aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1607.04149 | 2482076478 | In a combinatorial auction with item bidding, agents participate in multiple single-item second-price auctions at once. As some items might be substitutes, agents need to strategize in order to maximize their utilities. A number of results indicate that high welfare can be achieved this way, giving bounds on the welfare at equilibrium. Recently, however, criticism has been raised that equilibria are hard to compute and therefore unlikely to be attained. In this paper, we take a different perspective. We study simple best-response dynamics. That is, agents are activated one after the other and each activated agent updates his strategy myopically to a best response against the other agents' current strategies. Often these dynamics may take exponentially long before they converge or they may not converge at all. However, as we show, convergence is not even necessary for good welfare guarantees. Given that agents' bid updates are aggressive enough but not too aggressive, the game will remain in states of good welfare after each agent has updated his bid at least once. In more detail, we show that if agents have fractionally subadditive valuations, natural dynamics reach and remain in a state that provides a @math approximation to the optimal welfare after each agent has updated his bid at least once. For subadditive valuations, we can guarantee an @math approximation in case of @math items that applies after each agent has updated his bid at least once and at any point after that. The latter bound is complemented by a negative result, showing that no kind of best-response dynamics can guarantee more than an @math fraction of the optimal social welfare. | Lately, attempts at proving Price of Anarchy bounds for combinatorial auctions with item bidding have been criticized for not being constructive, in the sense that the computational complexity of finding an equilibrium remained open. @cite_19 , for example, showed that for subadditive valuations computing a pure Nash equilibrium requires exponential communication. Regarding fractionally subadditive valuations they concluded that if there exists an efficient algorithm that finds an equilibrium, it must use techniques that are very different from our current ones.'' Further negative findings were reported by Cai and Papadimitriou @cite_8 , who showed that computing a Bayes-Nash equilibrium is @math -hard. | {
"cite_N": [
"@cite_19",
"@cite_8"
],
"mid": [
"2949060750",
"2019385392"
],
"abstract": [
"We study combinatorial auctions where each item is sold separately but simultaneously via a second price auction. We ask whether it is possible to efficiently compute in this game a pure Nash equilibrium with social welfare close to the optimal one. We show that when the valuations of the bidders are submodular, in many interesting settings (e.g., constant number of bidders, budget additive bidders) computing an equilibrium with good welfare is essentially as easy as computing, completely ignoring incentives issues, an allocation with good welfare. On the other hand, for subadditive valuations, we show that computing an equilibrium requires exponential communication. Finally, for XOS (a.k.a. fractionally subadditive) valuations, we show that if there exists an efficient algorithm that finds an equilibrium, it must use techniques that are very different from our current ones.",
"Bayesian equilibria of simultaneous auctions for individual items have been explored recently [ 2008; Bhawalkar and Roughgarden 2011; 2011; 2013] as an alternative to the well-known complexity issues plaguing combinatorial auctions with incomplete information, and some strong positive results have been shown about their performance. We point out some very serious complexity obstacles to this approach: Computing a Bayesian equilibrium in such auctions is hard for PP --- a complexity class between the polynomial hierarchy and PSPACE --- and even finding an approximate such equilibrium is as hard as NP, for some small approximation ratio (additive or multiplicative); therefore, the assumption that such equilibria will be arrived at by rational agents is quite problematic. In fact, even recognizing a Bayesian Nash equilibrium is intractable. Furthermore, these results hold even if bidder valuations are quite benign: Only one bidder valuation in our construction is unit demand or monotone submodular, while all others are additive. We also explore the possibility of favorable price of anarchy results for no-regret dynamics of the Bayesian simultaneous auctions game, and identify complexity obstacles there as well."
]
} |
1607.04149 | 2482076478 | In a combinatorial auction with item bidding, agents participate in multiple single-item second-price auctions at once. As some items might be substitutes, agents need to strategize in order to maximize their utilities. A number of results indicate that high welfare can be achieved this way, giving bounds on the welfare at equilibrium. Recently, however, criticism has been raised that equilibria are hard to compute and therefore unlikely to be attained. In this paper, we take a different perspective. We study simple best-response dynamics. That is, agents are activated one after the other and each activated agent updates his strategy myopically to a best response against the other agents' current strategies. Often these dynamics may take exponentially long before they converge or they may not converge at all. However, as we show, convergence is not even necessary for good welfare guarantees. Given that agents' bid updates are aggressive enough but not too aggressive, the game will remain in states of good welfare after each agent has updated his bid at least once. In more detail, we show that if agents have fractionally subadditive valuations, natural dynamics reach and remain in a state that provides a @math approximation to the optimal welfare after each agent has updated his bid at least once. For subadditive valuations, we can guarantee an @math approximation in case of @math items that applies after each agent has updated his bid at least once and at any point after that. The latter bound is complemented by a negative result, showing that no kind of best-response dynamics can guarantee more than an @math fraction of the optimal social welfare. | Most recently, Daskalakis and Syrgkanis @cite_18 considered coarse correlated equilibria. They showed that even for unit-demand players (a strict subclass of submodular) there are no polynomial-time no-regret learning algorithms for finding such equilibria, unless @math , closing the last gap in the equilibrium landscape. However, they also proposed a novel solution concept to escape the hardness trap, no-envy learning, and gave a polynomial-time no-envy learning algorithm for XOS valuations and complemented this with a proof showing that for this class of valuations every no-envy outcome recovers at least @math of the optimal social welfare. | {
"cite_N": [
"@cite_18"
],
"mid": [
"2226238611"
],
"abstract": [
"A line of recent work provides welfare guarantees of simple combinatorial auction formats, such as selling m items via simultaneous second price auctions (SiSPAs) ( 2008, Bhawalkar and Roughgarden 2011, 2013). These guarantees hold even when the auctions are repeatedly executed and players use no-regret learning algorithms. Unfortunately, off-the-shelf no-regret algorithms for these auctions are computationally inefficient as the number of actions is exponential. We show that this obstacle is insurmountable: there are no polynomial-time no-regret algorithms for SiSPAs, unless RP @math NP, even when the bidders are unit-demand. Our lower bound raises the question of how good outcomes polynomially-bounded bidders may discover in such auctions. To answer this question, we propose a novel concept of learning in auctions, termed \"no-envy learning.\" This notion is founded upon Walrasian equilibrium, and we show that it is both efficiently implementable and results in approximately optimal welfare, even when the bidders have fractionally subadditive (XOS) valuations (assuming demand oracles) or coverage valuations (without demand oracles). No-envy learning outcomes are a relaxation of no-regret outcomes, which maintain their approximate welfare optimality while endowing them with computational tractability. Our results extend to other auction formats that have been studied in the literature via the smoothness paradigm. Our results for XOS valuations are enabled by a novel Follow-The-Perturbed-Leader algorithm for settings where the number of experts is infinite, and the payoff function of the learner is non-linear. This algorithm has applications outside of auction settings, such as in security games. Our result for coverage valuations is based on a novel use of convex rounding schemes and a reduction to online convex optimization."
]
} |
1607.04149 | 2482076478 | In a combinatorial auction with item bidding, agents participate in multiple single-item second-price auctions at once. As some items might be substitutes, agents need to strategize in order to maximize their utilities. A number of results indicate that high welfare can be achieved this way, giving bounds on the welfare at equilibrium. Recently, however, criticism has been raised that equilibria are hard to compute and therefore unlikely to be attained. In this paper, we take a different perspective. We study simple best-response dynamics. That is, agents are activated one after the other and each activated agent updates his strategy myopically to a best response against the other agents' current strategies. Often these dynamics may take exponentially long before they converge or they may not converge at all. However, as we show, convergence is not even necessary for good welfare guarantees. Given that agents' bid updates are aggressive enough but not too aggressive, the game will remain in states of good welfare after each agent has updated his bid at least once. In more detail, we show that if agents have fractionally subadditive valuations, natural dynamics reach and remain in a state that provides a @math approximation to the optimal welfare after each agent has updated his bid at least once. For subadditive valuations, we can guarantee an @math approximation in case of @math items that applies after each agent has updated his bid at least once and at any point after that. The latter bound is complemented by a negative result, showing that no kind of best-response dynamics can guarantee more than an @math fraction of the optimal social welfare. | Further relevant work comes from @cite_11 , who proposed an alternative to simultaneous second-price auctions, the so-called single-bid auction. This mechanism also admits a polynomial-time no-regret learning algorithm and by a result of @cite_17 achieves optimal Price of Anarchy bounds within a broader class of mechanisms. | {
"cite_N": [
"@cite_17",
"@cite_11"
],
"mid": [
"2950920780",
"2064287473"
],
"abstract": [
"We study the communication complexity of combinatorial auctions via interpolation mechanisms that interpolate between non-truthful and truthful protocols. Specifically, an interpolation mechanism has two phases. In the first phase, the bidders participate in some non-truthful protocol whose output is itself a truthful protocol. In the second phase, the bidders participate in the truthful protocol selected during phase one. Note that virtually all existing auctions have either a non-existent first phase (and are therefore truthful mechanisms), or a non-existent second phase (and are therefore just traditional protocols, analyzed via the Price of Anarchy Stability). The goal of this paper is to understand the benefits of interpolation mechanisms versus truthful mechanisms or traditional protocols, and develop the necessary tools to formally study them. Interestingly, we exhibit settings where interpolation mechanisms greatly outperform the optimal traditional and truthful protocols. Yet, we also exhibit settings where interpolation mechanisms are provably no better than truthful ones. Finally, we apply our new machinery to prove that the recent single-bid mechanism of Devanur et. al. DevanurMSW15 (the only pre-existing interpolation mechanism in the literature) achieves the optimal price of anarchy among a wide class of protocols, a claim that simply can't be addressed by appealing just to machinery from communication complexity or the study of truthful mechanisms.",
"We introduce single-bid auctions as a new format for combinatorial auctions. In single-bid auctions, each bidder submits a single real-valued bid for the right to buy items at a fixed price. Contrary to other simple auction formats, such as simultaneous or sequential single-item auctions, bidders can implement no-regret learning strategies for single-bid auctions in polynomial time. Price of anarchy bounds for correlated equilibria concepts in single-bid auctions therefore have more bite than their counterparts for auctions and equilibria for which learning is not known to be computationally tractable (or worse, known to be computationally intractable [Cai and Papadimitriou 2014; 2015] this end, we show that for any subadditive valuations the social welfare at equilibrium is an O(log m)-approximation to the optimal social welfare, where @math is the number of items. We also provide tighter approximation results for several subclasses. Our welfare guarantees hold for Nash equilibria and no-regret learning outcomes in both Bayesian and complete information settings via the smooth-mechanism framework. Of independent interest, our techniques show that in a combinatorial auction setting, efficiency guarantees of a mechanism via smoothness for a very restricted class of cardinality valuations extend, with a small degradation, to subadditive valuations, the largest complement-free class of valuations."
]
} |
1607.04149 | 2482076478 | In a combinatorial auction with item bidding, agents participate in multiple single-item second-price auctions at once. As some items might be substitutes, agents need to strategize in order to maximize their utilities. A number of results indicate that high welfare can be achieved this way, giving bounds on the welfare at equilibrium. Recently, however, criticism has been raised that equilibria are hard to compute and therefore unlikely to be attained. In this paper, we take a different perspective. We study simple best-response dynamics. That is, agents are activated one after the other and each activated agent updates his strategy myopically to a best response against the other agents' current strategies. Often these dynamics may take exponentially long before they converge or they may not converge at all. However, as we show, convergence is not even necessary for good welfare guarantees. Given that agents' bid updates are aggressive enough but not too aggressive, the game will remain in states of good welfare after each agent has updated his bid at least once. In more detail, we show that if agents have fractionally subadditive valuations, natural dynamics reach and remain in a state that provides a @math approximation to the optimal welfare after each agent has updated his bid at least once. For subadditive valuations, we can guarantee an @math approximation in case of @math items that applies after each agent has updated his bid at least once and at any point after that. The latter bound is complemented by a negative result, showing that no kind of best-response dynamics can guarantee more than an @math fraction of the optimal social welfare. | A final point of reference are truthful mechanisms for combinatorial auctions. While no mechanism can achieve a better than @math approximation for submodular valuations with valuation queries alone @cite_5 , Dobzinski @cite_22 recently managed to improve a long-standing approximation guarantee of @math for submodular valuations to @math for fractionally subadditive valuations, requiring access to both value and demand oracles. | {
"cite_N": [
"@cite_5",
"@cite_22"
],
"mid": [
"2333413293",
"2951660602"
],
"abstract": [
"A long-standing open question in algorithmic mechanism design is whether there exist computationally efficient truthful mechanisms for combinatorial auctions, with performance guarantees close to those possible without considerations of truthfulness. In this article, we answer this question negatively: the requirement of truthfulness can impact dramatically the ability of a mechanism to achieve a good approximation ratio for combinatorial auctions. More precisely, we show that every universally truthful randomized mechanism for combinatorial auctions with submodular valuations that approximates optimal social welfare within a factor of m1 2−e must use exponentially many value queries, where m is the number of items. Furthermore, we show that there exists a class of succinctly represented submodular valuation functions, for which the existence of a universally truthful polynomial-time mechanism that provides an m1 2−e-approximation would imply NP e RP. In contrast, ignoring truthfulness, there exist constant-factor approximation algorithms for this problem, and ignoring computational efficiency, the VCG mechanism is truthful and provides optimal social welfare. These are the first hardness results for truthful polynomial-time mechanisms for any type of combinatorial auctions, even for deterministic mechanisms. Our approach is based on a novel direct hardness technique that completely skips the notoriously hard step of characterizing truthful mechanisms. The characterization step was the main obstacle for proving impossibility results in algorithmic mechanism design so far.",
"We study a central problem in Algorithmic Mechanism Design: constructing truthful mechanisms for welfare maximization in combinatorial auctions with submodular bidders. Dobzinski, Nisan, and Schapira provided the first mechanism that guarantees a non-trivial approximation ratio of @math [STOC'06], where @math is the number of items. This was subsequently improved to @math [Dobzinski, APPROX'07] and then to @math [Krysta and Vocking, ICALP'12]. In this paper we develop the first mechanism that breaks the logarithmic barrier. Specifically, the mechanism provides an approximation ratio of @math . Similarly to previous constructions, our mechanism uses polynomially many value and demand queries, and in fact provides the same approximation ratio for the larger class of XOS (a.k.a. fractionally subadditive) valuations. We also develop a computationally efficient implementation of the mechanism for combinatorial auctions with budget additive bidders. Although in general computing a demand query is NP-hard for budget additive valuations, we observe that the specific form of demand queries that our mechanism uses can be efficiently computed when bidders are budget additive."
]
} |
1607.04342 | 2950593712 | GitHub is the largest source code repository in the world. It provides a git-based source code management platform and also many features inspired by social networks. For example, GitHub users can show appreciation to projects by adding stars to them. Therefore, the number of stars of a repository is a direct measure of its popularity. In this paper, we use multiple linear regressions to predict the number of stars of GitHub repositories. These predictions are useful both to repository owners and clients, who usually want to know how their projects are performing in a competitive open source development market. In a large-scale analysis, we show that the proposed models start to provide accurate predictions after being trained with the number of stars received in the last six months. Furthermore, specific models---generated using data from repositories that share the same growth trends---are recommended for repositories with slow growth and or for repositories with less stars. Finally, we evaluate the ability to predict not the number of stars of a repository but its rank among the GitHub repositories. We found a very strong correlation between predicted and real rankings (Spearman's rho greater than 0.95). | Our work was inspired by the vast literature on defect prediction. For example, a systematic literature review listed 208 defect prediction studies @cite_5 , which differ regarding the software metrics used for prediction, the modeling technique, the granularity of the independent variable, and the validation technique. As independent variables, the studies use source code metrics (size, cohesion, coupling, etc), change metrics, process metrics, code smells instances, etc. The modeling techniques vary with respect to linear regression, logistic regression, naive bayes, neural networks, etc. In this paper, instead of predicting the future number of defects of a system, we rely on multiple linear regressions to predict the number of stars of GitHub repositories. | {
"cite_N": [
"@cite_5"
],
"mid": [
"2151666086"
],
"abstract": [
"Background: The accurate prediction of where faults are likely to occur in code can help direct test effort, reduce costs, and improve the quality of software. Objective: We investigate how the context of models, the independent variables used, and the modeling techniques applied influence the performance of fault prediction models. Method: We used a systematic literature review to identify 208 fault prediction studies published from January 2000 to December 2010. We synthesize the quantitative and qualitative results of 36 studies which report sufficient contextual and methodological information according to the criteria we develop and apply. Results: The models that perform well tend to be based on simple modeling techniques such as Naive Bayes or Logistic Regression. Combinations of independent variables have been used by models that perform well. Feature selection has been applied to these combinations when models are performing particularly well. Conclusion: The methodology used to build models seems to be influential to predictive performance. Although there are a set of fault prediction studies in which confidence is possible, more studies are needed that use a reliable methodology and which report their context, methodology, and performance comprehensively."
]
} |
1607.04342 | 2950593712 | GitHub is the largest source code repository in the world. It provides a git-based source code management platform and also many features inspired by social networks. For example, GitHub users can show appreciation to projects by adding stars to them. Therefore, the number of stars of a repository is a direct measure of its popularity. In this paper, we use multiple linear regressions to predict the number of stars of GitHub repositories. These predictions are useful both to repository owners and clients, who usually want to know how their projects are performing in a competitive open source development market. In a large-scale analysis, we show that the proposed models start to provide accurate predictions after being trained with the number of stars received in the last six months. Furthermore, specific models---generated using data from repositories that share the same growth trends---are recommended for repositories with slow growth and or for repositories with less stars. Finally, we evaluate the ability to predict not the number of stars of a repository but its rank among the GitHub repositories. We found a very strong correlation between predicted and real rankings (Spearman's rho greater than 0.95). | @cite_30 record time-series information about popular Google Play apps and investigate how release frequency can affect an app's performance, as measured by rating, popularity and number of user reviews. They label as impactful releases'' the ones that caused a significant change on the app's popularity, as inferred by Causal Impact Analysis (a form of causal inference). They report that more mentions of features and fewer mentions of bug fixing increase the chance for a release to be impactful. @cite_10 follow a similar approach but to identify causal relationships between changes in internal measures of software quality (coupling, cohesion, complexity, etc) and the number of defects reported for a system. | {
"cite_N": [
"@cite_30",
"@cite_10"
],
"mid": [
"2546780486",
"2144746916"
],
"abstract": [
"App developers would like to understand the impact of their own and their competitors’ software releases. To address this we introduce Causal Impact Release Analysis for app stores, and our tool, CIRA, that implements this analysis. We mined 38,858 popular Google Play apps, over a period of 12 months. For these apps, we identified 26,339 releases for which there was adequate prior and posterior time series data to facilitate causal impact analysis. We found that 33 of these releases caused a statistically significant change in user ratings. We use our approach to reveal important characteristics that distinguish causal significance in Google Play. To explore the actionability of causal impact analysis, we elicited the opinions of app developers: 56 companies responded, 78 concurred with the causal assessment, of which 33 claimed that their company would consider changing its app release strategy as a result of our findings.",
"Abstract In this paper, we propose a defect prediction approach centered on more robust evidences towards causality between source code metrics (as predictors) and the occurrence of defects. More specifically, we rely on the Granger causality test to evaluate whether past variations in source code metrics values can be used to forecast changes in time series of defects. Our approach triggers alarms when changes made to the source code of a target system have a high chance of producing defects. We evaluated our approach in several life stages of four Java-based systems. We reached an average precision greater than 50 in three out of the four systems we evaluated. Moreover, by comparing our approach with baselines that are not based on causality tests, it achieved a better precision."
]
} |
1607.04342 | 2950593712 | GitHub is the largest source code repository in the world. It provides a git-based source code management platform and also many features inspired by social networks. For example, GitHub users can show appreciation to projects by adding stars to them. Therefore, the number of stars of a repository is a direct measure of its popularity. In this paper, we use multiple linear regressions to predict the number of stars of GitHub repositories. These predictions are useful both to repository owners and clients, who usually want to know how their projects are performing in a competitive open source development market. In a large-scale analysis, we show that the proposed models start to provide accurate predictions after being trained with the number of stars received in the last six months. Furthermore, specific models---generated using data from repositories that share the same growth trends---are recommended for repositories with slow growth and or for repositories with less stars. Finally, we evaluate the ability to predict not the number of stars of a repository but its rank among the GitHub repositories. We found a very strong correlation between predicted and real rankings (Spearman's rho greater than 0.95). | Several other studies examine the relationship between popularity of mobile apps and their code properties @cite_14 @cite_32 @cite_11 @cite_7 @cite_24 @cite_0 @cite_2 @cite_12 . investigate 28 factors along eight dimensions to understand how high-rated Android applications are different from low-rated ones @cite_15 . Their result shows that external factors, like number of promotional images, are the most influential factors. examine the relationship between the number of ad libraries and app's user ratings @cite_13 . They show that there is no relationship between the number of ad libraries in an app and its rating. analyse changes of Android API elements between releases and report that high app churn leads to lower user ratings @cite_1 . Linares-V 'a investigate how the fault- and change-proneness of Android API elements relate to applications' lack of success @cite_21 . They state that making heavy use of fault- and change-prone APIs can negatively impact the success of apps. | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_21",
"@cite_1",
"@cite_32",
"@cite_24",
"@cite_0",
"@cite_2",
"@cite_15",
"@cite_13",
"@cite_12",
"@cite_11"
],
"mid": [
"2056755818",
"",
"2154221125",
"2069442545",
"",
"",
"",
"",
"2162563790",
"2022990508",
"",
""
],
"abstract": [
"Given the enormity of the app market and the velocity with which new apps arrive, it is extremely challenging for apps to reach the intended audience, or any audience at all, in fact. An entire \"app promotion\" industry exists to help publishers achieve post-launch app success. This is done by understanding relationships between app success and a variety of app attributes (like category, price, etc.). In this paper, we study a dimension not addressed thus far - the timing of app launch. Specifically, we study a large data set to uncover relationships between app launch times and its subsequent commercial success, or lack thereof. A number of interesting findings are revealed in this study. Users are generally less price-sensitive around holiday seasons, especially around Christmas and New Year's, in stark contrast, they are extra price sensitive during weekends. Specifically, more expensive apps released on weekends tend to get a higher negative word-of-mouth (review valence) rating. In addition, our results indicate that apps released in the latter part of the week tend to fare better than do apps released earlier. Furthermore, Thursday is the optimal day to release an app when considering review sentiments. Finally, it does tend to get a higher number of reviews on weekends.",
"",
"During the recent years, the market of mobile software applications (apps) has maintained an impressive upward trajectory. Many small and large software development companies invest considerable resources to target available opportunities. As of today, the markets for such devices feature over 850K+ apps for Android and 900K+ for iOS. Availability, cost, functionality, and usability are just some factors that determine the success or lack of success for a given app. Among the other factors, reliability is an important criteria: users easily get frustrated by repeated failures, crashes, and other bugs; hence, abandoning some apps in favor of others. This paper reports a study analyzing how the fault- and change-proneness of APIs used by 7,097 (free) Android apps relates to applications' lack of success, estimated from user ratings. Results of this study provide important insights into a crucial issue: making heavy use of fault- and change-prone APIs can negatively impact the success of these apps.",
"User review is a crucial component of open mobile app markets such as the Google Play Store. How do we automatically summarize millions of user reviews and make sense out of them? Unfortunately, beyond simple summaries such as histograms of user ratings, there are few analytic tools that can provide insights into user reviews. In this paper, we propose Wiscom, a system that can analyze tens of millions user ratings and comments in mobile app markets at three different levels of detail. Our system is able to (a) discover inconsistencies in reviews; (b) identify reasons why users like or dislike a given app, and provide an interactive, zoomable view of how users' reviews evolve over time; and (c) provide valuable insights into the entire app market, identifying users' major concerns and preferences of different types of apps. Results using our techniques are reported on a 32GB dataset consisting of over 13 million user reviews of 171,493 Android apps in the Google Play Store. We discuss how the techniques presented herein can be deployed to help a mobile app market operator such as Google as well as individual app developers and end-users.",
"",
"",
"",
"",
"The tremendous rate of growth in the mobile app market over the past few years has attracted many developers to build mobile apps. However, while there is no shortage of stories of how lone developers have made great fortunes from their apps, the majority of developers are struggling to break even. For those struggling developers, knowing the “DNA” (i.e., characteristics) of high-rated apps is the first step towards successful development and evolution of their apps. In this paper, we investigate 28 factors along eight dimensions to understand how high-rated apps are different from low-rated apps. We also investigate what are the most influential factors by applying a random-forest classifier to identify high-rated apps. Through a case study on 1,492 high-rated and low-rated free apps mined from the Google Play store, we find that high-rated apps are statistically significantly different in 17 out of the 28 factors that we considered. Our experiment also shows that the size of an app, the number of promotional images that the app displays on its web store page, and the target SDK version of an app are the most influential factors.",
"One of the most popular ways to monetize a free app is by including advertisements in the app. Several advertising (ad) companies provide these ads to app developers through ad libraries that need to be integrated in the app. However, the demand for ads far exceeds the supply. This obstacle may lead app developers to integrate several ad libraries from different ad companies in their app to ensure they receive an ad with each request. However, no study has explored how many ad libraries are commonly integrated into apps. Additionally, no research to date has examined whether integrating many different ad libraries impacts an app's ratings. This article examines these two issues by empirically examining thousands of Android apps. The authors find that there are apps with as many as 28 ad libraries, but they find no evidence that the number of ad libraries in an app is related to its possible rating in the app store. However, integrating certain ad libraries can negatively impact an app's rating.",
"",
""
]
} |
1607.04318 | 2953119031 | In severe outbreaks such as Ebola, bird flu and SARS, people share news, and their thoughts and responses regarding the outbreaks on social media. Understanding how people perceive the severe outbreaks, what their responses are, and what factors affect these responses become important. In this paper, we conduct a comprehensive study of understanding and mining the spread of Ebola-related information on social media. In particular, we (i) conduct a large-scale data-driven analysis of geotagged social media messages to understand citizen reactions regarding Ebola; (ii) build information propagation models which measure locality of information; and (iii) analyze spatial, temporal and social properties of Ebola-related information. Our work provides new insights into Ebola outbreak by understanding citizen reactions and topic-based information propagation, as well as providing a foundation for analysis and response of future public health crises. | Ebola virus became a very serious problem in the world, researchers began studying Ebola-related information on social media. @cite_2 collected 1,217 Ebola-related images posted on Instagram and Flickr, and grouped the images by 9 themes. @cite_4 collected 2,155 tweets containing a hashtag #CDCchat, which were posted from the Centers for Disease Control and some Twitter users, and then found 8 topics from the tweets. @cite_11 analyzed how the frequency of Ebola-related tweets is correlated with the frequency of searches on Google. @cite_14 analyzed Google search queries related to the Ebola outbreak to understand the correlation between the number of web searches and the number of Ebola cases. | {
"cite_N": [
"@cite_14",
"@cite_11",
"@cite_4",
"@cite_2"
],
"mid": [
"2193397352",
"2134624036",
"2101612261",
"2136181417"
],
"abstract": [
"Background The 2014 Ebola epidemic in West Africa has attracted public interest worldwide, leading to millions of Ebola-related Internet searches being performed during the period of the epidemic. This study aimed to evaluate and interpret Google search queries for terms related to the Ebola outbreak both at the global level and in all countries where primary cases of Ebola occurred. The study also endeavoured to look at the correlation between the number of overall and weekly web searches and the number of overall and weekly new cases of Ebola.",
"In October 2014, during heightened news coverage about cases of Ebola in the USA, anecdotal observations suggested that many Americans were anxious about Ebola. Given the negligible risk of infection, their anxiety was arguably driven by perceived rather than actual risk. Exaggeration or reassurance from the media can infl ame or subdue people’s perceived risk of Ebola infection. Fear can also be acquired by observation of other people’s experiences, as expressed on social media. Thus, social media amplifi ed fear about the imported Ebola case. As discussed in The Lancet Editorial (Nov 8, 2014), Twitter traffi c shows an imbalance across the digital divide; there were more tweets about Ebola in the USA, where transmission was contained, than in Guinea, Liberia, and Sierra Leone, where there was and remains a continuing epidemic. Despite the risk to most Americans being negligible, many people expressed anxiety. The figure shows how worldwide traffi c on Twitter and Google about Ebola increased as news spread about the fi rst US case and how they compare with influenza (flu)-related searches and tweets. Similar peaks were observed when other news about Ebola was released. In a random sample of tweets, we observed that the frequency of Ebola-related tweets associated with negative emotions, anxiety, anger, swearing, and death, as well as discrepant thinking (eg, shouldn’t), were higher than those associated with infl uenza (see fi gure in appendix). Twitter data can provide public health practitioners with a quantitative indicator of anxiety, anger, or negative emotions in the general public where Twitter penetration is high. This indicator could help to alleviate anxiety and correctly communicate the risk associated with Ebola.",
"A diagnosis of Ebola on US soil triggered widespread panic. In response, the Centers for Disease Control and Prevention held a live Twitter chat to address public concerns. This study applied a textual analytics method to reveal insights from these tweets that can inform communication strategies. User-generated tweets were collected, sorted, and analyzed to reveal major themes. The public was concerned with symptoms and lifespan of the virus, disease transfer and contraction, safe travel, and protection of one's body.",
"Abstract Objective Social media have strongly influenced awareness and perceptions of public health emergencies, but a considerable amount of social media content is now carried through images, rather than just text. This study's objective is to explore how image-sharing platforms are used for information dissemination in public health emergencies. Study design Retrospective review of images posted on two popular image-sharing platforms to characterize public discourse about Ebola. Methods Using the keyword ‘#ebola’ we identified a 1 sample of images posted on Instagram and Flickr across two sequential weeks in November 2014. Images from both platforms were independently coded by two reviewers and characterized by themes. We reviewed 1217 images posted on Instagram and Flickr and identified themes. Results Nine distinct themes were identified. These included: images of health care workers and professionals [308 (25 )], West Africa [75 (6 )], the Ebola virus [59 (5 )], and artistic renderings of Ebola [64 (5 )]. Also identified were images with accompanying embedded text related to Ebola and associated: facts [68 (6 )], fears [40 (3 )], politics [46 (4 )], and jokes [284 (23 )]. Several [273 (22 )] images were unrelated to Ebola or its sequelae. Instagram images were primarily coded as jokes [255 (42 )] or unrelated [219 (36 )], while Flickr images primarily depicted health care workers and other professionals [281 (46 )] providing care or other services for prevention or treatment. Conclusion Image sharing platforms are being used for information exchange about public health crises, like Ebola. Use differs by platform and discerning these differences can help inform future uses for health care professionals and researchers seeking to assess public fears and misinformation or provide targeted education awareness interventions."
]
} |
1607.04315 | 2513651200 | 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. | One of the pioneering work that attempts to extend deep neural networks with an external memory is Neural Turing Machines (NTM) @cite_32 . NTM implements a centralized controller and a fixed-sized random access memory. The NTM memory is addressable by both content (i.e. soft attention) and location based access mechanisms. The authors evaluated NTM on algorithmic tasks such as copying and sorting sequences. | {
"cite_N": [
"@cite_32"
],
"mid": [
"2950527759"
],
"abstract": [
"We extend the capabilities of neural networks by coupling them to external memory resources, which they can interact with by attentional processes. The combined system is analogous to a Turing Machine or Von Neumann architecture but is differentiable end-to-end, allowing it to be efficiently trained with gradient descent. Preliminary results demonstrate that Neural Turing Machines can infer simple algorithms such as copying, sorting, and associative recall from input and output examples."
]
} |
1607.03881 | 2474757114 | Inspired by the work of [Kempe, Kleinberg, Oren, Slivkins, EC13] we introduce and analyze a model on opinion formation; the update rule of our dynamics is a simplified version of that of Kempe et. al. We assume that the population is partitioned into types whose interaction pattern is specified by a graph. Interaction leads to population mass moving from types of smaller mass to those of bigger. We show that starting uniformly at random over all population vectors on the simplex, our dynamics converges point-wise with probability one to an independent set. This settles an open problem of Kempe et. al., as applicable to our dynamics. We believe that our techniques can be used to settle the open problem for the Kempe et. al. dynamics as well. Next, we extend the model of Kempe et. al. by introducing the notion of birth and death of types, with the interaction graph evolving appropriately. Birth of types is determined by a Bernoulli process and types die when their population mass is less than a parameter @math . We show that if the births are infrequent, then there are long periods of "stability" in which there is no population mass that moves. Finally we show that even if births are frequent and "stability" is not attained, the total number of types does not explode: it remains logarithmic in @math . | Other works, including dynamical systems that show convergence to fixed points, are @cite_9 @cite_20 @cite_21 @cite_12 @cite_7 @cite_28 . @cite_28 focuses on quadratic dynamics and they show convergence in the limit. On the other hand @cite_0 shows that sampling from the distribution this dynamics induces at a given time step is PSPACE-complete. In @cite_9 @cite_20 , it is shown that replicator dynamics in linear congestion and 2-player coordination games converges to pure Nash equilibria, and in @cite_21 @cite_12 it is shown that gradient descent converges to local minima, avoiding saddle points even in the case where the fixed points are uncountably many. | {
"cite_N": [
"@cite_7",
"@cite_28",
"@cite_9",
"@cite_21",
"@cite_0",
"@cite_20",
"@cite_12"
],
"mid": [
"2342774186",
"2145818164",
"2493056695",
"2474090883",
"1995326443",
"1857453345",
"2467441529"
],
"abstract": [
"A new approach to understanding evolution [Val09], namely viewing it through the lens of computation, has already started yielding new insights, e.g., natural selection under sexual reproduction can be interpreted as the Multiplicative Weight Update (MWU) Algorithm in coordination games played among genes [CLPV14]. Using this machinery, we study the role of mutation in changing environments in the presence of sexual reproduction. Following [WVA05], we model changing environments via a Markov chain, with the states representing environments, each with its own fitness matrix. In this setting, we show that in the absence of mutation, the population goes extinct, but in the presence of mutation, the population survives with positive probability. On the way to proving the above theorem, we need to establish some facts about dynamics in games. We provide the first, to our knowledge, polynomial convergence bound for noisy MWU in a coordination game. Finally, we also show that in static environments, sexual evolution with mutation converges, for any level of mutation.",
"The paper promotes the study of computational aspects, primarily the convergence rate, of nonlinear dynamical systems from a combinatorial perspective. The authors identify the class of symmetric quadratic systems. Such systems have been widely used to model phenomena in the natural sciences, and also provide an appropriate framework for the study of genetic algorithms in combinatorial optimisation. They prove several fundamental general properties of these systems, notably that every trajectory converges to a fixed point. They go on to give a detailed analysis of a quadratic system defined in a natural way on probability distributions over the set of matchings in a graph. In particular, they prove that convergence to the limit requires only polynomial time when the graph is a tree. This result demonstrates that such systems, though nonlinear, are amenable to quantitative analysis. >",
"What does it mean to fully understand the behavior of a network of adaptive agents? The golden standard typically is the behavior of learning dynamics in potential games, where many evolutionary dynamics, e.g., replicator dynamics, are known to converge to sets of equilibria. Even in such classic settings many questions remain unanswered. We examine issues such as: Point-wise convergence: Does the system always equilibrate, even in the presence of continuums of equilibria? Computing regions of attraction: Given point-wise convergence can we compute the region of asymptotic stability of each equilibrium (e.g., estimate its volume, geometry)? System invariants: Invariant functions remain constant along every system trajectory. This notion is orthogonal to the game theoretic concept of a potential function, which always strictly increases decreases along system trajectories. Do dynamics in potential games exhibit invariant functions? If so, how many? How do these functions look like? Based on these geometric characterizations, we propose a novel quantitative framework for analyzing the efficiency of potential games with many equilibria. The predictions of different equilibria are weighted by their probability to arise under evolutionary dynamics given uniformly random initial conditions. This average case analysis is shown to offer novel insights in classic game theoretic challenges, including quantifying the risk dominance in stag-hunt games and allowing for more nuanced performance analysis in networked coordination and congestion games with large gaps between price of stability and price of anarchy. CCS Concepts: rTheory of computation! Solution concepts in game theory; Convergence and learning in games;",
"",
"Quadratic Dynamical Systems (QDS), whose definition extends that of Markov chains, are used to model phenomena in a variety of fields like statistical physics and natural evolution. Such systems also play a role in genetic algorithms, a widelyused class of heuristics that are notoriously hard to analyze. Recently took an important step in the study of QDS’s by showing, under some technical assumptions, that such systems converge to a stationary distribution (similar theorems for Markov Chains are well-known). We show, however, that the following sampling problem for QDS’s is PSPACE-hard: Given an initial distribution, produce a random sample from the t’th generation. The hardness result continues to hold for very restricted classes of QDS’s with very simple initial distributions, thus suggesting that QDS’s are intrinsically more complicated than Markov chains. ∗Supported by an IBM Graduate Fellowship and partly under NSF grant CCR-9310214. Email: arora@cs.berkeley.edu. †Work done while at ICSI, Berkeley, and supported in part by a Rothschild postdoctoral fellowship. Email: rabani@theory.lcs.mit.edu. ‡Supported by NSF grant CCR-9310214. Email: vazirani@cs.berkeley.edu.",
"In a recent series of papers a surprisingly strong connection was discovered between standard evolutionary models of natural selection and Multiplicative Weights Updates Algorithm, a ubiquitous model of online learning and optimization. These papers establish that, under specific assumptions, mathematical models of biological evolution can be reduced to studying discrete replicator dynamics, a close variant of MWUA, in coordination games. This connection allows for introducing insights from game theoretic dynamics into the field of mathematical biology. Using these results as a stepping stone, we show that mathematical models of haploid evolution imply the extinction of genetic diversity in the long term limit, a widely believed conjecture in genetics. In game theoretic terms we show that in the case of coordination games, under minimal genericity assumptions, discrete replicator dynamics converge to pure Nash equilibria for all but a zero measure of initial conditions. This result holds despite the fact that mixed Nash equilibria can be exponentially (or even uncountably) many, completely dominating in number the set of pure Nash equilibria. Thus, in haploid organisms the long term preservation of genetic diversity needs to be safeguarded by other evolutionary mechanisms such as mutations and speciation.",
"Given a non-convex twice differentiable cost function f , we prove that the set of initial conditions so that gradient descent converges to saddle points where ∇f has at least one strictly negative eigenvalue has (Lebesgue) measure zero, even for cost functions f with non-isolated critical points, answering an open question in [12]. Moreover, this result extends to forward-invariant convex subspaces, allowing for weak (non-globally Lipschitz) smoothness assumptions. Finally, we produce an upper bound on the allowable step-size."
