexity of exact deterministic algorithms, little is known about the average-case behavior of randomized search heuristics (RSHs). In this paper, two simple RSHs are studied on a simple scheduling problem. While it turns out that in the worst case, both RSHs need exponential time to create solutions being significantly better than 4 3-approximate, an average-case analysis for two input distributions reveals that one RSH is convergent to optimality in polynomial time. Moreover, it is shown that for both RSHs, parallel runs yield a PRAS.",
"We study the minimum s-t-cut problem in graphs with costs on the edges in the context of evolutionary algorithms. Minimum cut problems belong to the class of basic network optimization problems that occur as crucial subproblems in many real-world optimization problems and have a variety of applications in several different areas. We prove that there exist instances of the minimum s-t-cut problem that cannot be solved by standard single-objective evolutionary algorithms in reasonable time. On the other hand, we develop a bi-criteria approach based on the famous maximum-flow minimum-cut theorem that enables evolutionary algorithms to find an optimal solution in expected polynomial time.",
"Randomized search heuristics, among them randomized local search and evolutionary algorithms, are applied to problems whose structure is not well understood, as well as to problems in combinatorial optimization. The analysis of these randomized search heuristics has been started for some well-known problems, and this approach is followed here for the minimum spanning tree problem. After motivating this line of research, it is shown that randomized search heuristics find minimum spanning trees in expected polynomial time without employing the global technique of greedy algorithms.",
"Parameterized runtime analysis seeks to understand the influence of problem structure on algorithmic runtime. In this paper, we contribute to the theoretical understanding of evolutionary algorithms and carry out a parameterized analysis of evolutionary algorithms for the Euclidean traveling salesperson problem (Euclidean TSP). We investigate the structural properties in TSP instances that influence the optimization process of evolutionary algorithms and use this information to bound the runtime of simple evolutionary algorithms. Our analysis studies the runtime in dependence of the number of inner points @math and shows that @math evolutionary algorithms solve the Euclidean TSP in expected time @math where @math is a function of the minimum angle @math between any three points. Finally, our analysis provides insights into designing a mutation operator that improves the upper bound on expected runtime. We show that a mixed mutation strategy that incorporates both 2-opt moves and permutation jumps results in an upper bound of @math for the @math EA.",
"Randomized search heuristics like evolutionary algorithms are mostly applied to problems whose structure is not completely known but also to combinatorial optimization problems. Practitioners report surprising successes but almost no results with theoretically well-founded analyses exist. Such an analysis is started in this paper for a fundamental evolutionary algorithm and the well-known maximum matching problem. It is proven that the evolutionary algorithm is a polynomial-time randomized approximation scheme (PRAS) for this optimization problem, although the algorithm does not employ the idea of augmenting paths. Moreover, for very simple graphs it is proved that the expected optimization time of the algorithm is polynomially bounded and bipartite graphs are constructed where this time grows exponentially.",
"Recently, various randomized search heuristics have been studied for the solution of the minimum vertex cover problem, in particular for sparse random instances according to the G(n,c n) model, where c>0 is a constant. Methods from statistical physics suggest that the problem is easy if c"
]
} |
1703.04977 | 2950517871 | There are two major types of uncertainty one can model. Aleatoric uncertainty captures noise inherent in the observations. On the other hand, epistemic uncertainty accounts for uncertainty in the model -- uncertainty which can be explained away given enough data. Traditionally it has been difficult to model epistemic uncertainty in computer vision, but with new Bayesian deep learning tools this is now possible. We study the benefits of modeling epistemic vs. aleatoric uncertainty in Bayesian deep learning models for vision tasks. For this we present a Bayesian deep learning framework combining input-dependent aleatoric uncertainty together with epistemic uncertainty. We study models under the framework with per-pixel semantic segmentation and depth regression tasks. Further, our explicit uncertainty formulation leads to new loss functions for these tasks, which can be interpreted as learned attenuation. This makes the loss more robust to noisy data, also giving new state-of-the-art results on segmentation and depth regression benchmarks. | Existing approaches to Bayesian deep learning capture either epistemic uncertainty alone, or aleatoric uncertainty alone @cite_13 . These uncertainties are formalised as probability distributions over either the model parameters, or model outputs, respectively. Epistemic uncertainty is modeled by placing a prior distribution over a model's weights, and then trying to capture how much these weights vary given some data. Aleatoric uncertainty on the other hand is modeled by placing a distribution over the output of the model. For example, in regression our outputs might be modeled as corrupted with Gaussian random noise. In this case we are interested in learning the noise's variance as a function of different inputs (such noise can also be modeled with a constant value for all data points, but this is of less practical interest). These uncertainties, in the context of Bayesian deep learning, are explained in more detail in this section. | {
"cite_N": [
"@cite_13"
],
"mid": [
"2951266961"
],
"abstract": [
"We introduce a new, efficient, principled and backpropagation-compatible algorithm for learning a probability distribution on the weights of a neural network, called Bayes by Backprop. It regularises the weights by minimising a compression cost, known as the variational free energy or the expected lower bound on the marginal likelihood. We show that this principled kind of regularisation yields comparable performance to dropout on MNIST classification. We then demonstrate how the learnt uncertainty in the weights can be used to improve generalisation in non-linear regression problems, and how this weight uncertainty can be used to drive the exploration-exploitation trade-off in reinforcement learning."
]
} |
1703.04908 | 2602275733 | By capturing statistical patterns in large corpora, machine learning has enabled significant advances in natural language processing, including in machine translation, question answering, and sentiment analysis. However, for agents to intelligently interact with humans, simply capturing the statistical patterns is insufficient. In this paper we investigate if, and how, grounded compositional language can emerge as a means to achieve goals in multi-agent populations. Towards this end, we propose a multi-agent learning environment and learning methods that bring about emergence of a basic compositional language. This language is represented as streams of abstract discrete symbols uttered by agents over time, but nonetheless has a coherent structure that possesses a defined vocabulary and syntax. We also observe emergence of non-verbal communication such as pointing and guiding when language communication is unavailable. | Recent years have seen substantial progress in practical natural language applications such as machine translation @cite_12 @cite_18 , sentiment analysis @cite_0 , document summarization @cite_20 , and domain-specific dialogue @cite_23 . Much of this success is a result of intelligently designed statistical models trained on large static datasets. However, such approaches do not produce an understanding of language that can lead to productive cooperation with humans. | {
"cite_N": [
"@cite_18",
"@cite_0",
"@cite_23",
"@cite_12",
"@cite_20"
],
"mid": [
"2133564696",
"2251939518",
"2513380446",
"2949888546",
"2335367650"
],
"abstract": [
"Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.",
"Semantic word spaces have been very useful but cannot express the meaning of longer phrases in a principled way. Further progress towards understanding compositionality in tasks such as sentiment detection requires richer supervised training and evaluation resources and more powerful models of composition. To remedy this, we introduce a Sentiment Treebank. It includes fine grained sentiment labels for 215,154 phrases in the parse trees of 11,855 sentences and presents new challenges for sentiment compositionality. To address them, we introduce the Recursive Neural Tensor Network. When trained on the new treebank, this model outperforms all previous methods on several metrics. It pushes the state of the art in single sentence positive negative classification from 80 up to 85.4 . The accuracy of predicting fine-grained sentiment labels for all phrases reaches 80.7 , an improvement of 9.7 over bag of features baselines. Lastly, it is the only model that can accurately capture the effects of negation and its scope at various tree levels for both positive and negative phrases.",
"This paper proposes KB-InfoBot -- a multi-turn dialogue agent which helps users search Knowledge Bases (KBs) without composing complicated queries. Such goal-oriented dialogue agents typically need to interact with an external database to access real-world knowledge. Previous systems achieved this by issuing a symbolic query to the KB to retrieve entries based on their attributes. However, such symbolic operations break the differentiability of the system and prevent end-to-end training of neural dialogue agents. In this paper, we address this limitation by replacing symbolic queries with an induced \"soft\" posterior distribution over the KB that indicates which entities the user is interested in. Integrating the soft retrieval process with a reinforcement learner leads to higher task success rate and reward in both simulations and against real users. We also present a fully neural end-to-end agent, trained entirely from user feedback, and discuss its application towards personalized dialogue agents. The source code is available at this https URL",
"Deep Neural Networks (DNNs) are powerful models that have achieved excellent performance on difficult learning tasks. Although DNNs work well whenever large labeled training sets are available, they cannot be used to map sequences to sequences. In this paper, we present a general end-to-end approach to sequence learning that makes minimal assumptions on the sequence structure. Our method uses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to a vector of a fixed dimensionality, and then another deep LSTM to decode the target sequence from the vector. Our main result is that on an English to French translation task from the WMT'14 dataset, the translations produced by the LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's BLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did not have difficulty on long sentences. For comparison, a phrase-based SMT system achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM to rerank the 1000 hypotheses produced by the aforementioned SMT system, its BLEU score increases to 36.5, which is close to the previous best result on this task. The LSTM also learned sensible phrase and sentence representations that are sensitive to word order and are relatively invariant to the active and the passive voice. Finally, we found that reversing the order of the words in all source sentences (but not target sentences) improved the LSTM's performance markedly, because doing so introduced many short term dependencies between the source and the target sentence which made the optimization problem easier.",
"We present a discriminative model for single-document summarization that integrally combines compression and anaphoricity constraints. Our model selects textual units to include in the summary based on a rich set of sparse features whose weights are learned on a large corpus. We allow for the deletion of content within a sentence when that deletion is licensed by compression rules; in our framework, these are implemented as dependencies between subsentential units of text. Anaphoricity constraints then improve cross-sentence coherence by guaranteeing that, for each pronoun included in the summary, the pronoun's antecedent is included as well or the pronoun is rewritten as a full mention. When trained end-to-end, our final system outperforms prior work on both ROUGE as well as on human judgments of linguistic quality."
]
} |
1703.04908 | 2602275733 | By capturing statistical patterns in large corpora, machine learning has enabled significant advances in natural language processing, including in machine translation, question answering, and sentiment analysis. However, for agents to intelligently interact with humans, simply capturing the statistical patterns is insufficient. In this paper we investigate if, and how, grounded compositional language can emerge as a means to achieve goals in multi-agent populations. Towards this end, we propose a multi-agent learning environment and learning methods that bring about emergence of a basic compositional language. This language is represented as streams of abstract discrete symbols uttered by agents over time, but nonetheless has a coherent structure that possesses a defined vocabulary and syntax. We also observe emergence of non-verbal communication such as pointing and guiding when language communication is unavailable. | Aside from producing agents that can interact with humans through language, research in pragmatic language understanding can be informative to the fields of linguistics and cognitive science. Of particular interest in these fields has been the question of how syntax and compositional structure in language emerged, and why it is largely unique to human languages @cite_34 @cite_31 @cite_25 . Models such as Rational Speech Acts @cite_22 and Iterated Learning @cite_9 have been popular in cognitive science and evolutionary linguistics, but such approaches tend to rely on pre-specified procedures or models that limit their generality. | {
"cite_N": [
"@cite_22",
"@cite_9",
"@cite_31",
"@cite_34",
"@cite_25"
],
"mid": [
"1993979041",
"",
"1967449835",
"1523934301",
"27281439"
],
"abstract": [
"One of the most astonishing features of human language is its capacity to convey information efficiently in context. Many theories provide informal accounts of communicative inference, yet there have been few successes in making precise, quantitative predictions about pragmatic reasoning. We examined judgments about simple referential communication games, modeling behavior in these games by assuming that speakers attempt to be informative and that listeners use Bayesian inference to recover speakers’ intended referents. Our model provides a close, parameter-free fit to human judgments, suggesting that the use of information-theoretic tools to predict pragmatic reasoning may lead to more effective formal models of communication.",
"",
"Animal communication is typically non-syntactic, which means that signals refer to whole situations1,2,3,4,5,6,7. Human language is syntactic, and signals consist of discrete components that have their own meaning8. Syntax is a prerequisite for taking advantage of combinatorics, that is, “making infinite use of finite means”9,10,11. The vast expressive power of human language would be impossible without syntax, and the transition from non-syntactic to syntactic communication was an essential step in the evolution of human language12,13,14,15,16. We aim to understand the evolutionary dynamics of this transition and to analyse how natural selection can guide it. Here we present a model for the population dynamics of language evolution, define the basic reproductive ratio of words and calculate the maximum size of a lexicon. Syntax allows larger repertoires and the possibility to formulate messages that have not been learned beforehand. Nevertheless, according to our model natural selection can only favour the emergence of syntax if the number of required signals exceeds a threshold value. This result might explain why only humans evolved syntactic communication and hence complex language.",
"A new approach to the origins of syntax in human language is presented. Using computational models of populations of learners, it is shown that compositional, recursive mappings are inevitable end-states of a cultural process of linguistic transmission. This is true even if the starting state is no language at all. It is argued that the way that knowledge of language is transmitted through a learning bottleneck profoundly influences its emergent structure. This approach provides a radical alternative to one in which the structure of language is viewed as an innate, biological adaptation to communicative pressures.",
"Proceeding of Second International Symposium on the Emergence and Evolution of Linguistic Communication (EELC'05), celebrado en Hatfield (UK) del 12 al 15 de abril de 2005.-- All Content of the AISB Web Pages by the Society for the Study of Artificial Intelligence and Simulation of Behaviour (unless stated otherwise on the page or in the relevant document) are licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 2.0 UK: England & Wales License."