]
} |
1607.03777 | 2952266267 | Provable stable arbitrary order symmetric interior penalty discontinuous Galerkin (SIP) discretisations of variable viscosity, incompressible Stokes flow utilising @math -- @math elements and hierarchical Legendre basis polynomials are developed and investigated.For solving the resulting linear system, a block preconditioned iterative method is proposed. The nested viscous problem is solved by a @math -multilevel preconditioned Krylov subspace method. For the @math -coarsening, a twolevel method utilising element-block Jacobi preconditioned iterations as a smoother is employed. Piecewise bilinear ( @math ) and piecewise constant ( @math ) @math -coarse spaces are considered. Finally, Galerkin @math -coarsening is proposed and investigated for the two @math -coarse spaces considered. Through a number of numerical experiments, we demonstrate that utilising the @math coarse space results in the most robust @math -multigrid method for variable viscosity Stokes flow. Using this @math coarse space we observe that the convergence of the overall Stokes solver appears to be robust with respect to the jump in the viscosity and only mildly depending on the polynomial order @math . It is demonstrated and supported by theoretical results that the convergence of the SIP discretisations and the iterative methods rely on a sharp choice of the penalty parameter based on local values of the viscosity. | Inf-sup stable discontinuous Galerkin (DG) methods of the interior penalty type for the Stokes equations can be constructed by using the tensor product element pairs @math -- @math , and @math -- @math @cite_23 @cite_39 , as well as the @math -conforming Raviart-Thomas, Brezzi-Douglas-Marini, and Brezzi-Douglas-Fortin-Marini kind element pairs; see @cite_12 @cite_3 @cite_48 @cite_27 . | {
"cite_N": [
"@cite_48",
"@cite_3",
"@cite_39",
"@cite_27",
"@cite_23",
"@cite_12"
],
"mid": [
"1963859447",
"",
"",
"2225768769",
"1982359839",
"2157023312"
],
"abstract": [
"In this paper we construct Discontinuous Galerkin approximations of the Stokes problem where the velocity field is @math H ( div , Ω ) -conforming. This implies that the velocity solution is divergence-free in the whole domain. This property can be exploited to design a simple and effective preconditioner for the final linear system.",
"",
"",
"A multigrid method for the Stokes system discretized with an Hdiv-conforming discontinuous Galerkin method is presented. It acts on the combined velocity and pressure spaces and thus does not need a Schur complement approximation. The smoothers used are of overlapping Schwarz type and employ a local Helmholtz decomposition. Additionally, we use the fact that the discretization provides nested divergence free subspaces. We present the convergence analysis and numerical evidence that convergence rates are not only independent of mesh size, but also reasonably small.",
"We propose and analyze a discontinuous Galerkin approximation for the Stokes problem. The finite element triangulation employed is not required to be conforming and we use discontinuous pressures and velocities. No additional unknown fields need to be introduced, but only suitable bilinear forms defined on the interfaces between the elements, involving the jumps of the velocity and the average of the pressure. We consider hp approximations using ℚk′–ℚk velocity-pressure pairs with k′ = k + 2, k + 1, k. Our methods show better stability properties than the corresponding conforming ones. We prove that our first two choices of velocity spaces ensure uniform divergence stability with respect to the mesh size h. Numerical results show that they are uniformly stable with respect to the local polynomial degree k, a property that has no analog in the conforming case. An explicit bound in k which is not sharp is also proven. Numerical results show that if equal order approximation is chosen for the velocity and pressure, no spurious pressure modes are present but the method is not uniformly stable either with respect to h or k. We derive a priori error estimates generalizing the abstract theory of mixed methods. Optimal error estimates in h are proven. As for discontinuous Galerkin methods for scalar diffusive problems, half of the power of k is lost for p and hp pproximations independently of the divergence stability.",
"In this paper, the authors present two formulations for the Stokes problem which make use of the existing @math elements of the Raviart-Thomas type originally developed for the second-order elliptic problems. In addition, two new @math elements are constructed and analyzed particularly for the new formulations. Optimal-order error estimates are established for the corresponding finite element solutions in various Sobolev norms. The finite element solutions feature a full satisfaction of the continuity equation when existing Raviart-Thomas-type elements are employed in the numerical scheme."
]
} |
1607.03777 | 2952266267 | Provable stable arbitrary order symmetric interior penalty discontinuous Galerkin (SIP) discretisations of variable viscosity, incompressible Stokes flow utilising @math -- @math elements and hierarchical Legendre basis polynomials are developed and investigated.For solving the resulting linear system, a block preconditioned iterative method is proposed. The nested viscous problem is solved by a @math -multilevel preconditioned Krylov subspace method. For the @math -coarsening, a twolevel method utilising element-block Jacobi preconditioned iterations as a smoother is employed. Piecewise bilinear ( @math ) and piecewise constant ( @math ) @math -coarse spaces are considered. Finally, Galerkin @math -coarsening is proposed and investigated for the two @math -coarse spaces considered. Through a number of numerical experiments, we demonstrate that utilising the @math coarse space results in the most robust @math -multigrid method for variable viscosity Stokes flow. Using this @math coarse space we observe that the convergence of the overall Stokes solver appears to be robust with respect to the jump in the viscosity and only mildly depending on the polynomial order @math . It is demonstrated and supported by theoretical results that the convergence of the SIP discretisations and the iterative methods rely on a sharp choice of the penalty parameter based on local values of the viscosity. | Regarding the solution of the equation system arising from symmetric interior penalty DG (SIP) @cite_14 @cite_26 based discretisations of incompressible Stokes flow, we note that for @math -conforming discretisations, efficient preconditioners have been introduced very recently @cite_48 @cite_27 . Recent advances in developing efficient and robust solvers for interior penalty DG discretisations of second order elliptic problems with heterogeneous coefficients involve the algebraic multigrid preconditioner proposed in @cite_47 @cite_42 , as well as the twolevel methods proposed in @cite_24 and @cite_36 @cite_6 . | {
"cite_N": [
"@cite_14",
"@cite_26",
"@cite_36",
"@cite_48",
"@cite_42",
"@cite_6",
"@cite_24",
"@cite_27",
"@cite_47"
],
"mid": [
"1991817777",
"2130350269",
"",
"1963859447",
"",
"2067941796",
"2022811246",
"2225768769",
"2027582000"
],
"abstract": [
"A continuous interior penalty hp-finite element method that penalizes the jump of the discrete solution across mesh interfaces is introduced. Error estimates are obtained for first-order and advection-dominated transport operators. The analysis relies on three technical results that are of independent interest: an hp- inverse trace inequality, a local discontinuous to continuous hp-interpolation result, and hp-error estimates for continuous L2-orthogonal projections.",
"We provide a framework for the analysis of a large class of discontinuous methods for second-order elliptic problems. It allows for the understanding and comparison of most of the discontinuous Galerkin methods that have been proposed over the past three decades for the numerical treatment of elliptic problems.",
"",
"In this paper we construct Discontinuous Galerkin approximations of the Stokes problem where the velocity field is @math H ( div , Ω ) -conforming. This implies that the velocity solution is divergence-free in the whole domain. This property can be exploited to design a simple and effective preconditioner for the final linear system.",
"",
"We consider the behavior of the GMRES method for solving a linear system @math when @math is singular or nearly so, i.e., ill conditioned. The (near) singularity of @math may or may not affect the performance of GMRES, depending on the nature of the system and the initial approximate solution. For singular @math , we give conditions under which the GMRES iterates converge safely to a least-squares solution or to the pseudoinverse solution. These results also apply to any residual minimizing Krylov subspace method that is mathematically equivalent to GMRES. A practical procedure is outlined for efficiently and reliably detecting singularity or ill conditioning when it becomes a threat to the performance of GMRES.",
"This paper reviews some known and proposes some new preconditioning methods for a number of discontinuous Galerkin (or DG) finite element approximations for elliptic problems of second order. Nested hierarchy of meshes is generally assumed. Our approach utilizes a general two-level scheme, where the finite element space for the DG method is decomposed into a subspace (viewed as an auxiliary or ‘coarse’ space), plus a correction which can be handled by a standard smoothing procedure. We consider three different auxiliary subspaces, namely, piecewise linear C0-conforming functions, piecewise linear functions that are continuous at the centroids of the edges faces (Crouzeix-Raviart finite elements) and piecewise constant functions over the finite elements. To support the theoretical results, we also present numerical experiments for 3-D model problem showing uniform convergence of the constructed methods. Copyright © 2006 John Wiley & Sons, Ltd.",
"A multigrid method for the Stokes system discretized with an Hdiv-conforming discontinuous Galerkin method is presented. It acts on the combined velocity and pressure spaces and thus does not need a Schur complement approximation. The smoothers used are of overlapping Schwarz type and employ a local Helmholtz decomposition. Additionally, we use the fact that the discretization provides nested divergence free subspaces. We present the convergence analysis and numerical evidence that convergence rates are not only independent of mesh size, but also reasonably small.",
"Abstract The diapiric ascent of light rocks through a denser oberburden is simulated using a new numerical method based in part on finite differences. Rocks are modelled as viscous Newtonian liquids and the internal interfaces are located with the help of passive markers. These markers are relocated using a new algorithm equivalent to a cellular automaton rule. This allows the markers to be redistributed very rapidly. This new method appears to be well adapted to quickly resolving geological problems. Simulations of the growth of a single dome are performed under various viscosity ratios for the salt and the overburden. The effect of asymmetrical initial deformations of the salt layer is to produce diapiric structures that slowly return to symmetry as the diapir matures and rises in an isotropic medium. Simulations involving continuous sedimentation of the overburden are also performed. An initial geometry on a basin margin in which multiple diapirs eventually develop is studied. Diapirs grow relatively rapidly near the centre, while on the margins the salt layer flows upward following the basement. Simulations with salt that is more viscous than sediments show the development of previously unnoticed asymmetrical structures."
]
} |
1607.03791 | 2492967369 | The Weighted Tree Augmentation Problem (WTAP) is a fundamental well-studied problem in the field of network design. Given an undirected tree @math , an additional set of edges @math disjoint from @math called , and a cost vector @math , WTAP asks to find a minimum-cost set @math with the property that @math is @math -edge connected. The special case where @math for all @math is called the Tree Augmentation Problem (TAP). Both problems are known to be NP-hard. For the class of bounded cost vectors, we present a first improved approximation algorithm for WTAP since more than three decades. Concretely, for any @math and @math we present an LP based @math -approximation for WTAP restricted to cost vectors @math in @math for @math . For the special case of TAP we improve this factor to @math . Our results rely on a new LP, that significantly differs from existing LPs achieving improved bounds for TAP. We round a fractional solution in two phases. The first phase uses the fractional solution to decompose the tree and its fractional solution into so-called @math -simple pairs losing only an @math -factor in the objective function. We then show how to use the additional constraints in our LP combined with the @math -simple structure to round a fractional solution in each part of the decomposition. | The techniques used to achieve the latter improved algorithms for TAP are combinatorial in nature, and seem very hard to modify for an improved approximation of WTAP. In an effort to improve the approximation factor for WTAP, several algorithm with approximation factors better than @math have recently been developed for TAP, that are based on continuous relaxations of the problem. Along these lines, Kortsarz and Nutov @cite_8 recently showed an LP-based @math -approximation algorithm. Our @math -approximation for TAP improves on that factor for LP-based algorithms. Cheriyan and Gao @cite_12 presented an approximation with respect to a semidefinite program (SDP) obtained from Lasserre tightening of an LP relaxation with factor @math . A strong point of both papers compared to our result is that the mathematical program is used in the analysis, while the algorithms themselves are combinatorial. | {
"cite_N": [
"@cite_12",
"@cite_8"
],
"mid": [
"2963550906",
"2541541317"
],
"abstract": [
"In Part II, we study the unweighted tree augmentation problem (TAP) via the Lasserre (sum of squares) system. We prove that the integrality ratio of an SDP relaxation (the Lasserre tightening of an LP relaxation) is ( 3 2 + ), where ( >0 ) can be any small constant. We obtain this result by designing a polynomial-time algorithm for TAP that achieves an approximation guarantee of ( ( 3 2 + )) relative to the SDP relaxation. The algorithm is combinatorial and does not solve the SDP relaxation, but our analysis relies on the SDP relaxation. We generalize the combinatorial analysis of integral solutions from the previous literature to fractional solutions by identifying some properties of fractional solutions of the Lasserre system via the decomposition result of (Integer Programming and Combinatorial Optimization, LNCS, volume 6655, Springer, pp 301–314, 2011).",
"In the Tree Augmentation Problem (TAP) the goal is to augment a tree T by a minimum size edge set F from a given edge set E such that T+F is 2-edge-connected. The best approximation ratio known for TAP is 1.5. In the more general Weighted TAP problem, F should be of minimum weight. Weighted TAP admits several 2-approximation algorithms w.r.t. the standard cut-LP relaxation. The problem is equivalent to the problem of covering a laminar set family. Laminar set families play an important role in the design of approximation algorithms for connectivity network design problems. In fact, Weighted TAP is the simplest connectivity network design problem for which a ratio better than 2 is not known. Improving this \"natural\" ratio is a major open problem, which may have implications on many other network design problems. It seems that achieving this goal requires finding an LP-relaxation with integrality gap better than 2, which is an old open problem even for TAP. In this paper we introduce two different LP-relaxations, and for each of them give a simple algorithm that computes a feasible solution for TAP of size at most 7 4 times the optimal LP value. This gives some hope to break the ratio 2 for the weighted case."
]
} |
1607.03791 | 2492967369 | The Weighted Tree Augmentation Problem (WTAP) is a fundamental well-studied problem in the field of network design. Given an undirected tree @math , an additional set of edges @math disjoint from @math called , and a cost vector @math , WTAP asks to find a minimum-cost set @math with the property that @math is @math -edge connected. The special case where @math for all @math is called the Tree Augmentation Problem (TAP). Both problems are known to be NP-hard. For the class of bounded cost vectors, we present a first improved approximation algorithm for WTAP since more than three decades. Concretely, for any @math and @math we present an LP based @math -approximation for WTAP restricted to cost vectors @math in @math for @math . For the special case of TAP we improve this factor to @math . Our results rely on a new LP, that significantly differs from existing LPs achieving improved bounds for TAP. We round a fractional solution in two phases. The first phase uses the fractional solution to decompose the tree and its fractional solution into so-called @math -simple pairs losing only an @math -factor in the objective function. We then show how to use the additional constraints in our LP combined with the @math -simple structure to round a fractional solution in each part of the decomposition. | The bundle LP differs from all existing LPs that were recently proved to have integrality gap better than @math . Indeed, most existing such LPs include, on top of the constraints of the natural LP, constraints that exploit structural properties of feasible, or optimal solutions restricted to very specific structures, such as etc. (see @cite_8 for formal definitions of these and other related notions). In contrast, as we discussed before, the bundle LP attempts to uniformly cut off fractional solutions that have low cost due to sufficiently simple obstructions which have large integrality gap. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2541541317"
],
"abstract": [
"In the Tree Augmentation Problem (TAP) the goal is to augment a tree T by a minimum size edge set F from a given edge set E such that T+F is 2-edge-connected. The best approximation ratio known for TAP is 1.5. In the more general Weighted TAP problem, F should be of minimum weight. Weighted TAP admits several 2-approximation algorithms w.r.t. the standard cut-LP relaxation. The problem is equivalent to the problem of covering a laminar set family. Laminar set families play an important role in the design of approximation algorithms for connectivity network design problems. In fact, Weighted TAP is the simplest connectivity network design problem for which a ratio better than 2 is not known. Improving this \"natural\" ratio is a major open problem, which may have implications on many other network design problems. It seems that achieving this goal requires finding an LP-relaxation with integrality gap better than 2, which is an old open problem even for TAP. In this paper we introduce two different LP-relaxations, and for each of them give a simple algorithm that computes a feasible solution for TAP of size at most 7 4 times the optimal LP value. This gives some hope to break the ratio 2 for the weighted case."
]
} |
1607.03791 | 2492967369 | The Weighted Tree Augmentation Problem (WTAP) is a fundamental well-studied problem in the field of network design. Given an undirected tree @math , an additional set of edges @math disjoint from @math called , and a cost vector @math , WTAP asks to find a minimum-cost set @math with the property that @math is @math -edge connected. The special case where @math for all @math is called the Tree Augmentation Problem (TAP). Both problems are known to be NP-hard. For the class of bounded cost vectors, we present a first improved approximation algorithm for WTAP since more than three decades. Concretely, for any @math and @math we present an LP based @math -approximation for WTAP restricted to cost vectors @math in @math for @math . For the special case of TAP we improve this factor to @math . Our results rely on a new LP, that significantly differs from existing LPs achieving improved bounds for TAP. We round a fractional solution in two phases. The first phase uses the fractional solution to decompose the tree and its fractional solution into so-called @math -simple pairs losing only an @math -factor in the objective function. We then show how to use the additional constraints in our LP combined with the @math -simple structure to round a fractional solution in each part of the decomposition. | WTAP can also be interpreted as a problem of covering a laminar family with point-to-point links (see e.g. @cite_8 ). This observation implies that the problem of augmenting a @math -edge connected graph to a @math -edge connected graph can be reduced to WTAP, whenever @math is odd, due to the laminar structure of the family of minimum cuts, in this case. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2541541317"
],
"abstract": [
"In the Tree Augmentation Problem (TAP) the goal is to augment a tree T by a minimum size edge set F from a given edge set E such that T+F is 2-edge-connected. The best approximation ratio known for TAP is 1.5. In the more general Weighted TAP problem, F should be of minimum weight. Weighted TAP admits several 2-approximation algorithms w.r.t. the standard cut-LP relaxation. The problem is equivalent to the problem of covering a laminar set family. Laminar set families play an important role in the design of approximation algorithms for connectivity network design problems. In fact, Weighted TAP is the simplest connectivity network design problem for which a ratio better than 2 is not known. Improving this \"natural\" ratio is a major open problem, which may have implications on many other network design problems. It seems that achieving this goal requires finding an LP-relaxation with integrality gap better than 2, which is an old open problem even for TAP. In this paper we introduce two different LP-relaxations, and for each of them give a simple algorithm that computes a feasible solution for TAP of size at most 7 4 times the optimal LP value. This gives some hope to break the ratio 2 for the weighted case."
]
} |
1607.03914 | 2502626776 | We introduce multiforce, a force-directed layout for multiplex networks, where the nodes of the network are organized into multiple layers and both in-layer and inter-layer relationships among nodes are used to compute node coordinates. The proposed approach generalizes existing work, providing a range of intermediate layouts in-between the ones produced by known methods. Our experiments on real data show that multiforce can keep nodes well aligned across different layers without significantly affecting their internal layouts when the layers have similar or compatible topologies. As a consequence, multiforce enriches the benefits of force-directed layouts by also supporting the identification of topological correspondences between layers. | Many layouts have been designed to visualize monoplex networks. Here we briefly review the ones that are more relevant for our approach. For an extensive review, the reader may consult @cite_32 . | {
"cite_N": [
"@cite_32"
],
"mid": [
"2158453355"
],
"abstract": [
"The analysis of large graphs plays a prominent role in various fields of research and is relevant in many important application areas. Effective visual analysis of graphs requires appropriate visual presentations in combination with respective user interaction facilities and algorithmic graph analysis methods. How to design appropriate graph analysis systems depends on many factors, including the type of graph describing the data, the analytical task at hand and the applicability of graph analysis methods. The most recent surveys of graph visualization and navigation techniques cover techniques that had been introduced until 2000 or concentrate only on graph layouts published until 2002. Recently, new techniques have been developed covering a broader range of graph types, such as timevarying graphs. Also, in accordance with ever growing amounts of graph-structured data becoming available, the inclusion of algorithmic graph analysis and interaction techniques becomes increasingly important. In this State-of-the-Art Report, we survey available techniques for the visual analysis of large graphs. Our review first considers graph visualization techniques according to the type of graphs supported. The visualization techniques form the basis for the presentation of interaction approaches suitable for visual graph exploration. As an important component of visual graph analysis, we discuss various graph algorithmic aspects useful for the different stages of the visual graph analysis process. We also present main open research challenges in this field."
]
} |
1607.03914 | 2502626776 | We introduce multiforce, a force-directed layout for multiplex networks, where the nodes of the network are organized into multiple layers and both in-layer and inter-layer relationships among nodes are used to compute node coordinates. The proposed approach generalizes existing work, providing a range of intermediate layouts in-between the ones produced by known methods. Our experiments on real data show that multiforce can keep nodes well aligned across different layers without significantly affecting their internal layouts when the layers have similar or compatible topologies. As a consequence, multiforce enriches the benefits of force-directed layouts by also supporting the identification of topological correspondences between layers. | @cite_38 proposed a force-directed layout to change the position of nodes so that more graph-theoretically central nodes pushed towards the centre of the diagram. In this algorithm, an additional force called gravity is used to change the position of more central nodes. For each node @math in a graph @math the position of the node is influenced by the following force: where @math and @math are respectively repulsive and attractive forces, and @math is the gravity force, measured as: In this equation @math is the mass of node @math , which can be set according to the node degree, @math is the position of @math , @math is the gravitational parameter and @math is the centroid of all nodes. Notice that forces in the equations above are vectors. | {
"cite_N": [
"@cite_38"
],
"mid": [
"2952899446"
],
"abstract": [
"Force-directed layout algorithms produce graph drawings by resolving a system of emulated physical forces. We present techniques for using social gravity as an additional force in force-directed layouts, together with a scaling technique, to produce drawings of trees and forests, as well as more complex social networks. Social gravity assigns mass to vertices in proportion to their network centrality, which allows vertices that are more graph-theoretically central to be visualized in physically central locations. Scaling varies the gravitational force throughout the simulation, and reduces crossings relative to unscaled gravity. In addition to providing this algorithmic framework, we apply our algorithms to social networks produced by Mark Lombardi, and we show how social gravity can be incorporated into force-directed Lombardi-style drawings."
]
} |
1607.03914 | 2502626776 | We introduce multiforce, a force-directed layout for multiplex networks, where the nodes of the network are organized into multiple layers and both in-layer and inter-layer relationships among nodes are used to compute node coordinates. The proposed approach generalizes existing work, providing a range of intermediate layouts in-between the ones produced by known methods. Our experiments on real data show that multiforce can keep nodes well aligned across different layers without significantly affecting their internal layouts when the layers have similar or compatible topologies. As a consequence, multiforce enriches the benefits of force-directed layouts by also supporting the identification of topological correspondences between layers. | An extreme case of indirect methods, that we mention for completeness, consists in not visualizing nodes and edges at all but only indirect network properties, such as the degree of the nodes in the different layers or other summary measures @cite_3 @cite_28 @cite_2 . These approaches are complementary to graph drawing, and can also be used in combination with our proposal. Our method belongs to the slicing class, and is different from existing approaches because it allows a balancing of the effects of in-layer and inter-layer relationships. | {
"cite_N": [
"@cite_28",
"@cite_3",
"@cite_2"
],
"mid": [
"2963312008",
"2096822173",
"1559591149"
],
"abstract": [
"In this article we discuss visualisation strategies for multiplex networks. Since Moreno's early works on network analysis, visualisation has been one of the main ways to understand networks thanks ...",
"Multilayer relationships among entities and information about entities must be accompanied by the means to analyse, visualize and obtain insights from such data. We present open-source software (muxViz) that contains a collection of algorithms for the analysis of multilayer networks, which are an important way to represent a large variety of complex systems throughout science and engineering. We demonstrate the ability of muxViz to analyse and interactively visualize multilayer data using empirical genetic, neuronal and transportation networks. Our software is available at https: github.com manlius muxViz.",
"Recent advances in network science allows the modeling and analysis of complex inter-related entities. These entities often interact with each other in a number of different ways. Simple graphs fail to capture these multiple types of relationships requiring more sophisticated mathematical structures. One such structure is multigraph, where entities (or nodes) can be linked to each other through multiple edges. In this paper we describe a new method to manage multiple types of relationships existing in multigraphs. Our approach is based on the concept of pair of nodes (edges) and, in particular, we study how nodes on different layers interact which each other considering the edges they share. We propose a two level strategy that summarizes global local multigraph features. The global view helps us to gain knowledge related to the characteristics of layers and how they interact while the local view provides an analysis of individual layers highlighting edge properties such as cluster structure. Our proposal is complementary to standard node-link diagram and it can be coupled with such techniques in order to intelligently explore multigraphs. The proposed visualization is tested on a real world case study and the outcomes point out the ability of our proposal to discover patterns present in the data."
]
} |
1607.03516 | 2950790587 | In this paper, we propose a novel unsupervised domain adaptation algorithm based on deep learning for visual object recognition. Specifically, we design a new model called Deep Reconstruction-Classification Network (DRCN), which jointly learns a shared encoding representation for two tasks: i) supervised classification of labeled source data, and ii) unsupervised reconstruction of unlabeled target data.In this way, the learnt representation not only preserves discriminability, but also encodes useful information from the target domain. Our new DRCN model can be optimized by using backpropagation similarly as the standard neural networks. We evaluate the performance of DRCN on a series of cross-domain object recognition tasks, where DRCN provides a considerable improvement (up to 8 in accuracy) over the prior state-of-the-art algorithms. Interestingly, we also observe that the reconstruction pipeline of DRCN transforms images from the source domain into images whose appearance resembles the target dataset. This suggests that DRCN's performance is due to constructing a single composite representation that encodes information about both the structure of target images and the classification of source images. Finally, we provide a formal analysis to justify the algorithm's objective in domain adaptation context. | Domain adaptation is a large field of research, with related work under several names such as class imbalance @cite_14 , covariate shift @cite_46 , and sample selection bias @cite_18 . In @cite_47 , it is considered as a special case of transfer learning. Earlier work on domain adaptation focused on text document analysis and NLP @cite_27 @cite_13 . In recent years, it has gained a lot of attention in the computer vision community, mainly for object recognition application, see @cite_33 and references therein. The domain adaptation problem is often referred to as in computer vision @cite_0 . | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_33",
"@cite_46",
"@cite_0",
"@cite_27",
"@cite_47",
"@cite_13"
],
"mid": [
"",
"1941659294",
"1982696459",
"2034368206",
"",
"2158108973",
"2165698076",
"2120354757"
],
"abstract": [
"",
"In machine learning problems, differences in prior class probabilities -- or class imbalances -- have been reported to hinder the performance of some standard classifiers, such as decision trees. This paper presents a systematic study aimed at answering three different questions. First, we attempt to understand the nature of the class imbalance problem by establishing a relationship between concept complexity, size of the training set and class imbalance level. Second, we discuss several basic re-sampling or cost-modifying methods previously proposed to deal with the class imbalance problem and compare their effectiveness. The results obtained by such methods on artificial domains are linked to results in real-world domains. Finally, we investigate the assumption that the class imbalance problem does not only affect decision tree systems but also affects other classification systems such as Neural Networks and Support Vector Machines.",
"In pattern recognition and computer vision, one is often faced with scenarios where the training data used to learn a model have different distribution from the data on which the model is applied. Regardless of the cause, any distributional change that occurs after learning a classifier can degrade its performance at test time. Domain adaptation tries to mitigate this degradation. In this article, we provide a survey of domain adaptation methods for visual recognition. We discuss the merits and drawbacks of existing domain adaptation approaches and identify promising avenues for research in this rapidly evolving field.",
"Abstract A class of predictive densities is derived by weighting the observed samples in maximizing the log-likelihood function. This approach is effective in cases such as sample surveys or design of experiments, where the observed covariate follows a different distribution than that in the whole population. Under misspecification of the parametric model, the optimal choice of the weight function is asymptotically shown to be the ratio of the density function of the covariate in the population to that in the observations. This is the pseudo-maximum likelihood estimation of sample surveys. The optimality is defined by the expected Kullback–Leibler loss, and the optimal weight is obtained by considering the importance sampling identity. Under correct specification of the model, however, the ordinary maximum likelihood estimate (i.e. the uniform weight) is shown to be optimal asymptotically. For moderate sample size, the situation is in between the two extreme cases, and the weight function is selected by minimizing a variant of the information criterion derived as an estimate of the expected loss. The method is also applied to a weighted version of the Bayesian predictive density. Numerical examples as well as Monte-Carlo simulations are shown for polynomial regression. A connection with the robust parametric estimation is discussed.",
"",
"Discriminative learning methods are widely used in natural language processing. These methods work best when their training and test data are drawn from the same distribution. For many NLP tasks, however, we are confronted with new domains in which labeled data is scarce or non-existent. In such cases, we seek to adapt existing models from a resource-rich source domain to a resource-poor target domain. We introduce structural correspondence learning to automatically induce correspondences among features from different domains. We test our technique on part of speech tagging and show performance gains for varying amounts of source and target training data, as well as improvements in target domain parsing accuracy using our improved tagger.",
"A major assumption in many machine learning and data mining algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many real-world applications, this assumption may not hold. For example, we sometimes have a classification task in one domain of interest, but we only have sufficient training data in another domain of interest, where the latter data may be in a different feature space or follow a different data distribution. In such cases, knowledge transfer, if done successfully, would greatly improve the performance of learning by avoiding much expensive data-labeling efforts. In recent years, transfer learning has emerged as a new learning framework to address this problem. This survey focuses on categorizing and reviewing the current progress on transfer learning for classification, regression, and clustering problems. In this survey, we discuss the relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift. We also explore some potential future issues in transfer learning research.",
"We describe an approach to domain adaptation that is appropriate exactly in the case when one has enough “target” data to do slightly better than just using only “source” data. Our approach is incredibly simple, easy to implement as a preprocessing step (10 lines of Perl!) and outperforms stateof-the-art approaches on a range of datasets. Moreover, it is trivially extended to a multidomain adaptation problem, where one has data from a variety of different domains."
]
} |
1607.03516 | 2950790587 | In this paper, we propose a novel unsupervised domain adaptation algorithm based on deep learning for visual object recognition. Specifically, we design a new model called Deep Reconstruction-Classification Network (DRCN), which jointly learns a shared encoding representation for two tasks: i) supervised classification of labeled source data, and ii) unsupervised reconstruction of unlabeled target data.In this way, the learnt representation not only preserves discriminability, but also encodes useful information from the target domain. Our new DRCN model can be optimized by using backpropagation similarly as the standard neural networks. We evaluate the performance of DRCN on a series of cross-domain object recognition tasks, where DRCN provides a considerable improvement (up to 8 in accuracy) over the prior state-of-the-art algorithms. Interestingly, we also observe that the reconstruction pipeline of DRCN transforms images from the source domain into images whose appearance resembles the target dataset. This suggests that DRCN's performance is due to constructing a single composite representation that encodes information about both the structure of target images and the classification of source images. Finally, we provide a formal analysis to justify the algorithm's objective in domain adaptation context. | This paper is concerned with in which labeled data from the target domain is not available @cite_28 . A range of approaches along this line of research in object recognition have been proposed @cite_34 @cite_30 @cite_36 @cite_35 @cite_40 @cite_22 @cite_54 , most were designed specifically for small datasets such as the Office dataset @cite_57 . Furthermore, they usually operated on the SURF-based features @cite_48 extracted from the raw pixels. In essence, the unsupervised domain adaptation problem remains open and needs more powerful solutions that are useful for practical situations. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_22",
"@cite_28",
"@cite_36",
"@cite_54",
"@cite_48",
"@cite_57",
"@cite_40",
"@cite_34"
],
"mid": [
"2064447488",
"2963756240",
"2149466042",
"2311233238",
"2104068492",
"2057266281",
"",
"",
"2128053425",
"1910772337"
],
"abstract": [
"Domain-invariant representations are key to addressing the domain shift problem where the training and test examples follow different distributions. Existing techniques that have attempted to match the distributions of the source and target domains typically compare these distributions in the original feature space. This space, however, may not be directly suitable for such a comparison, since some of the features may have been distorted by the domain shift, or may be domain specific. In this paper, we introduce a Domain Invariant Projection approach: An unsupervised domain adaptation method that overcomes this issue by extracting the information that is invariant across the source and target domains. More specifically, we learn a projection of the data to a low-dimensional latent space where the distance between the empirical distributions of the source and target examples is minimized. We demonstrate the effectiveness of our approach on the task of visual object recognition and show that it outperforms state-of-the-art methods on a standard domain adaptation benchmark dataset.",
"This paper addresses classification tasks on a particular target domain in which labeled training data are only available from source domains different from (but related to) the target. Two closely related frameworks, domain adaptation and domain generalization, are concerned with such tasks, where the only difference between those frameworks is the availability of the unlabeled target data: domain adaptation can leverage unlabeled target information, while domain generalization cannot. We propose Scatter Component Analyis (SCA), a fast representation learning algorithm that can be applied to both domain adaptation and domain generalization. SCA is based on a simple geometrical measure, i.e., scatter , which operates on reproducing kernel Hilbert space . SCA finds a representation that trades between maximizing the separability of classes, minimizing the mismatch between domains, and maximizing the separability of data; each of which is quantified through scatter . The optimization problem of SCA can be reduced to a generalized eigenvalue problem, which results in a fast and exact solution. Comprehensive experiments on benchmark cross-domain object recognition datasets verify that SCA performs much faster than several state-of-the-art algorithms and also provides state-of-the-art classification accuracy in both domain adaptation and domain generalization. We also show that scatter can be used to establish a theoretical generalization bound in the case of domain adaptation.",
"In real-world applications of visual recognition, many factors — such as pose, illumination, or image quality — can cause a significant mismatch between the source domain on which classifiers are trained and the target domain to which those classifiers are applied. As such, the classifiers often perform poorly on the target domain. Domain adaptation techniques aim to correct the mismatch. Existing approaches have concentrated on learning feature representations that are invariant across domains, and they often do not directly exploit low-dimensional structures that are intrinsic to many vision datasets. In this paper, we propose a new kernel-based method that takes advantage of such structures. Our geodesic flow kernel models domain shift by integrating an infinite number of subspaces that characterize changes in geometric and statistical properties from the source to the target domain. Our approach is computationally advantageous, automatically inferring important algorithmic parameters without requiring extensive cross-validation or labeled data from either domain. We also introduce a metric that reliably measures the adaptability between a pair of source and target domains. For a given target domain and several source domains, the metric can be used to automatically select the optimal source domain to adapt and avoid less desirable ones. Empirical studies on standard datasets demonstrate the advantages of our approach over competing methods.",
"In supervised learning, it is typically assumed that the labeled training data comes from the same distribution as the test data to which the system will be applied. In recent years, machine-learning researchers have investigated methods to handle mismatch between the training and test domains, with the goal of building a classifier using the labeled data in the old domain that will perform well on the test data in the new domain. This problem is called domain adaptation or transfer learning, and it is a common scenario in speech processing applications. Labeled training data are often produced by an expensive hand-annotation process, and may consist of only one or two annotated corpora which are used to train virtually all systems regardless of the target domain. Often little or no labeled data is available for the new domain. In this work, we review the statistical machine learning literature dealing with the problem of “domain adaptation” or “transfer learning”. We focus on unsupervised domain adaptation methods, as opposed to model adaptation or supervised adaptation in which some labeled data is available from the test distribution. We consider four main classes of approaches in the literature: instance weighting for covariate shift; selflabeling methods; changes in feature representation; and cluster-based learning. Covariate shift methods re-weight training samples in the old domain to try to match the new domain, putting more weight on samples in populous regions in the new domain. Self-labeling methods incorporate unlabeled target domain examples into the training algorithm by making an initial guess about their labels and then iteratively refining the guesses or labeling more examples. Feature representation approaches try to find a new feature representation of the data, either to make the new and old distributions look similar, or to find an abstracted representation for domain-specific features. Cluster-based methods rely on the assumption that samples connected by high-density paths are likely to have the same label. Domain adaptation is a large area of research, with related work under several frameworks (and several names). A limited review from March 2008 can be found in [1], and one from Oct 2010 can be found in [2]. A recent book [3] investigates train test distribution mismatch in machine learning (particularly focused on covariate shift.) Some of the organization here roughly follows that in [1].",
"In this paper, we introduce a new domain adaptation (DA) algorithm where the source and target domains are represented by subspaces described by eigenvectors. In this context, our method seeks a domain adaptation solution by learning a mapping function which aligns the source subspace with the target one. We show that the solution of the corresponding optimization problem can be obtained in a simple closed form, leading to an extremely fast algorithm. We use a theoretical result to tune the unique hyper parameter corresponding to the size of the subspaces. We run our method on various datasets and show that, despite its intrinsic simplicity, it outperforms state of the art DA methods.",
"Visual domain adaptation, which learns an accurate classifier for a new domain using labeled images from an old domain, has shown promising value in computer vision yet still been a challenging problem. Most prior works have explored two learning strategies independently for domain adaptation: feature matching and instance reweighting. In this paper, we show that both strategies are important and inevitable when the domain difference is substantially large. We therefore put forward a novel Transfer Joint Matching (TJM) approach to model them in a unified optimization problem. Specifically, TJM aims to reduce the domain difference by jointly matching the features and reweighting the instances across domains in a principled dimensionality reduction procedure, and construct new feature representation that is invariant to both the distribution difference and the irrelevant instances. Comprehensive experimental results verify that TJM can significantly outperform competitive methods for cross-domain image recognition problems.",
"",
"",
"Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.",
"Domain adaptation (DA) has gained a lot of success in the recent years in computer vision to deal with situations where the learning process has to transfer knowledge from a source to a target domain. In this paper, we introduce a novel unsupervised DA approach based on both subspace alignment and selection of landmarks similarly distributed between the two domains. Those landmarks are selected so as to reduce the discrepancy between the domains and then are used to non linearly project the data in the same space where an efficient subspace alignment (in closed-form) is performed. We carry out a large experimental comparison in visual domain adaptation showing that our new method outperforms the most recent unsupervised DA approaches."
]
} |
1607.03516 | 2950790587 | In this paper, we propose a novel unsupervised domain adaptation algorithm based on deep learning for visual object recognition. Specifically, we design a new model called Deep Reconstruction-Classification Network (DRCN), which jointly learns a shared encoding representation for two tasks: i) supervised classification of labeled source data, and ii) unsupervised reconstruction of unlabeled target data.In this way, the learnt representation not only preserves discriminability, but also encodes useful information from the target domain. Our new DRCN model can be optimized by using backpropagation similarly as the standard neural networks. We evaluate the performance of DRCN on a series of cross-domain object recognition tasks, where DRCN provides a considerable improvement (up to 8 in accuracy) over the prior state-of-the-art algorithms. Interestingly, we also observe that the reconstruction pipeline of DRCN transforms images from the source domain into images whose appearance resembles the target dataset. This suggests that DRCN's performance is due to constructing a single composite representation that encodes information about both the structure of target images and the classification of source images. Finally, we provide a formal analysis to justify the algorithm's objective in domain adaptation context. | Deep learning now plays a major role in the advancement of domain adaptation. An early attempt addressed large-scale sentiment classification @cite_42 , where the concatenated features from fully connected layers of stacked denoising autoencoders have been found to be domain-adaptive @cite_16 . In visual recognition, a fully connected, shallow network pretrained by denoising autoencoders has shown a certain level of effectiveness @cite_50 . It is widely known that deep convolutional networks (ConvNets) @cite_1 are a more natural choice for visual recognition tasks and have achieved significant successes @cite_31 @cite_49 @cite_10 . More recently, ConvNets pretrained on a large-scale dataset, ImageNet, have been shown to be reasonably effective for domain adaptation @cite_49 . They provide significantly better performances than the SURF-based features on the Office dataset @cite_51 @cite_8 . An earlier approach on using a convolutional architecture without pretraining on ImageNet, DLID, has also been explored @cite_5 and performs better than the SURF-based features. | {
"cite_N": [
"@cite_8",
"@cite_42",
"@cite_1",
"@cite_50",
"@cite_49",
"@cite_51",
"@cite_5",
"@cite_31",
"@cite_16",
"@cite_10"
],
"mid": [
"2963449250",
"22861983",
"2310919327",
"2963168418",
"",
"2953360861",
"2186639548",
"2102605133",
"2145094598",
"1686810756"
],
"abstract": [
"",
"The exponential increase in the availability of online reviews and recommendations makes sentiment classification an interesting topic in academic and industrial research. Reviews can span so many different domains that it is difficult to gather annotated training data for all of them. Hence, this paper studies the problem of domain adaptation for sentiment classifiers, hereby a system is trained on labeled reviews from one source domain but is meant to be deployed on another. We propose a deep learning approach which learns to extract a meaningful representation for each review in an unsupervised fashion. Sentiment classifiers trained with this high-level feature representation clearly outperform state-of-the-art methods on a benchmark composed of reviews of 4 types of Amazon products. Furthermore, this method scales well and allowed us to successfully perform domain adaptation on a larger industrial-strength dataset of 22 domains.",
"",
"We propose a simple neural network model to deal with the domain adaptation problem in object recognition. Our model incorporates the Maximum Mean Discrepancy (MMD) measure as a regularization in the supervised learning to reduce the distribution mismatch between the source and target domains in the latent space. From experiments, we demonstrate that the MMD regularization is an effective tool to provide good domain adaptation models on both SURF features and raw image pixels of a particular image data set.",
"",
"We evaluate whether features extracted from the activation of a deep convolutional network trained in a fully supervised fashion on a large, fixed set of object recognition tasks can be re-purposed to novel generic tasks. Our generic tasks may differ significantly from the originally trained tasks and there may be insufficient labeled or unlabeled data to conventionally train or adapt a deep architecture to the new tasks. We investigate and visualize the semantic clustering of deep convolutional features with respect to a variety of such tasks, including scene recognition, domain adaptation, and fine-grained recognition challenges. We compare the efficacy of relying on various network levels to define a fixed feature, and report novel results that significantly outperform the state-of-the-art on several important vision challenges. We are releasing DeCAF, an open-source implementation of these deep convolutional activation features, along with all associated network parameters to enable vision researchers to be able to conduct experimentation with deep representations across a range of visual concept learning paradigms.",
"In many real world applications of machine learning, the distribution of the training data (on which the machine learning model is trained) is dierent from the distribution of the test data (where the learnt model is actually deployed). This is known as the problem of Domain Adaptation. We propose a novel deep learning model for domain adaptation which attempts to learn a predictively useful representation of the data by taking into account information from the distribution shift between the training and test data. Our key proposal is to successively learn multiple intermediate representations along an path\" between the train and test domains. Our experiments on a standard object recognition dataset show a signicant performance improvement over the state-of-the-art.",
"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 explore an original strategy for building deep networks, based on stacking layers of denoising autoencoders which are trained locally to denoise corrupted versions of their inputs. The resulting algorithm is a straightforward variation on the stacking of ordinary autoencoders. It is however shown on a benchmark of classification problems to yield significantly lower classification error, thus bridging the performance gap with deep belief networks (DBN), and in several cases surpassing it. Higher level representations learnt in this purely unsupervised fashion also help boost the performance of subsequent SVM classifiers. Qualitative experiments show that, contrary to ordinary autoencoders, denoising autoencoders are able to learn Gabor-like edge detectors from natural image patches and larger stroke detectors from digit images. This work clearly establishes the value of using a denoising criterion as a tractable unsupervised objective to guide the learning of useful higher level representations.",
"In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision."
]
} |
1607.03516 | 2950790587 | In this paper, we propose a novel unsupervised domain adaptation algorithm based on deep learning for visual object recognition. Specifically, we design a new model called Deep Reconstruction-Classification Network (DRCN), which jointly learns a shared encoding representation for two tasks: i) supervised classification of labeled source data, and ii) unsupervised reconstruction of unlabeled target data.In this way, the learnt representation not only preserves discriminability, but also encodes useful information from the target domain. Our new DRCN model can be optimized by using backpropagation similarly as the standard neural networks. We evaluate the performance of DRCN on a series of cross-domain object recognition tasks, where DRCN provides a considerable improvement (up to 8 in accuracy) over the prior state-of-the-art algorithms. Interestingly, we also observe that the reconstruction pipeline of DRCN transforms images from the source domain into images whose appearance resembles the target dataset. This suggests that DRCN's performance is due to constructing a single composite representation that encodes information about both the structure of target images and the classification of source images. Finally, we provide a formal analysis to justify the algorithm's objective in domain adaptation context. | To further improve the domain adaptation performance, the pretrained ConvNets can be under a particular constraint related to minimizing a domain discrepancy measure @cite_19 @cite_43 @cite_6 @cite_7 . Deep Domain Confusion (DDC) @cite_6 utilizes the maximum mean discrepancy (MMD) measure @cite_2 as an additional loss function for the fine-tuning to adapt the last fully connected layer. Deep Adaptation Network (DAN) @cite_43 fine-tunes not only the last fully connected layer, but also some convolutional and fully connected layers underneath, and outperforms DDC. Recently, the deep model proposed in @cite_7 extends the idea of DDC by adding a criterion to guarantee the class alignment between different domains. However, it is limited only to the adaptation setting, where a small number of target labels can be acquired. | {
"cite_N": [
"@cite_7",
"@cite_6",
"@cite_19",
"@cite_43",
"@cite_2"
],
"mid": [
"2953226914",
"1565327149",
"",
"2951670162",
"2164943005"
],
"abstract": [
"Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias. Fine-tuning deep models in a new domain can require a significant amount of labeled data, which for many applications is simply not available. We propose a new CNN architecture to exploit unlabeled and sparsely labeled target domain data. Our approach simultaneously optimizes for domain invariance to facilitate domain transfer and uses a soft label distribution matching loss to transfer information between tasks. Our proposed adaptation method offers empirical performance which exceeds previously published results on two standard benchmark visual domain adaptation tasks, evaluated across supervised and semi-supervised adaptation settings.",
"Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias on a standard benchmark. Fine-tuning deep models in a new domain can require a significant amount of data, which for many applications is simply not available. We propose a new CNN architecture which introduces an adaptation layer and an additional domain confusion loss, to learn a representation that is both semantically meaningful and domain invariant. We additionally show that a domain confusion metric can be used for model selection to determine the dimension of an adaptation layer and the best position for the layer in the CNN architecture. Our proposed adaptation method offers empirical performance which exceeds previously published results on a standard benchmark visual domain adaptation task.",
"",
"Recent studies reveal that a deep neural network can learn transferable features which generalize well to novel tasks for domain adaptation. However, as deep features eventually transition from general to specific along the network, the feature transferability drops significantly in higher layers with increasing domain discrepancy. Hence, it is important to formally reduce the dataset bias and enhance the transferability in task-specific layers. In this paper, we propose a new Deep Adaptation Network (DAN) architecture, which generalizes deep convolutional neural network to the domain adaptation scenario. In DAN, hidden representations of all task-specific layers are embedded in a reproducing kernel Hilbert space where the mean embeddings of different domain distributions can be explicitly matched. The domain discrepancy is further reduced using an optimal multi-kernel selection method for mean embedding matching. DAN can learn transferable features with statistical guarantees, and can scale linearly by unbiased estimate of kernel embedding. Extensive empirical evidence shows that the proposed architecture yields state-of-the-art image classification error rates on standard domain adaptation benchmarks.",
"Motivation: Many problems in data integration in bioinformatics can be posed as one common question: Are two sets of observations generated by the same distribution? We propose a kernel-based statistical test for this problem, based on the fact that two distributions are different if and only if there exists at least one function having different expectation on the two distributions. Consequently we use the maximum discrepancy between function means as the basis of a test statistic. The Maximum Mean Discrepancy (MMD) can take advantage of the kernel trick, which allows us to apply it not only to vectors, but strings, sequences, graphs, and other common structured data types arising in molecular biology. Results: We study the practical feasibility of an MMD-based test on three central data integration tasks: Testing cross-platform comparability of microarray data, cancer diagnosis, and data-content based schema matching for two different protein function classification schemas. In all of these experiments, including high-dimensional ones, MMD is very accurate in finding samples that were generated from the same distribution, and outperforms its best competitors. Conclusions: We have defined a novel statistical test of whether two samples are from the same distribution, compatible with both multivariate and structured data, that is fast, easy to implement, and works well, as confirmed by our experiments. Availability: Contact: [email protected]"
]
} |
1607.03780 | 2478536981 | Distributional semantics creates vector-space representations that capture many forms of semantic similarity, but their relation to semantic entailment has been less clear. We propose a vector-space model which provides a formal foundation for a distributional semantics of entailment. Using a mean-field approximation, we develop approximate inference procedures and entailment operators over vectors of probabilities of features being known (versus unknown). We use this framework to reinterpret an existing distributional-semantic model (Word2Vec) as approximating an entailment-based model of the distributions of words in contexts, thereby predicting lexical entailment relations. In both unsupervised and semi-supervised experiments on hyponymy detection, we get substantial improvements over previous results. | There has been a significant amount of work on using distributional-semantic vectors for hyponymy detection, using supervised, semi-supervised or unsupervised methods (e.g. @cite_4 @cite_16 @cite_3 @cite_17 @cite_0 @cite_22 ). Because our main concern is modelling entailment within a vector space, we do not do a thorough comparison to models which use measures computed outside the vector space (e.g. symmetric measures (LIN @cite_10 ), asymmetric measures (WeedsPrec @cite_18 @cite_9 , balAPinc @cite_23 , invCL @cite_15 ) and entropy-based measures (SLQS @cite_12 )), nor to models which encode hyponymy in the parameters of a vector-space operator or classifier @cite_0 @cite_5 @cite_24 ). We also limit our evaluation of lexical entailment to hyponymy, not including other related lexical relations (cf. @cite_17 @cite_3 @cite_20 @cite_14 ), leaving more complex cases to future work on compositional semantics. We are also not concerned with models or evaluations which require supervised learning about individual words, instead limiting ourselves to semi-supervised learning where the words in the training and test sets are disjoint. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_4",
"@cite_22",
"@cite_9",
"@cite_20",
"@cite_3",
"@cite_0",
"@cite_24",
"@cite_23",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_10",
"@cite_12",
"@cite_17"
],
"mid": [
"2101941521",
"2250318405",
"",
"2250916563",
"1966907789",
"2165897885",
"1944056442",
"2000216676",
"205765513",
"2131253837",
"1583820951",
"141602984",
"2155157567",
"2166776180",
"2251981022",
"1516501661"
],
"abstract": [
"We present a general framework for distributional similarity based on the concepts of precision and recall. Different parameter settings within this framework approximate different existing similarity measures as well as many more which have, until now, been unexplored. We show that optimal parameter settings outperform two existing state-of-the-art similarity measures on two evaluation tasks for high and low frequency nouns.",
"Open IE methods extract structured propositions from text. However, these propositions are neither consolidated nor generalized, and querying them may lead to insufficient or redundant information. This work suggests an approach to organize open IE propositions using entailment graphs. The entailment relation unifies equivalent propositions and induces a specific-to-general structure. We create a large dataset of gold-standard proposition entailment graphs, and provide a novel algorithm for automatically constructing them. Our analysis shows that predicate entailment is extremely context-sensitive, and that current lexical-semantic resources do not capture many of the lexical inferences induced by proposition entailment.",
"",
"The task of detecting and generating hyponyms is at the core of semantic understanding of language, and has numerous practical applications. We investigate how neural network embeddings perform on this task, compared to dependency-based vector space models, and evaluate a range of similarity measures on hyponym generation. A new asymmetric similarity measure and a combination approach are described, both of which significantly improve precision. We release three new datasets of lexical vector representations trained on the BNC and our evaluation dataset for hyponym generation.",
"This work investigates the variation in a word's distributionally nearest neighbours with respect to the similarity measure used. We identify one type of variation as being the relative frequency of the neighbour words with respect to the frequency of the target word. We then demonstrate a three-way connection between relative frequency of similar words, a concept of distributional gnerality and the semantic relation of hyponymy. Finally, we consider the impact that this has on one application of distributional similarity methods (judging the compositionality of collocations).",
"Inference in natural language often involves recognizing lexical entailment (RLE); that is, identifying whether one word entails another. For example, buy entails own. Two general strategies for RLE have been proposed: One strategy is to manually construct an asymmetric similarity measure for context vectors (directional similarity) and another is to treat RLE as a problem of learning to recognize semantic relations using supervised machine learning techniques (relation classification). In this paper, we experiment with two recent state-of-the-art representatives of the two general strategies. The first approach is an asymmetric similarity measure (an instance of the directional similarity strategy), designed to capture the degree to which the contexts of a word, a, form a subset of the contexts of another word, b. The second approach (an instance of the relation classification strategy) represents a word pair, a:b, with a feature vector that is the concatenation of the context vectors of a and b, and then applies supervised learning to a training set of labeled feature vectors. Additionally, we introduce a third approach that is a new instance of the relation classification strategy. The third approach represents a word pair, a : b, with a feature vector in which the features are the differences in the similarities of a and b to a set of reference words. All three approaches use vector space models (VSMs) of semantics, based on word–context matrices. We perform an extensive evaluation of the three approaches using three different datasets. The proposed new approach (similarity differences) performs significantly better than the other two approaches on some datasets and there is no dataset for which it is significantly worse. Along the way, we address some of the concerns raised in past research, regarding the treatment of RLE as a problem of semantic relation classification, and we suggest it is beneficial to make connections between the research in lexical entailment and the research in semantic relation classification.",
"Recent work on word embeddings has shown that simple vector subtraction over pre-trained embeddings is surprisingly effective at capturing different lexical relations, despite lacking explicit supervision. Prior work has evaluated this intriguing result using a word analogy prediction formulation and hand-selected relations, but the generality of the finding over a broader range of lexical relation types and different learning settings has not been evaluated. In this paper, we carry out such an evaluation in two learning settings: (1) spectral clustering to induce word relations, and (2) supervised learning to classify vector differences into relation types. We find that word embeddings capture a surprising amount of information, and that, under suitable supervised training, vector subtraction generalises well to a broad range of relations, including over unseen lexical items.",
"Semantic hierarchy construction aims to build structures of concepts linked by hypernym-hyponym (\"is-a\") relations. A major challenge for this task is the automatic discovery of such relations. This paper proposes a novel and effective method for the construction of semantic hierarchies based on continuous vector representation of words, named word embeddings, which can be used to measure the semantic relationship between words. We identify whether a candidate word pair has hypernym-hyponym relation by using the word-embedding-based semantic projections between words and their hypernyms. Our result, an F-score of 73.74 , outperforms the state-of-the-art methods on a manually labeled test dataset. Moreover, combining our method with a previous manually built hierarchy extension method can further improve F-score to 80.29 .",
"We introduce two ways to detect entailment using distributional semantic representations of phrases. Our first experiment shows that the entailment relation between adjective-noun constructions and their head nouns (big cat|= cat), once represented as semantic vector pairs, generalizes to lexical entailment among nouns (dog|= animal). Our second experiment shows that a classifier fed semantic vector pairs can similarly generalize the entailment relation among quantifier phrases (many dogs|= some dogs) to entailment involving unseen quantifiers (all cats|= several cats). Moreover, nominal and quantifier phrase entailment appears to be cued by different distributional correlates, as predicted by the type-based view of entailment in formal semantics.",
"Distributional word similarity is most commonly perceived as a symmetric relation. Yet, directional relations are abundant in lexical semantics and in many Natural Language Processing (NLP) settings that require lexical inference, making symmetric similarity measures less suitable for their identification. This paper investigates the nature of directional (asymmetric) similarity measures that aim to quantify distributional feature inclusion. We identify desired properties of such measures for lexical inference, specify a particular measure based on Average Precision that addresses these properties, and demonstrate the empirical benefit of directional measures for two different NLP datasets.",
"We test the Distributional Inclusion Hypothesis, which states that hypernyms tend to occur in a superset of contexts in which their hyponyms are found. We find that this hypothesis only holds when it is applied to relevant dimensions. We propose a robust supervised approach that achieves accuracies of .84 and .85 on two existing datasets and that can be interpreted as selecting the dimensions that are relevant for distributional inclusion.",
"In this paper we apply existing directional similarity measures to identify hypernyms with a state-of-the-art distributional semantic model. We also propose a new directional measure that achieves the best performance in hypernym identification.",
"The lexical semantic relationships between word pairs are key features for many NLP tasks. Most approaches for automatically classifying related word pairs are hindered by data sparsity because of their need to observe two words co-occurring in order to detect the lexical relation holding between them. Even when mining very large corpora, not every related word pair co-occurs. Using novel representations based on graphs and word embeddings, we present two systems that are able to predict relations between words, even when these are never found in the same sentence in a given corpus. In two experiments, we demonstrate superior performance of both approaches over the state of the art, achieving significant gains in recall.",
"Bootstrapping semantics from text is one of the greatest challenges in natural language learning. We first define a word similarity measure based on the distributional pattern of words. The similarity measure allows us to construct a thesaurus using a parsed corpus. We then present a new evaluation methodology for the automatically constructed thesaurus. The evaluation results show that the thesaurus is significantly closer to WordNet than Roget Thesaurus is.",
"In this paper, we introduce SLQS , a new entropy-based measure for the unsupervised identification of hypernymy and its directionality in Distributional Semantic Models (DSMs). SLQS is assessed through two tasks: (i.) identifying the hypernym in hyponym-hypernym pairs, and (ii.) discriminating hypernymy among various semantic relations. In both tasks, SLQS outperforms other state-of-the-art measures.",
"This work is concerned with distinguishing different semantic relations which exist between distributionally similar words. We compare a novel approach based on training a linear Support Vector Machine on pairs of feature vectors with state-of-the-art methods based on distributional similarity. We show that the new supervised approach does better even when there is minimal information about the target words in the training data, giving a 15 reduction in error rate over unsupervised approaches."
]
} |
1607.03780 | 2478536981 | Distributional semantics creates vector-space representations that capture many forms of semantic similarity, but their relation to semantic entailment has been less clear. We propose a vector-space model which provides a formal foundation for a distributional semantics of entailment. Using a mean-field approximation, we develop approximate inference procedures and entailment operators over vectors of probabilities of features being known (versus unknown). We use this framework to reinterpret an existing distributional-semantic model (Word2Vec) as approximating an entailment-based model of the distributions of words in contexts, thereby predicting lexical entailment relations. In both unsupervised and semi-supervised experiments on hyponymy detection, we get substantial improvements over previous results. | We base this evaluation on the Word2Vec @cite_13 @cite_21 distributional semantic model and its publicly available word embeddings. We choose it because it is popular, simple, fast, and its embeddings have been derived from a very large corpus. showed that it is closely related to the previous PMI-based distributional semantic models (e.g. @cite_8 ). | {
"cite_N": [
"@cite_21",
"@cite_13",
"@cite_8"
],
"mid": [
"2950133940",
"1614298861",
"1662133657"
],
"abstract": [
"The recently introduced continuous Skip-gram model is an efficient method for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. In this paper we present several extensions that improve both the quality of the vectors and the training speed. By subsampling of the frequent words we obtain significant speedup and also learn more regular word representations. We also describe a simple alternative to the hierarchical softmax called negative sampling. An inherent limitation of word representations is their indifference to word order and their inability to represent idiomatic phrases. For example, the meanings of \"Canada\" and \"Air\" cannot be easily combined to obtain \"Air Canada\". Motivated by this example, we present a simple method for finding phrases in text, and show that learning good vector representations for millions of phrases is possible.",
"",
"Computers understand very little of the meaning of human language. This profoundly limits our ability to give instructions to computers, the ability of computers to explain their actions to us, and the ability of computers to analyse and process text. Vector space models (VSMs) of semantics are beginning to address these limits. This paper surveys the use of VSMs for semantic processing of text. We organize the literature on VSMs according to the structure of the matrix in a VSM. There are currently three broad classes of VSMs, based on term-document, word-context, and pair-pattern matrices, yielding three classes of applications. We survey a broad range of applications in these three categories and we take a detailed look at a specific open source project in each category. Our goal in this survey is to show the breadth of applications of VSMs for semantics, to provide a new perspective on VSMs for those who are already familiar with the area, and to provide pointers into the literature for those who are less familiar with the field."
]
} |
1607.03780 | 2478536981 | Distributional semantics creates vector-space representations that capture many forms of semantic similarity, but their relation to semantic entailment has been less clear. We propose a vector-space model which provides a formal foundation for a distributional semantics of entailment. Using a mean-field approximation, we develop approximate inference procedures and entailment operators over vectors of probabilities of features being known (versus unknown). We use this framework to reinterpret an existing distributional-semantic model (Word2Vec) as approximating an entailment-based model of the distributions of words in contexts, thereby predicting lexical entailment relations. In both unsupervised and semi-supervised experiments on hyponymy detection, we get substantial improvements over previous results. | The most similar previous work, in terms of motivation and aims, is that of . They also model entailment directly using a vector space, without training a classifier. But instead of representing words as a point in a vector space (as in this work), they represent words as a Gaussian distribution over points in a vector space. This allows them to represent the extent to which a feature is known versus unknown as the amount of variance in the distribution for that feature's dimension. While nicely motivated theoretically, the model appears to be more computationally expensive than the one proposed here, particularly for inferring vectors. They do make unsupervised predictions of hyponymy relations with their learned vector distributions, using KL-divergence between the distributions for the two words. They evaluate their models on the hyponymy data from @cite_24 . As discussed further in section , our best models achieve non-significantly better average precision than their best models. | {
"cite_N": [
"@cite_24"
],
"mid": [
"205765513"
],
"abstract": [
"We introduce two ways to detect entailment using distributional semantic representations of phrases. Our first experiment shows that the entailment relation between adjective-noun constructions and their head nouns (big cat|= cat), once represented as semantic vector pairs, generalizes to lexical entailment among nouns (dog|= animal). Our second experiment shows that a classifier fed semantic vector pairs can similarly generalize the entailment relation among quantifier phrases (many dogs|= some dogs) to entailment involving unseen quantifiers (all cats|= several cats). Moreover, nominal and quantifier phrase entailment appears to be cued by different distributional correlates, as predicted by the type-based view of entailment in formal semantics."
]
} |
1607.03895 | 2491733261 | Gender bias is an increasingly important issue in sports journalism. In this work, we propose a language-model-based approach to quantify differences in questions posed to female vs. male athletes, and apply it to tennis post-match interviews. We find that journalists ask male players questions that are generally more focused on the game when compared with the questions they ask their female counterparts. We also provide a fine-grained analysis of the extent to which the salience of this bias depends on various factors, such as question type, game outcome or player rank. | In contrast with our work, prior investigations of bias in sport journalism rely on manual coding or are based on simple lists of manually defined keywords. These focus on bias with respect to race, nationality, and gender @cite_13 @cite_4 @cite_6 @cite_15 @cite_8 @cite_10 @cite_2 ; see van2010race for a review. | {
"cite_N": [
"@cite_4",
"@cite_8",
"@cite_10",
"@cite_6",
"@cite_2",
"@cite_15",
"@cite_13"
],
"mid": [
"2143209442",
"1573788826",
"1970564707",
"1965892432",
"2089478583",
"2039337535",
"2011404061"
],
"abstract": [
"This study used both qualitative and quantitative analyses to discern whether the narratives, metaphors, framing devices, and production practices in televised international athletic events differed by the race, ethnicity, or nationality of athletes. About 340 hours of videotapes of 7 televised international athletic events were used to study key aspects of production: (a) commentator descriptions of 161 athletes in 31 competitions, (b) 30 personal interviews drawn from 3 of the events under study, and (c) 5 opening and closing segments that commonly unify themes and metaphors and that produce the look of an event. Six major findings include the following: (a) efforts were made to provide fair treatment of athletes, (b) the treatment of race and ethnicity varied across productions, (c) little evidence of negative representations of Black athletes, (d) representations of Asian athletes drew on cultural stereotypes, (e) representations of Latino-Hispanic athletes were mixed, and (f) nationalistic bias was e...",
"1. Investigating the Biggest Show on Television 2. Meet the 'Framers': The Olympic Producers 3. Chronicling History: The Olympic Sportscasters 4. The Star-Spangled Games?: Nationalism and the Olympic Telecasts 5. Competing on the Same Stage: Gender and the Olympic Telecasts 6. Dialogue Differences in Black and White?: Ethnicity and the Olympic Telecasts 7. What do Americans Think Happened in Torino?: Examining Media Effects 8. Looking Forward by Looking Back: Reflections on the Olympic Telecasts",
"The purpose of this study was to determine any significant differences in how reporters for newspapers and online sites framed men's and women's tennis. Articles on the 2007 U.S. Open in The Los Angeles Times, The New York Times, USA Today, and online sites produced by ESPN, Fox Sports, and Sports Illustrated were examined. Results showed newspapers were more likely to minimize the athleticism of female athletes, thus strengthening hegemonic masculinity more than the newer medium of online journalism, which produced mixed results.",
"The words of sportscastersrepeated hundreds, even thousands, of times by different announcers in similar waysprovide a conceptual frame for the sports experience, and that mental frame has particular importance because fans often apply it to nonathletic situations. Contrary to assertions by some critics, analysis of 1,156 descriptors in sportscaster commentary during 66 televised men's and women's college basketball games showed no significant difference between the proportions of commentary and proportions of participating Black and White men players, but showed some overemphasis in comments about White women players. Predictably, Black men players tended to be stereotyped as naturally athletic, quick, and powerful, while White men players continued to be touted for their hard work, effort, and mental skill. The same racial stereotypes also appeared in the commentary about women basketball players, but few gender stereotypes emerged. Thus, increases in the numbers of Black and women game announcers may...",
"Television is a cultural service and its national renderings of the Olympic Games contribute to the viewers’ understanding of themselves and the world. This study examines the representation of nationality and gender within Slovenian broadcasts of the 2008 Beijing Olympics. Results show that evaluative commentary comprised as much as 44 of the dialogue. ‘Home’ athletes were given more prominence, while foreign athletes were largely portrayed through quantifiable features. Male athletes received more commentary than females. Announcers favoured analysing results and predicting outcomes for men, and resorted to personality and physicality depictions in women. A critical discourse analysis uncovered notions about gender and nationality that would be deemed inappropriate in many societies. Sports journalists and broadcasters at TV Slovenija do not have explicit editorial policies addressing chauvinistic dialogue. The absence of a policy is not perceived as an issue as individual announcers abide to their per...",
"Despite an emerging body of research on race and representation in televised sport, little is known about commentator practices and understandings, particularly in relation to the construction of racial difference. Based on interviews and analysis of commentary, the results of this study point to a complex interaction between embedded racist ideologies and media practices specific to live basketball coverage. Taken together, and despite the conscious intentions of commentators, these interactions appear to contribute to an ongoing ‘Othering’ of African Americans and men of color and the normalization of dominant white cultural understandings of difference.",
""
]
} |
1607.03895 | 2491733261 | Gender bias is an increasingly important issue in sports journalism. In this work, we propose a language-model-based approach to quantify differences in questions posed to female vs. male athletes, and apply it to tennis post-match interviews. We find that journalists ask male players questions that are generally more focused on the game when compared with the questions they ask their female counterparts. We also provide a fine-grained analysis of the extent to which the salience of this bias depends on various factors, such as question type, game outcome or player rank. | Much of the work on gender bias in sports reporting has focused on air-time'' @cite_16 @cite_14 . Other studies looked at stereotypical descriptions and framing @cite_7 @cite_5 @cite_3 @cite_1 . For surveys, see Knight:SexRoles or Kaskan:SexRoles:2014 , inter alia. Several studies have focused on the particular case of gender-correlated differences in tennis coverage @cite_12 @cite_11 @cite_10 . We extend this line of work by proposing an automatic way to quantify gender bias in sport journalism. | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_1",
"@cite_3",
"@cite_5",
"@cite_16",
"@cite_10",
"@cite_12",
"@cite_11"
],
"mid": [
"2145966799",
"2048945236",
"1993779339",
"2016489455",
"1485189459",
"2145996244",
"1970564707",
"2505191071",
"1980494983"
],
"abstract": [
"This study examined the televised coverage of the 1996 Summer Olympic Games in Atlanta to determine the amount of coverage time, quality of coverage devoted to men’s and women’s same sport activities, and to compare this coverage with that of the 1992 Summer Olympic Games. Of the more than 150 hours of NBC televised coverage, a total of 60 hours was then randomly selected as an appropriate sample. A content analysis was then performed on those segments that featured same-sport activities for men and women. This included quantitative (e.g., running time, length of segments, number of slow motion replays, and use of onscreen statistics) and qualitative (e.g., the narrative of the commentators) analyses. Although the findings suggest there have been notable improvements in the way female athletes were presented in the 1996 Olympic Games as compared to the 1992 Olympic Games, there are still many disparities in the coverage of women’s sports, especially those that traditionally appeal to the media audience.",
"This research compares and analyzes the verbal commentary of televised coverage of two women's and men's athletic events: the “final four” of the women's and men's 1989 National Collegiate Athletic Association (NCAA) basketball tournaments and the women's and men's singles, women's and men's doubles, and the mixed-doubles matches of the 1989 U.S. Open tennis tournament. Although we found less overtly sexist commentary than has been observed in past research, we did find two categories of difference: (1) gender marking and (2) a “hierarchy of naming” by gender and, to a certain extent, by race. These differences are described and analyzed in light of feminist analyses of gendered language. It is concluded that televised sports commentary contributes to the construction of gender and racial hierarchies by marking women's sports and women athletes as “other,” by infantilizing women athletes (and, to a certain extent, male athletes of color), and by framing the accomplishments of women athletes ambivalently.",
"The primary purpose of this exploratory study was to determine if gender-specific descriptors regularly found in television and newspaper sport coverage were present in two popular online sites from the emerging medium of Internet sport journalism. Descriptors given to players and coaches during the 2006 NCAA Division I women's and men's basketball tournaments by ESPN Internet and CBS SportsLine were examined. Results contradicted gender-specific descriptors found in previous studies on sport media coverage that scholars have argued help uphold hegemonic masculinity in sport.",
"This study used theories of agenda setting and framing to examine NBC’s Americanized telecast in the 2008 Beijing Olympics. Five sports (gymnastics, diving, swimming, track and field, and beach volleyball) received more than 90 of the prime-time coverage, which set an agenda about which sports were most relevant for Americans to watch. The limited scope within NBC’s televised agenda, in turn, facilitated the gendered framing of Olympians through sport commentator accounts. Gendered differences were statistically present in only four sports; diving had no significant differences, whereas beach volleyball contained the most differences. Implications and directions for future research are explored.",
"A content analysis of the ABC News Online website during the 2000 Olympic Games reveals a select few female role models were available to young audiences. One female athlete was 'news-privileged'. Cathy Freeman's exposure came at the expense of her Australian team mates, especially those women who won medals in team sports. While the results indicate an improvement in both the extent of women's sports coverage and the range of sports covered, stereotypical descriptions often characterised adult females as emotionally vulnerable, dependent adolescents. Male athletes were never infantilised and were far less likely to be described in emotive terms.",
"Comparison of the sportscasting on ESPN and CNN and sports reporting in The New York Times and USA Today revealed the very high degree of embedded favoritism toward men’s sports and men athletes, even at times when major women’s sporting events were peaking in newsworthiness. The quantity of gender bias was significantly greater on ESPN’s SportsCenter than on CNN’s Sports Tonight, perhaps because of the somewhat different audiences they target. In addition, the amount of gender bias—measured three different ways—in the respected The New York Times also far exceeded that of USA Today, a disheartening finding about America’s so-called newspaper of record. Week-by-week crossmedia comparisons demonstrated the much greater marginalization of women’s sports in the electronic media, suggesting that newspapers provide a somewhat more positive model for sports journalism.",
"The purpose of this study was to determine any significant differences in how reporters for newspapers and online sites framed men's and women's tennis. Articles on the 2007 U.S. Open in The Los Angeles Times, The New York Times, USA Today, and online sites produced by ESPN, Fox Sports, and Sports Illustrated were examined. Results showed newspapers were more likely to minimize the athleticism of female athletes, thus strengthening hegemonic masculinity more than the newer medium of online journalism, which produced mixed results.",
"An interpretive analysis of mass circulation magazine articles on leading male and female professional tennis players indicates that both groups are treated in terms of a “debunking motif” which reveals their imperfections and character flaws. The flaws identified among the women are closely associated with stereotypically feminine gender roles, while the flaws observed among the men are associated with stereotypically masculine gender roles. Thus, the articles reinforce the concept of professional sport as a male preserve, while suggesting an underlying traditionally feminine gender role for the female athletes. It is argued that this construction of the female athlete role derives from the commercial sponsorship of professional tennis.",
"This study examined 152 articles devoted to female tennis players competing in the 2000 Wimbledon Championships in The Times, Daily Mail, and The Sun, covering a 17-day period that coincided with the Wimbledon Championships fortnight. Based upon the theoretical framework of gender power relations, a qualitative textual analysis methodology was used to reveal recurring themes within the dominant discourse about female tennis players. Results indicate the narratives used by the predominantly male journalists devalued, marginalised, and trivialised the athleticism of the female tennis players. The British newspapers seemed equally infatuated with Anna Kournikova, who was portrayed as a kind of sporting Lolita, and her hyper-feminine peers, the 'heterosexual honeys'. In contrast, Serena and Venus Williams – the 'Amazons' – were subjected to racial bigotry. The racial distinctions suggested that the hegemonic standard of the media was a favourable bias towards White female athletes. The sociological implications of such coverage of professional female tennis players are also discussed."