]
} |
1703.04908 | 2602275733 | By capturing statistical patterns in large corpora, machine learning has enabled significant advances in natural language processing, including in machine translation, question answering, and sentiment analysis. However, for agents to intelligently interact with humans, simply capturing the statistical patterns is insufficient. In this paper we investigate if, and how, grounded compositional language can emerge as a means to achieve goals in multi-agent populations. Towards this end, we propose a multi-agent learning environment and learning methods that bring about emergence of a basic compositional language. This language is represented as streams of abstract discrete symbols uttered by agents over time, but nonetheless has a coherent structure that possesses a defined vocabulary and syntax. We also observe emergence of non-verbal communication such as pointing and guiding when language communication is unavailable. | The recent work that is most similar to ours is the application of reinforcement learning approaches towards the purposes of learning a communication protocol, as exemplified by @cite_17 @cite_10 @cite_33 @cite_13 . | {
"cite_N": [
"@cite_13",
"@cite_10",
"@cite_33",
"@cite_17"
],
"mid": [
"2950472486",
"2395575420",
"2402402867",
""
],
"abstract": [
"The current mainstream approach to train natural language systems is to expose them to large amounts of text. This passive learning is problematic if we are interested in developing interactive machines, such as conversational agents. We propose a framework for language learning that relies on multi-agent communication. We study this learning in the context of referential games. In these games, a sender and a receiver see a pair of images. The sender is told one of them is the target and is allowed to send a message from a fixed, arbitrary vocabulary to the receiver. The receiver must rely on this message to identify the target. Thus, the agents develop their own language interactively out of the need to communicate. We show that two networks with simple configurations are able to learn to coordinate in the referential game. We further explore how to make changes to the game environment to cause the \"word meanings\" induced in the game to better reflect intuitive semantic properties of the images. In addition, we present a simple strategy for grounding the agents' code into natural language. Both of these are necessary steps towards developing machines that are able to communicate with humans productively.",
"We consider the problem of multiple agents sensing and acting in environments with the goal of maximising their shared utility. In these environments, agents must learn communication protocols in order to share information that is needed to solve the tasks. By embracing deep neural networks, we are able to demonstrate end-to-end learning of protocols in complex environments inspired by communication riddles and multi-agent computer vision problems with partial observability. We propose two approaches for learning in these domains: Reinforced Inter-Agent Learning (RIAL) and Differentiable Inter-Agent Learning (DIAL). The former uses deep Q-learning, while the latter exploits the fact that, during learning, agents can backpropagate error derivatives through (noisy) communication channels. Hence, this approach uses centralised learning but decentralised execution. Our experiments introduce new environments for studying the learning of communication protocols and present a set of engineering innovations that are essential for success in these domains.",
"Many tasks in AI require the collaboration of multiple agents. Typically, the communication protocol between agents is manually specified and not altered during training. In this paper we explore a simple neural model, called CommNet, that uses continuous communication for fully cooperative tasks. The model consists of multiple agents and the communication between them is learned alongside their policy. We apply this model to a diverse set of tasks, demonstrating the ability of the agents to learn to communicate amongst themselves, yielding improved performance over non-communicative agents and baselines. In some cases, it is possible to interpret the language devised by the agents, revealing simple but effective strategies for solving the task at hand.",
""
]
} |
1703.04856 | 2600659099 | Source camera identification is still a hard task in forensics community, especially for the case of the small query image size. In this paper, we propose a solution to identify the source camera of the small-size images: content-adaptive fusion network. In order to learn better feature representation from the input data, content-adaptive convolutional neural networks(CA-CNN) are constructed. We add a convolutional layer in preprocessing stage. Moreover, with the purpose of capturing more comprehensive information, we parallel three CA-CNNs: CA3-CNN, CA5-CNN, CA7-CNN to get the content-adaptive fusion network. The difference of three CA-CNNs lies in the convolutional kernel size of pre-processing layer. The experimental results show that the proposed method is practicable and satisfactory. | In our previous work @cite_9 , we proposed the LCNN to detect recaptured images. The LCNN has better detection performance than the algorithms based on handcrafted features. The architecture of the LCNN is shown in Fig 1. There has a preprocessing layer, five convolutional layers and a softmax layer. Laplician filtering operation is put into preprocessing layer. The convolutional layer contains four operations: convolution, Batch-Normalization, ReLu, and average pooling. The numbers of feature maps in five convolutional layers are 8, 16, 32, 64, and 128, respectively. In order to avoid overfitting, we applied global average pooling to the last pooling layer and directly fed the output of global average pooling into softmax layer. What's more, the Batch-Normalization layer is used. It has been proved that it is an effective mode to accelerate convergence. Owing to the generalization of convolutional nerual networks, we also make use of the LCNN architecture in this work. | {
"cite_N": [
"@cite_9"
],
"mid": [
"2588403050"
],
"abstract": [
"Recapture image forensics has drawn much attention in public security forensics. Although some algorithms have been proposed to deal with it, there is still great challenge for small-size images. In this paper, we propose a generalized model for small-size recapture image forensics based on Laplacian Convolutional Neural Networks. Different from other Convolutional Neural Networks models, We put signal enhancement layer into Convolutional Neural Networks structure and Laplacian filter is used in the signal enhancement layer. We test the proposed method on four kinds of small-size image databases. The experimental results have demonstrate that the proposed algorithm is effective. The detection accuracies for different image size database are all above 95 ."
]
} |
1703.04823 | 2595133862 | Biological and cellular systems are often modeled as graphs in which vertices represent objects of interest (genes, proteins, drugs) and edges represent relational ties among these objects (binds-to, interacts-with, regulates). This approach has been highly successful owing to the theory, methodology and software that support analysis and learning on graphs. Graphs, however, often suffer from information loss when modeling physical systems due to their inability to accurately represent multiobject relationships. Hypergraphs, a generalization of graphs, provide a framework to mitigate information loss and unify disparate graph-based methodologies. In this paper, we present a hypergraph-based approach for modeling physical systems and formulate vertex classification, edge classification and link prediction problems on (hyper)graphs as instances of vertex classification on (extended, dual) hypergraphs in a semi-supervised setting. We introduce a novel kernel method on vertex- and edge-labeled (colored) hypergraphs for analysis and learning. The method is based on exact and inexact (via hypergraph edit distances) enumeration of small simple hypergraphs, referred to as hypergraphlets, rooted at a vertex of interest. We extensively evaluate this method and show its potential use in a positive-unlabeled setting to estimate the number of missing and false positive links in protein-protein interaction networks. | Under a supervised learning framework, Wachman and Khardon @cite_50 propose random walk-based hypergraph kernels on ordered hypergraphs, while Sun @cite_42 present a hypergraph spectral learning formulation for multi-label classification. More recently, Bai @cite_52 introduced a hypergraph kernel that transforms a hypergraph into a directed line graph and computes a Weisfeiler-Lehman isomorphism test between directed graphs. A major drawback of most such approaches is that no graph representation fully captures the hypergraph structure. For instance, Ihler @cite_56 have shown that it is impossible to have an exact representation of a hypergraph via a graph while still retaining its cut properties. Therefore, there is a need for a robust hypergraph-based methodology for learning directly on hypergraph data. | {
"cite_N": [
"@cite_42",
"@cite_52",
"@cite_50",
"@cite_56"
],
"mid": [
"",
"1976617626",
"2005683516",
"1973113780"
],
"abstract": [
"",
"In this paper, we present a hyper graph kernel computed using substructure isomorphism tests. Measuring the isomorphisms between hyper graphs straightforwardly tends to be elusive since a hyper graph may exhibit varying relational orders. We thus transform a hyper graph into a directed line graph. This not only accurately reflects the multiple relationships exhibited by the hyper graph but is also easier to manipulate isomorphism tests. To locate the isomorphisms between hyper graphs through their directed line graphs, we propose a new directed Weisfeiler-Lehman isomorphism test for directed graphs. The new isomorphism test precisely reflects the structure of the directed edges. By identifying the isomorphic substructures of directed graphs, the hyper graph kernel for a pair of hyper graphs is computed by counting the number of pair wise isomorphic substructures from their directed line graphs. We show that our kernel limits tottering that arises in the existing walk and sub tree based (hyper)graph kernels. Experiments on challenging (hyper)graph datasets demonstrate the effectiveness of our kernel.",
"The paper presents a kernel for learning from ordered hypergraphs, a formalization that captures relational data as used in Inductive Logic Programming (ILP). The kernel generalizes previous approaches to graph kernels in calculating similarity based on walks in the hypergraph. Experiments on challenging chemical datasets demonstrate that the kernel outperforms existing ILP methods, and is competitive with state-of-the-art graph kernels. The experiments also demonstrate that the encoding of graph data can affect performance dramatically, a fact that can be useful beyond kernel methods.",
"Abstract An elegant and general way to apply graph partitioning algorithms to hypergraphs would be to model hypergraphs by graphs and apply the graph algorithms to these models. Of course such models have to simulate the given hypergraphs with respect to their cut properties. An edge-weighted graph ( V , E ) is a cut-model for an edge-weighted hypergraph ( V , H ) if the weight of the edges cut by any bipartition of V in the graph is the same as the weight of the hyperedges cut by the same bipartition in the hypergraph. We show that there is no cut-model in general. Next we examine whether the addition of dummy vertices helps: An edge-weighted graph ( V ∪ D , E ) is a mincut-model for an edge-weighted hypergraph ( V, H ) if the weight of the hyperedges cut by a bipartition of the hypergraphs vertices is the same as the weight of a minimum cut separating the two parts in the graph. We construct such models using positive and negative weights. On the other hand, we show that there is no mincut-model in general if only positive weights are allowed."
]
} |
1703.04879 | 2595717484 | Named entity classification is the task of classifying text-based elements into various categories, including places, names, dates, times, and monetary values. A bottleneck in named entity classification, however, is the data problem of sparseness, because new named entities continually emerge, making it rather difficult to maintain a dictionary for named entity classification. Thus, in this paper, we address the problem of named entity classification using matrix factorization to overcome the problem of feature sparsity. Experimental results show that our proposed model, with fewer features and a smaller size, achieves competitive accuracy to state-of-the-art models. | A standard approach to named entity classification is to formulate a task as a sequence labeling problem and use a supervised method, such as conditional random fields @cite_7 @cite_9 @cite_1 . These studies heavily rely on feature templates for learning combinations of features; however, since combinations of features in conventional supervised learning are treated independently, this approach is not robust for named entities that do not appear in the training data. | {
"cite_N": [
"@cite_9",
"@cite_1",
"@cite_7"
],
"mid": [
"",
"2158188757",
"2147880316"
],
"abstract": [
"",
"We describe semi-Markov conditional random fields (semi-CRFs), a conditionally trained version of semi-Markov chains. Intuitively, a semi-CRF on an input sequence x outputs a \"segmentation\" of x, in which labels are assigned to segments (i.e., subsequences) of x rather than to individual elements xi of x. Importantly, features for semi-CRFs can measure properties of segments, and transitions within a segment can be non-Markovian. In spite of this additional power, exact learning and inference algorithms for semi-CRFs are polynomial-time—often only a small constant factor slower than conventional CRFs. In experiments on five named entity recognition problems, semi-CRFs generally outperform conventional CRFs.",
"We present conditional random fields , a framework for building probabilistic models to segment and label sequence data. Conditional random fields offer several advantages over hidden Markov models and stochastic grammars for such tasks, including the ability to relax strong independence assumptions made in those models. Conditional random fields also avoid a fundamental limitation of maximum entropy Markov models (MEMMs) and other discriminative Markov models based on directed graphical models, which can be biased towards states with few successor states. We present iterative parameter estimation algorithms for conditional random fields and compare the performance of the resulting models to HMMs and MEMMs on synthetic and natural-language data."
]
} |
1703.04879 | 2595717484 | Named entity classification is the task of classifying text-based elements into various categories, including places, names, dates, times, and monetary values. A bottleneck in named entity classification, however, is the data problem of sparseness, because new named entities continually emerge, making it rather difficult to maintain a dictionary for named entity classification. Thus, in this paper, we address the problem of named entity classification using matrix factorization to overcome the problem of feature sparsity. Experimental results show that our proposed model, with fewer features and a smaller size, achieves competitive accuracy to state-of-the-art models. | To address the task of unknown named entity classification, explored the use of sparse combinatorial features. They proposed a log-bilinear model that defines a score function considering interactions between features; the score function is regularized via a nuclear norm on a feature weight matrix. Further, heir method employs singular value decomposition (SVD)-based regularization to handle the combination of features. They reported that their regularization achieved higher accuracy than L1 and L2 regularization, frequently used in natural language processing @cite_4 . | {
"cite_N": [
"@cite_4"
],
"mid": [
"2017711997"
],
"abstract": [
"When linear classifiers cannot successfully classify data, we often add combination features, which are products of several original features. The searching for effective combination features, namely feature engineering, requires domain-specific knowledge and hard work. We present herein an efficient algorithm for learning an L1 regularized logistic regression model with combination features. We propose to use the grafting algorithm with efficient computation of gradients. This enables us to find optimal weights efficiently without enumerating all combination features. By using L1 regularization, the result we obtain is very compact and achieves very efficient inference. In experiments with NLP tasks, we show that the proposed method can extract effective combination features, and achieve high performance with very few features."
]
} |
1703.05019 | 2597322030 | In this paper, we present a method to initialize at a feasible point and unfailingly solve a non-convex optimization problem in which a set-point motion is planned for a multi-link manipulator under state and control constraints. We construct an initial feasible solution by analyzing the final time effect for feasibility problems of flatness based motion planning problems. More specifically, we first find a feasible time-optimal trajectory under state constraints without a control constraint by solving a linear programming problem. Then, we find a feasible trajectory under control constraints by scaling the trajectory. To evaluate the practical applicability of the proposed method, we did numerical experiments to solve a multi-link manipulator motion planning problem by combining the method with recursive inverse dynamics algorithms. | A method that deals with a state equation constraint by locating sample times in the control horizon and calculating the states by numerical integration is called a shooting method'' @cite_16 . There are many types of research that extend this family of methods and reduce their computational cost. We focus on a method that does not use numerical integration of a state equation because a state equation of a Multi-Link robot manipulator is difficult to numerically integrate both very accurately and quickly. | {
"cite_N": [
"@cite_16"
],
"mid": [
"1985249384"
],
"abstract": [
"Preface 1. Introduction to nonlinear programming 2. Large, sparse nonlinear programming 3. Optimal control preliminaries 4. The optimal control problem 5. Optimal control examples Appendix A. Software Bibliography, Index."
]
} |
1703.05019 | 2597322030 | In this paper, we present a method to initialize at a feasible point and unfailingly solve a non-convex optimization problem in which a set-point motion is planned for a multi-link manipulator under state and control constraints. We construct an initial feasible solution by analyzing the final time effect for feasibility problems of flatness based motion planning problems. More specifically, we first find a feasible time-optimal trajectory under state constraints without a control constraint by solving a linear programming problem. Then, we find a feasible trajectory under control constraints by scaling the trajectory. To evaluate the practical applicability of the proposed method, we did numerical experiments to solve a multi-link manipulator motion planning problem by combining the method with recursive inverse dynamics algorithms. | The differential flatness concept was proposed by @cite_11 . Flatness has been studied from the viewpoint of motion planning problems @cite_18 . @cite_8 developed a convex programming algorithm for a motion planning problem under semi-infinite inequality constraints by approximating a feasible region to a convex polytope. Van @cite_7 developed a method to approximate polynomial semi-infinite inequality constraints by using the convex hull property of B-splines @cite_15 . Furthermore, they applied the methods to a flexible robot arm planning problem @cite_9 . We follow this line of research further and tackle more practical problem settings on the basis of more complex dynamics. | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_8",
"@cite_9",
"@cite_15",
"@cite_11"
],
"mid": [
"2098022952",
"1508155872",
"",
"1553106265",
"",
"2065297540"
],
"abstract": [
"Flat systems, an important subclass of nonlinear control systems introduced via differential-algebraic methods, are defined in a differential geometric framework. We utilize the infinite dimensional geometry developed by Vinogradov and coworkers: a control system is a diffiety, or more precisely, an ordinary diffiety, i.e. a smooth infinite-dimensional manifold equipped with a privileged vector field. After recalling the definition of a Lie-Backlund mapping, we say that two systems are equivalent if they are related by a Lie-Backlund isomorphism. Flat systems are those systems which are equivalent to a controllable linear one. The interest of such an abstract setting relies mainly on the fact that the above system equivalence is interpreted in terms of endogenous dynamic feedback. The presentation is as elementary as possible and illustrated by the VTOL aircraft.",
"This research deals with the computation of optimal trajectories considering state and input constraints for linear and nonlinear systems that admit a polynomial representation through differential flatness. Based on a polynomial spline parameterization of the flat output an optimization problem in terms of the B-spline coefficients is derived that guarantees constraint satisfaction over the entire time horizon whereas classical approaches in the literature only impose the constraints on a finite time grid. As the proposed constraints are only sufficient conditions, a novel method is presented that effectively reduces their conservatism. Two numerical examples, a linear benchmark tracking problem and an optimal quadrotor maneuver, illustrate the efficiency and practicality of the presented method.",
"",
"Abstract. When optimizing the performance of constrained robotic system, the motion trajectory plays a crucial role. In this research the motion planning problem for systems that admit a polynomial description of the system dynamics through differential flatness is tackled by parameterizing the system's so-called flat output as a polynomial spline. Using basic properties of B-splines, sufficient conditions on the spline coefficients are derived ensuring satisfaction of the operating constraints over the entire time horizon. Furthermore, an intuitive relaxation is proposed to tackle conservatism and a supporting software package is released. Finally, to illustrate the overall approach and potential, a numerical benchmark of a flexible link manipulator is discussed.",
"",
"We introduce flat systems, which are equivalent to linear ones via a special type of feedback called endogenous. Their physical properties are subsumed by a linearizing output and they might be regarded as providing another nonlinear extension of Kalman's controllability. The distance to flatness is measured by a non-negative integer, the defect. We utilize differential algebra where flatness- and defect are best defined without distinguishing between input, state, output and other variables. Many realistic classes of examples are flat. We treat two popular ones: the crane and the car with n trailers, the motion planning of which is obtained via elementary properties of plane curves. The three non-flat examples, the simple, double and variable length pendulums, are borrowed from non-linear physics. A high frequency control strategy is proposed such that the averaged systems become flat."