]
} |
1607.03202 | 2479255631 | Predicting and improving player retention is crucial to the success of mobile Free-to-Play games. This paper explores the problem of rapid retention prediction in this context. Heuristic modeling approaches are introduced as a way of building simple rules for predicting short-term retention. Compared to common classification algorithms, our heuristic-based approach achieves reasonable and comparable performance using information from the first session, day, and week of player activity. | Unlike previous work, the focus here is on the problem of rapid prediction of players by considering heuristic approaches owing to their ease of implementation and interpretation. Heuristics are strategies derived from experience with similar problems, using readily accessible information to control problem solving. They can be likened to rules of thumbs. They are often associated with the concept of satisfacing from economic decision-making @cite_11 . When finding an optimal solution is impossible or impractical, heuristic methods can be used for a satisfactory solution. They are used in a similar fashion in computer science, when the computational burden of complex methods is excessive. @cite_0 present a comprehensive review of their use in forecasting and prediction. @cite_4 ) empirically investigate their viability for use in database marketing. @cite_15 detail their application in management more broadly. The work presented here can be viewed as a special case and extension of the previous authors' contributions. | {
"cite_N": [
"@cite_0",
"@cite_15",
"@cite_4",
"@cite_11"
],
"mid": [
"2099549054",
"2144108520",
"1990627282",
"2157581667"
],
"abstract": [
"Simple statistical forecasting rules, which are usually simplifications of classical models, have been shown to make better predictions than more complex rules, especially when the future values of a criterion are highly uncertain. In this article, we provide evidence that some of the fast and frugal heuristics that people use intuitively are able to make forecasts that are as good as or better than those of knowledge-intensive procedures. We draw from research on the adaptive toolbox and ecological rationality to demonstrate the power of using intuitive heuristics for forecasting in various domains including sport, business, and crime.",
"Summary In the management literature, heuristics are often conceived of as a source of systematic error, whereas logic and statistics are regarded as the sine qua non of good decision making. Yet, this view can be incorrect for decisions made under uncertainty, as opposed to risk. Research on fast and frugal heuristics shows that simple heuristics can be successful in complex, uncertain environments and also when and why this is the case. This article describes the conceptual framework of heuristics as adaptive decision strategies and connects it with the managerial literature. We review five classes of heuristics, analyze their common building blocks, and show how these are applied in managerial decision making. We conclude by highlighting some prominent opportunities for future research in the field. In the uncertain world of management, simple heuristics can lead to better and faster decisions than complex statistical procedures. Copyright © 2014 John Wiley & Sons, Ltd.",
"Abstract Recently, academics have shown interest and enthusiasm in the development and implementation of stochastic customer base analysis models, such as the Pareto NBD model and the BG NBD model. Using the information these models provide, customer managers should be able to (1) distinguish active customers from inactive customers, (2) generate transaction forecasts for individual customers and determine future best customers, and (3) predict the purchase volume of the entire customer base. However, there is also a growing frustration among academics insofar as these models have not found their way into wide managerial application. To present arguments in favor of or against the use of these models in practice, the authors compare the quality of these models when applied to managerial decision making with the simple heuristics that firms typically use. The authors find that the simple heuristics perform at least as well as the stochastic models with regard to all managerially relevant areas, except for ...",
"Heuristics are efficient cognitive processes that ignore information. In contrast to the widely held view that less processing reduces accuracy, the study of heuristics shows that less information, computation, and time can in fact improve accuracy. We review the major progress made so far: (a) the discovery of less-is-more effects; (b) the study of the ecological rationality of heuristics, which examines in which environments a given strategy succeeds or fails, and why; (c) an advancement from vague labels to computational models of heuristics; (d) the development of a systematic theory of heuristics that identifies their building blocks and the evolved capacities they exploit, and views the cognitive system as relying on an ‘‘adaptive toolbox;’’ and (e) the development of an empirical methodology that accounts for individual differences, conducts competitive tests, and has provided evidence for people’s adaptive use of heuristics. Homo heuristicus has a biased mind and ignores part of the available information, yet a biased mind can handle uncertainty more efficiently and robustly than an unbiased mind relying on more resource-intensive and general-purpose processing strategies."
]
} |
1607.03250 | 2495425901 | State-of-the-art neural networks are getting deeper and wider. While their performance increases with the increasing number of layers and neurons, it is crucial to design an efficient deep architecture in order to reduce computational and memory costs. Designing an efficient neural network, however, is labor intensive requiring many experiments, and fine-tunings. In this paper, we introduce network trimming which iteratively optimizes the network by pruning unimportant neurons based on analysis of their outputs on a large dataset. Our algorithm is inspired by an observation that the outputs of a significant portion of neurons in a large network are mostly zero, regardless of what inputs the network received. These zero activation neurons are redundant, and can be removed without affecting the overall accuracy of the network. After pruning the zero activation neurons, we retrain the network using the weights before pruning as initialization. We alternate the pruning and retraining to further reduce zero activations in a network. Our experiments on the LeNet and VGG-16 show that we can achieve high compression ratio of parameters without losing or even achieving higher accuracy than the original network. | Significant redundancy has been demonstrated in several deep learning models @cite_1 and such redundancy is mainly caused by the overwhelming amount of parameters in deep neural networks. An over-parameterized model not only wastes memory and computation, but also leads to serious overfitting problem. Therefore, reducing the number of parameters has been studied by many researchers in this field. However, there is little work directly addressing the optimization of the number of neurons. Most previous works on improving network architectures fall in two main categories; one concentrates on the high level architectural design and the other focuses on low level weight pruning. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2952899695"
],
"abstract": [
"We demonstrate that there is significant redundancy in the parameterization of several deep learning models. Given only a few weight values for each feature it is possible to accurately predict the remaining values. Moreover, we show that not only can the parameter values be predicted, but many of them need not be learned at all. We train several different architectures by learning only a small number of weights and predicting the rest. In the best case we are able to predict more than 95 of the weights of a network without any drop in accuracy."
]
} |
1607.03250 | 2495425901 | State-of-the-art neural networks are getting deeper and wider. While their performance increases with the increasing number of layers and neurons, it is crucial to design an efficient deep architecture in order to reduce computational and memory costs. Designing an efficient neural network, however, is labor intensive requiring many experiments, and fine-tunings. In this paper, we introduce network trimming which iteratively optimizes the network by pruning unimportant neurons based on analysis of their outputs on a large dataset. Our algorithm is inspired by an observation that the outputs of a significant portion of neurons in a large network are mostly zero, regardless of what inputs the network received. These zero activation neurons are redundant, and can be removed without affecting the overall accuracy of the network. After pruning the zero activation neurons, we retrain the network using the weights before pruning as initialization. We alternate the pruning and retraining to further reduce zero activations in a network. Our experiments on the LeNet and VGG-16 show that we can achieve high compression ratio of parameters without losing or even achieving higher accuracy than the original network. | On the high level side, some researchers invented new layers or modules to substitute main bottleneck components in conventional neural networks. Two famous examples of this kind are the global average pooling in Network in Network @cite_2 invented to replace the extremely dense paramaterized fully connected layer and the inception module employed by GoogLeNet @cite_18 to avoid explosion in computational complexity at later stage. Both methods achieve state-of-the-art results on several benchmarks with much less memory and computation consumption. More recently, SqueezeNet @cite_6 used a Fire module together with other strategies to achieve AlexNet-level accuracy with @math less parameters. | {
"cite_N": [
"@cite_18",
"@cite_6",
"@cite_2"
],
"mid": [
"2950179405",
"2279098554",
""
],
"abstract": [
"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.",
"Recent research on deep neural networks has focused primarily on improving accuracy. For a given accuracy level, it is typically possible to identify multiple DNN architectures that achieve that accuracy level. With equivalent accuracy, smaller DNN architectures offer at least three advantages: (1) Smaller DNNs require less communication across servers during distributed training. (2) Smaller DNNs require less bandwidth to export a new model from the cloud to an autonomous car. (3) Smaller DNNs are more feasible to deploy on FPGAs and other hardware with limited memory. To provide all of these advantages, we propose a small DNN architecture called SqueezeNet. SqueezeNet achieves AlexNet-level accuracy on ImageNet with 50x fewer parameters. Additionally, with model compression techniques we are able to compress SqueezeNet to less than 0.5MB (510x smaller than AlexNet).",
""
]
} |
1607.03250 | 2495425901 | State-of-the-art neural networks are getting deeper and wider. While their performance increases with the increasing number of layers and neurons, it is crucial to design an efficient deep architecture in order to reduce computational and memory costs. Designing an efficient neural network, however, is labor intensive requiring many experiments, and fine-tunings. In this paper, we introduce network trimming which iteratively optimizes the network by pruning unimportant neurons based on analysis of their outputs on a large dataset. Our algorithm is inspired by an observation that the outputs of a significant portion of neurons in a large network are mostly zero, regardless of what inputs the network received. These zero activation neurons are redundant, and can be removed without affecting the overall accuracy of the network. After pruning the zero activation neurons, we retrain the network using the weights before pruning as initialization. We alternate the pruning and retraining to further reduce zero activations in a network. Our experiments on the LeNet and VGG-16 show that we can achieve high compression ratio of parameters without losing or even achieving higher accuracy than the original network. | On the low level side, different methods have been explored to reduce number of connections and weights in neural networks. Some late 20th century methods, such as magnitude-based approach @cite_9 and Hessian matrix based approach @cite_12 , prune weights basing on numerical properties of the weights and loss functions without any external data involved. recently proposed an iterative method @cite_8 to prune connections in deep architectures, together with an external compression by quantization and encoding @cite_16 . The network is first pruned by removing low weights connections. Then, learned mapping of similar weights to fixed bits are used to perform quantization of weights after pruning, which facilitates the Huffman coding compression in the last stage to reduce bits for storage. When all three techniques used in pipeline, the number of parameters in the network can be reduced by around @math . | {
"cite_N": [
"@cite_9",
"@cite_16",
"@cite_12",
"@cite_8"
],
"mid": [
"",
"2119144962",
"2125389748",
"2963674932"
],
"abstract": [
"",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.",
"We investigate the use of information from all second order derivatives of the error function to perform network pruning (i.e., removing unimportant weights from a trained network) in order to improve generalization, simplify networks, reduce hardware or storage requirements, increase the speed of further training, and in some cases enable rule extraction. Our method, Optimal Brain Surgeon (OBS), is Significantly better than magnitude-based methods and Optimal Brain Damage [Le Cun, Denker and Solla, 1990], which often remove the wrong weights. OBS permits the pruning of more weights than other methods (for the same error on the training set), and thus yields better generalization on test data. Crucial to OBS is a recursion relation for calculating the inverse Hessian matrix H-1 from training data and structural information of the net. OBS permits a 90 , a 76 , and a 62 reduction in weights over backpropagation with weight decay on three benchmark MONK's problems [, 1991]. Of OBS, Optimal Brain Damage, and magnitude-based methods, only OBS deletes the correct weights from a trained XOR network in every case. Finally, whereas Sejnowski and Rosenberg [1987] used 18,000 weights in their NETtalk network, we used OBS to prune a network to just 1560 weights, yielding better generalization.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the total number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy."
]
} |
1607.03343 | 2513748609 | In this paper, we propose a novel encoder-decoder neural network model referred to as DeepBinaryMask for video compressive sensing. In video compressive sensing one frame is acquired using a set of coded masks (sensing matrix) from which a number of video frames is reconstructed, equal to the number of coded masks. The proposed framework is an end-to-end model where the sensing matrix is trained along with the video reconstruction. The encoder learns the binary elements of the sensing matrix and the decoder is trained to recover the unknown video sequence. The reconstruction performance is found to improve when using the trained sensing mask from the network as compared to other mask designs such as random, across a wide variety of compressive sensing reconstruction algorithms. Finally, our analysis and discussion offers insights into understanding the characteristics of the trained mask designs that lead to the improved reconstruction quality. | Differently from these works, our proposed approach does not impose any mask constraints but instead generates mask patterns learnt from the training data. To the best of our knowledge this is the first study that investigates the construction of an optimized binary CS mask through DNNs. DNNs for image recovery. The capabilities of deep architectures have been investigated in image recovery problems such as deconvolution @cite_30 @cite_46 @cite_29 , denoising @cite_24 @cite_26 @cite_0 , inpainting @cite_21 , and super-resolution @cite_3 @cite_16 @cite_20 @cite_40 @cite_14 . Deep architectures have also been proposed for CS of still images. @cite_11 , stacked denoising auto-encoders (SDAs) were employed to learn a mapping between the CS measurements and image blocks. A similar approach was also utilized in @cite_7 but instead of SDAs, convolutional neural networks (CNNs) were used. | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_14",
"@cite_7",
"@cite_29",
"@cite_21",
"@cite_3",
"@cite_24",
"@cite_0",
"@cite_40",
"@cite_46",
"@cite_16",
"@cite_20",
"@cite_11"
],
"mid": [
"1973567017",
"",
"",
"2950577421",
"",
"",
"",
"2037642501",
"2145094598",
"2184360182",
"1916935112",
"1885185971",
"2320725294",
"2219727625"
],
"abstract": [
"Image deconvolution is the ill-posed problem of recovering a sharp image, given a blurry one generated by a convolution. In this work, we deal with space-invariant non-blind deconvolution. Currently, the most successful methods involve a regularized inversion of the blur in Fourier domain as a first step. This step amplifies and colors the noise, and corrupts the image information. In a second (and arguably more difficult) step, one then needs to remove the colored noise, typically using a cleverly engineered algorithm. However, the methods based on this two-step approach do not properly address the fact that the image information has been corrupted. In this work, we also rely on a two-step procedure, but learn the second step on a large dataset of natural images, using a neural network. We will show that this approach outperforms the current state-of-the-art on a large dataset of artificially blurred images. We demonstrate the practical applicability of our method in a real-world example with photographic out-of-focus blur.",
"",
"",
"The goal of this paper is to present a non-iterative and more importantly an extremely fast algorithm to reconstruct images from compressively sensed (CS) random measurements. To this end, we propose a novel convolutional neural network (CNN) architecture which takes in CS measurements of an image as input and outputs an intermediate reconstruction. We call this network, ReconNet. The intermediate reconstruction is fed into an off-the-shelf denoiser to obtain the final reconstructed image. On a standard dataset of images we show significant improvements in reconstruction results (both in terms of PSNR and time complexity) over state-of-the-art iterative CS reconstruction algorithms at various measurement rates. Further, through qualitative experiments on real data collected using our block single pixel camera (SPC), we show that our network is highly robust to sensor noise and can recover visually better quality images than competitive algorithms at extremely low sensing rates of 0.1 and 0.04. To demonstrate that our algorithm can recover semantically informative images even at a low measurement rate of 0.01, we present a very robust proof of concept real-time visual tracking application.",
"",
"",
"",
"Image denoising can be described as the problem of mapping from a noisy image to a noise-free image. The best currently available denoising methods approximate this mapping with cleverly engineered algorithms. In this work we attempt to learn this mapping directly with a plain multi layer perceptron (MLP) applied to image patches. While this has been done before, we will show that by training on large image databases we are able to compete with the current state-of-the-art image denoising methods. Furthermore, our approach is easily adapted to less extensively studied types of noise (by merely exchanging the training data), for which we achieve excellent results as well.",
"We explore an original strategy for building deep networks, based on stacking layers of denoising autoencoders which are trained locally to denoise corrupted versions of their inputs. The resulting algorithm is a straightforward variation on the stacking of ordinary autoencoders. It is however shown on a benchmark of classification problems to yield significantly lower classification error, thus bridging the performance gap with deep belief networks (DBN), and in several cases surpassing it. Higher level representations learnt in this purely unsupervised fashion also help boost the performance of subsequent SVM classifiers. Qualitative experiments show that, contrary to ordinary autoencoders, denoising autoencoders are able to learn Gabor-like edge detectors from natural image patches and larger stroke detectors from digit images. This work clearly establishes the value of using a denoising criterion as a tractable unsupervised objective to guide the learning of useful higher level representations.",
"Super resolving a low-resolution video is usually handled by either single-image super-resolution (SR) or multi-frame SR. Single-Image SR deals with each video frame independently, and ignores intrinsic temporal dependency of video frames which actually plays a very important role in video super-resolution. Multi-Frame SR generally extracts motion information, e.g., optical flow, to model the temporal dependency, which often shows high computational cost. Considering that recurrent neural networks (RNNs) can model long-term contextual information of temporal sequences well, we propose a bidirectional recurrent convolutional network for efficient multi-frame SR. Different from vanilla RNNs, 1) the commonly-used recurrent full connections are replaced with weight-sharing convolutional connections and 2) conditional convolutional connections from previous input layers to the current hidden layer are added for enhancing visual-temporal dependency modelling. With the powerful temporal dependency modelling, our model can super resolve videos with complex motions and achieve state-of-the-art performance. Due to the cheap convolution operations, our model has a low computational complexity and runs orders of magnitude faster than other multi-frame methods.",
"In this paper, we address the problem of estimating and removing non-uniform motion blur from a single blurry image. We propose a deep learning approach to predicting the probabilistic distribution of motion blur at the patch level using a convolutional neural network (CNN). We further extend the candidate set of motion kernels predicted by the CNN using carefully designed image rotations. A Markov random field model is then used to infer a dense non-uniform motion blur field enforcing motion smoothness. Finally, motion blur is removed by a non-uniform deblurring model using patch-level image prior. Experimental evaluations show that our approach can effectively estimate and remove complex non-uniform motion blur that is not handled well by previous approaches.",
"We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low high-resolution images. The mapping is represented as a deep convolutional neural network (CNN) that takes the low-resolution image as the input and outputs the high-resolution one. We further show that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network. But unlike traditional methods that handle each component separately, our method jointly optimizes all layers. Our deep CNN has a lightweight structure, yet demonstrates state-of-the-art restoration quality, and achieves fast speed for practical on-line usage. We explore different network structures and parameter settings to achieve trade-offs between performance and speed. Moreover, we extend our network to cope with three color channels simultaneously, and show better overall reconstruction quality.",
"Convolutional neural networks (CNN) are a special type of deep neural networks (DNN). They have so far been successfully applied to image super-resolution (SR) as well as other image restoration tasks. In this paper, we consider the problem of video super-resolution. We propose a CNN that is trained on both the spatial and the temporal dimensions of videos to enhance their spatial resolution. Consecutive frames are motion compensated and used as input to a CNN that provides super-resolved video frames as output. We investigate different options of combining the video frames within one CNN architecture. While large image databases are available to train deep neural networks, it is more challenging to create a large video database of sufficient quality to train neural nets for video restoration. We show that by using images to pretrain our model, a relatively small video database is sufficient for the training of our model to achieve and even improve upon the current state-of-the-art. We compare our proposed approach to current video as well as image SR algorithms.",
"In this paper, we develop a new framework for sensing and recovering structured signals. In contrast to compressive sensing (CS) systems that employ linear measurements, sparse representations, and computationally complex convex greedy algorithms, we introduce a deep learning framework that supports both linear and mildly nonlinear measurements, that learns a structured representation from training data, and that efficiently computes a signal estimate. In particular, we apply a stacked denoising autoencoder (SDA), as an unsupervised feature learner. SDA enables us to capture statistical dependencies between the different elements of certain signals and improve signal recovery performance as compared to the CS approach."
]
} |
1607.03476 | 2951785524 | We present a method for training CNN-based object class detectors directly using mean average precision (mAP) as the training loss, in a truly end-to-end fashion that includes non-maximum suppression (NMS) at training time. This contrasts with the traditional approach of training a CNN for a window classification loss, then applying NMS only at test time, when mAP is used as the evaluation metric in place of classification accuracy. However, mAP following NMS forms a piecewise-constant structured loss over thousands of windows, with gradients that do not convey useful information for gradient descent. Hence, we define new, general gradient-like quantities for piecewise constant functions, which have wide applicability. We describe how to calculate these efficiently for mAP following NMS, enabling to train a detector based on Fast R-CNN directly for mAP. This model achieves equivalent performance to the standard Fast R-CNN on the PASCAL VOC 2007 and 2012 datasets, while being conceptually more appealing as the very same model and loss are used at both training and test time. | Nearly all works on object class detection train a window classifier, and ignore NMS and mAP at training time. Earlier approaches @cite_9 @cite_17 @cite_11 @cite_14 apply the classifier to all windows in a dense regular grid, while more recently, object proposal methods @cite_24 @cite_18 have been used to greatly reduce the number of windows @cite_10 @cite_25 . Below we review the few works that try to either train for AP or other structured losses, or include NMS at training time. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_9",
"@cite_17",
"@cite_24",
"@cite_10",
"@cite_25",
"@cite_11"
],
"mid": [
"7746136",
"2164598857",
"2168356304",
"",
"2128715914",
"",
"2102605133",
""
],
"abstract": [
"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 describes a machine learning approach for visual object detection which is capable of processing images extremely rapidly and achieving high detection rates. This work is distinguished by three key contributions. The first is the introduction of a new image representation called the \"integral image\" which allows the features used by our detector to be computed very quickly. The second is a learning algorithm, based on AdaBoost, which selects a small number of critical visual features from a larger set and yields extremely efficient classifiers. The third contribution is a method for combining increasingly more complex classifiers in a \"cascade\" which allows background regions of the image to be quickly discarded while spending more computation on promising object-like regions. The cascade can be viewed as an object specific focus-of-attention mechanism which unlike previous approaches provides statistical guarantees that discarded regions are unlikely to contain the object of interest. In the domain of face detection the system yields detection rates comparable to the best previous systems. Used in real-time applications, the detector runs at 15 frames per second without resorting to image differencing or skin color detection.",
"We describe an object detection system based on mixtures of multiscale deformable part models. Our system is able to represent highly variable object classes and achieves state-of-the-art results in the PASCAL object detection challenges. While deformable part models have become quite popular, their value had not been demonstrated on difficult benchmarks such as the PASCAL data sets. Our system relies on new methods for discriminative training with partially labeled data. We combine a margin-sensitive approach for data-mining hard negative examples with a formalism we call latent SVM. A latent SVM is a reformulation of MI--SVM in terms of latent variables. A latent SVM is semiconvex, and the training problem becomes convex once latent information is specified for the positive examples. This leads to an iterative training algorithm that alternates between fixing latent values for positive examples and optimizing the latent SVM objective function.",
"",
"We present a generic objectness measure, quantifying how likely it is for an image window to contain an object of any class. We explicitly train it to distinguish objects with a well-defined boundary in space, such as cows and telephones, from amorphous background elements, such as grass and road. The measure combines in a Bayesian framework several image cues measuring characteristics of objects, such as appearing different from their surroundings and having a closed boundary. This includes an innovative cue measuring the closed boundary characteristic. In experiments on the challenging PASCAL VOC 07 dataset, we show this new cue to outperform a state-of-the-art saliency measure [17], and the combined measure to perform better than any cue alone. Finally, we show how to sample windows from an image according to their objectness distribution and give an algorithm to employ them as location priors for modern class-specific object detectors. In experiments on PASCAL VOC 07 we show this greatly reduces the number of windows evaluated by class-specific object detectors.",
"",
"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.",
""
]
} |
1607.03476 | 2951785524 | We present a method for training CNN-based object class detectors directly using mean average precision (mAP) as the training loss, in a truly end-to-end fashion that includes non-maximum suppression (NMS) at training time. This contrasts with the traditional approach of training a CNN for a window classification loss, then applying NMS only at test time, when mAP is used as the evaluation metric in place of classification accuracy. However, mAP following NMS forms a piecewise-constant structured loss over thousands of windows, with gradients that do not convey useful information for gradient descent. Hence, we define new, general gradient-like quantities for piecewise constant functions, which have wide applicability. We describe how to calculate these efficiently for mAP following NMS, enabling to train a detector based on Fast R-CNN directly for mAP. This model achieves equivalent performance to the standard Fast R-CNN on the PASCAL VOC 2007 and 2012 datasets, while being conceptually more appealing as the very same model and loss are used at both training and test time. | @cite_28 formulate object detection as a structured prediction problem, outputing a binary indicator for object presence and a set of bounding-box coordinates. This is trained using a structured SVM, with a task loss that aims for correct classification and maximal IoU of predicted and ground-truth boxes in images containing the target class. Like our method, this is a structured loss involving IoU of detections and ground-truth objects; however, it does not correspond to maximising AP, and only a single detection is returned in each image, so there is no NMS. More recently, @cite_3 uses the same structured SVM loss, but with a CNN in place of a kernelised linear model over SURF features @cite_28 . This work directly optimises the structured SVM loss via gradient descent, allowing backpropagation to update the nonlinear CNN layers. | {
"cite_N": [
"@cite_28",
"@cite_3"
],
"mid": [
"1525954826",
"1960182310"
],
"abstract": [
"Sliding window classifiers are among the most successful and widely applied techniques for object localization. However, training is typically done in a way that is not specific to the localization task. First a binary classifier is trained using a sample of positive and negative examples, and this classifier is subsequently applied to multiple regions within test images. We propose instead to treat object localization in a principled way by posing it as a problem of predicting structured data: we model the problem not as binary classification, but as the prediction of the bounding box of objects located in images. The use of a joint-kernelframework allows us to formulate the training procedure as a generalization of an SVM, which can be solved efficiently. We further improve computational efficiency by using a branch-and-bound strategy for localization during both training and testing. Experimental evaluation on the PASCAL VOC and TU Darmstadt datasets show that the structured training procedure improves performance over binary training as well as the best previously published scores.",
"Object detection systems based on the deep convolutional neural network (CNN) have recently made ground-breaking advances on several object detection benchmarks. While the features learned by these high-capacity neural networks are discriminative for categorization, inaccurate localization is still a major source of error for detection. Building upon high-capacity CNN architectures, we address the localization problem by 1) using a search algorithm based on Bayesian optimization that sequentially proposes candidate regions for an object bounding box, and 2) training the CNN with a structured loss that explicitly penalizes the localization inaccuracy. In experiments, we demonstrate that each of the proposed methods improves the detection performance over the baseline method on PASCAL VOC 2007 and 2012 datasets. Furthermore, two methods are complementary and significantly outperform the previous state-of-the-art when combined."
]
} |
1607.03385 | 2497710620 | Networks are difficult to configure correctly, and tricky to debug. These problems are accentuated by temporal and stateful behavior. Static verification, while useful, is ineffectual for detecting behavioral deviations induced by hardware faults, security failures, and so on, so dynamic property monitoring is also valuable. Unfortunately, existing monitoring and runtime verification for networks largely focuses on properties about individual packets (such as connectivity) or requires a digest of all network events be sent to a server, incurring enormous cost. We present a network monitoring system that avoids these problems. Because traces of network events correspond well to temporal logic, we use a subset of Metric First-Order Temporal Logic as the query language. These queries are compiled down to execute completely on the network switches. This vastly reduces network load, improves the precision of queries, and decreases detection latency. We show the practical feasibility of our work by extending a widely-used software switch and deploying it on networks. Our work also suggests improvements to network instruction sets to better support temporal monitoring. | Narayana, et al @cite_10 compile path queries'' about packet forwarding to switches. Path queries, by their nature, are distributed across multiple switches, whereas we compile stateful queries only for single switches. Path queries are restricted to regular expressions over possible forwarding paths, and so cannot express the array of stateful properties that targets, but do not require extensions to OpenFlow. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2303191924"
],
"abstract": [
"Measuring the flow of traffic along network paths is crucial for many management tasks, including traffic engineering, diagnosing congestion, and mitigating DDoS attacks. We introduce a declarative query language for efficient path-based traffic monitoring. Path queries are specified as regular expressions over predicates on packet locations and header values, with SQL-like \"groupby\" constructs for aggregating results anywhere along a path. A run-time system compiles queries into a deterministic finite automaton. The automaton's transition function is then partitioned, compiled into match-action rules, and distributed over the switches. Switches stamp packets with automaton states to track the progress towards fulfilling a query. Only when packets satisfy a query are the packets counted, sampled, or sent to collectors for further analysis. By processing queries in the data plane, users \"pay as they go\", as data-collection overhead is limited to exactly those packets that satisfy the query. We implemented our system on top of the Pyretic SDN controller and evaluated its performance on a campus topology. Our experiments indicate that the system can enable \"interactive debugging\"-- compiling multiple queries in a few seconds--while fitting rules comfortably in modern switch TCAMs and the automaton state into two bytes (e.g., a VLAN header)."
]
} |
1607.03385 | 2497710620 | Networks are difficult to configure correctly, and tricky to debug. These problems are accentuated by temporal and stateful behavior. Static verification, while useful, is ineffectual for detecting behavioral deviations induced by hardware faults, security failures, and so on, so dynamic property monitoring is also valuable. Unfortunately, existing monitoring and runtime verification for networks largely focuses on properties about individual packets (such as connectivity) or requires a digest of all network events be sent to a server, incurring enormous cost. We present a network monitoring system that avoids these problems. Because traces of network events correspond well to temporal logic, we use a subset of Metric First-Order Temporal Logic as the query language. These queries are compiled down to execute completely on the network switches. This vastly reduces network load, improves the precision of queries, and decreases detection latency. We show the practical feasibility of our work by extending a widely-used software switch and deploying it on networks. Our work also suggests improvements to network instruction sets to better support temporal monitoring. | NetSight @cite_24 captures packet trajectories by capturing digests for each packet, which are processed off-switch. As our work focuses on compiling temporal, multi-packet queries to switches, it is largely orthogonal, but our extensions to switch rules potentially provide NetSight with a means to reduce the number of packet digests produced. Moreover, NetSight differentiates packets by hashing, whereas our approach allows a more fine-grained distinction. | {
"cite_N": [
"@cite_24"
],
"mid": [
"1408671314"
],
"abstract": [
"The complexity of networks has outpaced our tools to debug them; today, administrators use manual tools to diagnose problems. In this paper, we show how packet histories--the full stories of every packet's journey through the network--can simplify network diagnosis. To demonstrate the usefulness of packet histories and the practical feasibility of constructing them, we built NetSight, an extensible platform that captures packet histories and enables applications to concisely and flexibly retrieve packet histories of interest. Atop NetSight, we built four applications that illustrate its flexibility: an interactive network debugger, a live invariant monitor, a path-aware history logger, and a hierarchical network profiler. On a single modern multi-core server, NetSight can process packet histories for the traffic of multiple 10 Gb s links. For larger networks, NetSight scales linearly with additional servers and scales even further with straightforward additions to hardware- and hypervisor-based switches."
]
} |
1607.03385 | 2497710620 | Networks are difficult to configure correctly, and tricky to debug. These problems are accentuated by temporal and stateful behavior. Static verification, while useful, is ineffectual for detecting behavioral deviations induced by hardware faults, security failures, and so on, so dynamic property monitoring is also valuable. Unfortunately, existing monitoring and runtime verification for networks largely focuses on properties about individual packets (such as connectivity) or requires a digest of all network events be sent to a server, incurring enormous cost. We present a network monitoring system that avoids these problems. Because traces of network events correspond well to temporal logic, we use a subset of Metric First-Order Temporal Logic as the query language. These queries are compiled down to execute completely on the network switches. This vastly reduces network load, improves the precision of queries, and decreases detection latency. We show the practical feasibility of our work by extending a widely-used software switch and deploying it on networks. Our work also suggests improvements to network instruction sets to better support temporal monitoring. | Numerous other tools (e.g., @cite_12 @cite_0 @cite_7 ) perform runtime monitoring, but either lack compilation to switches or monitor only single-packet paths through the network. Others (e.g., @cite_39 @cite_37 @cite_25 ) focus on data analysis after the fact, rather than real-time monitoring. | {
"cite_N": [
"@cite_37",
"@cite_7",
"@cite_39",
"@cite_0",
"@cite_25",
"@cite_12"
],
"mid": [
"2286028193",
"2146012756",
"2147236343",
"2146191242",
"2408119291",
"2137845741"
],
"abstract": [
"Effective analysis of raw data from networked systems requires bridging the semantic gap between the data and the user's high-level understanding of the system. The raw data represents facts about the system state and analysis involves identifying a set of semantically relevant behaviors, which represent \"interesting\" relationships between these facts. Current analysis tools, such as wireshark and splunk, restrict analysis to the low-level of individual facts and provide limited constructs to aid users in bridging the semantic gap. Our objective is to enable semantic analysis at a level closer to the user's understanding of the system or process. The key to our approach is the introduction of a logic-based formulation of high-level behavior abstractions as a sequence or a group of related facts. This allows treating behavior representations as fundamental analysis primitives, elevating analysis to a higher semantic-level of abstraction. In this paper, we propose a behavior-based semantic analysis framework which provides: (a) a formal language for modeling high-level assertions over networked systems data as behavior models, (b) an analysis engine for extracting instances of user-specified behavior models from raw data. Our approach emphasizes reuse, composibility and extensibility of abstractions. We demonstrate the effectiveness of our approach by applying it to five analyses tasks; modeling a hypothesis on traffic traces, modeling experiment behavior, modeling a security threat, modeling dynamic change and composing higher-level models. Finally, we discuss the performance of our framework in terms of behavior complexity and number of input records.",
"Debugging faults in complex networks often requires capturing and analyzing traffic at the packet level. In this task, datacenter networks (DCNs) present unique challenges with their scale, traffic volume, and diversity of faults. To troubleshoot faults in a timely manner, DCN administrators must a) identify affected packets inside large volume of traffic; b) track them across multiple network components; c) analyze traffic traces for fault patterns; and d) test or confirm potential causes. To our knowledge, no tool today can achieve both the specificity and scale required for this task. We present Everflow, a packet-level network telemetry system for large DCNs. Everflow traces specific packets by implementing a powerful packet filter on top of \"match and mirror\" functionality of commodity switches. It shuffles captured packets to multiple analysis servers using load balancers built on switch ASICs, and it sends \"guided probes\" to test or confirm potential faults. We present experiments that demonstrate Everflow's scalability, and share experiences of troubleshooting network faults gathered from running it for over 6 months in Microsoft's DCNs.",
"Software bugs are inevitable in software-defined networking control software, and troubleshooting is a tedious, time-consuming task. In this paper we discuss how to improve control software troubleshooting by presenting a technique for automatically identifying a minimal sequence of inputs responsible for triggering a given bug, without making assumptions about the language or instrumentation of the software under test. We apply our technique to five open source SDN control platforms---Floodlight, NOX, POX, Pyretic, ONOS---and illustrate how the minimal causal sequences our system found aided the troubleshooting process.",
"Software-defined networking provides flexibility in designing networks by allowing distributed network state to be managed by logically centralized control programs. However, this flexibility brings added complexity, which requires new debugging tools that can provide insights into network behavior. We propose a tool, SDN traceroute, that can query the current path taken by any packet through an SDN-enabled network. The path is traced by using the actual forwarding mechanisms at each SDN-enabled device without changing the forwarding rules being measured. This enables administrators to discover the forwarding behavior for arbitrary Ethernet packets, as well as debug problems in both switch and controller logic. Our prototype implementation requires only a few high-priority rules per device, runs on commodity hardware using only the required features of the OpenFlow 1.0 specification, and can generate traces in about one millisecond per hop.",
"Debugging operational networks can be a daunting task, due to their size, distributed state, and the presence of black box components such as commercial routers and switches, which are poorly instrumentable and only coarsely configurable. The debugging tool set available to administrators is limited, and provides only aggregated statistics (SNMP), sampled data (NetFlow sFlow), or local measurements on single hosts (tcpdump). In this paper, we leverage split forwarding architectures such as OpenFlow to add record and replay debugging capabilities to networks - a powerful, yet currently lacking approach. We present the design of OFRewind, which enables scalable, multi-granularity, temporally consistent recording and coordinated replay in a network, with fine-grained, dynamic, centrally orchestrated control over both record and replay. Thus, OFRewind helps operators to reproduce software errors, identify datapath limitations, or locate configuration errors.",
"Software-defined networks facilitate rapid and open innovation at the network control layer by providing a programmable network infrastructure for computing flow policies on demand. However, the dynamism of programmable networks also introduces new security challenges that demand innovative solutions. A critical challenge is efficient detection and reconciliation of potentially conflicting flow rules imposed by dynamic OpenFlow (OF) applications. To that end, we introduce FortNOX, a software extension that provides role-based authorization and security constraint enforcement for the NOX OpenFlow controller. FortNOX enables NOX to check flow rule contradictions in real time, and implements a novel analysis algorithm that is robust even in cases where an adversarial OF application attempts to strategically insert flow rules that would otherwise circumvent flow rules imposed by OF security applications. We demonstrate the utility of FortNOX through a prototype implementation and use it to examine performance and efficiency aspects of the proposed framework."
]
} |
1607.03385 | 2497710620 | Networks are difficult to configure correctly, and tricky to debug. These problems are accentuated by temporal and stateful behavior. Static verification, while useful, is ineffectual for detecting behavioral deviations induced by hardware faults, security failures, and so on, so dynamic property monitoring is also valuable. Unfortunately, existing monitoring and runtime verification for networks largely focuses on properties about individual packets (such as connectivity) or requires a digest of all network events be sent to a server, incurring enormous cost. We present a network monitoring system that avoids these problems. Because traces of network events correspond well to temporal logic, we use a subset of Metric First-Order Temporal Logic as the query language. These queries are compiled down to execute completely on the network switches. This vastly reduces network load, improves the precision of queries, and decreases detection latency. We show the practical feasibility of our work by extending a widely-used software switch and deploying it on networks. Our work also suggests improvements to network instruction sets to better support temporal monitoring. | FAST @cite_14 , OpenState @cite_1 , and POF @cite_38 conservatively extend the power of OpenFlow rules. In all of these, switches maintain additional per-flow'' state. A flow is defined by matching a set of header fields, giving a fixed equivalence relation on traffic: equivalent packets use the same index into the state table. While some of our queries could be compiled to these extensions, in general our equivalence relation needs to vary with the current state---one observation may match on IP source address, another on TCP port. FAST and OpenState both assume a deterministic automaton for state evolution, and so using them would prevent us from leveraging non-determinism as we do. The P4 @cite_3 language also allows limited state on switches, in the form of persistent registers. Our state is more self contained since rules can only learn fresh rules or delete their own table ( sec:compile ), rather than affecting other rules---as is possible via shared registers. | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_1",
"@cite_3"
],
"mid": [
"2040678819",
"1983720905",
"2062596448",
"1994926493"
],
"abstract": [
"A flexible and programmable forwarding plane is essential to maximize the value of Software-Defined Networks (SDN). In this paper, we propose Protocol-Oblivious Forwarding (POF) as a key enabler for highly flexible and programmable SDN. Our goal is to remove any dependency on protocol-specific configurations on the forwarding elements and enhance the data-path with new stateful instructions to support genuine software defined networking behavior. A generic flow instruction set (FIS) is defined to fulfill this purpose. POF helps to lower network cost by using commodity forwarding elements and to create new value by enabling numerous innovative network services. We built both hardware-based and open source software-based prototypes to demonstrate the feasibility and advantages of POF. We report the preliminary evaluation results and the insights we learnt from the experiments. POF is future-proof and expressive. We believe it represents a promising direction to evolve the OpenFlow protocol and the future SDN forwarding elements.",
"In software-defined networking, the controller installs flow-based rules at switches either proactively or reactively. The reactive approach allows controller applications to make dynamic decisions about incoming traffic, but performs worse than the proactive one due to the controller involvement. To support dynamic applications with better performance, we propose FAST (Flow-level State Transitions) as a new switch primitive for software-defined networks. With FAST, the controller simply preinstalls a state machine and switches can automatically record flow state transitions by matching incoming packets to installed filters. FAST can support a variety of dynamic applications, and can be readily implemented with today's commodity switch components and software switches.",
"Software Defined Networking envisions smart centralized controllers governing the forwarding behavior of dumb low-cost switches. But are \"dumb\" switches an actual strategic choice, or (at least to some extent) are they a consequence of the lack of viable alternatives to OpenFlow as programmatic data plane forwarding interface? Indeed, some level of (programmable) control logic in the switches might be beneficial to offload logically centralized controllers (de facto complex distributed systems) from decisions just based on local states (versus network-wide knowledge), which could be handled at wire speed inside the device itself. Also, it would reduce the amount of flow processing tasks currently delegated to specialized middleboxes. The underlying challenge is: can we devise a stateful data plane programming abstraction (versus the stateless OpenFlow match action table) which still entails high performance and remains consistent with the vendors' preference for closed platforms? We posit that a promising answer revolves around the usage of extended finite state machines, as an extension (super-set) of the OpenFlow match action abstraction. We concretely turn our proposed abstraction into an actual table-based API, and, perhaps surprisingly, we show how it can be supported by (mostly) reusing core primitives already implemented in OpenFlow devices.",
"P4 is a high-level language for programming protocol-independent packet processors. P4 works in conjunction with SDN control protocols like OpenFlow. In its current form, OpenFlow explicitly specifies protocol headers on which it operates. This set has grown from 12 to 41 fields in a few years, increasing the complexity of the specification while still not providing the flexibility to add new headers. In this paper we propose P4 as a strawman proposal for how OpenFlow should evolve in the future. We have three goals: (1) Reconfigurability in the field: Programmers should be able to change the way switches process packets once they are deployed. (2) Protocol independence: Switches should not be tied to any specific network protocols. (3) Target independence: Programmers should be able to describe packet-processing functionality independently of the specifics of the underlying hardware. As an example, we describe how to use P4 to configure a switch to add a new hierarchical label."
]
} |
1607.03385 | 2497710620 | Networks are difficult to configure correctly, and tricky to debug. These problems are accentuated by temporal and stateful behavior. Static verification, while useful, is ineffectual for detecting behavioral deviations induced by hardware faults, security failures, and so on, so dynamic property monitoring is also valuable. Unfortunately, existing monitoring and runtime verification for networks largely focuses on properties about individual packets (such as connectivity) or requires a digest of all network events be sent to a server, incurring enormous cost. We present a network monitoring system that avoids these problems. Because traces of network events correspond well to temporal logic, we use a subset of Metric First-Order Temporal Logic as the query language. These queries are compiled down to execute completely on the network switches. This vastly reduces network load, improves the precision of queries, and decreases detection latency. We show the practical feasibility of our work by extending a widely-used software switch and deploying it on networks. Our work also suggests improvements to network instruction sets to better support temporal monitoring. | Purely static tools, such as network configuration analyzers @cite_34 @cite_28 @cite_11 @cite_2 @cite_9 @cite_19 and languages built for analysis (e.g., @cite_13 @cite_35 ) are powerful but not sufficient for robust network testing. Veriflow @cite_29 statically analyzes to (stateless) OpenFlow switch rules in real time. This hybrid approach is nevertheless limited to switches governed by an SDN controller. Since our approach only on SDN switches, it can be used even in a hybrid network. @cite_20 enhance SDN programs with annotations that reference program state directly. These compile to new VeriFlow assertions as the program state changes, and are therefore limited to analyzing switch-rule updates. | {
"cite_N": [
"@cite_35",
"@cite_28",
"@cite_9",
"@cite_29",
"@cite_19",
"@cite_2",
"@cite_34",
"@cite_13",
"@cite_20",
"@cite_11"
],
"mid": [
"",
"2096244038",
"",
"2122695394",
"",
"2115526539",
"273646509",
"",
"2116206868",
"158224344"
],
"abstract": [
"",
"Firewalls are crucial elements in network security, and have been widely deployed in most businesses and institutions for securing private networks. The function of a firewall is to examine each incoming and outgoing packet and decide whether to accept or to discard the packet based on its policy. Due to the lack of tools for analyzing firewall policies, most firewalls on the Internet have been plagued with policy errors. A firewall policy error either creates security holes that will allow malicious traffic to sneak into a private network or blocks legitimate traffic and disrupts normal business processes, which in turn could lead to irreparable, if not tragic, consequences. Because a firewall may have a large number of rules and the rules often conflict, understanding and analyzing the function of a firewall has been known to be notoriously difficult. An effective way to assist firewall administrators to understand and analyze the function of their firewalls is by issuing queries. An example of a firewall query is \"Which computers in the private network can receive packets from a known malicious host in the outside Internet?rdquo Two problems need to be solved in order to make firewall queries practically useful: how to describe a firewall query and how to process a firewall query. In this paper, we first introduce a simple and effective SQL-like query language, called the Structured Firewall Query Language (SFQL), for describing firewall queries. Second, we give a theorem, called the Firewall Query Theorem, as the foundation for developing firewall query processing algorithms. Third, we present an efficient firewall query processing algorithm, which uses decision diagrams as its core data structure. Fourth, we propose methods for optimizing firewall query results. Finally, we present methods for performing the union, intersect, and minus operations on firewall query results. Our experimental results show that our firewall query processing algorithm is very efficient: it takes less than 10 milliseconds to process a query over a firewall that has up to 10,000 rules.",
"",
"Networks are complex and prone to bugs. Existing tools that check configuration files and data-plane state operate offline at timescales of seconds to hours, and cannot detect or prevent bugs as they arise. Is it possible to check network-wide invariants in real time, as the network state evolves? The key challenge here is to achieve extremely low latency during the checks so that network performance is not affected. In this paper, we present a preliminary design, VeriFlow, which suggests that this goal is achievable. VeriFlow is a layer between a software-defined networking controller and network devices that checks for network-wide invariant violations dynamically as each forwarding rule is inserted. Based on an implementation using a Mininet OpenFlow network and Route Views trace data, we find that VeriFlow can perform rigorous checking within hundreds of microseconds per rule insertion.",
"",
"Diagnosing problems in networks is a time-consuming and error-prone process. Existing tools to assist operators primarily focus on analyzing control plane configuration. Configuration analysis is limited in that it cannot find bugs in router software, and is harder to generalize across protocols since it must model complex configuration languages and dynamic protocol behavior. This paper studies an alternate approach: diagnosing problems through static analysis of the data plane. This approach can catch bugs that are invisible at the level of configuration files, and simplifies unified analysis of a network across many protocols and implementations. We present Anteater, a tool for checking invariants in the data plane. Anteater translates high-level network invariants into boolean satisfiability problems (SAT), checks them against network state using a SAT solver, and reports counterexamples if violations have been found. Applied to a large university network, Anteater revealed 23 bugs, including forwarding loops and stale ACL rules, with only five false positives. Nine of these faults are being fixed by campus network operators.",
"Instances of router models and filter models respectively are populated with configuration data from routers and filters in a network. A route advertising graph is derived from the router model instances. The route advertising graph indicates propagation of routes between the ones of the real-world devices serving as routers according to routing protocols implemented by the ones of the real-world devices serving as routers. Consolidated routing data is determined for the ones of the real-world devices serving as routers. In this process, the propagation of routes indicated by the route advertising graph is iterated to stability. For a destination node in the network, a respective route graph indicating available paths to the destination node from each source node in the network is constructed from the consolidated routing data. Services between each source node and the destination node are classified based on a full traversal of the route advertising graph.",
"",
"Software Defined Networking (SDN) provides opportunities for network verification and debugging by offering centralized visibility of the data plane. This has enabled both offline and online data-plane verification. However, little work has gone into the verification of time-varying properties (e.g., dynamic access control), where verification conditions change dynamically in response to application logic, network events, and external stimulus (e.g., operator requests). This paper introduces an assertion language to support verifying and debugging SDN applications with dynamically changing verification conditions. The language allows programmers to annotate controller applications with C-style assertions about the data plane. Assertions consist of regular expressions on paths to describe path properties for classes of packets, and universal and existential quantifiers that range over programmer-defined sets of hosts, switches, or other network entities. As controller programs dynamically add and remove elements from these sets, they generate new verification conditions that the existing data plane must satisfy. This work proposes an incremental data structure together with an underlying verification engine, to avoid naively re-verifying the entire data plane as these verification conditions change. To validate our ideas, we have implemented a debugging library on top of a modified version of VeriFlow, which is easily integrated into existing controller systems with minimal changes. Using this library, we have verified correctness properties for applications on several controller platforms.",
"Network state may change rapidly in response to customer demands, load conditions or configuration changes. But the network must also ensure correctness conditions such as isolating tenants from each other and from critical services. Existing policy checkers cannot verify compliance in real time because of the need to collect \"state\" from the entire network and the time it takes to analyze this state. SDNs provide an opportunity in this respect as they provide a logically centralized view from which every proposed change can be checked for compliance with policy. But there remains the need for a fast compliance checker. Our paper introduces a real time policy checking tool called NetPlumber based on Header Space Analysis (HSA) [8]. Unlike HSA, however, NetPlumber incrementally checks for compliance of state changes, using a novel set of conceptual tools that maintain a dependency graph between rules. While NetPlumber is a natural fit for SDNs, its abstract intermediate form is conceptually applicable to conventional networks as well. We have tested NetPlumber on Google's SDN, the Stanford backbone and Internet 2. With NetPlumber, checking the compliance of a typical rule update against a single policy on these networks takes 50-500µs on average."
]
} |
1607.03420 | 2510811505 | The wireless communications in complex environments, such as underground and underwater, can enable various applications in the environmental, industrial, homeland security, law enforcement, and military fields. However, conventional electromagnetic wave-based techniques do not work due to the lossy media and complicated structures. Magnetic induction (MI) has been proved to achieve reliable communication in such environments. However, due to the small antenna size, the communication range of MI is still very limited, especially for the portable mobile devices. To this end, Metamaterial-enhanced MI (M2I) communication has been proposed, where the theoretical results predict that it can significantly increase the data rate and range. Nevertheless, there exists a significant gap between the theoretical prediction and the practical realization of M2I; the theoretical model relies on an ideal spherical metamaterial, while it does not exist in nature. In this paper, a practical design is proposed by leveraging a spherical coil array to realize M2I communication. The full-wave simulation is conducted to validate the design objectives. By using the spherical coil array-based M2I communication, the communication range can be significantly extended, exactly as we predicted in the ideal M2I model. Finally, the proposed M2I communication is implemented and tested in various environments. | Motivated by this, M @math I communication is introduced to complex environments in @cite_3 to increase the mutual coupling between transceivers and extend the communication range. Different from @cite_41 , the effect of the lossy medium on M @math I antenna is analyzed and simulated. Although an initial prototype is provided in @cite_3 , the size of the prototype is large and an analytical analysis to connect the ideal EM analysis with the prototype is missing. Also, there is a lack of practical implementation guideline on M @math I communication to achieve the promising results in @cite_3 . In this paper, we propose a spherical coil array to realize the metamaterial sphere and M @math I communication. Different from @cite_3 , which is based on EM analysis and simulation, this paper fills the gap between ideal model and real implementation by using equivalent circuit analysis to provide more intuitive understandings. There are various approaches to fabricate metamaterials in 2D and 3D @cite_29 . We adopt a 3D spherical structure with coil arrays to realize the proposed M @math I communication. The unique characteristics of MI and M @math I are also summarized in Table . | {
"cite_N": [
"@cite_41",
"@cite_29",
"@cite_3"
],
"mid": [
"2118503032",
"2755061929",
"2246276066"
],
"abstract": [
"The effect of surrounding an electrically small dipole antenna with a shell of double negative (DNG) material ( spl epsiv sub r <0 and spl mu sub r <0) has been investigated both analytically and numerically. The problem of an infinitesimal electric dipole embedded in a homogeneous DNG medium is treated; its analytical solution shows that this electrically small antenna acts inductively rather than capacitively as it would in free space. It is then shown that a properly designed dipole-DNG shell combination increases the real power radiated by more than an order of magnitude over the corresponding free space case. The reactance of the antenna is shown to have a corresponding decrease. Analysis of the reactive power within this dipole-DNG shell system indicates that the DNG shell acts as a natural matching network for the dipole. An equivalent circuit model is introduced that confirms this explanation. Several cases are presented to illustrate these results. The difficult problem of interpreting the energy stored in this dipole-DNG shell system when the DNG medium is frequency independent and, hence, of calculating the radiation Q is discussed from several points of view.",
"Metamaterials are compound materials exhibiting electromagnetic properties not readily found in nature. Novel antennas, couplers, imaging systems and methods for the reduction of the radar cross-section based on these properties have been proposed in literature. In this thesis a systematic synthesis technique for metamaterials is presented. The foundation of this approach is that metamaterials consist of unit cells that are small compared to the wavelength. The topology of the unit cell's network representation can be derived from first-order space-discretising schemes in 1D, 2D, and 3D. Using this technique two novel isotropic 3D metamaterials are synthesised. Maximally-symmetric and planar physical realisations are proposed and verified by full-wave simulation and experiment, respectively.",
"Magnetic induction (MI) communication technique has shown great potentials in complex and RF-challenging environments, such as underground and underwater, due to its advantage over EM wave-based techniques in penetrating lossy medium. However, the transmission distance of MI techniques is limited since magnetic field attenuates very fast in the near field. To this end, this paper proposes metamaterial-enhanced magnetic induction ( @math ) communication mechanism, where an MI coil antenna is enclosed by a metamaterial shell that can enhance the magnetic fields around the MI transceivers. As a result, the @math communication system can achieve tens of meters communication range by using pocket-sized antennas. In this paper, an analytical channel model is developed to explore the fundamentals of the @math mechanism, in the aspects of communication range and channel capacity, and the susceptibility to various hostile and complex environments. The theoretical model is validated through the finite element simulation software, Comsol Multiphysics. Proof-of-concept experiments are also conducted to validate the feasibility of @math ."
]
} |
1607.02715 | 2950932290 | This paper employs case-based reasoning (CBR) to capture the personal styles of individual artists and generate the human facial portraits from photos accordingly. For each human artist to be mimicked, a series of cases are firstly built-up from her his exemplars of source facial photo and hand-drawn sketch, and then its stylization for facial photo is transformed as a style-transferring process of iterative refinement by looking-for and applying best-fit cases in a sense of style optimization. Two models, fitness evaluation model and parameter estimation model, are learned for case retrieval and adaptation respectively from these cases. The fitness evaluation model is to decide which case is best-fitted to the sketching of current interest, and the parameter estimation model is to automate case adaptation. The resultant sketch is synthesized progressively with an iterative loop of retrieval and adaptation of candidate cases until the desired aesthetic style is achieved. To explore the effectiveness and advantages of the novel approach, we experimentally compare the sketch portraits generated by the proposed method with that of a state-of-the-art example-based facial sketch generation algorithm as well as a couple commercial software packages. The comparisons reveal that our CBR based synthesis method for facial portraits is superior both in capturing and reproducing artists' personal illustration styles to the peer methods. | A multitude of image-based artistic rendering (IB-AR) techniques have been proposed. @cite_28 gave an in depth survey of IB-AR techniques. Here we merely focus on example-based artistic rendering (EBAR), which can be roughly categorized into two classes: model-based and model-free methods. | {
"cite_N": [
"@cite_28"
],
"mid": [
"2109253138"
],
"abstract": [
"This paper surveys the field of nonphotorealistic rendering (NPR), focusing on techniques for transforming 2D input (images and video) into artistically stylized renderings. We first present a taxonomy of the 2D NPR algorithms developed over the past two decades, structured according to the design characteristics and behavior of each technique. We then describe a chronology of development from the semiautomatic paint systems of the early nineties, through to the automated painterly rendering systems of the late nineties driven by image gradient analysis. Two complementary trends in the NPR literature are then addressed, with reference to our taxonomy. First, the fusion of higher level computer vision and NPR, illustrating the trends toward scene analysis to drive artistic abstraction and diversity of style. Second, the evolution of local processing approaches toward edge-aware filtering for real-time stylization of images and video. The survey then concludes with a discussion of open challenges for 2D NPR identified in recent NPR symposia, including topics such as user and aesthetic evaluation."
]
} |
1607.02715 | 2950932290 | This paper employs case-based reasoning (CBR) to capture the personal styles of individual artists and generate the human facial portraits from photos accordingly. For each human artist to be mimicked, a series of cases are firstly built-up from her his exemplars of source facial photo and hand-drawn sketch, and then its stylization for facial photo is transformed as a style-transferring process of iterative refinement by looking-for and applying best-fit cases in a sense of style optimization. Two models, fitness evaluation model and parameter estimation model, are learned for case retrieval and adaptation respectively from these cases. The fitness evaluation model is to decide which case is best-fitted to the sketching of current interest, and the parameter estimation model is to automate case adaptation. The resultant sketch is synthesized progressively with an iterative loop of retrieval and adaptation of candidate cases until the desired aesthetic style is achieved. To explore the effectiveness and advantages of the novel approach, we experimentally compare the sketch portraits generated by the proposed method with that of a state-of-the-art example-based facial sketch generation algorithm as well as a couple commercial software packages. The comparisons reveal that our CBR based synthesis method for facial portraits is superior both in capturing and reproducing artists' personal illustration styles to the peer methods. | Model-based methods acquire prior knowledge of artistic rendering styles from examples and accordingly built-up the stylization models. Besides the model-based approaches in @cite_40 @cite_50 (see ), @cite_49 modeled color style of a source image through the means and standard deviations along each of the three axes in l @math color space, and then imposed the means and standard deviations onto the target image, transferring the color style to target image. @cite_23 models facial pexaggeration style through analyzing the correlation between the image caricature pairs using partial least-squares (PLS). The model-based hatching in @cite_15 trained a mapping from the features of input 3D object to hatching properties. A new illustration was generated in target style according to predicted properties. The aforementioned model-based methods can generate diverse new illustrations with parametric styles through generalization on exemplars. However subtle and unique artistic characteristics related to individual artist are somewhat lost while modelling the stylization. | {
"cite_N": [
"@cite_40",
"@cite_23",
"@cite_49",
"@cite_50",
"@cite_15"
],
"mid": [
"2039755782",
"2153381665",
"2129112648",
"2136407193",
"2153968339"
],
"abstract": [
"We propose a new system to produce pencil drawing from natural images. The results contain various natural strokes and patterns, and are structurally representative. They are accomplished by novelly combining the tone and stroke structures, which complement each other in generating visually constrained results. Prior knowledge on pencil drawing is also incorporated, making the two basic functions robust against noise, strong texture, and significant illumination variation. In light of edge, shadow, and shading information conveyance, our pencil drawing system establishes a style that artists use to understand visual data and draw them. Meanwhile, it lets the results contain rich and well-ordered lines to vividly express the original scene.",
"In this paper, we present a system that automatically generates caricatures from input face images. From example caricatures drawn by an artist, our caricature system learns how an artist draws caricatures. In our approach, we decouple the process of caricature generation into two parts, i.e., shape exaggeration and texture style transferring. The exaggeration of a caricature is accomplished by a prototype-based method that captures the artist's understanding of what are distinctive features of a face and the exaggeration style. Such prototypes are learnt by analyzing the correlation between the image caricature pairs using partial least-squares (PLS). Experimental results demonstrate the effectiveness of our system.",
"We use a simple statistical analysis to impose one image's color characteristics on another. We can achieve color correction by choosing an appropriate source image and apply its characteristic to another image.",
"Automatically locating multiple feature points (i.e., the shape) in a facial image and then synthesizing the corresponding facial sketch are highly challenging since facial images typically exhibit a wide range of poses, expressions, and scales, and have differing degrees of illumination and or occlusion. When the facial sketches are to be synthesized in the unique sketching style of a particular artist, the problem becomes even more complex. To resolve these problems, this paper develops an automatic facial sketch synthesis system based on a novel direct combined model (DCM) algorithm. The proposed system executes three cascaded procedures, namely, (1) synthesis of the facial shape from the input texture information (i.e., the facial image); (2) synthesis of the exaggerated facial shape from the synthesized facial shape; and (3) synthesis of a sketch from the original input image and the synthesized exaggerated shape. Previous proposals for reconstructing facial shapes and synthesizing the corresponding facial sketches are heavily reliant on the quality of the texture reconstruction results, which, in turn, are highly sensitive to occlusion and lighting effects in the input image. However, the DCM approach proposed in this paper accurately reconstructs the facial shape and then produces lifelike synthesized facial sketches without the need to recover occluded feature points or to restore the texture information lost as a result of unfavorable lighting conditions. Moreover, the DCM approach is capable of synthesizing facial sketches from input images with a wide variety of facial poses, gaze directions, and facial expressions even when such images are not included within the original training data set.",
"This article presents an algorithm for learning hatching styles from line drawings. An artist draws a single hatching illustration of a 3D object. Her strokes are analyzed to extract the following per-pixel properties: hatching level (hatching, cross-hatching, or no strokes), stroke orientation, spacing, intensity, length, and thickness. A mapping is learned from input geometric, contextual, and shading features of the 3D object to these hatching properties, using classification, regression, and clustering techniques. Then, a new illustration can be generated in the artist's style, as follows. First, given a new view of a 3D object, the learned mapping is applied to synthesize target stroke properties for each pixel. A new illustration is then generated by synthesizing hatching strokes according to the target properties."
]
} |
1607.02720 | 2498318846 | Deep convolutional neural network (DCNN) has achieved remarkable performance on object detection and speech recognition in recent years. However, the excellent performance of a DCNN incurs high computational complexity and large memory requirement. In this paper, an equal distance nonuniform quantization (ENQ) scheme and a K-means clustering nonuniform quantization (KNQ) scheme are proposed to reduce the required memory storage when low complexity hardware or software implementations are considered. For the VGG-16 and the AlexNet, the proposed nonuniform quantization schemes reduce the number of required memory storage by approximately 50 while achieving almost the same or even better classification accuracy compared to the state-of-the-art quantization method. Compared to the ENQ scheme, the proposed KNQ scheme provides a better tradeoff when higher accuracy is required. | The quantization of DCNNs has been widely discussed in the open literatures. These works can be categorized into weight compression and activation quantization. Many previous works focus on weight compression of convolutional and fully-connect layers. A partial pruning algorithm was proposed in @cite_1 to reduce the memory required by all weights. A compression method, which consists of pruning, quantization and Huffman coding, was proposed in @cite_21 to compress the weights of a DCNN. The coding technology @cite_14 was also employed to perform the quantization of DCNN as well. @cite_11 , it was showed that the weight of fully-connected layers can be compressed by truncated singular value decomposition. A hash trick was proposed in @cite_15 to compress the DCNN. Moreover, binary weights were employed in @cite_3 @cite_19 to reduce both computational complexity and storage requirement at the cost of certain accuracy loss. On the other hand, the quantization of activations has rarely been discussed. @cite_8 , the signal-to-quantization noise ratio (SQNR) metric was employed to compute the number of quantization bits for each activation. The quantization of activations in the fully-connect layer was discussed in @cite_10 . | {
"cite_N": [
"@cite_14",
"@cite_8",
"@cite_21",
"@cite_1",
"@cite_3",
"@cite_19",
"@cite_15",
"@cite_10",
"@cite_11"
],
"mid": [
"1724438581",
"2952936791",
"2119144962",
"2952826672",
"2963114950",
"",
"2952432176",
"",
"2167215970"
],
"abstract": [
"Deep convolutional neural networks (CNN) has become the most promising method for object recognition, repeatedly demonstrating record breaking results for image classification and object detection in recent years. However, a very deep CNN generally involves many layers with millions of parameters, making the storage of the network model to be extremely large. This prohibits the usage of deep CNNs on resource limited hardware, especially cell phones or other embedded devices. In this paper, we tackle this model storage issue by investigating information theoretical vector quantization methods for compressing the parameters of CNNs. In particular, we have found in terms of compressing the most storage demanding dense connected layers, vector quantization methods have a clear gain over existing matrix factorization methods. Simply applying k-means clustering to the weights or conducting product quantization can lead to a very good balance between model size and recognition accuracy. For the 1000-category classification task in the ImageNet challenge, we are able to achieve 16-24 times compression of the network with only 1 loss of classification accuracy using the state-of-the-art CNN.",
"In recent years increasingly complex architectures for deep convolution networks (DCNs) have been proposed to boost the performance on image recognition tasks. However, the gains in performance have come at a cost of substantial increase in computation and model storage resources. Fixed point implementation of DCNs has the potential to alleviate some of these complexities and facilitate potential deployment on embedded hardware. In this paper, we propose a quantizer design for fixed point implementation of DCNs. We formulate and solve an optimization problem to identify optimal fixed point bit-width allocation across DCN layers. Our experiments show that in comparison to equal bit-width settings, the fixed point DCNs with optimized bit width allocation offer >20 reduction in the model size without any loss in accuracy on CIFAR-10 benchmark. We also demonstrate that fine-tuning can further enhance the accuracy of fixed point DCNs beyond that of the original floating point model. In doing so, we report a new state-of-the-art fixed point performance of 6.78 error-rate on CIFAR-10 benchmark.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.",
"Real time application of deep learning algorithms is often hindered by high computational complexity and frequent memory accesses. Network pruning is a promising technique to solve this problem. However, pruning usually results in irregular network connections that not only demand extra representation efforts but also do not fit well on parallel computation. We introduce structured sparsity at various scales for convolutional neural networks, which are channel wise, kernel wise and intra kernel strided sparsity. This structured sparsity is very advantageous for direct computational resource savings on embedded computers, parallel computing environments and hardware based systems. To decide the importance of network connections and paths, the proposed method uses a particle filtering approach. The importance weight of each particle is assigned by computing the misclassification rate with corresponding connectivity pattern. The pruned network is re-trained to compensate for the losses due to pruning. While implementing convolutions as matrix products, we particularly show that intra kernel strided sparsity with a simple constraint can significantly reduce the size of kernel and feature map matrices. The pruned network is finally fixed point optimized with reduced word length precision. This results in significant reduction in the total storage size providing advantages for on-chip memory based implementations of deep neural networks.",
"Deep Neural Networks (DNN) have achieved state-of-the-art results in a wide range of tasks, with the best results obtained with large training sets and large models. In the past, GPUs enabled these breakthroughs because of their greater computational speed. In the future, faster computation at both training and test time is likely to be crucial for further progress and for consumer applications on low-power devices. As a result, there is much interest in research and development of dedicated hardware for Deep Learning (DL). Binary weights, i.e., weights which are constrained to only two possible values (e.g. -1 or 1), would bring great benefits to specialized DL hardware by replacing many multiply-accumulate operations by simple accumulations, as multipliers are the most space and power-hungry components of the digital implementation of neural networks. We introduce BinaryConnect, a method which consists in training a DNN with binary weights during the forward and backward propagations, while retaining precision of the stored weights in which gradients are accumulated. Like other dropout schemes, we show that BinaryConnect acts as regularizer and we obtain near state-of-the-art results with BinaryConnect on the permutation-invariant MNIST, CIFAR-10 and SVHN.",
"",
"As deep nets are increasingly used in applications suited for mobile devices, a fundamental dilemma becomes apparent: the trend in deep learning is to grow models to absorb ever-increasing data set sizes; however mobile devices are designed with very little memory and cannot store such large models. We present a novel network architecture, HashedNets, that exploits inherent redundancy in neural networks to achieve drastic reductions in model sizes. HashedNets uses a low-cost hash function to randomly group connection weights into hash buckets, and all connections within the same hash bucket share a single parameter value. These parameters are tuned to adjust to the HashedNets weight sharing architecture with standard backprop during training. Our hashing procedure introduces no additional memory overhead, and we demonstrate on several benchmark data sets that HashedNets shrink the storage requirements of neural networks substantially while mostly preserving generalization performance.",
"",
"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."
]
} |
1607.03057 | 2949882214 | In this work, we tackle the problem of predicting entity popularity on Twitter based on the news cycle. We apply a supervised learn- ing approach and extract four types of features: (i) signal, (ii) textual, (iii) sentiment and (iv) semantic, which we use to predict whether the popularity of a given entity will be high or low in the following hours. We run several experiments on six different entities in a dataset of over 150M tweets and 5M news and obtained F1 scores over 0.70. Error analysis indicates that news perform better on predicting entity popularity on Twitter when they are the primary information source of the event, in opposition to events such as live TV broadcasts, political debates or football matches. | In recent years, a number of research works have studied the relationship and predictive behavior of user response to the publication of online media items, such as, commenting news articles, playing Youtube videos, sharing URLs or retweeting patterns @cite_11 @cite_15 @cite_17 @cite_13 . The first attempt to predict the volume of user comments for online news articles used both metadata from the news articles and linguistic features @cite_17 . The prediction was divided in two binary classification problems: if an article would get any comments and if it would be high or low number of comments. Similarly other works, found that shallow linguistic features (e.g. TF-IDF or sentiment) and named entities have good predictive power @cite_18 @cite_2 . | {
"cite_N": [
"@cite_18",
"@cite_11",
"@cite_2",
"@cite_15",
"@cite_13",
"@cite_17"
],
"mid": [
"1595917920",
"",
"2184410296",
"2112056172",
"2036254771",
""
],
"abstract": [
"Online user comments contain valuable user opinions. Comments vary greatly in quality and detecting high quality comments is a subtask of opinion mining and summarization research. Finding attentive comments that provide some reasoning is highly valuable in understanding the user’s opinion particularly in sociopolitical opinion mining and aids policy makers, social organizations or government sectors in decision making. In this paper we study the problem of detecting thoughtful comments. We empirically study various textual features, discourse relations and relevance features to predict thoughtful comments. We use logistic regression model and test on the datasets related to sociopolitical content. We found that the most useful features include the discourse relations and relevance features along with basic textual features to predict the comment quality in terms of thoughtfulness. In our experiments on two different datasets, we could achieve a prediction score of 79.37 and 73.47 in terms of F-measure on the two data sets, respectively.",
"",
"Great writing is rare and highly admired. Readers seek out articles that are beautifully written, informative and entertaining. Yet information-access technologies lack capabilities for predicting article quality at this level. In this paper we present first experiments on article quality prediction in the science journalism domain. We introduce a corpus of great pieces of science journalism, along with typical articles from the genre. We implement features to capture aspects of great writing, including surprising, visual and emotional content, as well as general features related to discourse organization and sentence structure. We show that the distinction between great and typical articles can be detected fairly accurately, and that the entire spectrum of our features contribute to the distinction.",
"Online content exhibits rich temporal dynamics, and diverse realtime user generated content further intensifies this process. However, temporal patterns by which online content grows and fades over time, and by which different pieces of content compete for attention remain largely unexplored. We study temporal patterns associated with online content and how the content's popularity grows and fades over time. The attention that content receives on the Web varies depending on many factors and occurs on very different time scales and at different resolutions. In order to uncover the temporal dynamics of online content we formulate a time series clustering problem using a similarity metric that is invariant to scaling and shifting. We develop the K-Spectral Centroid (K-SC) clustering algorithm that effectively finds cluster centroids with our similarity measure. By applying an adaptive wavelet-based incremental approach to clustering, we scale K-SC to large data sets. We demonstrate our approach on two massive datasets: a set of 580 million Tweets, and a set of 170 million blog posts and news media articles. We find that K-SC outperforms the K-means clustering algorithm in finding distinct shapes of time series. Our analysis shows that there are six main temporal shapes of attention of online content. We also present a simple model that reliably predicts the shape of attention by using information about only a small number of participants. Our analyses offer insight into common temporal patterns of the content on theWeb and broaden the understanding of the dynamics of human attention.",
"In the current Web 2.0 era, the popularity of Web resources fluctuates ephemerally, based on trends and social interest. As a result, content-based relevance signals are insufficient to meet users' constantly evolving information needs in searching for Web 2.0 items. Incorporating future popularity into ranking is one way to counter this. However, predicting popularity as a third party (as in the case of general search engines) is difficult in practice, due to their limited access to item view histories. To enable popularity prediction externally without excessive crawling, we propose an alternative solution by leveraging user comments, which are more accessible than view counts. Due to the sparsity of comments, traditional solutions that are solely based on view histories do not perform well. To deal with this sparsity, we mine comments to recover additional signal, such as social influence. By modeling comments as a time-aware bipartite graph, we propose a regularization-based ranking algorithm that accounts for temporal, social influence and current popularity factors to predict the future popularity of items. Experimental results on three real-world datasets --- crawled from YouTube, Flickr and Last.fm --- show that our method consistently outperforms competitive baselines in several evaluation tasks.",
""
]
} |
1607.03057 | 2949882214 | In this work, we tackle the problem of predicting entity popularity on Twitter based on the news cycle. We apply a supervised learn- ing approach and extract four types of features: (i) signal, (ii) textual, (iii) sentiment and (iv) semantic, which we use to predict whether the popularity of a given entity will be high or low in the following hours. We run several experiments on six different entities in a dataset of over 150M tweets and 5M news and obtained F1 scores over 0.70. Error analysis indicates that news perform better on predicting entity popularity on Twitter when they are the primary information source of the event, in opposition to events such as live TV broadcasts, political debates or football matches. | Research work more in line with ours, tries to predict the popularity of news articles shares (url sharing) on Twitter based on content features @cite_7 . Authors considered the news source, the article's category, the article's author, the subjectivity of the language in the article, and number of named entities in the article as features. Recently, there was a large study of the life cycle of news articles in terms of distribution of visits, tweets and shares over time across different sections of the publisher @cite_8 . Their work was able to improve, for some content type, the prediction of web visits using data from social media after ten to twenty minutes of publication. | {
"cite_N": [
"@cite_7",
"@cite_8"
],
"mid": [
"2097343308",
"1995358441"
],
"abstract": [
"News articles are extremely time sensitive by nature. There is also intense competition among news items to propagate as widely as possible. Hence, the task of predicting the popularity of news items on the social web is both interesting and challenging. Prior research has dealt with predicting eventual online popularity based on early popularity. It is most desirable, however, to predict the popularity of items prior to their release, fostering the possibility of appropriate decision making to modify an article and the manner of its publication. In this paper, we construct a multi-dimensional feature space derived from properties of an article and evaluate the efficacy of these features to serve as predictors of online popularity. We examine both regression and classification algorithms and demonstrate that despite randomness in human behavior, it is possible to predict ranges of popularity on twitter with an overall 84 accuracy. Our study also serves to illustrate the differences between traditionally prominent sources and those immensely popular on the social web.",
"This paper presents a study of the life cycle of news articles posted online. We describe the interplay between website visitation patterns and social media reactions to news content. We show that we can use this hybrid observation method to characterize distinct classes of articles. We also find that social media reactions can help predict future visitation patterns early and accurately. We validate our methods using qualitative analysis as well as quantitative analysis on data from a large international news network, for a set of articles generating more than 3,000,000 visits and 200,000 social media reactions. We show that it is possible to model accurately the overall traffic articles will ultimately receive by observing the first ten to twenty minutes of social media reactions. Achieving the same prediction accuracy with visits alone would require to wait for three hours of data. We also describe significant improvements on the accuracy of the early prediction of shelf-life for news stories."