]
} |
1703.05019 | 2597322030 | In this paper, we present a method to initialize at a feasible point and unfailingly solve a non-convex optimization problem in which a set-point motion is planned for a multi-link manipulator under state and control constraints. We construct an initial feasible solution by analyzing the final time effect for feasibility problems of flatness based motion planning problems. More specifically, we first find a feasible time-optimal trajectory under state constraints without a control constraint by solving a linear programming problem. Then, we find a feasible trajectory under control constraints by scaling the trajectory. To evaluate the practical applicability of the proposed method, we did numerical experiments to solve a multi-link manipulator motion planning problem by combining the method with recursive inverse dynamics algorithms. | Their methods are based on the Lie group formulation of multi-link robot manipulator modeling and effective calculation of the inverse dynamics of a manipulator. The theory of the Lie group formulation of robot dynamics is described in detail elsewhere @cite_0 , @cite_6 . A recursive algorithm for the Lie group formulation of robot dynamics was developed @cite_6 and then analyzed more deeply and extended @cite_1 @cite_12 . Our method references Kim and Polland's implementation @cite_13 . | {
"cite_N": [
"@cite_1",
"@cite_6",
"@cite_0",
"@cite_13",
"@cite_12"
],
"mid": [
"",
"2136621952",
"1564897360",
"2053162587",
"2061271987"
],
"abstract": [
"",
"In this article we present a unified geometric treatment of robot dynamics. Using standard ideas from Lie groups and Rieman nian geometry, we formulate the equations of motion for an open chain manipulator both recursively and in closed form. The recursive formulation leads to an O(n) algorithm that ex presses the dynamics entirely in terms of coordinate-free Lie algebraic operations. The Lagrangian formulation also ex presses the dynamics in terms of these Lie algebraic operations and leads to a particularly simple set of closed-form equations, in which the kinematic and inertial parameters appear explic itly and independently of each other. The geometric approach permits a high-level, coordinate-free view of robot dynamics that shows explicitly some of the connections with the larger body of work in mathematics and physics. At the same time the resulting equations are shown to be computationally ef fective and easily differentiated and factored with respect to any of the robot parameters. This latter fe...",
"INTRODUCTION: Brief History. Multifingered Hands and Dextrous Manipulation. Outline of the Book. Bibliography. RIGID BODY MOTION: Rigid Body Transformations. Rotational Motion in R3. Rigid Motion in R3. Velocity of a Rigid Body. Wrenches and Reciprocal Screws. MANIPULATOR KINEMATICS: Introduction. Forward Kinematics. Inverse Kinematics. The Manipulator Jacobian. Redundant and Parallel Manipulators. ROBOT DYNAMICS AND CONTROL: Introduction. Lagrange's Equations. Dynamics of Open-Chain Manipulators. Lyapunov Stability Theory. Position Control and Trajectory Tracking. Control of Constrained Manipulators. MULTIFINGERED HAND KINEMATICS: Introduction to Grasping. Grasp Statics. Force-Closure. Grasp Planning. Grasp Constraints. Rolling Contact Kinematics. HAND DYNAMICS AND CONTROL: Lagrange's Equations with Constraints. Robot Hand Dynamics. Redundant and Nonmanipulable Robot Systems. Kinematics and Statics of Tendon Actuation. Control of Robot Hands. NONHOLONOMIC BEHAVIOR IN ROBOTIC SYSTEMS: Introduction. Controllability and Frobenius' Theorem. Examples of Nonholonomic Systems. Structure of Nonholonomic Systems. NONHOLONOMIC MOTION PLANNING: Introduction. Steering Model Control Systems Using Sinusoids. General Methods for Steering. Dynamic Finger Repositioning. FUTURE PROSPECTS: Robots in Hazardous Environments. Medical Applications for Multifingered Hands. Robots on a Small Scale: Microrobotics. APPENDICES: Lie Groups and Robot Kinematics. A Mathematica Package for Screw Calculus. Bibliography. Index Each chapter also includes a Summary, Bibliography, and Exercises",
"We propose a fast physically-based simulation system for skeleton-driven deformable body characters. Our system can generate realistic motions of self-propelled deformable body characters by considering the two-way interactions among the skeleton, the deformable body, and the environment in the dynamic simulation. It can also compute the passive jiggling behavior of a deformable body driven by a kinematic skelet al motion. We show that a well-coordinated combination of: (1) a reduced deformable body model with nonlinear finite elements, (2) a linear-time algorithm for skeleton dynamics, and (3) explicit integration can boost simulation speed to orders of magnitude faster than existing methods, while preserving modeling accuracy as much as possible. Parallel computation on the GPU has also been implemented to obtain an additional speedup for complicated characters. Detailed discussions of our engineering decisions for speed and accuracy of the simulation system are presented in the article. We tested our approach with a variety of skeleton-driven deformable body characters, and the tested characters were simulated in real time or near real time.",
"In this work an efficient dynamics algorithm is developed, which is applicable to a wide range of multibody systems, including underactuated systems, branched or tree-topology systems, robots, and walking machines. The dynamics algorithm is differentiated with respect to the input parameters in order to form sensitivity equations. The algorithm makes use of techniques and notation from the theory of Lie groups and Lie algebras, which is reviewed briefly. One of the strengths of our formulation is the ability to easily differentiate the dynamics algorithm with respect to parameters of interest. We demonstrate one important use of our dynamics and sensitivity algorithms by using them to solve difficult optimal control problems for underactuated systems. The algorithms in this paper have been implemented in a software package named Cstorm (Computer simulation tool for the optimization of robot manipulators), which runs from within Matlab and Simulink. It can be downloaded from the website http: www.eng.uci.edu bobrow @DOI: 10.1115 1.1376121#"
]
} |
1703.05019 | 2597322030 | In this paper, we present a method to initialize at a feasible point and unfailingly solve a non-convex optimization problem in which a set-point motion is planned for a multi-link manipulator under state and control constraints. We construct an initial feasible solution by analyzing the final time effect for feasibility problems of flatness based motion planning problems. More specifically, we first find a feasible time-optimal trajectory under state constraints without a control constraint by solving a linear programming problem. Then, we find a feasible trajectory under control constraints by scaling the trajectory. To evaluate the practical applicability of the proposed method, we did numerical experiments to solve a multi-link manipulator motion planning problem by combining the method with recursive inverse dynamics algorithms. | @cite_4 solved a problem by formulating control constraints as costs, which are soft constraints. We deal with control constraints as hard constraints to avoid the weight of control constraints, which vary depending on cost weight tuning. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2161137845"
],
"abstract": [
"Although the dynamic equations of motion of open-chained robot systems are well known, they are seldom taken into account during the planning of motions. In this work, we show that the dynamics of a robot can be used to produce motions that extend the payload capability beyond the limit set by traditional methods. In particular, we develop a point-to-point weightlifting motion planner for open-chained robots. The governing optimal control problem is converted into a direct, SQP parameter optimization in which the gradient is determined analytically. The joint trajectories are defined by B-spline polynomials along with a time-scale factor. The algorithm is applied to a Puma 762 robot, with its physical limitations incorporated into the formulation. The torque limits are formulated as soft constraints added into the objective function while the position and velocity limits are formulated as hard, linear inequality constraints, on the parameters. The solutions obtained with our algorithm extend the robot's payload capability while reducing the joint torques. Interestingly, nearly all the trajectories found pass through singular configurations, where large internal forces from the robot are applied to the payload and little torque is needed from the motors."
]
} |
1703.05082 | 2950671542 | Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation. | The closest work to ours is on active search. The goal of active search is to uncover as many nodes of a target class as possible in a network where the topology is known @cite_7 @cite_15 @cite_6 @cite_2 . Like , active search considers situations where only members of a target class (e.g., malicious class) are sought. Since obtaining labels is associated with a cost (time or money), it is paramount to avoid spending resources on nodes that are unlikely to be targets. Unlike our problem, active search assumes the network topology is known and that any node can be queried at any time. | {
"cite_N": [
"@cite_15",
"@cite_6",
"@cite_7",
"@cite_2"
],
"mid": [
"2952562730",
"2069790826",
"",
"2402688996"
],
"abstract": [
"We consider two active binary-classification problems with atypical objectives. In the first, active search, our goal is to actively uncover as many members of a given class as possible. In the second, active surveying, our goal is to actively query points to ultimately predict the proportion of a given class. Numerous real-world problems can be framed in these terms, and in either case typical model-based concerns such as generalization error are only of secondary importance. We approach these problems via Bayesian decision theory; after choosing natural utility functions, we derive the optimal policies. We provide three contributions. In addition to introducing the active surveying problem, we extend previous work on active search in two ways. First, we prove a novel theoretical result, that less-myopic approximations to the optimal policy can outperform more-myopic approximations by any arbitrary degree. We then derive bounds that for certain models allow us to reduce (in practice dramatically) the exponential search space required by a naive implementation of the optimal policy, enabling further lookahead while still ensuring that optimal decisions are always made.",
"Active search is an increasingly important learning problem in which we use a limited budget of label queries to discover as many members of a certain class as possible. Numerous real-world applications may be approached in this manner, including fraud detection, product recommendation, and drug discovery. Active search has model learning and exploration exploitation features similar to those encountered in active learning and bandit problems, but algorithms for those problems do not fit active search. Previous work on the active search problem [5] showed that the optimal algorithm requires a lookahead evaluation of expected utility that is exponential in the number of selections to be made and proposed a truncated lookahead heuristic. Inspired by the success of myopic methods for active learning and bandit problems, we propose a myopic method for active search on graphs. We suggest selecting points by maximizing a score considering the potential impact of selecting a node, meant to emulate lookahead while avoiding exponential search. We test the proposed algorithm empirically on real-world graphs and show that it outperforms popular approaches for active learning and bandit problems as well as truncated lookahead of a few steps.",
"",
"Many modern information access problems involve highly complex patterns that cannot be handled by traditional keyword based search. Active Search is an emerging paradigm that helps users quickly find relevant information by efficiently collecting and learning from user feedback. We consider active search on graphs, where the nodes represent the set of instances users want to search over and the edges encode pairwise similarity among the instances. Existing active search algorithms are either short of theoretical guarantees or inadequate for graph data. Motivated by recent advances in active learning on graphs, namely the Σ-optimality selection criterion, we propose new active search algorithms suitable for graphs with theoretical guarantees and demonstrate their effectiveness on several real-world datasets. We relate our active search setting to multi-armed bandits whose rewards are binary values indicating search hits or misses and arms cannot be pulled more than once. We also discussed theoretical guarantees for applying Σ-optimality as the exploration term for bandits on graphs."
]
} |
1703.05082 | 2950671542 | Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation. | In @cite_29 a problem similar to is investigated and a learning-based method called Active Exploration (AE) is proposed. Unlike in , border nodes attributes are assumed to be observable. Since node attributes often carry considerable information about the node's label, AE is not directly comparable with other methods. Our solution differs from AE in that it leverages heuristics in addition to base learners and is applicable to a wider range of applications. | {
"cite_N": [
"@cite_29"
],
"mid": [
"2134092622"
],
"abstract": [
"Many interesting domains in machine learning can be viewed as networks, with relationships (e.g., friendships) connecting items (e.g., individuals). The Active Exploration (AE) task is to identify all items in a network with a desired trait (i.e., positive labels) given only partial information about the network. The AE process iteratively queries for labels or network structure within a limited budget; thus, accurate predictions prior to making each query is critical to maximizing the number of positives gathered. However, the targeted AE query process produces partially observed networks that can create difficulties for predictive modeling. In particular, we demonstrate that these partial networks can exhibit extreme label correlation bias, which makes it difficult for conventional relational learning methods to accurately estimate relational parameters. To overcome this issue, we model the joint distribution of possible edges and labels to improve learning and inference. Our proposed method, Probabilistic Relational Expectation Maximization (PR-EM), is the first AE approach to accurately learn the complex dependencies between attributes, labels, and structure to improve predictions. PR-EM utilizes collective inference over the missing relationships in the partial network to jointly infer unknown item traits. Further, we develop a linear inference algorithm to facilitate efficient use of PR-EM in large networks. We test our approach on four real world networks, showing that AE with PR-EM gathers significantly more positive items compared to state-of-the-art methods."
]
} |
1703.05082 | 2950671542 | Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation. | Similarly to , active learning is an interactive framework for deciding what data points to collect in order to train a classifier or a regression model. Unlike active search, (i) its main objective is to improve the generalization performance of a model with as few label queries as possible, and (ii) the set of unlabeled points does not grow based on the collected points. A slew of active learning techniques have been proposed for non-relational data settings, including some tailored for logistic regression @cite_30 , for dealing with streamed data @cite_8 and for the case of extreme class imbalance @cite_19 . Although the retrieval of target nodes can benefit from an accurate model, it is unlikely that active learning heuristics (e.g., uncertainty sampling @cite_21 ) for training a single classifier can be used for without sacrificing performance. However, it may be possible to adapt active learning techniques proposed for training classifier ensembles (e.g., query by committee @cite_4 ) in such a way that, at the same time we collect points on which many classifiers disagree, we ensure that promising candidates among border nodes are queried before the sampling budget is exhausted. | {
"cite_N": [
"@cite_30",
"@cite_4",
"@cite_8",
"@cite_21",
"@cite_19"
],
"mid": [
"2021367230",
"2080021732",
"2090051261",
"2903158431",
"2188134654"
],
"abstract": [
"Which active learning methods can we expect to yield good performance in learning binary and multi-category logistic regression classifiers? Addressing this question is a natural first step in providing robust solutions for active learning across a wide variety of exponential models including maximum entropy, generalized linear, log-linear, and conditional random field models. For the logistic regression model we re-derive the variance reduction method known in experimental design circles as A-optimality.' We then run comparisons against different variations of the most widely used heuristic schemes: query by committee and uncertainty sampling, to discover which methods work best for different classes of problems and why. We find that among the strategies tested, the experimental design methods are most likely to match or beat a random sample baseline. The heuristic alternatives produced mixed results, with an uncertainty sampling variant called margin sampling and a derivative method called QBB-MM providing the most promising performance at very low computational cost. Computational running times of the experimental design methods were a bottleneck to the evaluations. Meanwhile, evaluation of the heuristic methods lead to an accumulation of negative results. We explore alternative evaluation design parameters to test whether these negative results are merely an artifact of settings where experimental design methods can be applied. The results demonstrate a need for improved active learning methods that will provide reliable performance at a reasonable computational cost.",
"We propose an algorithm called query by commitee , in which a committee of students is trained on the same data set. The next query is chosen according to the principle of maximal disagreement . The algorithm is studied for two toy models: the high-low game and perceptron learning of another perceptron. As the number of queries goes to infinity, the committee algorithm yields asymptotically finite information gain. This leads to generalization error that decreases exponentially with the number of examples. This in marked contrast to learning from randomly chosen inputs, for which the information gain approaches zero and the generalization error decreases with a relatively slow inverse power law. We suggest that asymptotically finite information gain may be an important characteristic of good query algorithms.",
"We present a generalized framework for active inference, the selective acquisition of labels for cases at prediction time in lieu of using the estimated labels of a predictive model. We develop techniques within this framework for classifying in an online setting, for example, for classifying the stream of web pages where online advertisements are being served. Stream applications present novel complications because (i) at the time of label acquisition, we don't know the set of instances that we will eventually see, (ii) instances repeat based on some unknown (and possibly skewed) distribution. We combine ideas from decision theory, cost-sensitive learning, and online density estimation. We also introduce a method for on-line estimation of the utility distribution, which allows us to manage the budget over the stream. The resulting model tells which instances to label so that by the end of each budget period, the budget is best spent (in expectation). The main results show that: (1) our proposed approach to active inference on streams can indeed reduce error costs substantially over alternative approaches, (2) more sophisticated online estimations achieve larger reductions in error. We next discuss simultaneously conducting active inference and active learning. We show that our expected-utility active inference strategy also selects good examples for learning. We close by pointing out that our utility-distribution estimation strategy can also be applied to convert pool-based active learning techniques into budget-sensitive online active learning techniques.",
"",
"Extreme class skew is a hurdle in many machine learning tasks. In such skewed settings, traditional methods for procuring labeled examples, including random sampling and active learning, are often ineective| they struggle to nd representative minority examples. The framework of Dual Supervision, which incorporates feature-based background information into traditional supervised learning, provides one avenue to combat this problem. However, active learning for feature information (feature labeling), like active learning, is often not resilient to extreme class skew. In this work, we present an alternative to active feature labeling, Guided Feature Labeling. In this paradigm, human domain experts are tasked with"
]
} |
1703.05082 | 2950671542 | Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation. | Despite these differences, there is an interesting parallel between with many models and a body of research on active learning with a set of active learners (or heuristics). Both problems can be cast as MABs, where border nodes are analogous to unlabeled data points. In active learning, a reward is indirectly related to the collected point: it is computed as some proxy for or estimate of the model's performance on a test set, when fit to all points collected up to a given step. In contrast, rewards in are simply the node labels. Like , active learning can either map heuristics directly as arms @cite_18 or map heuristics as experts that give recommendations on how to choose the unlabeled points @cite_25 . In both cases it has been observed that combining heuristics may often outperform the single best heuristic. While these works apply algorithms for adversarial bandits to active learning, we find that Dynamic Thompson Sampling for stochastic bandits with non-stationary rewards seem to exploit better the fact that arms rewards are slowly changing in . | {
"cite_N": [
"@cite_18",
"@cite_25"
],
"mid": [
"2140679654",
"603830301"
],
"abstract": [
"This work is concerned with the question of how to combine online an ensemble of active learners so as to expedite the learning progress in pool-based active learning. We develop an active-learning master algorithm, based on a known competitive algorithm for the multi-armed bandit problem. A major challenge in successfully choosing top performing active learners online is to reliably estimate their progress during the learning session. To this end we propose a simple maximum entropy criterion that provides effective estimates in realistic settings. We study the performance of the proposed master algorithm using an ensemble containing two of the best known active-learning algorithms as well as a new algorithm. The resulting active-learning master algorithm is empirically shown to consistently perform almost as well as and sometimes outperform the best algorithm in the ensemble on a range of classification problems.",
"Pool-based active learning is an important technique that helps reduce labeling efforts within a pool of unlabeled instances. Currently, most pool-based active learning strategies are constructed based on some human-designed philosophy; that is, they reflect what human beings assume to be \"good labeling questions.\" However, while such human-designed philosophies can be useful on specific data sets, it is often difficult to establish the theoretical connection of those philosophies to the true learning performance of interest. In addition, given that a single human-designed philosophy is unlikely to work on all scenarios, choosing and blending those strategies under different scenarios is an important but challenging practical task. This paper tackles this task by letting the machines adaptively \"learn\" from the performance of a set of given strategies on a particular data set. More specifically, we design a learning algorithm that connects active learning with the well-known multi-armed bandit problem. Further, we postulate that, given an appropriate choice for the multi-armed bandit learner, it is possible to estimate the performance of different strategies on the fly. Extensive empirical studies of the resulting ALBL algorithm confirm that it performs better than state-of-the-art strategies and a leading blending algorithm for active learning, all of which are based on human-designed philosophy."
]
} |
1703.05082 | 2950671542 | Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation. | Last, another variant of active learning considers the task of learning an ensemble of models @cite_13 or finding a low risk hypothesis @math @cite_27 @cite_9 while labeling as few points as possible. Since the labeled points are biased by the collection process, estimating the models' generalization performances requires either building an uniformly random validation set, or sampling probabilistically at every step and then using importance weighted estimates. In , however, the models relative performances can be directly measured from the queried nodes payoffs. Moreover, building a random validation set is bound to degrade performance in scenarios where target nodes are scarce. | {
"cite_N": [
"@cite_9",
"@cite_27",
"@cite_13"
],
"mid": [
"",
"2122458503",
"2133860205"
],
"abstract": [
"",
"In this paper we address the problem of pool based active learning, and provide an algorithm, called UPAL, that works by minimizing the unbiased estimator of the risk of a hypothesis in a given hypothesis space. For the space of linear classifiers and the squared loss we show that UPAL is equivalent to an exponentially weighted average forecaster. Exploiting some recent results regarding the spectra of random matrices allows us to establish consistency of UPAL when the true hypothesis is a linear hypothesis. Empirical comparison with an active learner implementation in Vowpal Wabbit, and a previously proposed pool based active learner implementation show good empirical performance and better scalability.",
"Most active learning methods avoid model selection by training models of one type (SVMs, boosted trees, etc.) using one pre-defined set of model hyperparameters. We propose an algorithm that actively samples data to simultaneously train a set of candidate models (different model types and or different hyperparameters) and also select the best model from this set. The algorithm actively samples points for training that are most likely to improve the accuracy of the more promising candidate models, and also samples points for model selection-- all samples count against the same labeling budget. This exposes a natural trade-off between the focused active sampling that is most effective for training models, and the unbiased sampling that is better for model selection. We empirically demonstrate on six test problems that this algorithm is nearly as effective as an active learning oracle that knows the optimal model in advance."
]
} |
1703.04877 | 2953272347 | Acquiring the accurate 3-D position of a target person around a robot provides fundamental and valuable information that is applicable to a wide range of robotic tasks, including home service, navigation and entertainment. This paper presents a real-time robotic 3-D human tracking system which combines a monocular camera with an ultrasonic sensor by the extended Kalman filter (EKF). The proposed system consists of three sub-modules: monocular camera sensor tracking model, ultrasonic sensor tracking model and multi-sensor fusion. An improved visual tracking algorithm is presented to provide partial location estimation (2-D). The algorithm is designed to overcome severe occlusions, scale variation, target missing and achieve robust re-detection. The scale accuracy is further enhanced by the estimated 3-D information. An ultrasonic sensor array is employed to provide the range information from the target person to the robot and Gaussian Process Regression is used for partial location estimation (2-D). EKF is adopted to sequentially process multiple, heterogeneous measurements arriving in an asynchronous order from the vision sensor and the ultrasonic sensor separately. In the experiments, the proposed tracking system is tested in both simulation platform and actual mobile robot for various indoor and outdoor scenes. The experimental results show the superior performance of the 3-D tracking system in terms of both the accuracy and robustness. | An exhaustive analysis is beyond this work. Thus we recommend @cite_0 and @cite_7 for a full understanding about this problem. | {
"cite_N": [
"@cite_0",
"@cite_7"
],
"mid": [
"2158592639",
"2130026429"
],
"abstract": [
"Object tracking has been one of the most important and active research areas in the field of computer vision. A large number of tracking algorithms have been proposed in recent years with demonstrated success. However, the set of sequences used for evaluation is often not sufficient or is sometimes biased for certain types of algorithms. Many datasets do not have common ground-truth object positions or extents, and this makes comparisons among the reported quantitative results difficult. In addition, the initial conditions or parameters of the evaluated tracking algorithms are not the same, and thus, the quantitative results reported in literature are incomparable or sometimes contradictory. To address these issues, we carry out an extensive evaluation of the state-of-the-art online object-tracking algorithms with various evaluation criteria to understand how these methods perform within the same framework. In this work, we first construct a large dataset with ground-truth object positions and extents for tracking and introduce the sequence attributes for the performance analysis. Second, we integrate most of the publicly available trackers into one code library with uniform input and output formats to facilitate large-scale performance evaluation. Third, we extensively evaluate the performance of 31 algorithms on 100 sequences with different initialization settings. By analyzing the quantitative results, we identify effective approaches for robust tracking and provide potential future research directions in this field.",
"The Visual Object Tracking challenge 2015, VOT2015, aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance. Results of 62 trackers are presented. The number of tested trackers makes VOT 2015 the largest benchmark on short-term tracking to date. For each participating tracker, a short description is provided in the appendix. Features of the VOT2015 challenge that go beyond its VOT2014 predecessor are: (i) a new VOT2015 dataset twice as large as in VOT2014 with full annotation of targets by rotated bounding boxes and per-frame attribute, (ii) extensions of the VOT2014 evaluation methodology by introduction of a new performance measure. The dataset, the evaluation kit as well as the results are publicly available at the challenge website."
]
} |
1703.04877 | 2953272347 | Acquiring the accurate 3-D position of a target person around a robot provides fundamental and valuable information that is applicable to a wide range of robotic tasks, including home service, navigation and entertainment. This paper presents a real-time robotic 3-D human tracking system which combines a monocular camera with an ultrasonic sensor by the extended Kalman filter (EKF). The proposed system consists of three sub-modules: monocular camera sensor tracking model, ultrasonic sensor tracking model and multi-sensor fusion. An improved visual tracking algorithm is presented to provide partial location estimation (2-D). The algorithm is designed to overcome severe occlusions, scale variation, target missing and achieve robust re-detection. The scale accuracy is further enhanced by the estimated 3-D information. An ultrasonic sensor array is employed to provide the range information from the target person to the robot and Gaussian Process Regression is used for partial location estimation (2-D). EKF is adopted to sequentially process multiple, heterogeneous measurements arriving in an asynchronous order from the vision sensor and the ultrasonic sensor separately. In the experiments, the proposed tracking system is tested in both simulation platform and actual mobile robot for various indoor and outdoor scenes. The experimental results show the superior performance of the 3-D tracking system in terms of both the accuracy and robustness. | Ultrasonic sensors have been used extensively as time-of-flight range sensors in localizing the tracking targets @cite_20 @cite_25 @cite_11 . However, one disadvantage of this type of sensor is that when the target moves at the vertical direction of the sonar beam, the calculated locations are usually inaccurate. Another problem with sonar sensors is the reflection from obstacles in the environment will usually cause invalid and incorrect results. Furthermore, relying on the time-of-flight measurement only, the receiver is unlikely able to discriminate multiple sources which means that the system does not work when multiple targets present. | {
"cite_N": [
"@cite_25",
"@cite_20",
"@cite_11"
],
"mid": [
"2536519148",
"1981289574",
"2081793425"
],
"abstract": [
"In the intelligent transportation system, various accident avoidance techniques have been applied. Among them, one of the most common issues is the collision, which is yet unsolved problem. To this end, we develop collision warning and avoidance system (CWAS), which is implemented in the wheeled mobile robot. Likewise, path planning is a crucial problem in the mobile robots to perform a given task correctly. Here, a tracking system is presented for the mobile robot, which follows an object. Thus, we have implemented an integrated CWAS and tracking system in the mobile robot. Both systems can be activated independently. In the CWAS, the robot is controlled through a remotely controlled device, and collision prediction and avoidance functions are performed. In the tracking system, the robot performs tasks autonomously, where the robot maintains a constant distance from the followed object. The surrounding information is obtained through the range sensors, and the control functions are performed through the microcontroller. The front, left, and right sensors are activated to track the object, and all the sensors are used for the CWAS. Two algorithms based on the sensory information are developed with the distance control approach. The proposed system is tested using the binary logic controller and the fuzzy logic controller (FLC). The comparison of both controllers is also described by preferring time delay and complexity. The efficiency of the robot is improved by increasing smoothness in motion using the FLC, achieving accuracy in tracking, and advancements in the CWAS. Finally, simulation and experimental outcomes have displayed the authenticity of the system.",
"A blind spot detection device for protection against misshapenness such as automobiles collisions, obstacles, and accident that leads to great loss of human lives and can have disastrous results.Technology used for this purpose worked by detecting the other automobiles, obstacles and bystanders. Upon detecting, the device triggers a timer that delays the activation of alarm circuitry for a brief period of time.This time delay is instituted to minimize the triggers of nuisance alarm by a momentary intrusion in the hazard zone. If the obstaclepsilas presence is still detected after the delay time, LED's and audible alarms are triggered to alert the system operator of the dangerous situation. The alarms remain activated for a time period, allowing the operator to clear the hazard zone.",
"An non-linear Bayesian regression engine for robotic tracking based on an ultrasonic RF sensor unit is presented in this paper. The proposed system is able to maintain systematic tracking of a leading human in indoor outdoor settings with minimalistic instrumentation. Compared to popular camera based localization system the sonar array RF based system has the advantage of being insensitive to background light intensity changes, a primary concern in outdoor environments. In contrast to single-plane laser range finder based tracking the proposed scheme is able to better adapt to small terrain variations, while at the same time being a significantly more affordable proposition for tracking with a robotic unit. A key novelty in this work is the utilisation of Gaussian Process Regression (GPR) to build a model for the sensor unit, which is shown to compare favourably against traditional linear triangulation approaches. The covariance function yield by the GPR sensor model also provides the additional benefit of outlier rejection. We present experimental results of indoors and outdoors tracking by mounting the sensor unit on a Garden Utility Transportation System (GUTS) robot and compare the proposed approach with linear triangulation which clearly show the inference engine capability to generalise relative localisation of human and a marked improvement in tracking accuracy and robustness."