]
} |
1607.03050 | 2461198954 | Naive Bayes Nearest Neighbour (NBNN) is a simple and effective framework which addresses many of the pitfalls of K-Nearest Neighbour (KNN) classification. It has yielded competitive results on several computer vision benchmarks. Its central tenet is that during NN search, a query is not compared to every example in a database, ignoring class information. Instead, NN searches are performed within each class, generating a score per class. A key problem with NN techniques, including NBNN, is that they fail when the data representation does not capture perceptual (e.g. class-based) similarity. NBNN circumvents this by using independent engineered descriptors (e.g. SIFT). To extend its applicability outside of image-based domains, we propose to learn a metric which captures perceptual similarity. Similar to how Neighbourhood Components Analysis optimizes a differentiable form of KNN classification, we propose "Class Conditional" metric learning (CCML), which optimizes a soft form of the NBNN selection rule. Typical metric learning algorithms learn either a global or local metric. However, our proposed method can be adjusted to a particular level of locality by tuning a single parameter. An empirical evaluation on classification and retrieval tasks demonstrates that our proposed method clearly outperforms existing learned distance metrics across a variety of image and non-image datasets. | Salakhutdinov and Hinton extended NCA to the non-linear setting by incorporating a multi-layer neural network into the embedding and training it with backpropagation @cite_18 . They incorporated unsupervised pre-training @cite_31 into their formulation and regularized with an auto-encoder objective. In a similar manner, we demonstrate the extension of our method to the non-linear setting by utilizing an embedding based on convolutional neural networks. | {
"cite_N": [
"@cite_18",
"@cite_31"
],
"mid": [
"205159212",
"2136922672"
],
"abstract": [
"A dental model trimmer having an easily replaceable abrasive surfaced member. The abrasive surfaced member is contained within a housing and is releasably coupled onto a back plate assembly which is driven by a drive motor. The housing includes a releasably coupled cover plate providing access to the abrasive surfaced member. An opening formed in the cover plate exposes a portion of the abrasive surface so that a dental model workpiece can be inserted into the opening against the abrasive surface to permit work on the dental model workpiece. A tilting work table beneath the opening supports the workpiece during the operation. A stream of water is directed through the front cover onto the abrasive surface and is redirected against this surface by means of baffles positioned inside the cover plate. The opening includes a beveled boundary and an inwardly directed lip permitting angular manipulation of the workpiece, better visibility of the workpiece and maximum safety.",
"We show how to use \"complementary priors\" to eliminate the explaining-away effects that make inference difficult in densely connected belief nets that have many hidden layers. Using complementary priors, we derive a fast, greedy algorithm that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory. The fast, greedy algorithm is used to initialize a slower learning procedure that fine-tunes the weights using a contrastive version of the wake-sleep algorithm. After fine-tuning, a network with three hidden layers forms a very good generative model of the joint distribution of handwritten digit images and their labels. This generative model gives better digit classification than the best discriminative learning algorithms. The low-dimensional manifolds on which the digits lie are modeled by long ravines in the free-energy landscape of the top-level associative memory, and it is easy to explore these ravines by using the directed connections to display what the associative memory has in mind."
]
} |
1607.03050 | 2461198954 | Naive Bayes Nearest Neighbour (NBNN) is a simple and effective framework which addresses many of the pitfalls of K-Nearest Neighbour (KNN) classification. It has yielded competitive results on several computer vision benchmarks. Its central tenet is that during NN search, a query is not compared to every example in a database, ignoring class information. Instead, NN searches are performed within each class, generating a score per class. A key problem with NN techniques, including NBNN, is that they fail when the data representation does not capture perceptual (e.g. class-based) similarity. NBNN circumvents this by using independent engineered descriptors (e.g. SIFT). To extend its applicability outside of image-based domains, we propose to learn a metric which captures perceptual similarity. Similar to how Neighbourhood Components Analysis optimizes a differentiable form of KNN classification, we propose "Class Conditional" metric learning (CCML), which optimizes a soft form of the NBNN selection rule. Typical metric learning algorithms learn either a global or local metric. However, our proposed method can be adjusted to a particular level of locality by tuning a single parameter. An empirical evaluation on classification and retrieval tasks demonstrates that our proposed method clearly outperforms existing learned distance metrics across a variety of image and non-image datasets. | One drawback of the NCA criterion is that it only finds the best metric for a 1-NN classifier since it stochastically selects a single neighbour per point. address this issue by extending to stochastic @math -neighbourhood selection @cite_32 . Though this method is theoretically sound, in practice, optimization is complicated. Moreover, the technique is memory-intensive in the case of large datasets, and as a result, it is not as straightforward to extend to non-linear mappings. | {
"cite_N": [
"@cite_32"
],
"mid": [
"2161627023"
],
"abstract": [
"Neighborhood Components Analysis (NCA) is a popular method for learning a distance metric to be used within a k-nearest neighbors (kNN) classifier. A key assumption built into the model is that each point stochastically selects a single neighbor, which makes the model well-justified only for kNN with k = 1. However, kNN classifiers with k > 1 are more robust and usually preferred in practice. Here we present kNCA, which generalizes NCA by learning distance metrics that are appropriate for kNN with arbitrary k. The main technical contribution is showing how to efficiently compute and optimize the expected accuracy of a kNN classifier. We apply similar ideas in an unsupervised setting to yield kSNE and kt-SNE, generalizations of Stochastic Neighbor Embedding (SNE, t-SNE) that operate on neighborhoods of size k, which provide an axis of control over embeddings that allow for more homogeneous and interpretable regions. Empirically, we show that kNCA often improves classification accuracy over state of the art methods, produces qualitative differences in the embeddings as k is varied, and is more robust with respect to label noise."
]
} |
1607.03050 | 2461198954 | Naive Bayes Nearest Neighbour (NBNN) is a simple and effective framework which addresses many of the pitfalls of K-Nearest Neighbour (KNN) classification. It has yielded competitive results on several computer vision benchmarks. Its central tenet is that during NN search, a query is not compared to every example in a database, ignoring class information. Instead, NN searches are performed within each class, generating a score per class. A key problem with NN techniques, including NBNN, is that they fail when the data representation does not capture perceptual (e.g. class-based) similarity. NBNN circumvents this by using independent engineered descriptors (e.g. SIFT). To extend its applicability outside of image-based domains, we propose to learn a metric which captures perceptual similarity. Similar to how Neighbourhood Components Analysis optimizes a differentiable form of KNN classification, we propose "Class Conditional" metric learning (CCML), which optimizes a soft form of the NBNN selection rule. Typical metric learning algorithms learn either a global or local metric. However, our proposed method can be adjusted to a particular level of locality by tuning a single parameter. An empirical evaluation on classification and retrieval tasks demonstrates that our proposed method clearly outperforms existing learned distance metrics across a variety of image and non-image datasets. | Another well-known metric learning algorithm is Large Margin Nearest Neighbour (LMNN) @cite_22 , which also learns a Mahalanobis distance for @math -NN by maximizing a hinge loss function. The optimization is formulated as a semi-definite programming problem, which is convex. However, it does not permit non-linear embeddings as in @cite_18 or the present work. | {
"cite_N": [
"@cite_18",
"@cite_22"
],
"mid": [
"205159212",
"2106053110"
],
"abstract": [
"A dental model trimmer having an easily replaceable abrasive surfaced member. The abrasive surfaced member is contained within a housing and is releasably coupled onto a back plate assembly which is driven by a drive motor. The housing includes a releasably coupled cover plate providing access to the abrasive surfaced member. An opening formed in the cover plate exposes a portion of the abrasive surface so that a dental model workpiece can be inserted into the opening against the abrasive surface to permit work on the dental model workpiece. A tilting work table beneath the opening supports the workpiece during the operation. A stream of water is directed through the front cover onto the abrasive surface and is redirected against this surface by means of baffles positioned inside the cover plate. The opening includes a beveled boundary and an inwardly directed lip permitting angular manipulation of the workpiece, better visibility of the workpiece and maximum safety.",
"The accuracy of k-nearest neighbor (kNN) classification depends significantly on the metric used to compute distances between different examples. In this paper, we show how to learn a Mahalanobis distance metric for kNN classification from labeled examples. The Mahalanobis metric can equivalently be viewed as a global linear transformation of the input space that precedes kNN classification using Euclidean distances. In our approach, the metric is trained with the goal that the k-nearest neighbors always belong to the same class while examples from different classes are separated by a large margin. As in support vector machines (SVMs), the margin criterion leads to a convex optimization based on the hinge loss. Unlike learning in SVMs, however, our approach requires no modification or extension for problems in multiway (as opposed to binary) classification. In our framework, the Mahalanobis distance metric is obtained as the solution to a semidefinite program. On several data sets of varying size and difficulty, we find that metrics trained in this way lead to significant improvements in kNN classification. Sometimes these results can be further improved by clustering the training examples and learning an individual metric within each cluster. We show how to learn and combine these local metrics in a globally integrated manner."
]
} |
1607.02737 | 2558718173 | A human action can be seen as transitions between one's body poses over time, where the transition depicts a temporal relation between two poses. Recognizing actions thus involves learning a classifier sensitive to these pose transitions as well as to static poses. In this paper, we introduce a novel method called transitions forests, an ensemble of decision trees that both learn to discriminate static poses and transitions between pairs of two independent frames. During training, node splitting is driven by alternating two criteria: the standard classification objective that maximizes the discrimination power in individual frames, and the proposed one in pairwise frame transitions. Growing the trees tends to group frames that have similar associated transitions and share same action label incorporating temporal information that was not available otherwise. Unlike conventional decision trees where the best split in a node is determined independently of other nodes, the transition forests try to find the best split of nodes jointly (within a layer) for incorporating distant node transitions. When inferring the class label of a new frame, it is passed down the trees and the prediction is made based on previous frame predictions and the current one in an efficient and online manner. We apply our method on varied skeleton action recognition and online detection datasets showing its suitability over several baselines and state-of-the-art approaches. | Generative models such as Hidden Markov Models (HMM) have been proposed with the disadvantages of being difficult to estimate model parameters and time consuming learning and inference stages. Discriminative approaches have been widely adopted due to their superior performance and efficiency. For instance, extracts local features from body joints captures temporal dynamics using Fourier Temporal Pyramids (FTP), further classifying the sequence using Support Vector Machines (SVM). Similarly, represents the whole skeletons as points in a Lie group before temporally aligning sequences using Dynamic Time Warping (DTW) and applying FTP. proposes a Moving Pose descriptor (MP) using both pose and atomic motion information and then temporally mining key frames using a k-NN aproach in contrast to that uses DTW. Using key frames or key motion units has been also studied by @cite_22 @cite_11 @cite_13 showing good performance revealing that static information is important to recognize actions. Recently, deep models using Recurrent Neural Networks (RNN) and Long-Short Term Memory (LSTM) @cite_2 @cite_16 have been proposed to model temporal dependencies, but showed inferior performance than recent (offline) models that explicitly exploit static information @cite_11 @cite_25 or well-suited time-series mining . Our forest learns bost static per-frame and temporal information in a discriminative way. | {
"cite_N": [
"@cite_22",
"@cite_2",
"@cite_16",
"@cite_13",
"@cite_25",
"@cite_11"
],
"mid": [
"2155042682",
"2951208315",
"2952587893",
"2275713145",
"2519758495",
"2143267104"
],
"abstract": [
"Recognizing human actions or analyzing human behaviors from 3D videos is an important problem currently investigated in many research domains. The high complexity of human motions and the variability of gesture combinations make this task challenging. Local (over time) analysis of a sequence is often necessary in order to have a more accurate and thorough understanding of what the human is doing. In this paper, we propose a method based on the combination of pose-based and segment-based approaches in order to segment an action sequence into motion units (MUs). We jointly analyze the shape of the human pose and the shape of its motion using a shape analysis framework that represents and compares shapes in a Riemannian manifold. On one hand, this allows us to detect periodic MUs and thus perform action segmentation. On another hand, we can remove repetitions of gestures in order to handle with failure cases for the task of action recognition. Experiments are performed on three representative datasets for the task of action segmentation and action recognition. Competitive results with state-of-the-art methods are obtained in both the tasks.",
"The long short-term memory (LSTM) neural network is capable of processing complex sequential information since it utilizes special gating schemes for learning representations from long input sequences. It has the potential to model any sequential time-series data, where the current hidden state has to be considered in the context of the past hidden states. This property makes LSTM an ideal choice to learn the complex dynamics of various actions. Unfortunately, the conventional LSTMs do not consider the impact of spatio-temporal dynamics corresponding to the given salient motion patterns, when they gate the information that ought to be memorized through time. To address this problem, we propose a differential gating scheme for the LSTM neural network, which emphasizes on the change in information gain caused by the salient motions between the successive frames. This change in information gain is quantified by Derivative of States (DoS), and thus the proposed LSTM model is termed as differential Recurrent Neural Network (dRNN). We demonstrate the effectiveness of the proposed model by automatically recognizing actions from the real-world 2D and 3D human action datasets. Our study is one of the first works towards demonstrating the potential of learning complex time-series representations via high-order derivatives of states.",
"Skeleton based action recognition distinguishes human actions using the trajectories of skeleton joints, which provide a very good representation for describing actions. Considering that recurrent neural networks (RNNs) with Long Short-Term Memory (LSTM) can learn feature representations and model long-term temporal dependencies automatically, we propose an end-to-end fully connected deep LSTM network for skeleton based action recognition. Inspired by the observation that the co-occurrences of the joints intrinsically characterize human actions, we take the skeleton as the input at each time slot and introduce a novel regularization scheme to learn the co-occurrence features of skeleton joints. To train the deep LSTM network effectively, we propose a new dropout algorithm which simultaneously operates on the gates, cells, and output responses of the LSTM neurons. Experimental results on three human action recognition datasets consistently demonstrate the effectiveness of the proposed model.",
"Taking fully into consideration the fact that one human action can be intuitively considered as a sequence of key poses and atomic motions in a particular order, a human action recognition method using multi-layer codebooks of key poses and atomic motions is proposed in this paper. Inspired by the dynamics models of human joints, normalized relative orientations are computed as features for each limb of human body. In order to extract key poses and atomic motions precisely, feature sequences are segmented into pose feature segments and motion feature segments dynamically, based on the potential differences of feature sequences. Multi-layer codebooks of each human action are constructed with the key poses extracted from pose feature segments and the atomic motions extracted from motion feature segments associated with each two key poses. The multi-layer codebooks represent action patterns of each human action, which can be used to recognize human actions with the proposed pattern-matching method. Three classification methods are employed for action recognition based on the multi-layer codebooks. Two public action datasets, i.e., CAD-60 and MSRC-12 datasets, are used to demonstrate the advantages of the proposed method. The experimental results show that the proposed method can obtain a comparable or better performance compared with the state-of-the-art methods. Human actions are modeled by a sequence of key poses and atomic motions.Normalized relative orientations are computed as features for each limb.Feature sequences are segmented into pose and motion feature segments dynamically.Multi-layer codebooks which constructed with extracted key poses and atomic motions.A pattern-matching method is proposed and integrated with traditional classifiers.",
"Most of existing skeleton-based representations for action recognition can not effectively capture the spatio-temporal motion characteristics of joints and are not robust enough to noise from depth sensors and estimation errors of joints. In this paper, we propose a novel low-level representation for the motion of each joint through tracking its trajectory and segmenting it into several semantic parts called motionlets. During this process, the disturbance of noise is reduced by trajectory fitting, sampling and segmentation. Then we construct an undirected complete labeled graph to represent a video by combining these motionlets and their spatio-temporal correlations. Furthermore, a new graph kernel called subgraph-pattern graph kernel (SPGK) is proposed to measure the similarity between graphs. Finally, the SPGK is directly used as the kernel of SVM to classify videos. In order to evaluate our method, we perform a series of experiments on several public datasets and our approach achieves a comparable performance to the state-of-the-art approaches.",
"Human action recognition is an important yet challenging task. The recently developed commodity depth sensors open up new possibilities of dealing with this problem but also present some unique challenges. The depth maps captured by the depth cameras are very noisy and the 3D positions of the tracked joints may be completely wrong if serious occlusions occur, which increases the intra-class variations in the actions. In this paper, an actionlet ensemble model is learnt to represent each action and to capture the intra-class variance. In addition, novel features that are suitable for depth data are proposed. They are robust to noise, invariant to translational and temporal misalignments, and capable of characterizing both the human motion and the human-object interactions. The proposed approach is evaluated on two challenging action recognition datasets captured by commodity depth cameras, and another dataset captured by a MoCap system. The experimental evaluations show that the proposed approach achieves superior performance to the state of the art algorithms."
]
} |
1607.02630 | 2470193492 | Selective inference is a recent research topic that tries to perform valid inference after using the data to select a reasonable statistical model. We propose MAGIC, a new method for selective inference that is general, powerful and tractable. MAGIC is a method for selective inference after solving a convex optimization problem with smooth loss and @math penalty. Randomization is incorporated into the optimization problem to boost statistical power. Through reparametrization, MAGIC reduces the problem into a sampling problem with simple constraints. MAGIC applies to many @math penalized optimization problem including the Lasso, logistic Lasso and neighborhood selection in graphical models, all of which we consider in this paper. | In the post selection literature, the authors in @cite_3 proposed the PoSI approach, which reduce the problem to a simultaneous inference problem. Because of the simultaneity, it prevents data snooping from any selection procedure, but also results in more conservative inference. In addition, the PoSI method has extremely high computational cost, and is only applicable when the dimension @math or for very sparse models. The authors proposed a method for computing p-values that controls false discovery rate (FDR) among all variables. This is quite different from the hypothesis testing framework of this work, as the hypotheses tested in selective inference are chosen as a function of the data. Hence, the hypotheses tested are not directly comparable. Furthermore, compared with @cite_7 , MAGIC has the advantage of being able to construct confidence intervals for the selected variables. | {
"cite_N": [
"@cite_7",
"@cite_3"
],
"mid": [
"2950190315",
"2009462809"
],
"abstract": [
"High-dimensional statistical inference deals with models in which the the number of parameters p is comparable to or larger than the sample size n. Since it is usually impossible to obtain consistent procedures unless @math , a line of recent work has studied models with various types of low-dimensional structure, including sparse vectors, sparse and structured matrices, low-rank matrices and combinations thereof. In such settings, a general approach to estimation is to solve a regularized optimization problem, which combines a loss function measuring how well the model fits the data with some regularization function that encourages the assumed structure. This paper provides a unified framework for establishing consistency and convergence rates for such regularized M-estimators under high-dimensional scaling. We state one main theorem and show how it can be used to re-derive some existing results, and also to obtain a number of new results on consistency and convergence rates, in both @math -error and related norms. Our analysis also identifies two key properties of loss and regularization functions, referred to as restricted strong convexity and decomposability, that ensure corresponding regularized M-estimators have fast convergence rates and which are optimal in many well-studied cases.",
"It is common practice in statistical data analysis to perform data-driven variable selection and derive statistical inference from the resulting model. Such inference enjoys none of the guarantees that classical statistical theory provides for tests and confidence intervals when the model has been chosen a priori. We propose to produce valid post-selection inference'' by reducing the problem to one of simultaneous inference and hence suitably widening conventional confidence and retention intervals. Simultaneity is required for all linear functions that arise as coefficient estimates in all submodels. By purchasing simultaneity insurance'' for all possible submodels, the resulting post-selection inference is rendered universally valid under all possible model selection procedures. This inference is therefore generally conservative for particular selection procedures, but it is always less conservative than full Scheffe protection. Importantly it does not depend on the truth of the selected submodel, and hence it produces valid inference even in wrong models. We describe the structure of the simultaneous inference problem and give some asymptotic results."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | Many researchers have investigated automatic license plate recognition and its subtasks. Since this work focuses on the license plate character segmentation (LPCS) and on evaluation datasets, this section briefly reviews related works. We refer the reader to works such as @cite_20 @cite_40 for further information on the ALPR problem. The remaining of this section focuses on two aspects. First, we review works related to techniques of character segmentation. Then, we present works that propose character segmentation evaluation datasets used in different contexts. Finally, we present the works describing the techniques used as baselines in this work | {
"cite_N": [
"@cite_40",
"@cite_20"
],
"mid": [
"2102608210",
"2135449683"
],
"abstract": [
"Automatic license plate recognition (ALPR) is the extraction of vehicle license plate information from an image or a sequence of images. The extracted information can be used with or without a database in many applications, such as electronic payment systems (toll payment, parking fee payment), and freeway and arterial monitoring systems for traffic surveillance. The ALPR uses either a color, black and white, or infrared camera to take images. The quality of the acquired images is a major factor in the success of the ALPR. ALPR as a real-life application has to quickly and successfully process license plates under different environmental conditions, such as indoors, outdoors, day or night time. It should also be generalized to process license plates from different nations, provinces, or states. These plates usually contain different colors, are written in different languages, and use different fonts; some plates may have a single color background and others have background images. The license plates can be partially occluded by dirt, lighting, and towing accessories on the car. In this paper, we present a comprehensive review of the state-of-the-art techniques for ALPR. We categorize different ALPR techniques according to the features they used for each stage, and compare them in terms of pros, cons, recognition accuracy, and processing speed. Future forecasts of ALPR are given at the end.",
"License plate recognition (LPR) algorithms in images or videos are generally composed of the following three processing steps: 1) extraction of a license plate region; 2) segmentation of the plate characters; and 3) recognition of each character. This task is quite challenging due to the diversity of plate formats and the nonuniform outdoor illumination conditions during image acquisition. Therefore, most approaches work only under restricted conditions such as fixed illumination, limited vehicle speed, designated routes, and stationary backgrounds. Numerous techniques have been developed for LPR in still images or video sequences, and the purpose of this paper is to categorize and assess them. Issues such as processing time, computational power, and recognition rate are also addressed, when available. Finally, this paper offers to researchers a link to a public image database to define a common reference point for LPR algorithmic assessment."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | Besides license plates, there are works that propose character segmentation on various contexts. Some of them focus on handwritten text segmentation, such as in @cite_22 that proposed two methods using non-linear clustering methods, and in @cite_19 , that uses seven convolutional networks executing in GPUs (Graphics Processing Unit) for this purpose. In @cite_23 and Neumann & Matas @cite_38 , the authors propose character segmentation methods to handle digital documents and real scenes. The main goal of those works is to present an approach to segment characters on other contexts. However, they do not present promising effectiveness on license plate segmentation because they do not explore the contextual information found in licenses plates. | {
"cite_N": [
"@cite_19",
"@cite_22",
"@cite_23",
"@cite_38"
],
"mid": [
"2033154814",
"2033088582",
"2024603080",
"2061802763"
],
"abstract": [
"In 2010, after many years of stagnation, the MNIST handwriting recognition benchmark record dropped from 0.40 error rate to 0.35 . Here we report 0.27 for a committee of seven deep CNNs trained on graphics cards, narrowing the gap to human performance. We also apply the same architecture to NIST SD 19, a more challenging dataset including lower and upper case letters. A committee of seven CNNs obtains the best results published so far for both NIST digits and NIST letters. The robustness of our method is verified by analyzing 78125 different 7-net committees.",
"In handwritten character recognition, it is a significant step to segment a text line into characters. The unsupervised clustering is a common approach for this task. However, due to the strong overlapping and touch among characters, the separation boundaries between two characters are usually nonlinear, which leads to the failure of the widely used clustering methods such as k-means. To tackle this problem, this paper proposes a new handwritten character segmentation method based on nonlinear clustering methods. In the proposed approach, we first segment the entire text line into strokes, the similarity matrix of which is computed according to stroke gravities. Then, the nonlinear clustering methods are performed on this similarity matrix to obtain cluster labels for these strokes. According to the obtained cluster labels, the strokes are combined to form characters. In this paper, we consider two nonlinear clustering methods, namely, spectral clustering based on Normalized cut (Ncut) and kernel clustering based on Conscience On-Line Learning (COLL). Whereby, two segmentation approaches are proposed with the one using Ncut termed SegNcut, and the one using COLL termed SegCOLL. Experiments on four databases are conducted to demonstrate the effectiveness of our SegNcut and SegCOLL approaches.",
"The touching character segmentation problem becomes complex when touching strings are multi-oriented. Moreover in graphical documents sometimes characters in a single-touching string have different orientations. Segmentation of such complex touching is more challenging. In this paper, we present a scheme towards the segmentation of English multi-oriented touching strings into individual characters. When two or more characters touch, they generate a big cavity region in the background portion. Based on the convex hull information, at first, we use this background information to find some initial points for segmentation of a touching string into possible primitives (a primitive consists of a single character or part of a character). Next, the primitives are merged to get optimum segmentation. A dynamic programming algorithm is applied for this purpose using the total likelihood of characters as the objective function. A SVM classifier is used to find the likelihood of a character. To consider multi-oriented touching strings the features used in the SVM are invariant to character orientation. Experiments were performed in different databases of real and synthetic touching characters and the results show that the method is efficient in segmenting touching characters of arbitrary orientations and sizes.",
"An end-to-end real-time scene text localization and recognition method is presented. The real-time performance is achieved by posing the character detection problem as an efficient sequential selection from the set of Extremal Regions (ERs). The ER detector is robust to blur, illumination, color and texture variation and handles low-contrast text. In the first classification stage, the probability of each ER being a character is estimated using novel features calculated with O(1) complexity per region tested. Only ERs with locally maximal probability are selected for the second stage, where the classification is improved using more computationally expensive features. A highly efficient exhaustive search with feedback loops is then applied to group ERs into words and to select the most probable character segmentation. Finally, text is recognized in an OCR stage trained using synthetic fonts. The method was evaluated on two public datasets. On the ICDAR 2011 dataset, the method achieves state-of-the-art text localization results amongst published methods and it is the first one to report results for end-to-end text recognition. On the more challenging Street View Text dataset, the method achieves state-of-the-art recall. The robustness of the proposed method against noise and low contrast of characters is demonstrated by “false positives” caused by detected watermark text in the dataset."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | The LPCS can be seen as a challenging task that must be performed by ALPR systems once the acquisition of plate images usually is affected by problems such as skew, shadows, perspective projection and blurring. In an attempt to reduce that, the majority of works segment characters in manually cropped images to evaluate the effectiveness of the recognition methods. However, the license plates must be cropped automatically when applied on real applications, tackling the aforementioned problems @cite_26 @cite_25 . | {
"cite_N": [
"@cite_26",
"@cite_25"
],
"mid": [
"1965661050",
"1540278691"
],
"abstract": [
"In this paper, a novel algorithm is proposed for segmentation of touching characters on the license plate. In our method, rough and precise segmentation of characters proceeds in sequence. At first, characters on the license plate are roughly classified as segmented and touching characters by vertical projection on the edge map. Next, segmentation points of touching characters are evaluated using fixed ratio relations of width, interval and height of characters. Finally, the touching characters are segmented by a path created by A* pathfinding algorithm. The proposed method is tested on 238 license plate images and the successful segmentation rate of 97.06 is achieved with only 62ms. And the experiment result demonstrates the effectiveness and efficient of our suggestion.",
"This paper presents a novel algorithm for license plate detection and license plate character segmentation problems by using the Gabor transform in detection and local vector quantization in segmentation. As of our knowledge this is the first application of Gabor filters to license plate segmentation problem. Even though much of the research efforts are devoted to the edge or global thresholding-based approaches, it is more practical and efficient to analyze the image in certain directions and scales utilizing the Gabor transform instead of error-prone edge detection or thresholding. Gabor filter response only gives a rough estimate of the plate boundary. Then binary split tree is used for vector quantization in order to extract the exact boundary and segment the plate region into disjoint characters which become ready for the optical character recognition."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | @cite_40 classify the license plate character segmentation techniques into five main categories: based on pixel connectivity, pixel projection, prior knowledge of the characters, characters contours and based on the combination of these features. | {
"cite_N": [
"@cite_40"
],
"mid": [
"2102608210"
],
"abstract": [
"Automatic license plate recognition (ALPR) is the extraction of vehicle license plate information from an image or a sequence of images. The extracted information can be used with or without a database in many applications, such as electronic payment systems (toll payment, parking fee payment), and freeway and arterial monitoring systems for traffic surveillance. The ALPR uses either a color, black and white, or infrared camera to take images. The quality of the acquired images is a major factor in the success of the ALPR. ALPR as a real-life application has to quickly and successfully process license plates under different environmental conditions, such as indoors, outdoors, day or night time. It should also be generalized to process license plates from different nations, provinces, or states. These plates usually contain different colors, are written in different languages, and use different fonts; some plates may have a single color background and others have background images. The license plates can be partially occluded by dirt, lighting, and towing accessories on the car. In this paper, we present a comprehensive review of the state-of-the-art techniques for ALPR. We categorize different ALPR techniques according to the features they used for each stage, and compare them in terms of pros, cons, recognition accuracy, and processing speed. Future forecasts of ALPR are given at the end."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | Many works employ approaches based on Connected Component Analysis (CCA) and pixel projection techniques to tackle the LPCS problem. A study of the effectiveness of the CCA technique concluded that vertical projection can segment characters effectively @cite_33 . The segmentation approach proposed by Shapiro & Gluhchev @cite_18 utilizes an adaptive iterative thresholding approach to binarize the image and then segment the plate characters by employing a connected component analysis. @cite_3 proposed a method that uses ten samples of the same plate, binarize them, select the best one and then segment it using vertical projections. The work described in @cite_9 segments characters counting the black pixels in the horizontal and vertical direction of each license plate region. @cite_7 proposed a technique to segment the characters using CCA and evaluate it in databases manually and automatically cropped. | {
"cite_N": [
"@cite_18",
"@cite_33",
"@cite_7",
"@cite_9",
"@cite_3"
],
"mid": [
"1991882200",
"2170386550",
"",
"2183272078",
"2142922789"
],
"abstract": [
"Image-based car license plate recognition (CLPR) systems provide an inexpensive automatic solution for remote vehicle identification. Localization stage of the CLPR yields a gray-scale plate clip with printed characters. This paper describes the method of plate clip segmentation into isolated characters, feature extraction and classification. The method is independent on character size, thickness, illumination and is capable of handling plates from various countries. The method uses extensively the gray-scale information and is robust to breaks in character connectivity. It is tolerant to character deformations, such as shear and skew. Promising results have been obtained on Israeli and Bulgarian plates.",
"With the development of economy, more and more people can afford private cars. However, the traffic situation becomes more and more terrible because of the increase of cars in the city. To a large extent, Intelligent transport system (ITS) is proved to resolve this problem. License plate automatic recognition (LPR) is the key part of the ITS. Plate location, character segmentation and character identification are the three parts of LPR. Character segmentation is one of the important technologies of the License Plate Recognition. The paper mainly introduces the character segmentation algorithm based on vertical projection. We firstly analyses the principle of this method, and then verify the algorithm according to simulation in Matlab. The result of simulation shows that the algorithm can divide the character of plate effectively.",
"",
"License Plate Detection and Recognition System is an image processing technique used to identify a vehicle by its license plate. Here we propose an accurate and robust method of license plate detection and recognition from an image using contour analysis. The system is composed of two phases: the detection of the license plate, and the character recognition. The license plate detection is performed for obtaining the candidate region of the vehicle license plate and determined using the edge based text detection technique. In the recognition phase, the contour analysis is used to recognize the characters after segmenting each character. The performance of the proposed system has been tested on various images and provides better results.",
"Intelligent Traffic Systems (ITS) is an integral component of modern day road transportation networks. Identification of vehicles is one of the most important challenges to be addressed in the design of any intelligent traffic system. License plate being a unique identity for any registered vehicle, License Plate Recognition (LPR) systems have been used as the means to resolve the issue of identification of vehicles. Intelligent systems involving LPR has been widely applied in several applications such as intelligent traffic rule enforcement, prevention of car theft, intelligent traffic emergency accident handling, monitoring of vehicle traffic and flow control, intelligent parking, automated toll payment, intelligent surveillance and security enforcement etc. A typical LPR process consists of three stages viz; License Plate Extraction, Character Segmentation and Character Recognition. This paper address the Character Segmentation problem. The character segmentation method using horizontal and vertical projection and with dynaic thresholding is proposed in this paper. From the experimental studies made, it is learned that the proposed technique works reasonably well in real world scenarios."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | Some works focus on other techniques to segment the license plates characters. For instance, @cite_25 employed a Gabor transform and vector quantization to LPCS. In Xing-lin & Yun-lou @cite_4 , the authors proposed a technique to segment the characters using prior knowledge regarding the shape and the license plate font considering English (Latin) and Chinese characters. In addition, there are works that employ additional techniques to improve the quality of the results. For instance, @cite_26 use two segmentation techniques in sequence and @cite_13 applies super-resolution techniques. | {
"cite_N": [
"@cite_13",
"@cite_26",
"@cite_4",
"@cite_25"
],
"mid": [
"1984798566",
"1965661050",
"2356061219",
"1540278691"
],
"abstract": [
"",
"In this paper, a novel algorithm is proposed for segmentation of touching characters on the license plate. In our method, rough and precise segmentation of characters proceeds in sequence. At first, characters on the license plate are roughly classified as segmented and touching characters by vertical projection on the edge map. Next, segmentation points of touching characters are evaluated using fixed ratio relations of width, interval and height of characters. Finally, the touching characters are segmented by a path created by A* pathfinding algorithm. The proposed method is tested on 238 license plate images and the successful segmentation rate of 97.06 is achieved with only 62ms. And the experiment result demonstrates the effectiveness and efficient of our suggestion.",
"The license plate character segmentation is one of the three key technologies of the automatic license plate recognition system and the character segmentation is the foundation of the character recognition.The large amount of calculations and a long time for processing are the drawbacks of traditional character segmentation algorithm based on the connected domain character segmentation algorithm,so this paper presents an improved algorithm,which makes full use of priori knowledge for character initial segmentation,and then realizes complete character segmentation based on the connected domain,meanwhile,this paper improves the traditional iteration binaryzation threshold algorithm by the priori knowledge,reducing iteration times.The experimental results show that the character segmentation algorithm provided by this paper greatly cuts down the processing time and meets real time requirement under the premise of accurate extraction of license plate character.",
"This paper presents a novel algorithm for license plate detection and license plate character segmentation problems by using the Gabor transform in detection and local vector quantization in segmentation. As of our knowledge this is the first application of Gabor filters to license plate segmentation problem. Even though much of the research efforts are devoted to the edge or global thresholding-based approaches, it is more practical and efficient to analyze the image in certain directions and scales utilizing the Gabor transform instead of error-prone edge detection or thresholding. Gabor filter response only gives a rough estimate of the plate boundary. Then binary split tree is used for vector quantization in order to extract the exact boundary and segment the plate region into disjoint characters which become ready for the optical character recognition."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | Some statistical-based and machine learning approaches also have been employed on LPCS. For instance, while @cite_30 used likelihood maximization to find the best parameters values of the license plate features and its characters and Franc at al. @cite_41 proposed a technique using Hidden Markov Models to create a relationship between the license plate input and the correct segmentation of its characters. Nagare @cite_2 and @cite_11 employed supervised machine learning techniques to aid the character segmentation phase of the ALPR. | {
"cite_N": [
"@cite_30",
"@cite_41",
"@cite_11",
"@cite_2"
],
"mid": [
"2277804901",
"1831353325",
"2123725958",
"2120072352"
],
"abstract": [
"A method determines a license plate layout configuration. The method includes generating at least one model representing a license plate layout configuration. The generating includes segmenting training images each defining a license plate to extract characters and logos from the training images. The segmenting includes calculating values corresponding to parameters of the license plate and features of the characters and logos. The segmenting includes estimating a likelihood function specified by the features using the values. The likelihood function measures deviations between an observed plate and the model. The method includes storing a layout structure and the distributions for each of the at least one model. The method includes receiving as input an observed image including a plate region. The method includes segmenting the plate region and determining a license plate layout configuration of the observed plate by comparing the segmented plate region to the at least one model.",
"We propose a method for segmentation of a line of characters in a noisy low resolution image of a car license plate. The Hidden Markov Chains are used to model a stochastic relation between an input image and a corresponding character segmentation. The segmentation problem is expressed as the maximum a posteriori estimation from a set of admissible segmentations. The proposed method exploits a specific prior knowledge available for the application at hand. Namely, the number of characters is known and its is also known that the characters can be segmented to sectors with equal but unknown width. The efficient algorithm for estimation based on dynamic programming is derived. The proposed method was successfully tested on data from a real life license plate recognition system.",
"License plate localization (LPL) and character segmentation (CS) play key roles in the license plate (LP) recognition system. In this paper, we dedicate ourselves to these two issues. In LPL, histogram equalization is employed to solve the low-contrast and dynamic-range problems; the texture properties, e.g., aspect ratio, and color similarity are used to locate the LP; and the Hough transform is adopted to correct the rotation problem. In CS, the hybrid binarization technique is proposed to effectively segment the characters in the dirt LP. The feedback self-learning procedure is also employed to adjust the parameters in the system. As documented in the experiments, good localization and segmentation results are achieved with the proposed algorithms.",
"Intelligent Transportation System (ITS) has become an integral part of the Transportation Industry these days and it consists of License Plate Recognition (LPR) System. License Plate Recognition is also called Car Plate Recognition (CPR) or Automatic Number Plate Recognition (ANPR) System. In LPR System, when a vehicle steps over magnetic loop detector it senses car and takes image of the car, following image preprocessing operations for improvement in the quality of car image. From this enhanced image, license plate region is recognized and extracted. Then character fragmentation segmentation is performed on extracted License Plate and these segmented characters are recognized using Neural Network in this paper. General Terms Image Processing, Neural Network."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | There are works proposing datasets to evaluate several aspects of text recognition and document analysis. For instance, @cite_39 proposed a dataset to evaluate techniques of document layout analysis. That dataset contains @math images from websites, newspaper pages, magazines pages. The UNIPEN dataset was proposed in @cite_14 and is composed by over @math images of words and handwritten characters. @cite_32 proposed a dataset of real images to evaluate approaches to perform text detection containing 500 real images in various sizes. | {
"cite_N": [
"@cite_14",
"@cite_32",
"@cite_39"
],
"mid": [
"2147510700",
"1972065312",
"2151765755"
],
"abstract": [
"We report the status of the UNIPEN project of data exchange and recognizer benchmarks started two years ago at the initiative of the International Association of Pattern Recognition (Technical Committee 11). The purpose of the project is to propose and implement solutions to the growing need of handwriting samples for online handwriting recognizers used by pen-based computers. Researchers from several companies and universities have agreed on a data format, a platform of data exchange and a protocol for recognizer benchmarks. The online handwriting data of concern may include handprint and cursive from various alphabets (including Latin and Chinese), signatures and pen gestures. These data will be compiled and distributed by the Linguistic Data Consortium. The benchmarks will be arbitrated the US National Institute of Standards and Technologies. We give a brief introduction to the UNIPEN format. We explain the protocol of data exchange and benchmarks.",
"With the increasing popularity of practical vision systems and smart phones, text detection in natural scenes becomes a critical yet challenging task. Most existing methods have focused on detecting horizontal or near-horizontal texts. In this paper, we propose a system which detects texts of arbitrary orientations in natural images. Our algorithm is equipped with a two-level classification scheme and two sets of features specially designed for capturing both the intrinsic characteristics of texts. To better evaluate our algorithm and compare it with other competing algorithms, we generate a new dataset, which includes various texts in diverse real-world scenarios; we also propose a protocol for performance evaluation. Experiments on benchmark datasets and the proposed dataset demonstrate that our algorithm compares favorably with the state-of-the-art algorithms when handling horizontal texts and achieves significantly enhanced performance on texts of arbitrary orientations in complex natural scenes.",
"There is a significant need for a realistic dataset on which to evaluate layout analysis methods and examine their performance in detail. This paper presents a new dataset (and the methodology used to create it) based on a wide range of contemporary documents. Strong emphasis is placed on comprehensive and detailed representation of both complex and simple layouts, and on colour originals. In-depth information is recorded both at the page and region level. Ground truth is efficiently created using a new semi-automated tool and stored in a new comprehensive XML representation, the PAGE format. The dataset can be browsed and searched via a web-based front end to the underlying database and suitable subsets (relevant to specific evaluation goals) can be selected and downloaded."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | There are also datasets to evaluate ALPR approaches. Two UIUC datasets were proposed in Agarwal & Roth @cite_0 and @cite_12 composed of @math cars to single-scale approaches and @math cars to multi-scale approaches, respectively. The Caltech dataset @cite_1 provides @math rear images of cars and is commonly used to vehicle recognition and license plates detection. In @cite_43 , the authors collected @math images of cars from the websites such as Flickr, Google and Bing. The BIT Dataset @cite_42 contains images of @math cars and aims at evaluating techniques to recognize the vehicle type. In addition, there are other datasets @cite_21 @cite_15 designed to evaluate tasks such as vehicle pose estimation and vehicle detection. | {
"cite_N": [
"@cite_42",
"@cite_1",
"@cite_21",
"@cite_0",
"@cite_43",
"@cite_15",
"@cite_12"
],
"mid": [
"2153795334",
"",
"2151973304",
"2160225842",
"780950768",
"2128942651",
"2012330712"
],
"abstract": [
"In this paper, we propose an appearance-based vehicle type classification method from vehicle frontal view images. Unlike other methods using hand-crafted visual features, our method is able to automatically learn good features for vehicle type classification by using a convolutional neural network. In order to capture rich and discriminative information of vehicles, the network is pre-trained by the sparse filtering which is an unsupervised learning method. Besides, the network is with layer-skipping to ensure that final features contain both high-level global and low-level local features. After the final features are obtained, the soft max regression is used to classify vehicle types. We build a challenging vehicle dataset called BIT-Vehicle dataset to evaluate the performance of our method. Experimental results on a public dataset and our own dataset demonstrate that our method is quite effective in classifying vehicle types.",
"",
"We address the problem of identifying specific instances of a class (cars) from a set of images all belonging to that class. Although we cannot build a model for any particular instance (as we may be provided with only one \"training\" example of it), we can use information extracted from observing other members of the class. We pose this task as a learning problem, in which the learner is given image pairs, labeled as matching or not, and must discover which image features are most consistent for matching instances and discriminative for mismatches. We explore a patch based representation, where we model the distributions of similarity measurements defined on the patches. Finally, we describe an algorithm that selects the most salient patches based on a mutual information criterion. This algorithm performs identification well for our challenging dataset of car images, after matching only a few, well chosen patches.",
"We present an approach for learning to detect objects in still gray images, that is based on a sparse, part-based representation of objects. A vocabulary of information-rich object parts is automatically constructed from a set of sample images of the object class of interest. Images are then represented using parts from this vocabulary, along with spatial relations observed among them. Based on this representation, a feature-efficient learning algorithm is used to learn to detect instances of the object class. The framework developed can be applied to any object with distinguishable parts in a relatively fixed spatial configuration. We report experiments on images of side views of cars. Our experiments show that the method achieves high detection accuracy on a difficult test set of real-world images, and is highly robust to partial occlusion and background variation.In addition, we discuss and offer solutions to several methodological issues that are significant for the research community to be able to evaluate object detection approaches.",
"In this work we introduce a large-scale, fine-grained dataset of cars. This dataset, consisting of 197 classes and 16,185 images, represents an order of magnitude increase in size over the only existing fine-grained car dataset [7] (14 classes, 1,904 images) and is comparable in size to the largest fine-grained datasets publicly available [9, 3]. The goals of this work are twofold: 1) to describe the difficulties encountered when collecting such a dataset and 2) to present baseline performance for two state-of-the-art methods.",
"We propose an approach to overcome the two main challenges of 3D multiview object detection and localization: The variation of object features due to changes in the viewpoint and the variation in the size and aspect ratio of the object. Our approach proceeds in three steps. Given an initial bounding box of fixed size, we first refine its aspect ratio and size. We can then predict the viewing angle, under the hypothesis that the bounding box actually contains an object instance. Finally, a classifier tuned to this particular viewpoint checks the existence of an instance. As a result, we can find the object instances and estimate their poses, without having to search over all window sizes and potential orientations. We train and evaluate our method on a new object database specifically tailored for this task, containing real-world objects imaged over a wide range of smoothly varying viewpoints and significant lighting changes. We show that the successive estimations of the bounding box and the viewpoint lead to better localization results.",
"We study the problem of detecting objects in still, gray-scale images. Our primary focus is the development of a learning-based approach to the problem that makes use of a sparse, part-based representation. A vocabulary of distinctive object parts is automatically constructed from a set of sample images of the object class of interest; images are then represented using parts from this vocabulary, together with spatial relations observed among the parts. Based on this representation, a learning algorithm is used to automatically learn to detect instances of the object class in new images. The approach can be applied to any object with distinguishable parts in a relatively fixed spatial configuration; it is evaluated here on difficult sets of real-world images containing side views of cars, and is seen to successfully detect objects in varying conditions amidst background clutter and mild occlusion. In evaluating object detection approaches, several important methodological issues arise that have not been satisfactorily addressed in the previous work. A secondary focus of this paper is to highlight these issues, and to develop rigorous evaluation standards for the object detection problem. A critical evaluation of our approach under the proposed standards is presented."