]
} |
1703.04877 | 2953272347 | Acquiring the accurate 3-D position of a target person around a robot provides fundamental and valuable information that is applicable to a wide range of robotic tasks, including home service, navigation and entertainment. This paper presents a real-time robotic 3-D human tracking system which combines a monocular camera with an ultrasonic sensor by the extended Kalman filter (EKF). The proposed system consists of three sub-modules: monocular camera sensor tracking model, ultrasonic sensor tracking model and multi-sensor fusion. An improved visual tracking algorithm is presented to provide partial location estimation (2-D). The algorithm is designed to overcome severe occlusions, scale variation, target missing and achieve robust re-detection. The scale accuracy is further enhanced by the estimated 3-D information. An ultrasonic sensor array is employed to provide the range information from the target person to the robot and Gaussian Process Regression is used for partial location estimation (2-D). EKF is adopted to sequentially process multiple, heterogeneous measurements arriving in an asynchronous order from the vision sensor and the ultrasonic sensor separately. In the experiments, the proposed tracking system is tested in both simulation platform and actual mobile robot for various indoor and outdoor scenes. The experimental results show the superior performance of the 3-D tracking system in terms of both the accuracy and robustness. | Combining the sonar sensors with cameras is another popular research direction @cite_12 @cite_15 @cite_22 . They usually use the sonar sensors to detect the regions that might contain the target in the sonar's field of view. Corresponding regions in the images are then used as additional measurements.This method may be invalid when the ultrasonic sensors lose the target, leading to the fact that the target is beyond the view of the camera. | {
"cite_N": [
"@cite_15",
"@cite_22",
"@cite_12"
],
"mid": [
"2082023026",
"2073385733",
"81273923"
],
"abstract": [
"In this paper, we propose a heterogeneous multisensor fusion algorithm for mapping in dynamic environments. The algorithm synergistically integrates the information obtained from an uncalibrated camera and sonar sensors to facilitate mapping and tracking. The sonar data is mainly used to build a weighted line-based map via the fuzzy clustering technique. The line weight, with confidence corresponding to the moving object, is determined by both sonar and vision data. The motion tracking is primarily accomplished by vision data using particle filtering and the sonar vectors originated from moving objects are used to modulate the sample weighting. A fuzzy system is implemented to fuse the two sensor data features. Additionally, in order to build a consistent global map and maintain reliable tracking of moving objects, the well-known extended Kalman filter is applied to estimate the states of robot pose and map features. Thus, more robust performance in mapping as well as tracking are achieved. The empirical ...",
"In this paper we present the development and implementation of a patrol robot for indoor environments. First, we propose an indoor patrol strategy by using ultrasonic sensors. According to the gathered range information of the environment, we apply the wall-following strategy to design of patrol rules for determining the navigation commands of the patrol robot. Then we design a steering controller based on the potential field method to drive the patrol robot moving along the wall baselines. In addition, we propose another navigation strategy by using ultrasonic and vision data fusion to improve the accuracy and robustness of the sensing system. We implement a visual navigation controller to successfully steer the patrol robot following the wall and keeping a certain distance to the wall.",
"Service robots intended to interact with people must be able to localize and continuously track their users. A method is described which integrates information from visual and sonar based tracking path- ways while updating hypotheses about the position of the robot's human user. Each tracking method uses information from the other to generate a more robust measure of the user's position, and thus a more robust behavior generation is achieved."
]
} |
1703.04529 | 2949444583 | With the increasing popularity of machine learning techniques, it has become common to see prediction algorithms operating within some larger process. However, the criteria by which we train these algorithms often differ from the ultimate criteria on which we evaluate them. This paper proposes an end-to-end approach for learning probabilistic machine learning models in a manner that directly captures the ultimate task-based objective for which they will be used, within the context of stochastic programming. We present three experimental evaluations of the proposed approach: a classical inventory stock problem, a real-world electrical grid scheduling task, and a real-world energy storage arbitrage task. We show that the proposed approach can outperform both traditional modeling and purely black-box policy optimization approaches in these applications. | In cases where the underlying probability distribution is known but the objective cannot be solved analytically, it is common to use Monte Carlo sample average approximation methods, which draw multiple iid samples from the underlying probability distribution and then use deterministic optimization methods to solve the resultant problems @cite_0 . In cases where the underlying distribution is not known, it is common to learn or estimate some model from observed samples @cite_19 . | {
"cite_N": [
"@cite_0",
"@cite_19"
],
"mid": [
"2000257769",
"1969007958"
],
"abstract": [
"We investigate the quality of solutions obtained from sample-average approximations to two-stage stochastic linear programs with recourse. We use a recently developed software tool executing on a computational grid to solve many large instances of these problems, allowing us to obtain high-quality solutions and to verify optimality and near-optimality of the computed solutions in various ways.",
"A common approach in coping with multiperiod optimization problems under uncertainty where statistical information is not really enough to support a stochastic programming model, has been to set up and analyze a number of scenarios. The aim then is to identify trends and essential features on which a robust decision policy can be based. This paper develops for the first time a rigorous algorithmic procedure for determining such a policy in response to any weighting of the scenarios. The scenarios are bundled at various levels to reflect the availability of information, and iterative adjustments are made to the decision policy to adapt to this structure and remove the dependence on hindsight."
]
} |
1703.04529 | 2949444583 | With the increasing popularity of machine learning techniques, it has become common to see prediction algorithms operating within some larger process. However, the criteria by which we train these algorithms often differ from the ultimate criteria on which we evaluate them. This paper proposes an end-to-end approach for learning probabilistic machine learning models in a manner that directly captures the ultimate task-based objective for which they will be used, within the context of stochastic programming. We present three experimental evaluations of the proposed approach: a classical inventory stock problem, a real-world electrical grid scheduling task, and a real-world energy storage arbitrage task. We show that the proposed approach can outperform both traditional modeling and purely black-box policy optimization approaches in these applications. | Recent years have seen a dramatic increase in the number of systems building on so-called end-to-end'' learning. Generally speaking, this term refers to systems where the end goal of the machine learning process is directly predicted from raw inputs [e.g.][] lecun2005off,thomas2006cognitive . In the context of deep learning systems, the term now traditionally refers to architectures where, for example, there is no explicit encoding of hand-tuned features on the data, but the system directly predicts what the image, text, etc. is from the raw inputs @cite_5 @cite_4 @cite_8 @cite_23 @cite_20 . The context in which we use the term end-to-end is similar, but slightly more in line with its older usage: instead of (just) attempting to learn an output (with known and typically straightforward loss functions), we are specifically attempting to learn a model based upon an end-to-end that the user is ultimately trying to accomplish. We feel that this concept--of describing the entire closed-loop performance of the system as evaluated on the real task at hand--is beneficial to add to the notion of end-to-end learning. | {
"cite_N": [
"@cite_4",
"@cite_8",
"@cite_23",
"@cite_5",
"@cite_20"
],
"mid": [
"2194775991",
"1607307044",
"2102113734",
"1998042868",
"2949640717"
],
"abstract": [
"Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.",
"Full end-to-end text recognition in natural images is a challenging problem that has received much attention recently. Traditional systems in this area have relied on elaborate models incorporating carefully hand-engineered features or large amounts of prior knowledge. In this paper, we take a different route and combine the representational power of large, multilayer neural networks together with recent developments in unsupervised feature learning, which allows us to use a common framework to train highly-accurate text detector and character recognizer modules. Then, using only simple off-the-shelf methods, we integrate these two modules into a full end-to-end, lexicon-driven, scene text recognition system that achieves state-of-the-art performance on standard benchmarks, namely Street View Text and ICDAR 2003.",
"This paper presents a speech recognition system that directly transcribes audio data with text, without requiring an intermediate phonetic representation. The system is based on a combination of the deep bidirectional LSTM recurrent neural network architecture and the Connectionist Temporal Classification objective function. A modification to the objective function is introduced that trains the network to minimise the expectation of an arbitrary transcription loss function. This allows a direct optimisation of the word error rate, even in the absence of a lexicon or language model. The system achieves a word error rate of 27.3 on the Wall Street Journal corpus with no prior linguistic information, 21.9 with only a lexicon of allowed words, and 8.2 with a trigram language model. Combining the network with a baseline system further reduces the error rate to 6.7 .",
"This paper focuses on the problem of word detection and recognition in natural images. The problem is significantly more challenging than reading text in scanned documents, and has only recently gained attention from the computer vision community. Sub-components of the problem, such as text detection and cropped image word recognition, have been studied in isolation [7, 4, 20]. However, what is unclear is how these recent approaches contribute to solving the end-to-end problem of word recognition. We fill this gap by constructing and evaluating two systems. The first, representing the de facto state-of-the-art, is a two stage pipeline consisting of text detection followed by a leading OCR engine. The second is a system rooted in generic object recognition, an extension of our previous work in [20]. We show that the latter approach achieves superior performance. While scene text recognition has generally been treated with highly domain-specific methods, our results demonstrate the suitability of applying generic computer vision methods. Adopting this approach opens the door for real world scene text recognition to benefit from the rapid advances that have been taking place in object recognition.",
"We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech--two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, resulting in a 7x speedup over our previous system. Because of this efficiency, experiments that previously took weeks now run in days. This enables us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale."
]
} |
1703.04529 | 2949444583 | With the increasing popularity of machine learning techniques, it has become common to see prediction algorithms operating within some larger process. However, the criteria by which we train these algorithms often differ from the ultimate criteria on which we evaluate them. This paper proposes an end-to-end approach for learning probabilistic machine learning models in a manner that directly captures the ultimate task-based objective for which they will be used, within the context of stochastic programming. We present three experimental evaluations of the proposed approach: a classical inventory stock problem, a real-world electrical grid scheduling task, and a real-world energy storage arbitrage task. We show that the proposed approach can outperform both traditional modeling and purely black-box policy optimization approaches in these applications. | There has been a great deal of work in recent years on using machine learning procedures to optimize different loss criteria than those naturally'' optimized by the algorithm. For example, and propose methods for optimizing loss criteria in structured prediction that are from the inference procedure of the prediction algorithm; this work has also recently been extended to deep networks . Recent work has also explored using auxiliary prediction losses to satisfy multiple objectives @cite_22 , learning dynamics models that maximize control performance in Bayesian optimization @cite_15 , and learning adaptive predictive models via differentiation through a meta-learning optimization objective @cite_9 . | {
"cite_N": [
"@cite_9",
"@cite_15",
"@cite_22"
],
"mid": [
"2951775809",
"2953084784",
"2950872548"
],
"abstract": [
"We propose an algorithm for meta-learning that is model-agnostic, in the sense that it is compatible with any model trained with gradient descent and applicable to a variety of different learning problems, including classification, regression, and reinforcement learning. The goal of meta-learning is to train a model on a variety of learning tasks, such that it can solve new learning tasks using only a small number of training samples. In our approach, the parameters of the model are explicitly trained such that a small number of gradient steps with a small amount of training data from a new task will produce good generalization performance on that task. In effect, our method trains the model to be easy to fine-tune. We demonstrate that this approach leads to state-of-the-art performance on two few-shot image classification benchmarks, produces good results on few-shot regression, and accelerates fine-tuning for policy gradient reinforcement learning with neural network policies.",
"Real-world robots are becoming increasingly complex and commonly act in poorly understood environments where it is extremely challenging to model or learn their true dynamics. Therefore, it might be desirable to take a task-specific approach, wherein the focus is on explicitly learning the dynamics model which achieves the best control performance for the task at hand, rather than learning the true dynamics. In this work, we use Bayesian optimization in an active learning framework where a locally linear dynamics model is learned with the intent of maximizing the control performance, and used in conjunction with optimal control schemes to efficiently design a controller for a given task. This model is updated directly based on the performance observed in experiments on the physical system in an iterative manner until a desired performance is achieved. We demonstrate the efficacy of the proposed approach through simulations and real experiments on a quadrotor testbed.",
"Deep reinforcement learning agents have achieved state-of-the-art results by directly maximising cumulative reward. However, environments contain a much wider variety of possible training signals. In this paper, we introduce an agent that also maximises many other pseudo-reward functions simultaneously by reinforcement learning. All of these tasks share a common representation that, like unsupervised learning, continues to develop in the absence of extrinsic rewards. We also introduce a novel mechanism for focusing this representation upon extrinsic rewards, so that learning can rapidly adapt to the most relevant aspects of the actual task. Our agent significantly outperforms the previous state-of-the-art on Atari, averaging 880 expert human performance, and a challenging suite of first-person, three-dimensional tasks leading to a mean speedup in learning of 10 @math and averaging 87 expert human performance on Labyrinth."
]
} |
1703.04594 | 2950909246 | High-performance computing systems are more and more often based on accelerators. Computing applications targeting those systems often follow a host-driven approach in which hosts offload almost all compute-intensive sections of the code onto accelerators; this approach only marginally exploits the computational resources available on the host CPUs, limiting performance and energy efficiency. The obvious step forward is to run compute-intensive kernels in a concurrent and balanced way on both hosts and accelerators. In this paper we consider exactly this problem for a class of applications based on Lattice Boltzmann Methods, widely used in computational fluid-dynamics. Our goal is to develop just one program, portable and able to run efficiently on several different combinations of hosts and accelerators. To reach this goal, we define common data layouts enabling the code to exploit efficiently the different parallel and vector options of the various accelerators, and matching the possibly different requirements of the compute-bound and memory-bound kernels of the application. We also define models and metrics that predict the best partitioning of workloads among host and accelerator, and the optimally achievable overall performance level. We test the performance of our codes and their scaling properties using as testbeds HPC clusters incorporating different accelerators: Intel Xeon-Phi many-core processors, NVIDIA GPUs and AMD GPUs. | Over the years, LB codes have been written and optimized for large clusters of commodity CPUs ( @cite_18 ) and for application-specific machines ( @cite_22 @cite_5 @cite_31 ). More recent work has focused on exploiting the parallelism of powerful traditional many-core processors ( @cite_15 ), and of power-efficient accelerators such as GP-GPU ( @cite_4 @cite_11 @cite_1 ) and Xeon-Phi processors ( @cite_0 ), and even FPGAs ( @cite_26 ). | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_4",
"@cite_22",
"@cite_15",
"@cite_1",
"@cite_0",
"@cite_5",
"@cite_31",
"@cite_11"
],
"mid": [
"2097495737",
"2107076945",
"",
"1583193221",
"2009392786",
"",
"2077575837",
"",
"166372336",
""
],
"abstract": [
"Computationally intensive programs with moderate communication requirements such as CFD codes suffer from the standard slow interconnects of commodity \"off the shelf\" (COTS) hardware. We will introduce different large-scale applications of the Lattice Boltzmann Method (LBM) in fluid dynamics, material science, and chemical engineering and present results of the parallel performance on different architectures. It will be shown that a high speed communication network in combination with an efficient CPU is mandatory in order to achieve the required performance. An estimation of the necessary CPU count to meet the performance of 1 TFlop s will be given as well as a prediction as to which architecture is the most suitable for LBM. Finally, ratios of costs to application performance for tailored HPC systems and COTS architectures will be presented.",
"This paper presents an FPGA-based streaming computation for the lattice Boltzmann method (LBM) to simulate fluid flow with floating-point calculations. LBM is suitable for streaming computation because of its parallelism and regularity. We optimize the equations of LBM, and then formulate a streaming computation. To design an efficient data-path for throughput and hardware resource utilization, we introduce multiple cycle inputs and computing-unit sharing to the streaming data-path. The streaming accelerator implemented on a Virtex-4 FPGA with PCTExpress x8 interface achieves 2.93 and 2.46 times faster computation than a 3.4 GHz Pentium4 processor and a 2.2 GHz Opteron processor, respectively, for 2-dimensional time-dependent fluid dynamics problems.",
"",
"Computational experiments are one of the most used and flexible investigation tools in fluid dynamics. The Lattice Boltzmann Equation is a well established computational method particularly promising for multi-phase flows at micro and macro scales. Here we present preliminary results on performances of the LBE method on the Cell Broadband Engine platform.",
"Abstract Performances on recent processor architectures heavily rely on the ability of applications and compilers to exploit a more and more diverse and large set of parallel features. In this paper we focus on issues related to the efficient programming of multi-core processors based on the Sandy Bridge micro-architecture recently introduced by Intel. As a test-case application we use a D2Q37 Lattice Boltzmann algorithm, which accurately reproduces the thermo-hydrodynamics of a 2D-fluid obeying the equations of state of a perfect gas. The regular structure and the high degree of parallelism available in this class of applications make it relatively easy to exploit several processor features relevant for performance, such as, for example, the new Advanced Vector Extension (AVX) SIMD instructions set. However the main challenge is how to efficiently map the application onto the hardware structure of the processor. In this paper we present the implementation of our Lattice Boltzmann code on the Sandy Bridge processor, and assess the efficiency of several programming strategies and data-structure organizations, both in terms of memory access and computing performance. We also compare our results with that obtained on previous generation Intel processors, and with recent NVIDIA GP-GPU computing systems.",
"",
"Abstract In this paper we report on our early experience on porting, optimizing and benchmarking a Lattice Boltzmann (LB) code on the Xeon-Phi co-processor, the first generally available version of the new Many Integrated Core (MIC) architecture, developed by Intel. We consider as a test-bed a state-of-the-art LB model, that accurately reproduces the thermo-hydrodynamics of a 2D- fluid obeying the equations of state of a perfect gas. The regular structure of LB algorithms makes it relatively easy to identify a large degree of available parallelism. However, mapping a large fraction of this parallelism onto this new class of processors is not straightforward. The D2Q37 LB algorithm considered in this paper is an appropriate test-bed for this architecture since the critical computing kernels require high performances both in terms of memory bandwidth for sparse memory access patterns and number crunching capability. We describe our implementation of the code, that builds on previous experience made on other (simpler) many-core processors and GPUs, present benchmark results and measure performances, and finally compare with the results obtained by previous implementations developed on state-of-the-art classic multi-core CPUs and GP-GPUs.",
"",
"In this paper we describe an optimized implementation of a Lattice Boltzmann (LB) code on the BlueGene Q system, the latest generation massively parallel system of the BlueGene family. We consider a state-of-art LB code, that accurately reproduces the thermo-hydrodynamics of a 2D-fluid obeying the equations of state of a perfect gas. The regular structure of LB algorithms offers several levels of algorithmic parallelism that can be matched by a massively parallel computer architecture. However the complex memory access patterns associated to our LB model make it not trivial to efficiently exploit all available parallelism. We describe our implementation strategies, based on previous experience made on clusters of many-core processors and GPUs, present results and analyze and compare performances.",
""
]
} |
1703.04706 | 2950964646 | In the domain of sequence modelling, Recurrent Neural Networks (RNN) have been capable of achieving impressive results in a variety of application areas including visual question answering, part-of-speech tagging and machine translation. However this success in modelling short term dependencies has not successfully transitioned to application areas such as trajectory prediction, which require capturing both short term and long term relationships. In this paper, we propose a Tree Memory Network (TMN) for modelling long term and short term relationships in sequence-to-sequence mapping problems. The proposed network architecture is composed of an input module, controller and a memory module. In contrast to related literature, which models the memory as a sequence of historical states, we model the memory as a recursive tree structure. This structure more effectively captures temporal dependencies across both short term and long term sequences using its hierarchical structure. We demonstrate the effectiveness and flexibility of the proposed TMN in two practical problems, aircraft trajectory modelling and pedestrian trajectory modelling in a surveillance setting, and in both cases we outperform the current state-of-the-art. Furthermore, we perform an in depth analysis on the evolution of the memory module content over time and provide visual evidence on how the proposed TMN is able to map both long term and short term relationships efficiently via a hierarchical structure. | Deep learning models such as Recurrent Neural Networks (RNN) have been applied extensively for many sequence-to-sequence modelling problems and have been capable of producing state-of-the-art results. A number of approaches ( @cite_45 @cite_37 @cite_8 @cite_38 ) have also utilised what are termed memory modules'', to aid prediction. The memory stores important facts from historical inputs and then generates the future predictions based on the stored knowledge. A sample architecture with an input module, controller and an external memory is shown in Fig. . Firstly the input module generates a vector representation @math for the input @math at time instance @math . The controller then triggers a memory read operation. The memory module, with an attention process, searches the history and outputs relevant facts. The final output is generated by merging @math with the memory output. Finally, the controller triggers a memory update operation where the memory @math is updated with @math . | {
"cite_N": [
"@cite_38",
"@cite_37",
"@cite_45",
"@cite_8"
],
"mid": [
"2949626814",
"2131494463",
"",
"2951619830"
],
"abstract": [
"Despite the recent achievements in machine learning, we are still very far from achieving real artificial intelligence. In this paper, we discuss the limitations of standard deep learning approaches and show that some of these limitations can be overcome by learning how to grow the complexity of a model in a structured way. Specifically, we study the simplest sequence prediction problems that are beyond the scope of what is learnable with standard recurrent networks, algorithmically generated sequences which can only be learned by models which have the capacity to count and to memorize sequences. We show that some basic algorithms can be learned from sequential data using a recurrent network associated with a trainable memory.",
"Most tasks in natural language processing can be cast into question answering (QA) problems over language input. We introduce the dynamic memory network (DMN), a neural network architecture which processes input sequences and questions, forms episodic memories, and generates relevant answers. Questions trigger an iterative attention process which allows the model to condition its attention on the inputs and the result of previous iterations. These results are then reasoned over in a hierarchical recurrent sequence model to generate answers. The DMN can be trained end-to-end and obtains state-of-the-art results on several types of tasks and datasets: question answering (Facebook's bAbI dataset), text classification for sentiment analysis (Stanford Sentiment Treebank) and sequence modeling for part-of-speech tagging (WSJ-PTB). The training for these different tasks relies exclusively on trained word vector representations and input-question-answer triplets.",
"",
"We propose a method for automatically answering questions about images by bringing together recent advances from natural language processing and computer vision. We combine discrete reasoning with uncertain predictions by a multi-world approach that represents uncertainty about the perceived world in a bayesian framework. Our approach can handle human questions of high complexity about realistic scenes and replies with range of answer like counts, object classes, instances and lists of them. The system is directly trained from question-answer pairs. We establish a first benchmark for this task that can be seen as a modern attempt at a visual turing test."