]
} |
1607.02937 | 2470672561 | Automatic license plate recognition (ALPR) has been the focus of many researches in the past years. In general, ALPR is divided into the following problems: detection of on-track vehicles, license plate detection, segmentation of license plate characters, and optical character recognition (OCR). Even though commercial solutions are available for controlled acquisition conditions, e.g., the entrance of a parking lot, ALPR is still an open problem when dealing with data acquired from uncontrolled environments, such as roads and highways when relying only on imaging sensors. Due to the multiple orientations and scales of the license plates captured by the camera, a very challenging task of the ALPR is the license plate character segmentation (LPCS) step, because its effectiveness is required to be (near) optimal to achieve a high recognition rate by the OCR. To tackle the LPCS problem, this work proposes a benchmark composed of a dataset designed to focus specifically on the character segmentation step of the ALPR within an evaluation protocol. Furthermore, we propose the Jaccard-centroid coefficient, an evaluation measure more suitable than the Jaccard coefficient regarding the location of the bounding box within the ground-truth annotation. The dataset is composed of 2000 Brazilian license plates consisting of 14000 alphanumeric symbols and their corresponding bounding box annotations. We also present a straightforward approach to perform LPCS efficiently. Finally, we provide an experimental evaluation for the dataset based on five LPCS approaches and demonstrate the importance of character segmentation for achieving an accurate OCR. | This subsection describes four LPCS techniques chosen to be our baselines. We considered approaches based on three methods available in the literature, in which the first aims at improving the quality of degraded images of words @cite_16 and counting the blacks pixels of the image (); the second performs segmentation by find connected components in a binarized license plate @cite_18 (); and the third performs a pixel counting as well as the first technique, but utilize a license plate binarized by a method of Iterative Global Threshold (IGT) @cite_27 (. In addition, a simple technique that employs prior knowledge regarding the license plate layout and its number of characters was used as a fourth approach and will be described in . | {
"cite_N": [
"@cite_27",
"@cite_18",
"@cite_16"
],
"mid": [
"2098287947",
"1991882200",
"2036723249"
],
"abstract": [
"In this paper, we present a binarization technique specifically designed for historical document images. Existing methods for this problem focus on either finding a good global threshold or adapting the threshold for each area so that to remove smear, strains, uneven illumination etc. We propose a hybrid approach that first applies a global thresholding method and, then, identifies the image areas that are more likely to still contain noise. Each of these areas is re-processed separately to achieve better quality of binarization. We evaluate the proposed approach for different kinds of degradation problems. The results show that our method can handle hard cases while documents already in good condition are not affected drastically.",
"Image-based car license plate recognition (CLPR) systems provide an inexpensive automatic solution for remote vehicle identification. Localization stage of the CLPR yields a gray-scale plate clip with printed characters. This paper describes the method of plate clip segmentation into isolated characters, feature extraction and classification. The method is independent on character size, thickness, illumination and is capable of handling plates from various countries. The method uses extensively the gray-scale information and is robust to breaks in character connectivity. It is tolerant to character deformations, such as shear and skew. Promising results have been obtained on Israeli and Bulgarian plates.",
"This paper presents a novel preprocessing method based on mathematical morphology techniques to improve the subsequent thresholding quality of raw degraded word images. The raw degraded word images contain undesirable shapes called critical shadows on the background that cause noise in binary images. This noise constitutes obstacles to posterior segmentation of characters. Direct application of a thresholding method produces inadequate binary versions of these degraded word images. Our preprocessing method called Shadow Location and Lightening (SL*L) adaptively, accurately and without manual fine-tuning of parameters locates these critical shadows on grayscale degraded images using morphological operations, and lightens them before applying eventual thresholding process. In this way, enhanced binary images without unpredictable and inappropriate noise can be provided to subsequent segmentation of characters. Then, adequate binary characters can be segmented and extracted as input data to optical character recognition (OCR) applications saving computational effort and increasing recognition rate. The proposed method is experimentally tested with a set of several raw degraded images extracted from real photos acquired by unsophisticated imaging systems. A qualitative analysis of experimental results led to conclusions that the thresholding result quality was significantly improved with the proposed preprocessing method. Also, a quantitative evaluation using a testing data of 1194 degraded word images showed the essentiality and effectiveness of the proposed preprocessing method to increase segmentation and recognition rates of their characters. Furthermore, an advantage of the proposed method is that Otsu's method as a simple and easily implementable global thresholding technique can be sufficient to reducing computational load."
]
} |
1607.02646 | 2481170768 | Efficient processing of large-scale graphs in distributed environments has been an increasingly popular topic of research in recent years. Inter-connected data that can be modeled as graphs appear in application domains such as machine learning, recommendation, web search, and social network analysis. Writing distributed graph applications is inherently hard and requires programming models that can cover a diverse set of problems, including iterative refinement algorithms, graph transformations, graph aggregations, pattern matching, ego-network analysis, and graph traversals. Several high-level programming abstractions have been proposed and adopted by distributed graph processing systems and big data platforms. Even though significant work has been done to experimentally compare distributed graph processing frameworks, no qualitative study and comparison of graph programming abstractions has been conducted yet. In this survey, we review and analyze the most prevalent high-level programming models for distributed graph processing, in terms of their semantics and applicability. We review 34 distributed graph processing systems with respect to the graph processing models they implement and we survey applications that appear in recent distributed graph systems papers. Finally, we discuss trends and open research questions in the area of distributed graph processing. | The work that is closest to ours is a recent survey of vertex-centric frameworks for graph processing @cite_56 . It presents a extensive study of frameworks that implement the vertex-centric programming model and compares them in terms of system design characteristics, such as scheduling, partitioning, fault-tolerance, and scalability. Moreover, it briefly introduces subgraph-centric frameworks, as an optimization to vertex-centric implementations. While there is some overlap between this work and ours, the objective of our study is to present the first comprehensive comparison of distributed graph processing abstractions, regardless of the specifics of their implementations. While in @cite_56 the discussion revolves around certain frameworks, in our work, we first consider the programming models decoupled from the systems, then build a taxonomy of systems based on their implemented model. | {
"cite_N": [
"@cite_56"
],
"mid": [
"2080098453"
],
"abstract": [
"The vertex-centric programming model is an established computational paradigm recently incorporated into distributed processing frameworks to address challenges in large-scale graph processing. Billion-node graphs that exceed the memory capacity of commodity machines are not well supported by popular Big Data tools like MapReduce, which are notoriously poor performing for iterative graph algorithms such as PageRank. In response, a new type of framework challenges one to “think like a vertex” (TLAV) and implements user-defined programs from the perspective of a vertex rather than a graph. Such an approach improves locality, demonstrates linear scalability, and provides a natural way to express and compute many iterative graph algorithms. These frameworks are simple to program and widely applicable but, like an operating system, are composed of several intricate, interdependent components, of which a thorough understanding is necessary in order to elicit top performance at scale. To this end, the first comprehensive survey of TLAV frameworks is presented. In this survey, the vertex-centric approach to graph processing is overviewed, TLAV frameworks are deconstructed into four main components and respectively analyzed, and TLAV implementations are reviewed and categorized."
]
} |
1607.02524 | 2466557364 | This paper considers the fundamental limit of compressed sensing for i.i.d. signal distributions and i.i.d. Gaussian measurement matrices. Its main contribution is a rigorous characterization of the asymptotic mutual information (MI) and minimum mean-square error (MMSE) in this setting. Under mild technical conditions, our results show that the limiting MI and MMSE are equal to the values predicted by the replica method from statistical physics. This resolves a well-known problem that has remained open for over a decade. | The replica method was developed originally to study mean-field approximations in spin glasses @cite_13 @cite_14 . It was first applied to linear estimation problems in the context of CDMA wireless communication @cite_45 @cite_23 @cite_28 , with subsequent work focusing on the compressed sensing directly @cite_26 @cite_38 @cite_11 @cite_43 @cite_25 @cite_39 . | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_26",
"@cite_28",
"@cite_39",
"@cite_43",
"@cite_45",
"@cite_23",
"@cite_13",
"@cite_25",
"@cite_11"
],
"mid": [
"",
"2566505556",
"2550925785",
"",
"",
"",
"2116394223",
"",
"1967084746",
"2071108209",
""
],
"abstract": [
"",
"This book presents a unified approach to a rich and rapidly evolving research domain at the interface between statistical physics, theoretical computer science discrete mathematics, and coding information theory. It is accessible to graduate students and researchers without a specific training in any of these fields. The selected topics include spin glasses, error correcting codes, satisfiability, and are central to each field. The approach focuses on large random instances, adopting a common probabilistic formulation in terms of graphical models. It presents message passing algorithms like belief propagation and survey propagation, and their use in decoding and constraint satisfaction solving. It also explains analysis techniques like density evolution and the cavity method, and uses them to study phase transitions.",
"Compressed sensing deals with the reconstruction of a high-dimensional signal from far fewer linear measurements, where the signal is known to admit a sparse representation in a certain linear space. The asymptotic scaling of the number of measurements needed for reconstruction as the dimension of the signal increases has been studied extensively. This work takes a fundamental perspective on the problem of inferring about individual elements of the sparse signal given the measurements, where the dimensions of the system become increasingly large. Using the replica method, the outcome of inferring about any fixed collection of signal elements is shown to be asymptotically decoupled, i.e., those elements become independent conditioned on the measurements. Furthermore, the problem of inferring about each signal element admits a single-letter characterization in the sense that the posterior distribution of the element, which is a sufficient statistic, becomes asymptotically identical to the posterior of inferring about the same element in scalar Gaussian noise. The result leads to simple characterization of all other elemental metrics of the compressed sensing problem, such as the mean squared error and the error probability for reconstructing the support set of the sparse signal. Finally, the single-letter characterization is rigorously justified in the special case of sparse measurement matrices where belief propagation becomes asymptotically optimal.",
"",
"",
"",
"We present a theory, based on statistical mechanics, to evaluate analytically the performance of uncoded, fully synchronous, randomly spread code-division multiple-access (CDMA) multiuser detectors with additive white Gaussian noise (AWGN) channel, under perfect power control, and in the large-system limit. Application of the replica method, a tool developed in the literature of statistical mechanics, allows us to derive analytical expressions for the bit-error rate, as well as the multiuser efficiency, of the individually optimum (IO) and jointly optimum (JO) multiuser detectors over the whole range of noise levels. The information-theoretic capacity of the randomly spread CDMA channel and the performance of decorrelating and linear minimum mean-square error (MMSE) detectors are also derived in the same replica formulation, thereby demonstrating validity of the statistical-mechanical approach.",
"",
"A new theory of the class of dilute magnetic alloys, called the spin glasses, is proposed which offers a simple explanation of the cusp found experimentally in the susceptibility. The argument is that because the interaction between the spins dissolved in the matrix oscillates in sign according to distance, there will be no mean ferro- or antiferromagnetism, but there will be a ground state with the spins aligned in definite directions, even if these directions appear to be at random. At the critical temperature the existence of these preferred directions affects the orientation of the spins, leading to a cusp in the susceptibility. This cusp is smoothed by an external field. Although the behaviour at low t needs a quantum mechanical treatment, it is interesting to complete the classical calculations down to t=0. Classically the susceptibility tends to a constant value at t=0, and the specific heat to a constant value.",
"In recent work, two different methods have been used to characterize the fundamental limits of compressed sensing. On the one hand are rigorous bounds based on information-theoretic arguments or the analysis of specific algorithms. On the other hand are exact but heuristic predictions made using the replica method from statistical physics. In this paper, it is shown that, for certain problem settings, these bounds are in agreement, and thus provide a rigorous and accurate characterization of the compressed sensing problem. This characterization shows that the limits of sparse recovery can be quantified succinctly in terms of an effective signal-to-interference-plus-noise ratio, that depends on the number of measurements and the behavior of the sparse components themselves. Connections with the MMSE dimension by Wu and Verdu and minimax behavior of approximate message passing by are discussed.",
""
]
} |
1607.02524 | 2466557364 | This paper considers the fundamental limit of compressed sensing for i.i.d. signal distributions and i.i.d. Gaussian measurement matrices. Its main contribution is a rigorous characterization of the asymptotic mutual information (MI) and minimum mean-square error (MMSE) in this setting. Under mild technical conditions, our results show that the limiting MI and MMSE are equal to the values predicted by the replica method from statistical physics. This resolves a well-known problem that has remained open for over a decade. | Within the context of compressed sensing, the results of the replica method have been proven rigorously in a number of settings. One example is given by message passing on matrices with special structure, such as sparsity @cite_3 @cite_24 @cite_22 or spatial coupling @cite_40 @cite_19 @cite_31 . However, in the case of i.i.d. matrices, the results are limited to signal distributions with a unique fixed point @cite_32 @cite_21 (e.g., Gaussian inputs @cite_41 @cite_44 ). For the special case of i.i.d. matrices with binary inputs, it has also been shown that the replica prediction provides an upper bound for the asymptotic mutual information @cite_2 . Bounds on the locations of discontinuities in the MMSE with sparse priors have also been obtain by analyzing the problem of approximate support recovery @cite_17 @cite_29 @cite_39 . | {
"cite_N": [
"@cite_22",
"@cite_41",
"@cite_29",
"@cite_21",
"@cite_32",
"@cite_3",
"@cite_39",
"@cite_24",
"@cite_19",
"@cite_40",
"@cite_44",
"@cite_2",
"@cite_31",
"@cite_17"
],
"mid": [
"",
"",
"",
"",
"2082029531",
"2949400297",
"",
"",
"",
"2952440034",
"2127995963",
"",
"",
"2085493178"
],
"abstract": [
"",
"",
"",
"",
"Compressed sensing aims to undersample certain high-dimensional signals yet accurately reconstruct them by exploiting signal characteristics. Accurate reconstruction is possible when the object to be recovered is sufficiently sparse in a known basis. Currently, the best known sparsity–undersampling tradeoff is achieved when reconstructing by convex optimization, which is expensive in important large-scale applications. Fast iterative thresholding algorithms have been intensively studied as alternatives to convex optimization for large-scale problems. Unfortunately known fast algorithms offer substantially worse sparsity–undersampling tradeoffs than convex optimization. We introduce a simple costless modification to iterative thresholding making the sparsity–undersampling tradeoff of the new algorithms equivalent to that of the corresponding convex optimization procedures. The new iterative-thresholding algorithms are inspired by belief propagation in graphical models. Our empirical measurements of the sparsity–undersampling tradeoff for the new algorithms agree with theoretical calculations. We show that a state evolution formalism correctly derives the true sparsity–undersampling tradeoff. There is a surprising agreement between earlier calculations based on random convex polytopes and this apparently very different theoretical formalism.",
"We consider the CDMA (code-division multiple-access) multi-user detection problem for binary signals and additive white gaussian noise. We propose a spreading sequences scheme based on random sparse signatures, and a detection algorithm based on belief propagation (BP) with linear time complexity. In the new scheme, each user conveys its power onto a finite number of chips l, in the large system limit. We analyze the performances of BP detection and prove that they coincide with the ones of optimal (symbol MAP) detection in the l-> limit. In the same limit, we prove that the information capacity of the system converges to Tanaka's formula for random dense' signatures, thus providing the first rigorous justification of this formula. Apart from being computationally convenient, the new scheme allows for optimization in close analogy with irregular low density parity check code ensembles.",
"",
"",
"",
"Recently, it was observed that spatially-coupled LDPC code ensembles approach the Shannon capacity for a class of binary-input memoryless symmetric (BMS) channels. The fundamental reason for this was attributed to a \"threshold saturation\" phenomena derived by Kudekar, Richardson and Urbanke. In particular, it was shown that the belief propagation (BP) threshold of the spatially coupled codes is equal to the maximum a posteriori (MAP) decoding threshold of the underlying constituent codes. In this sense, the BP threshold is saturated to its maximum value. Moreover, it has been empirically observed that the same phenomena also occurs when transmitting over more general classes of BMS channels. In this paper, we show that the effect of spatial coupling is not restricted to the realm of channel coding. The effect of coupling also manifests itself in compressed sensing. Specifically, we show that spatially-coupled measurement matrices have an improved sparsity to sampling threshold for reconstruction algorithms based on verification decoding. For BP-based reconstruction algorithms, this phenomenon is also tested empirically via simulation. At the block lengths accessible via simulation, the effect is quite small and it seems that spatial coupling is not providing the gains one might expect. Based on the threshold analysis, however, we believe this warrants further study.",
"Multiuser receivers improve the performance of spread-spectrum and antenna-array systems by exploiting the structure of the multiaccess interference when demodulating the signal of a user. Much of the previous work on the performance analysis of multiuser receivers has focused on their ability to reject worst case interference. Their performance in a power-controlled network and the resulting user capacity are less well-understood. We show that in a large system with each user using random spreading sequences, the limiting interference effects under several linear multiuser receivers can be decoupled, such that each interferer can be ascribed a level of effective interference that it provides to the user to be demodulated. Applying these results to the uplink of a single power-controlled cell, we derive an effective bandwidth characterization of the user capacity: the signal-to-interference requirements of all the users can be met if and only if the sum of the effective bandwidths of the users is less than the total number of degrees of freedom in the system. The effective bandwidth of a user depends only on its own SIR requirement, and simple expressions are derived for three linear receivers: the conventional matched filter, the decorrelator, and the MMSE receiver. The effective bandwidths under the three receivers serve as a basis for performance comparison.",
"",
"",
"Recovery of the sparsity pattern (or support) of an unknown sparse vector from a limited number of noisy linear measurements is an important problem in compressed sensing. In the high-dimensional setting, it is known that recovery with a vanishing fraction of errors is impossible if the measurement rate and the per-sample signal-to-noise ratio (SNR) are finite constants, independent of the vector length. In this paper, it is shown that recovery with an arbitrarily small but constant fraction of errors is, however, possible, and that in some cases computationally simple estimators are near-optimal. Bounds on the measurement rate needed to attain a desired fraction of errors are given in terms of the SNR and various key parameters of the unknown vector for several different recovery algorithms. The tightness of the bounds, in a scaling sense, as a function of the SNR and the fraction of errors, is established by comparison with existing information-theoretic necessary bounds. Near optimality is shown for a wide variety of practically motivated signal models."
]
} |
1607.02524 | 2466557364 | This paper considers the fundamental limit of compressed sensing for i.i.d. signal distributions and i.i.d. Gaussian measurement matrices. Its main contribution is a rigorous characterization of the asymptotic mutual information (MI) and minimum mean-square error (MMSE) in this setting. Under mild technical conditions, our results show that the limiting MI and MMSE are equal to the values predicted by the replica method from statistical physics. This resolves a well-known problem that has remained open for over a decade. | Recent work by Huleihel and Merhav @cite_18 addresses the validity of the replica MMSE directly in the case of Gaussian mixture models, using tools from statistical physics and random matrix theory @cite_5 @cite_30 . | {
"cite_N": [
"@cite_30",
"@cite_5",
"@cite_18"
],
"mid": [
"2168671066",
"2117093731",
"2090842051"
],
"abstract": [
"In continuation to a recent work on the statistical-mechanical analysis of minimum mean square error (MMSE) estimation in Gaussian noise via its relation to the mutual information (the I-MMSE relation), here we propose a simple and more direct relationship between optimum estimation and certain information measures (e.g., the information density and the Fisher information), which can be viewed as partition functions and hence are amenable to analysis using statistical-mechanical techniques. The proposed approach has several advantages, most notably, its applicability to general sources and channels, as opposed to the I-MMSE relation and its variants which hold only for certain classes of channels (e.g., additive white Gaussian noise channels). We then demonstrate the derivation of the conditional mean estimator and the MMSE in a few examples. Two of these examples turn out to be generalizable to a fairly wide class of sources and channels. For this class, the proposed approach is shown to yield an approximate conditional mean estimator and an MMSE formula that has the flavor of a single-letter expression. We also show how our approach can easily be generalized to situations of mismatched estimation.",
"We consider the problem of signal estimation (denoising) from a statistical mechanical perspective, using a relationship between the minimum mean square error (MMSE), of estimating a signal, and the mutual information between this signal and its noisy version. The paper consists of essentially two parts. In the first, we derive several statistical-mechanical relationships between a few important quantities in this problem area, such as the MMSE, the differential entropy, the Fisher information, the free energy, and a generalized notion of temperature. We also draw analogies and differences between certain relations pertaining to the estimation problem and the parallel relations in thermodynamics and statistical physics. In the second part of the paper, we provide several application examples, where we demonstrate how certain analysis tools that are customary in statistical physics, prove useful in the analysis of the MMSE. In most of these examples, the corresponding statistical-mechanical systems turn out to consist of strong interactions that cause phase transitions, which in turn are reflected as irregularities and discontinuities (similar to threshold effects) in the behavior of the MMSE.",
"The replica method is a nonrigorous but well-known technique from statistical physics used in the asymptotic analysis of large, random, nonlinear problems. This paper applies the replica method, under the assumption of replica symmetry, to study estimators that are maximum a posteriori (MAP) under a postulated prior distribution. It is shown that with random linear measurements and Gaussian noise, the replica-symmetric prediction of the asymptotic behavior of the postulated MAP estimate of an -dimensional vector “decouples” as scalar postulated MAP estimators. The result is based on applying a hardening argument to the replica analysis of postulated posterior mean estimators of Tanaka and of Guo and Verdu. The replica-symmetric postulated MAP analysis can be readily applied to many estimators used in compressed sensing, including basis pursuit, least absolute shrinkage and selection operator (LASSO), linear estimation with thresholding, and zero norm-regularized estimation. In the case of LASSO estimation, the scalar estimator reduces to a soft-thresholding operator, and for zero norm-regularized estimation, it reduces to a hard threshold. Among other benefits, the replica method provides a computationally tractable method for precisely predicting various performance metrics including mean-squared error and sparsity pattern recovery probability."
]
} |
1607.02562 | 2949311108 | We propose extensions to the Dolev-Yao attacker model to make it suitable for arguments about security of Cyber-Physical Systems. The Dolev-Yao attacker model uses a set of rules to define potential actions by an attacker with respect to messages (i.e. information) exchanged between parties during a protocol execution. The model can be used to argue about the security of such protocols (e.g., using model checking). As the traditional Dolev-Yao model considers only information (exchanged over a channel controlled by the attacker), the model cannot directly be used to argue about the security of cyber-physical systems where physical-layer interactions are possible. In this work, we propose a cyber-physical Dolev-Yao (CPDY) attacker model, a general extension of the Dolev-Yao model to allow additional orthogonal interaction channels between the parties. In particular, such orthogonal channels can be used to model physical-layer mechanical, chemical, or electrical interactions between components. In addition, we discuss the inclusion of physical properties such as location or distance in the rule set. We present an example set of additional rules for the Dolev-Yao attacker, using those we are able to formally discover physical attacks that previously could only be found by empirical methods or detailed physical process models. | The formal verification of security properties of CPS is a non trivial task, as CPS introduce physical properties to the system under analysis. SAT SMT solvers used by security analysis tools (e.g., @cite_28 ) do not support such properties. In order to overcome this limitation, one could simulate the process (e.g., @cite_7 ) or adapt the level of abstraction of CPS components. @cite_9 , the author presents a formal definition of an attacker model for CPS. The attacker is defined as a set of pairs representing locations and capabilities. Capabilities are defined as a set of tuples expressing actions, cost (energy time) and range (with respect to the topology) of the attacker. The attacker is assumed to perform two types of attacks: , against a device and against the communications; where the first requires physical access while the second proximity to the node. The actions of the attacker are (remove, read write, reveal, reprogram, starve) and (block, eavesdrop, inject). | {
"cite_N": [
"@cite_28",
"@cite_9",
"@cite_7"
],
"mid": [
"1782799247",
"",
"2952719762"
],
"abstract": [
"The AVANTSSAR Platform is an integrated toolset for the formal specification and automated validation of trust and security of service-oriented architectures and other applications in the Internet of Services. The platform supports application-level specification languages (such as BPMN and our custom languages) and features three validation backends (CL-AtSe, OFMC, and SATMC), which provide a range of complementary automated reasoning techniques (including service orchestration, compositional reasoning, model checking, and abstract interpretation). We have applied the platform to a large number of industrial case studies, collected into the AVANTSSAR Library of validated problem cases. In doing so, we unveiled a number of problems and vulnerabilities in deployed services. These include, most notably, a serious flaw in the SAML-based Single Sign-On for Google Apps (now corrected by Google as a result of our findings). We also report on the migration of the platform to industry.",
"",
"In recent years, tremendous effort has been spent to modernizing communication infrastructure in Cyber-Physical Systems (CPS) such as Industrial Control Systems (ICS) and related Supervisory Control and Data Acquisition (SCADA) systems. While a great amount of research has been conducted on network security of office and home networks, recently the security of CPS and related systems has gained a lot of attention. Unfortunately, real-world CPS are often not open to security researchers, and as a result very few reference systems and topologies are available. In this work, we present MiniCPS, a CPS simulation toolbox intended to alleviate this problem. The goal of MiniCPS is to create an extensible, reproducible research environment targeted to communications and physical-layer interactions in CPS. MiniCPS builds on Mininet to provide lightweight real-time network emulation, and extends Mininet with tools to simulate typical CPS components such as programmable logic controllers, which use industrial protocols (Ethernet IP, Modbus TCP). In addition, MiniCPS defines a simple API to enable physical-layer interaction simulation. In this work, we demonstrate applications of MiniCPS in two example scenarios, and show how MiniCPS can be used to develop attacks and defenses that are directly applicable to real systems."
]
} |
1607.02562 | 2949311108 | We propose extensions to the Dolev-Yao attacker model to make it suitable for arguments about security of Cyber-Physical Systems. The Dolev-Yao attacker model uses a set of rules to define potential actions by an attacker with respect to messages (i.e. information) exchanged between parties during a protocol execution. The model can be used to argue about the security of such protocols (e.g., using model checking). As the traditional Dolev-Yao model considers only information (exchanged over a channel controlled by the attacker), the model cannot directly be used to argue about the security of cyber-physical systems where physical-layer interactions are possible. In this work, we propose a cyber-physical Dolev-Yao (CPDY) attacker model, a general extension of the Dolev-Yao model to allow additional orthogonal interaction channels between the parties. In particular, such orthogonal channels can be used to model physical-layer mechanical, chemical, or electrical interactions between components. In addition, we discuss the inclusion of physical properties such as location or distance in the rule set. We present an example set of additional rules for the Dolev-Yao attacker, using those we are able to formally discover physical attacks that previously could only be found by empirical methods or detailed physical process models. | @cite_29 @cite_15 , the authors present a formalization to reason on security properties of wireless networks (including a considerations of physical properties related to those networks). The authors present an attacker model as a variation of the DY attacker model. The attacker is a malicious agent of the network who cannot break cryptography. He has a fixed location, while the usual DY controls the entire network, a set of transmitters and receivers, an initial knowledge with his private public keys which can use to create and analyze messages. The authors also consider constraints on the distance of communicating parties. An attacker can only intercept messages at his location and colluding attackers do not instantaneously exchange knowledge, they are constrained by the network topology. | {
"cite_N": [
"@cite_29",
"@cite_15"
],
"mid": [
"2594004772",
"2011329056"
],
"abstract": [
"We present a formal model for modeling and reasoning about security protocols. Our model extends standard, inductive, trace-based, symbolic approaches with a formalization of physical properties of the environment, namely communication, location, and time. In particular, communication is subject to physical constraints, for example, message transmission takes time determined by the communication medium used and the distance traveled. All agents, including intruders, are subject to these constraints and this results in a distributed intruder with restricted, but more realistic, communication capabilities than the standard Dolev-Yao intruder. We have formalized our model in Isabelle HOL and used it to verify protocols for authenticated ranging, distance bounding, and broadcast authentication based on delayed key disclosure.",
"Traditional security protocols are mainly concerned with authentication and key establishment and rely on predistributed keys and properties of cryptographic operators. In contrast, new application areas are emerging that establish and rely on properties of the physical world. Examples include protocols for secure localization, distance bounding, and secure time synchronization. We present a formal model for modeling and reasoning about such physical security protocols. Our model extends standard, inductive, trace-based, symbolic approaches with a formalization of physical properties of the environment, namely communication, location, and time. In particular, communication is subject to physical constraints, for example, message transmission takes time determined by the communication medium used and the distance between nodes. All agents, including intruders, are subject to these constraints and this results in a distributed intruder with restricted, but more realistic, communication capabilities than those of the standard Dolev-Yao intruder. We have formalized our model in Isabelle HOL and have used it to verify protocols for authenticated ranging, distance bounding, broadcast authentication based on delayed key disclosure, and time synchronization."