]
} |
1703.04706 | 2950964646 | In the domain of sequence modelling, Recurrent Neural Networks (RNN) have been capable of achieving impressive results in a variety of application areas including visual question answering, part-of-speech tagging and machine translation. However this success in modelling short term dependencies has not successfully transitioned to application areas such as trajectory prediction, which require capturing both short term and long term relationships. In this paper, we propose a Tree Memory Network (TMN) for modelling long term and short term relationships in sequence-to-sequence mapping problems. The proposed network architecture is composed of an input module, controller and a memory module. In contrast to related literature, which models the memory as a sequence of historical states, we model the memory as a recursive tree structure. This structure more effectively captures temporal dependencies across both short term and long term sequences using its hierarchical structure. We demonstrate the effectiveness and flexibility of the proposed TMN in two practical problems, aircraft trajectory modelling and pedestrian trajectory modelling in a surveillance setting, and in both cases we outperform the current state-of-the-art. Furthermore, we perform an in depth analysis on the evolution of the memory module content over time and provide visual evidence on how the proposed TMN is able to map both long term and short term relationships efficiently via a hierarchical structure. | @cite_45 have utilised a memory module to improve the accuracy on natural language processing problems. Their proposed memory architecture is not fully extendible considering the usage of an offline feature engineering process using a bag-of-words approach. In similar works, such as @cite_38 and @cite_30 for image caption generation, and @cite_8 and @cite_41 for visual question answering, the authors have extensively applied the notion of external memory. The memory architecture, episodic memory'', proposed in @cite_37 has been shown to be capable of outperforming the other external memory architectures proposed above in terms of accuracy . | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_37",
"@cite_8",
"@cite_41",
"@cite_45"
],
"mid": [
"2173051530",
"2949626814",
"2131494463",
"2951619830",
"2122180654",
""
],
"abstract": [
"Learning an algorithm from examples is a fundamental problem that has been widely studied. Recently it has been addressed using neural networks, in particular by Neural Turing Machines (NTMs). These are fully differentiable computers that use backpropagation to learn their own programming. Despite their appeal NTMs have a weakness that is caused by their sequential nature: they are not parallel and are are hard to train due to their large depth when unfolded. We present a neural network architecture to address this problem: the Neural GPU. It is based on a type of convolutional gated recurrent unit and, like the NTM, is computationally universal. Unlike the NTM, the Neural GPU is highly parallel which makes it easier to train and efficient to run. An essential property of algorithms is their ability to handle inputs of arbitrary size. We show that the Neural GPU can be trained on short instances of an algorithmic task and successfully generalize to long instances. We verified it on a number of tasks including long addition and long multiplication of numbers represented in binary. We train the Neural GPU on numbers with upto 20 bits and observe no errors whatsoever while testing it, even on much longer numbers. To achieve these results we introduce a technique for training deep recurrent networks: parameter sharing relaxation. We also found a small amount of dropout and gradient noise to have a large positive effect on learning and generalization.",
"Despite the recent achievements in machine learning, we are still very far from achieving real artificial intelligence. In this paper, we discuss the limitations of standard deep learning approaches and show that some of these limitations can be overcome by learning how to grow the complexity of a model in a structured way. Specifically, we study the simplest sequence prediction problems that are beyond the scope of what is learnable with standard recurrent networks, algorithmically generated sequences which can only be learned by models which have the capacity to count and to memorize sequences. We show that some basic algorithms can be learned from sequential data using a recurrent network associated with a trainable memory.",
"Most tasks in natural language processing can be cast into question answering (QA) problems over language input. We introduce the dynamic memory network (DMN), a neural network architecture which processes input sequences and questions, forms episodic memories, and generates relevant answers. Questions trigger an iterative attention process which allows the model to condition its attention on the inputs and the result of previous iterations. These results are then reasoned over in a hierarchical recurrent sequence model to generate answers. The DMN can be trained end-to-end and obtains state-of-the-art results on several types of tasks and datasets: question answering (Facebook's bAbI dataset), text classification for sentiment analysis (Stanford Sentiment Treebank) and sequence modeling for part-of-speech tagging (WSJ-PTB). The training for these different tasks relies exclusively on trained word vector representations and input-question-answer triplets.",
"We propose a method for automatically answering questions about images by bringing together recent advances from natural language processing and computer vision. We combine discrete reasoning with uncertain predictions by a multi-world approach that represents uncertainty about the perceived world in a bayesian framework. Our approach can handle human questions of high complexity about realistic scenes and replies with range of answer like counts, object classes, instances and lists of them. The system is directly trained from question-answer pairs. We establish a first benchmark for this task that can be seen as a modern attempt at a visual turing test.",
"In this paper we explore the bi-directional mapping between images and their sentence-based descriptions. We propose learning this mapping using a recurrent neural network. Unlike previous approaches that map both sentences and images to a common embedding, we enable the generation of novel sentences given an image. Using the same model, we can also reconstruct the visual features associated with an image given its visual description. We use a novel recurrent visual memory that automatically learns to remember long-term visual concepts to aid in both sentence generation and visual feature reconstruction. We evaluate our approach on several tasks. These include sentence generation, sentence retrieval and image retrieval. State-of-the-art results are shown for the task of generating novel image descriptions. When compared to human generated captions, our automatically generated captions are preferred by humans over @math of the time. Results are better than or comparable to state-of-the-art results on the image and sentence retrieval tasks for methods using similar visual features.",
""
]
} |
1703.04706 | 2950964646 | In the domain of sequence modelling, Recurrent Neural Networks (RNN) have been capable of achieving impressive results in a variety of application areas including visual question answering, part-of-speech tagging and machine translation. However this success in modelling short term dependencies has not successfully transitioned to application areas such as trajectory prediction, which require capturing both short term and long term relationships. In this paper, we propose a Tree Memory Network (TMN) for modelling long term and short term relationships in sequence-to-sequence mapping problems. The proposed network architecture is composed of an input module, controller and a memory module. In contrast to related literature, which models the memory as a sequence of historical states, we model the memory as a recursive tree structure. This structure more effectively captures temporal dependencies across both short term and long term sequences using its hierarchical structure. We demonstrate the effectiveness and flexibility of the proposed TMN in two practical problems, aircraft trajectory modelling and pedestrian trajectory modelling in a surveillance setting, and in both cases we outperform the current state-of-the-art. Furthermore, we perform an in depth analysis on the evolution of the memory module content over time and provide visual evidence on how the proposed TMN is able to map both long term and short term relationships efficiently via a hierarchical structure. | Fig. (a) depicts the episodic memory model proposed in @cite_37 . The authors model the episodic memory'' as a hierarchical recurrent sequence model utilising the sequential nature of the memory. The authors propose a generalised neural sequential module with recurrent LSTM memory cells for sequence encoding, memory mechanism and response generation. The above work is further extended with a shared memory architecture in @cite_5 . Even with the exemplary results for short term dependency modelling problems, none of the above stated architectures are capable of handling sequences with long term relationships. | {
"cite_N": [
"@cite_5",
"@cite_37"
],
"mid": [
"2214429195",
"2131494463"
],
"abstract": [
"Deep neural networks have achieved impressive supervised classification performance in many tasks including image recognition, speech recognition, and sequence to sequence learning. However, this success has not been translated to applications like question answering that may involve complex arithmetic and logic reasoning. A major limitation of these models is in their inability to learn even simple arithmetic and logic operations. For example, it has been shown that neural networks fail to learn to add two binary numbers reliably. In this work, we propose Neural Programmer, an end-to-end differentiable neural network augmented with a small set of basic arithmetic and logic operations. Neural Programmer can call these augmented operations over several steps, thereby inducing compositional programs that are more complex than the built-in operations. The model learns from a weak supervision signal which is the result of execution of the correct program, hence it does not require expensive annotation of the correct program itself. The decisions of what operations to call, and what data segments to apply to are inferred by Neural Programmer. Such decisions, during training, are done in a differentiable fashion so that the entire network can be trained jointly by gradient descent. We find that training the model is difficult, but it can be greatly improved by adding random noise to the gradient. On a fairly complex synthetic table-comprehension dataset, traditional recurrent networks and attentional models perform poorly while Neural Programmer typically obtains nearly perfect accuracy.",
"Most tasks in natural language processing can be cast into question answering (QA) problems over language input. We introduce the dynamic memory network (DMN), a neural network architecture which processes input sequences and questions, forms episodic memories, and generates relevant answers. Questions trigger an iterative attention process which allows the model to condition its attention on the inputs and the result of previous iterations. These results are then reasoned over in a hierarchical recurrent sequence model to generate answers. The DMN can be trained end-to-end and obtains state-of-the-art results on several types of tasks and datasets: question answering (Facebook's bAbI dataset), text classification for sentiment analysis (Stanford Sentiment Treebank) and sequence modeling for part-of-speech tagging (WSJ-PTB). The training for these different tasks relies exclusively on trained word vector representations and input-question-answer triplets."
]
} |
1703.04706 | 2950964646 | In the domain of sequence modelling, Recurrent Neural Networks (RNN) have been capable of achieving impressive results in a variety of application areas including visual question answering, part-of-speech tagging and machine translation. However this success in modelling short term dependencies has not successfully transitioned to application areas such as trajectory prediction, which require capturing both short term and long term relationships. In this paper, we propose a Tree Memory Network (TMN) for modelling long term and short term relationships in sequence-to-sequence mapping problems. The proposed network architecture is composed of an input module, controller and a memory module. In contrast to related literature, which models the memory as a sequence of historical states, we model the memory as a recursive tree structure. This structure more effectively captures temporal dependencies across both short term and long term sequences using its hierarchical structure. We demonstrate the effectiveness and flexibility of the proposed TMN in two practical problems, aircraft trajectory modelling and pedestrian trajectory modelling in a surveillance setting, and in both cases we outperform the current state-of-the-art. Furthermore, we perform an in depth analysis on the evolution of the memory module content over time and provide visual evidence on how the proposed TMN is able to map both long term and short term relationships efficiently via a hierarchical structure. | In approaches such as the memory is a composed of a single layer of memory units. The memory update mechanism in can be written as, where @math is a score value that quantifies the relevance of the content of the memory module ( @math ) at time @math to the current context, @math (see (a)); where as in @cite_19 the authors completely update the content of the memory locations based on @math . | {
"cite_N": [
"@cite_19"
],
"mid": [
"2513651200"
],
"abstract": [
"We present a memory augmented neural network for natural language understanding: Neural Semantic Encoders (NSE). NSE has a variable sized encoding memory that evolves over time and maintains the understanding of input sequences through read, compose and write operations. NSE can access multiple and shared memories depending on the complexity of a task. We demonstrated the effectiveness and the flexibility of NSE on five different natural language tasks, natural language inference, question answering, sentence classification, document sentiment analysis and machine translation where NSE achieved state-of-the-art performance when evaluated on publically available benchmarks. For example, our shared-memory model showed an encouraging result on neural machine translation, improving an attention-based baseline by approximately 1.0 BLEU."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.