]
} |
1607.02466 | 2460961953 | In this paper, we investigate the possibility of improvement of the widely-used filtering algorithm for the linear constraints in constraint satisfaction problems in the presence of the alldifferent constraints. In many cases, the fact that the variables in a linear constraint are also constrained by some alldifferent constraints may help us to calculate stronger bounds of the variables, leading to a stronger constraint propagation. We propose an improved filtering algorithm that targets such cases. We provide a detailed description of the proposed algorithm and prove its correctness. We evaluate the approach on five different problems that involve combinations of the linear and the alldifferent constraints. We also compare our algorithm to other relevant approaches. The experimental results show a great potential of the proposed improvement. | Finally, the idea of combining different types of global constraints in order to improve the constraint propagation is also employed in case of other constraints. For instance, in @cite_1 , the combination of the with two linear constraints is considered. On the other hand, in @cite_22 , the authors consider a combination of an constraint with the . | {
"cite_N": [
"@cite_1",
"@cite_22"
],
"mid": [
"1484088481",
"2952340783"
],
"abstract": [
"We introduce a new global constraint which combines together the lexicographic ordering constraint with some sum constraints. Lexicographic ordering constraints are frequently used to break symmetry, whilst sum constraints occur in many problems involving capacity or partitioning. Our results show that this global constraint is useful when there is a very large space to explore, such as when the problem is unsatisfiable, or when the search strategy is poor or conflicts with the symmetry breaking constraints. By studying in detail when combining lexicographical ordering with other constraints is useful, we propose a new heuristic for deciding when to combine constraints together.",
"We propose AllDiffPrecedence, a new global constraint that combines together an AllDifferent constraint with precedence constraints that strictly order given pairs of variables. We identify a number of applications for this global constraint including instruction scheduling and symmetry breaking. We give an efficient propagation algorithm that enforces bounds consistency on this global constraint. We show how to implement this propagator using a decomposition that extends the bounds consistency enforcing decomposition proposed for the AllDifferent constraint. Finally, we prove that enforcing domain consistency on this global constraint is NP-hard in general."
]
} |
1607.02552 | 2464501940 | We consider the problem of power allocation over a time-varying channel with unknown distribution in energy harvesting communication systems. In this problem, the transmitter has to choose the transmit power based on the amount of stored energy in its battery with the goal of maximizing the average rate obtained over time. We model this problem as a Markov decision process (MDP) with the transmitter as the agent, the battery status as the state, the transmit power as the action and the rate obtained as the reward. The average reward maximization problem over the MDP can be solved by a linear program (LP) that uses the transition probabilities for the state-action pairs and their reward values to choose a power allocation policy. Since the rewards associated the state-action pairs are unknown, we propose two online learning algorithms: UCLP and Epoch-UCLP that learn these rewards and adapt their policies along the way. The UCLP algorithm solves the LP at each step to decide its current policy using the upper confidence bounds on the rewards, while the Epoch-UCLP algorithm divides the time into epochs, solves the LP only at the beginning of the epochs and follows the obtained policy in that epoch. We prove that the reward losses or regrets incurred by both these algorithms are upper bounded by constants. Epoch-UCLP incurs a higher regret compared to UCLP, but reduces the computational requirements substantially. We also show that the presented algorithms work for online learning in cost minimization problems like the packet scheduling with power-delay tradeoff with minor changes. | Our problem is also closely related to the reinforcement learning problem over MDPs from @cite_22 @cite_32 @cite_4 . The objective for these problems is to maximize the average undiscounted reward over time. In @cite_22 @cite_32 , the agent is unaware of the transition probabilities and the mean rewards corresponding to the state-action pairs. In @cite_4 , the agent knows the mean rewards, but the transition probabilities are still unknown. In our problem, the mean rewards are unknown, while the transition probabilities of the MDP can be inferred from the knowledge of the arrival distribution and the action taken from each state. In contrast to the works above, for our problem motivated by the practical application in energy harvesting communications, we show that the learning incurs a constant regret in the single channel case. | {
"cite_N": [
"@cite_32",
"@cite_22",
"@cite_4"
],
"mid": [
"",
"2097931172",
"2145049547"
],
"abstract": [
"",
"We present a learning algorithm for undiscounted reinforcement learning. Our interest lies in bounds for the algorithm's online performance after some finite number of steps. In the spirit of similar methods already successfully applied for the exploration-exploitation tradeoff in multi-armed bandit problems, we use upper confidence bounds to show that our UCRL algorithm achieves logarithmic online regret in the number of steps taken with respect to an optimal policy.",
"We present an algorithm called Optimistic Linear Programming (OLP) for learning to optimize average reward in an irreducible but otherwise unknown Markov decision process (MDP). OLP uses its experience so far to estimate the MDP. It chooses actions by optimistically maximizing estimated future rewards over a set of next-state transition probabilities that are close to the estimates, a computation that corresponds to solving linear programs. We show that the total expected reward obtained by OLP up to time T is within C(P) log T of the reward obtained by the optimal policy, where C(P) is an explicit, MDP-dependent constant. OLP is closely related to an algorithm proposed by Burnetas and Katehakis with four key differences: OLP is simpler, it does not require knowledge of the supports of transition probabilities, the proof of the regret bound is simpler, but our regret bound is a constant factor larger than the regret of their algorithm. OLP is also similar in flavor to an algorithm recently proposed by Auer and Ortner. But OLP is simpler and its regret bound has a better dependence on the size of the MDP."
]
} |
1607.02537 | 2464704181 | Context in image is crucial for scene labeling while existing methods only exploit local context generated from a small surrounding area of an image patch or a pixel, by contrast long-range and global contextual information is ignored. To handle this issue, we in this work propose a novel approach for scene labeling by exploring multi-level contextual recurrent neural networks (ML-CRNNs). Specifically, we encode three kinds of contextual cues, i.e., local context, global context and image topic context in structural recurrent neural networks (RNNs) to model long-range local and global dependencies in image. In this way, our method is able to see' the image in terms of both long-range local and holistic views, and make a more reliable inference for image labeling. Besides, we integrate the proposed contextual RNNs into hierarchical convolutional neural networks (CNNs), and exploit dependence relationships in multiple levels to provide rich spatial and semantic information. Moreover, we novelly adopt an attention model to effectively merge multiple levels and show that it outperforms average- or max-pooling fusion strategies. Extensive experiments demonstrate that the proposed approach achieves new state-of-the-art results on the CamVid, SiftFlow and Stanford-background datasets. | As one of the most fundamental problems in computer vision, image labeling has attracted increasing attention in recent years. Several previous non-parametric approaches try to transfer the labels of training data to the query images and perform label inference in a probabilistic graphical model (PGM). Liu @cite_0 propose a non-parametric image parsing method by estimating SIFT Flow' between images, and infer the labels of pixels in a markov random field (MRF). In @cite_10 , Tighe introduce a superparsing method to classify superpixels by comparing @math -nearest neighbors in a retrieval dataset, and infer their labels with MRF. Yang @cite_13 suggest to incorporate context information to improve image retrieval and superpixel classification, and develop a four-connected pairwise MRF for semantic labeling. | {
"cite_N": [
"@cite_0",
"@cite_10",
"@cite_13"
],
"mid": [
"2125849446",
"1542723449",
"2051458493"
],
"abstract": [
"In this paper we propose a novel nonparametric approach for object recognition and scene parsing using dense scene alignment. Given an input image, we retrieve its best matches from a large database with annotated images using our modified, coarse-to-fine SIFT flow algorithm that aligns the structures within two images. Based on the dense scene correspondence obtained from the SIFT flow, our system warps the existing annotations, and integrates multiple cues in a Markov random field framework to segment and recognize the query image. Promising experimental results have been achieved by our nonparametric scene parsing system on a challenging database. Compared to existing object recognition approaches that require training for each object category, our system is easy to implement, has few parameters, and embeds contextual information naturally in the retrieval alignment procedure.",
"This paper presents a simple and effective nonparametric approach to the problem of image parsing, or labeling image regions (in our case, superpixels produced by bottom-up segmentation) with their categories. This approach requires no training, and it can easily scale to datasets with tens of thousands of images and hundreds of labels. It works by scene-level matching with global image descriptors, followed by superpixel-level matching with local features and efficient Markov random field (MRF) optimization for incorporating neighborhood context. Our MRF setup can also compute a simultaneous labeling of image regions into semantic classes (e.g., tree, building, car) and geometric classes (sky, vertical, ground). Our system outperforms the state-of-the-art non-parametric method based on SIFT Flow on a dataset of 2,688 images and 33 labels. In addition, we report per-pixel rates on a larger dataset of 15,150 images and 170 labels. To our knowledge, this is the first complete evaluation of image parsing on a dataset of this size, and it establishes a new benchmark for the problem.",
"This paper presents a scalable scene parsing algorithm based on image retrieval and superpixel matching. We focus on rare object classes, which play an important role in achieving richer semantic understanding of visual scenes, compared to common background classes. Towards this end, we make two novel contributions: rare class expansion and semantic context description. First, considering the long-tailed nature of the label distribution, we expand the retrieval set by rare class exemplars and thus achieve more balanced superpixel classification results. Second, we incorporate both global and local semantic context information through a feedback based mechanism to refine image retrieval and superpixel matching. Results on the SIFTflow and LMSun datasets show the superior performance of our algorithm, especially on the rare classes, without sacrificing overall labeling accuracy."
]
} |
1607.02537 | 2464704181 | Context in image is crucial for scene labeling while existing methods only exploit local context generated from a small surrounding area of an image patch or a pixel, by contrast long-range and global contextual information is ignored. To handle this issue, we in this work propose a novel approach for scene labeling by exploring multi-level contextual recurrent neural networks (ML-CRNNs). Specifically, we encode three kinds of contextual cues, i.e., local context, global context and image topic context in structural recurrent neural networks (RNNs) to model long-range local and global dependencies in image. In this way, our method is able to see' the image in terms of both long-range local and holistic views, and make a more reliable inference for image labeling. Besides, we integrate the proposed contextual RNNs into hierarchical convolutional neural networks (CNNs), and exploit dependence relationships in multiple levels to provide rich spatial and semantic information. Moreover, we novelly adopt an attention model to effectively merge multiple levels and show that it outperforms average- or max-pooling fusion strategies. Extensive experiments demonstrate that the proposed approach achieves new state-of-the-art results on the CamVid, SiftFlow and Stanford-background datasets. | The recent deep CNNs @cite_28 , which demonstrates powerfulness in extracting high-level feature representation @cite_12 , have been successfully applied to scene labeling. In @cite_17 , Farabet propose to learn hierarchical features with CNNs for scene labeling. To incorporate rich context, this method stacks surrounding contextual windows from different scales. Long @cite_6 introduce the fully conventional networks for semantic labeling. Shuai @cite_23 adopt CNNs as parametric model to learn discriminative features and integrate it with a non-parametric model to infer pixel labels. Pinheiro @cite_39 utilize CNNs in a recurrent way to model spatial dependencies in image by attaching raw input with the output of CNNs. In @cite_36 , Liang suggest to model the relationships among intermediate convolutional layers with RNNs for scene labeling. However, they do not consider inner structure among image units, thus the long-range dependencies in image are not captured. | {
"cite_N": [
"@cite_28",
"@cite_36",
"@cite_6",
"@cite_39",
"@cite_23",
"@cite_12",
"@cite_17"
],
"mid": [
"2147800946",
"2186025005",
"2952632681",
"2951277909",
"1897260080",
"1686810756",
"2022508996"
],
"abstract": [
"The ability of learning networks to generalize can be greatly enhanced by providing constraints from the task domain. This paper demonstrates how such constraints can be integrated into a backpropagation network through the architecture of the network. This approach has been successfully applied to the recognition of handwritten zip code digits provided by the U.S. Postal Service. A single network learns the entire recognition operation, going from the normalized image of the character to the final classification.",
"Scene labeling is a challenging computer vision task. It requires the use of both local discriminative features and global context information. We adopt a deep recurrent convolutional neural network (RCNN) for this task, which is originally proposed for object recognition. Different from traditional convolutional neural networks (CNN), this model has intra-layer recurrent connections in the convolutional layers. Therefore each convolutional layer becomes a two-dimensional recurrent neural network. The units receive constant feed-forward inputs from the previous layer and recurrent inputs from their neighborhoods. While recurrent iterations proceed, the region of context captured by each unit expands. In this way, feature extraction and context modulation are seamlessly integrated, which is different from typical methods that entail separate modules for the two steps. To further utilize the context, a multi-scale RCNN is proposed. Over two benchmark datasets, Standford Background and Sift Flow, the model outperforms many state-of-the-art models in accuracy and efficiency.",
"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 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.",
"We adopt Convolutional Neural Networks (CNN) as our parametric model to learn discriminative features and classifiers for local patch classification. As visually similar pixels are indistinguishable from local context, we alleviate such ambiguity by introducing a global scene constraint. We estimate the global potential in a non-parametric framework. Furthermore, a large margin based CNN metric learning method is proposed for better global potential estimation. The final pixel class prediction is performed by integrating local and global beliefs. Even without any post-processing, we achieve state-of-the-art performance on SiftFlow and competitive results on Stanford Background benchmark.",
"In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.",
"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."
]
} |
1607.02555 | 2464674920 | We present a dataset for evaluating the tracking accuracy of monocular visual odometry and SLAM methods. It contains 50 real-world sequences comprising more than 100 minutes of video, recorded across dozens of different environments -- ranging from narrow indoor corridors to wide outdoor scenes. All sequences contain mostly exploring camera motion, starting and ending at the same position. This allows to evaluate tracking accuracy via the accumulated drift from start to end, without requiring ground truth for the full sequence. In contrast to existing datasets, all sequences are photometrically calibrated. We provide exposure times for each frame as reported by the sensor, the camera response function, and dense lens attenuation factors. We also propose a novel, simple approach to non-parametric vignette calibration, which requires minimal set-up and is easy to reproduce. Finally, we thoroughly evaluate two existing methods (ORB-SLAM and DSO) on the dataset, including an analysis of the effect of image resolution, camera field of view, and the camera motion direction. | Many approaches exist to calibrate and remove vignetting artefacts and account for non-linear response functions. Early work focuses on image stitching and mosaicking, where the required calibration parameters need to be estimated from a small set of overlapping images @cite_14 @cite_11 @cite_13 . Since the available data is limited, such methods attempt to find low-dimensional (parametric) function representations, like radially symmetric polynomial representations for vignetting. More recent work @cite_17 @cite_15 has shown that such representations may not be sufficiently expressive to capture the complex nature of real-world lenses and hence advocate non-parametric -- dense -- vignetting calibration. In contrast to @cite_17 @cite_15 , our formulation however does not require a uniformly lit white paper , simplifying the required calibration set-up. | {
"cite_N": [
"@cite_14",
"@cite_11",
"@cite_15",
"@cite_13",
"@cite_17"
],
"mid": [
"2159133693",
"2123315723",
"2067127947",
"2126060993",
""
],
"abstract": [
"We discuss calibration and removal of \"vignetting\" (radial falloff) and exposure (gain) variations from sequences of images. Unique solutions for vignetting, exposure and scene radiances are possible when the response curve is known. When the response curve is unknown, an exponential ambiguity prevents us from recovering these parameters uniquely. However, the vignetting and exposure variations can nonetheless be removed from the images without resolving this ambiguity. Applications include panoramic image mosaics, photometry for material reconstruction, image-based rendering, and preprocessing for correlation-based vision algorithms.",
"In many computer vision systems, it is assumed that the image brightness of a point directly reflects the scene radiance of the point. However, the assumption does not hold in most cases due to nonlinear camera response function, exposure changes, and vignetting. The effects of these factors are most visible in image mosaics and textures of 3D models where colors look inconsistent and notable boundaries exist. In this paper, we propose a full radiometric calibration algorithm that includes robust estimation of the radiometric response function, exposures, and vignetting. By decoupling the effect of vignetting from the response function estimation, we approach each process in a manner that is robust to noise and outliers. We verify our algorithm with both synthetic and real data, which shows significant improvement compared to existing methods. We apply our estimation results to radiometrically align images for seamless mosaics and 3D model textures. We also use our method to create high dynamic range (HDR) mosaics that are more representative of the scene than normal mosaics.",
"Creating textured 3D scans of indoor environments has experienced a large boost with the advent of cheap commodity depth sensors. However, the quality of the acquired 3D models is often impaired by color seams in the reconstruction due to varying illumination (e.g., Shadows or highlights) and object surfaces whose brightness and color vary with the viewpoint of the camera. In this paper, we propose a direct and simple method to estimate the pure albedo of the texture, which allows us to remove illumination effects from IR and color images. Our approach first computes the illumination-independent albedo in the IR domain, which we subsequently transfer to the color albedo. As shadows and highlights lead to over- and underexposed image regions with little or no color information, we apply an advanced optimization scheme to infer color information in the color albedo from neigh boring image regions. We demonstrate the applicability of our approach to various real-world scenes.",
"This paper concerns the problem of fully automated panoramic image stitching. Though the 1D problem (single axis of rotation) is well studied, 2D or multi-row stitching is more difficult. Previous approaches have used human input or restrictions on the image sequence in order to establish matching images. In this work, we formulate stitching as a multi-image matching problem, and use invariant local features to find matches between all of the images. Because of this our method is insensitive to the ordering, orientation, scale and illumination of the input images. It is also insensitive to noise images that are not part of a panorama, and can recognise multiple panoramas in an unordered image dataset. In addition to providing more detail, this paper extends our previous work in the area (Brown and Lowe, 2003) by introducing gain compensation and automatic straightening steps.",
""
]
} |
1607.02555 | 2464674920 | We present a dataset for evaluating the tracking accuracy of monocular visual odometry and SLAM methods. It contains 50 real-world sequences comprising more than 100 minutes of video, recorded across dozens of different environments -- ranging from narrow indoor corridors to wide outdoor scenes. All sequences contain mostly exploring camera motion, starting and ending at the same position. This allows to evaluate tracking accuracy via the accumulated drift from start to end, without requiring ground truth for the full sequence. In contrast to existing datasets, all sequences are photometrically calibrated. We provide exposure times for each frame as reported by the sensor, the camera response function, and dense lens attenuation factors. We also propose a novel, simple approach to non-parametric vignette calibration, which requires minimal set-up and is easy to reproduce. Finally, we thoroughly evaluate two existing methods (ORB-SLAM and DSO) on the dataset, including an analysis of the effect of image resolution, camera field of view, and the camera motion direction. | For response function estimation, a well-known and straight-forward method is that of Debevec and Malik @cite_10 , which -- like our approach -- recovers a @math -valued lookup table for the inverse response from two or more images of a static scene at different exposures. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2069281566"
],
"abstract": [
"We present a method of recovering high dynamic range radiance maps from photographs taken with conventional imaging equipment. In our method, multiple photographs of the scene are taken with different amounts of exposure. Our algorithm uses these differently exposed photographs to recover the response function of the imaging process, up to factor of scale, using the assumption of reciprocity. With the known response function, the algorithm can fuse the multiple photographs into a single, high dynamic range radiance map whose pixel values are proportional to the true radiance values in the scene. We demonstrate our method on images acquired with both photochemical and digital imaging processes. We discuss how this work is applicable in many areas of computer graphics involving digitized photographs, including image-based modeling, image compositing, and image processing. Lastly, we demonstrate a few applications of having high dynamic range radiance maps, such as synthesizing realistic motion blur and simulating the response of the human visual system."
]
} |
1607.02497 | 2462399439 | The predicted reduced resiliency of next-generation high performance computers means that it will become necessary to take into account the effects of randomly occurring faults on numerical methods. Further, in the event of a hard fault occurring, a decision has to be made as to what remedial action should be taken in order to resume the execution of the algorithm. The action that is chosen can have a dramatic effect on the performance and characteristics of the scheme. Ideally, the resulting algorithm should be subjected to the same kind of mathematical analysis that was applied to the original, deterministic variant. The purpose of this work is to provide an analysis of the behaviour of the multigrid algorithm in the presence of faults. Multigrid is arguably the method of choice for the solution of large-scale linear algebra problems arising from discretization of partial differential equations and it is of considerable importance to anticipate its behaviour on an exascale machine. The analysis of resilience of algorithms is in its infancy and the current work is perhaps the first to provide a mathematical model for faults and analyse the behaviour of a state-of-the-art algorithm under the model. It is shown that the Two Grid Method fails to be resilient to faults. Attention is then turned to identifying the minimal necessary remedial action required to restore the rate of convergence to that enjoyed by the ideal fault-free method. | Different techniques have been previously employed in order to achieve fault resilience for iterative methods in general and multigrid in particular. Replication was used in @cite_26 @cite_7 , and checkpoint-restart in @cite_20 . Stoyanov and Webster @cite_22 proposed a method based on selective reliability for fixed point methods. Finally, @cite_8 proposed a recovery method to mitigate the effect of hard faults. | {
"cite_N": [
"@cite_26",
"@cite_22",
"@cite_7",
"@cite_8",
"@cite_20"
],
"mid": [
"2951188726",
"1455915656",
"",
"2277193644",
"2295110215"
],
"abstract": [
"As we stride toward the exascale era, due to increasing complexity of supercomputers, hard and soft errors are causing more and more problems in high-performance scientific and engineering computation. In order to improve reliability (increase the mean time to failure) of computing systems, a lot of efforts have been devoted to developing techniques to forecast, prevent, and recover from errors at different levels, including architecture, application, and algorithm. In this paper, we focus on algorithmic error resilient iterative linear solvers and introduce a redundant subspace correction method. Using a general framework of redundant subspace corrections, we construct iterative methods, which have the following properties: (1) Maintain convergence when error occurs assuming it is detectable; (2) Introduce low computational overhead when no error occurs; (3) Require only small amount of local (point-to-point) communication compared to traditional methods and maintain good load balance; (4) Improve the mean time to failure. With the proposed method, we can improve reliability of many scientific and engineering applications. Preliminary numerical experiments demonstrate the efficiency and effectiveness of the new subspace correction method.",
"The exponential growth of computational power of the extreme scale machines over the past few decades has led to a corresponding decrease in reliability and a sharp increase of the frequency of hardware faults. Our research focuses on the mathematical challenges presented by the silent hardware faults; i.e., faults that can perturb the result of computations in an inconspicuous way. Using the approach of selective reliability, we present an analytic fault mode that can be used to study the resilience properties of a numerical algorithm. We apply our approach to the classical fixed point iteration and demonstrate that in the presence of hardware faults, the classical method fails to converge in expectation. We preset a modified resilient algorithm that detects and rejects faults resulting in error with large magnitude, while small faults are negated by the natural self-correcting properties of the algorithm. We show that our method is convergent (in first and second statistical moments) even in the presenc...",
"",
"Fault tolerant algorithms for the numerical approximation of elliptic partial differential equations on modern supercomputers play a more and more important role in the future design of exa-scale enabled iterative solvers. Here, we combine domain partitioning with highly scalable geometric multigrid schemes to obtain fast and fault-robust solvers in three dimensions. The recovery strategy is based on a hierarchical hybrid concept where the values on lower dimensional primitives such as faces are stored redundantly and thus can be recovered easily in case of a failure. The lost volume unknowns in the faulty region are re-computed approximately with multigrid cycles by solving a local Dirichlet problem on the faulty subdomain. Different strategies are compared and evaluated with respect to performance, computational cost, and speed up. Especially effective are strategies in which the local recovery in the faulty region is executed in parallel with global solves and when the local recovery is additionally accelerated. This results in an asynchronous multigrid iteration that can fully compensate faults. Excellent parallel performance on a current peta-scale system is demonstrated.",
"The effectiveness of sparse, linear solvers is typically studied in terms of their convergence properties and computational complexity, while their ability to handle transient hardware errors, such as bit-flips that lead to silent data corruption (SDC), has received less attention. As supercomputers continue to add more cores to increase performance, they are also becoming more susceptible to SDC. Consequently, understanding the impact of SDC on algorithms and common applications is an important component of solver analysis. In this paper, we investigate algebraic multigrid (AMG) in an environment exposed to corruptions through bit-flips. We propose an algorithmic based detection and recovery scheme that maintains the numerical properties of AMG, while maintaining high convergence rates in this environment. We also introduce a performance model and numerical results in support of the methodology."
]
} |
1607.02434 | 2949567618 | As the use of automotive radar increases, performance limitations associated with radar-to-radar interference will become more significant. In this paper we employ tools from stochastic geometry to characterize the statistics of radar interference. Specifically, using two different models for vehicle spacial distributions, namely, a Poisson point process and a Bernoulli lattice process, we calculate for each case the interference statistics and obtain analytical expressions for the probability of successful range estimation. Our study shows that the regularity of the geometrical model appears to have limited effect on the interference statistics, and so it is possible to obtain tractable tight bounds for worst case performance. A technique is proposed for designing the duty cycle for random spectrum access which optimizes the total performance. This analytical framework is verified using Monte-Carlo simulations. | The bandwidth requirement of the long range category is planned to be 1 GHz, with maximum allowed Equivalent Isotropic Radiated Power (EIRP) of 55 dBm. The medium short range category has less power allowance, and a wider spectrum bandwidth to support higher range resolution for close targets, based on the standard resolution and bandwidth relation @cite_0 , @math , where @math is the range resolution, @math is the speed of light and @math is the used bandwidth. | {
"cite_N": [
"@cite_0"
],
"mid": [
"1987860578"
],
"abstract": [
"This paper presents a 77-GHz long-range automotive radar transceiver with the function of reducing mutual interference. The proposed frequency-hopping random chirp FMCW technique reconfigures the chirp sweep frequency and time every cycle to result in noise-like frequency response for mutual interference after the received signal is down-converted and demodulated. Thus, the false alarm rate can be reduced significantly. The transceiver IC is fully integrated in TSMC 1P9M 65-nm digital CMOS technology. The chip including pads occupies a silicon area of 1.03 mm × 0.94 mm. The transceiver consumes totally 275 mW of power, and the measured transmitting power and receiver noise figure are 6.4 dBm and 14.8 dB, respectively. To the authors' knowledge, this is the first integrated 77-GHz automotive radar transceiver with the feature of anti-interference."
]
} |
1607.02434 | 2949567618 | As the use of automotive radar increases, performance limitations associated with radar-to-radar interference will become more significant. In this paper we employ tools from stochastic geometry to characterize the statistics of radar interference. Specifically, using two different models for vehicle spacial distributions, namely, a Poisson point process and a Bernoulli lattice process, we calculate for each case the interference statistics and obtain analytical expressions for the probability of successful range estimation. Our study shows that the regularity of the geometrical model appears to have limited effect on the interference statistics, and so it is possible to obtain tractable tight bounds for worst case performance. A technique is proposed for designing the duty cycle for random spectrum access which optimizes the total performance. This analytical framework is verified using Monte-Carlo simulations. | Further analytic attempts to investigate automotive radar interference can be found in @cite_9 which studies the desired-to-undesired signal power ratio in ultra wideband automotive radar, also in @cite_7 and @cite_26 utilizing Frequency Modulation Continuous Wave (FMCW) as a modulation scheme. Simulation approaches can be found in @cite_19 and @cite_25 , mainly based on ray tracing with scenario specific simulation environments. To summarize, our understanding of the available literature on automotive radar interference, we list the following points: The majority of the literature is based on simulation and empirical approaches. Some analytic approaches investigate the interference in simple scenarios consisting of two vehicles. Simulation approaches investigate interference based on complex ray-tracing and stochastic environments. Most of the literature uses simulations and analysis, namely FMCW, and pulse radar. | {
"cite_N": [
"@cite_26",
"@cite_7",
"@cite_9",
"@cite_19",
"@cite_25"
],
"mid": [
"2219056947",
"",
"2080217539",
"2151000402",
"2299968745"
],
"abstract": [
"In this paper a method for interference detection and cancellation for automotive radar systems is proposed. With the growing amount of vehicles equipped with radar sensors, interference mitigation techniques are getting more and more important to maintain good interoperability. Based on the time domain signal of a 76 GHz chirp sequence radar the interfering signals of FMCW radar sensors are identified. This is performed by image processing methods applied to the time-frequency-image. With the maximally stable extremal regions algorithm the interference pattern in the signal is identified. Once the disturbed samples are known they are zeroed. To avoid any ringing effects in the processed radar image the neighborhood of affected samples is smoothed using a raised cosine window. The effectiveness of the proposed method is demonstrated on real world measurements. The method reveals weak scattering centers of the vehicle, which are occluded by interference otherwise.",
"",
"Ultra wideband (UWB) automotive radars, less expensive than conventional millimeter-wave radars, have attracted attention from the viewpoint of reducing traffic accidents. The performance of automotive radars is degraded by interference from nearby radars operating at the same frequency. We assumed a scenario where two cars pass each other on a road. The desired-to-undesired signal power ratio (DUR) was found to vary approximately from −10 to 30 dB when employing cross polarization. Allocation of different maximum length sequences to different radars facilitates suppression of interference from other radars. Probabilities of false alarm (P fa ) and detection of the passing car (P d ) were evaluated by simulation. It was found that P d = 0.995 and 0.993 for P fa = 10−2 and 10−4, respectively, when DUR = −10 dB (the worst prediction).",
"This paper provides qualitative and quantitative values for the received interference power at the antenna ports of automotive radars as well as the probability of their occurrence for actual and future, not yet measurable traffic scenarios on main roads. The influence of the environment, the road traffic behavior, and the radar penetration rate for a defined antenna configuration can be observed. The basis for the analyses are ray-tracing based simulations in order to achieve adequate predictions for the received power levels due to multipaths. The results show that for a radar penetration rate of 100 , the difference between the strongest overall incoherent received interference power level and the level that is received in 90 of the time is up to 7 dB, dependent on the antenna placement and the environment.",
"Radar is an essential element of state of the art advanced driver assistance systems. In the foreseeable future, radar will be an indispensable sensor for the use in affordable, automated driven cars. Simulation tools are the key for an efficient development process and hence will lower the price of sophisticated driver assistance systems. Therefore, the development of adequate simulators is important for suppliers, car makers, and final consumers. This paper introduces the concept of such a simulator for multi-user automotive radar scenarios and presents selected simulation results for a use case of radar interference."
]
} |
1607.02480 | 2507064736 | Much of the worlds data is streaming, time-series data, where anomalies give significant information in critical situations. Yet detecting anomalies in streaming data is a difficult task, requiring detectors to process data in real-time, and learn while simultaneously making predictions. We present a novel anomaly detection technique based on an on-line sequence memory algorithm called Hierarchical Temporal Memory (HTM). We show results from a live application that detects anomalies in financial metrics in real-time. We also test the algorithm on NAB, a published benchmark for real-time anomaly detection, where our algorithm achieves best-in-class results. | Anomaly detection in time-series is a heavily studied area, dating back to @cite_10 . Some techniques, like classification-based methods, are supervised or semi-supervised. While labelled data can be used to improve results, supervised techniques are typically unsuitable for anomaly detection @cite_27 . Figure illustrates the need for continuous learning, which is not typically possible with supervised algorithms. | {
"cite_N": [
"@cite_27",
"@cite_10"
],
"mid": [
"2143559571",
"4012559"
],
"abstract": [
"Anomaly detection is being regarded as an unsupervised learning task as anomalies stem from adversarial or unlikely events with unknown distributions. However, the predictive performance of purely unsupervised anomaly detection often fails to match the required detection rates in many tasks and there exists a need for labeled data to guide the model generation. Our first contribution shows that classical semi-supervised approaches, originating from a supervised classifier, are inappropriate and hardly detect new and unknown anomalies. We argue that semi-supervised anomaly detection needs to ground on the unsupervised learning paradigm and devise a novel algorithm that meets this requirement. Although being intrinsically non-convex, we further show that the optimization problem has a convex equivalent under relatively mild assumptions. Additionally, we propose an active learning strategy to automatically filter candidates for labeling. In an empirical study on network intrusion detection data, we observe that the proposed learning methodology requires much less labeled data than the state-of-the-art, while achieving higher detection accuracies.",
"THE detection of outliers has mainly been considered for single random samples, although some recent work deals also with standard linear models; see, for example, Anscombe (1960) and Kruskal (1960). Essentially similar problems arise in time series (Burman, 1965) but there seems no published work taking into account correlations between successive observations. In the past, the search for outliers in time series has been based on the assumption that the observations are independently and identically normally distributed. This assumption leads to analyses which will be called random sample procedures. Two types of outlier that may occur in a time series are considered in this paper. A Type I outlier corresponds to the situation in which a gross error of observation or recording error affects a single observation. A Type II outlier corresponds to the situation in which a single \"innovation\" is extreme. This will affect not only the particular observation but also subsequent observations. For the development of tests and the interpretation of outliers, it is necessary to distinguish among the types of outlier likely to be contained in the process. The present approach is based on four possible formulations of the problem: the outliers are all of Type I; the outliers are all of Type II; the outliers are all of the same type but whether they are of Type I or of Type II is not known; and the outliers are a mixture of the two types. Since more practical solutions than those given by likelihood ratio methods are often obtained from simplifications of likelihood ratio criteria, some simpler criteria are derived. These criteria are of the form &2a, where A is the estimated error in the observation tested and ^ is the estimated standard error of A. Throughout this paper, trend and seasonal components are assumed either negligible or to have been eliminated. The method adopted to remove these components might affect the results in some way."
]
} |
1607.02480 | 2507064736 | Much of the worlds data is streaming, time-series data, where anomalies give significant information in critical situations. Yet detecting anomalies in streaming data is a difficult task, requiring detectors to process data in real-time, and learn while simultaneously making predictions. We present a novel anomaly detection technique based on an on-line sequence memory algorithm called Hierarchical Temporal Memory (HTM). We show results from a live application that detects anomalies in financial metrics in real-time. We also test the algorithm on NAB, a published benchmark for real-time anomaly detection, where our algorithm achieves best-in-class results. | There are other algorithms capable of detecting temporal anomalies in complex scenarios. ARIMA is a general purpose technique for modeling temporal data with seasonality @cite_17 . It is effective at detecting anomalies in data with regular daily or weekly patterns. It is not capable of dynamically determining the period of seasonality, although extensions have been developed for doing so @cite_21 . A technique for applying ARIMA to multivariate data has also been studied @cite_16 . Bayesian change point detection methods are a natural approach to segmenting time series and can be used for online anomaly detection @cite_15 @cite_7 . Some additional techniques for general purpose anomaly detection on streaming data include @cite_19 @cite_23 . | {
"cite_N": [
"@cite_7",
"@cite_21",
"@cite_19",
"@cite_23",
"@cite_15",
"@cite_16",
"@cite_17"
],
"mid": [
"2046660258",
"2116512828",
"2952321802",
"",
"1483365869",
"2000145235",
"2053125529"
],
"abstract": [
"We consider the problem of efficient on-line anomaly detection in computer network traffic. The problem is approached statistically, as that of sequential (quickest) changepoint detection. A multi-cyclic setting of quickest change detection is a natural fit for this problem. We propose a novel score-based multi-cyclic detection algorithm. The algorithm is based on the so-called Shiryaev-Roberts procedure. This procedure is as easy to employ in practice and as computationally inexpensive as the popular Cumulative Sum chart and the Exponentially Weighted Moving Average scheme. The likelihood ratio based Shiryaev-Roberts procedure has appealing optimality properties, particularly it is exactly optimal in a multi-cyclic setting geared to detect a change occurring at a far time horizon. It is therefore expected that an intrusion detection algorithm based on the Shiryaev-Roberts procedure will perform better than other detection schemes. This is confirmed experimentally for real traces. We also discuss the possibility of complementing our anomaly detection algorithm with a spectral-signature intrusion detection system with false alarm filtering and true attack confirmation capability, so as to obtain a synergistic system.",
"Automatic forecasts of large numbers of univariate time series are often needed in business and other contexts. We describe two automatic forecasting algorithms that have been implemented in the forecast package for R. The first is based on innovations state space models that underly exponential smoothing methods. The second is a step-wise algorithm for forecasting with ARIMA models. The algorithms are applicable to both seasonal and non-seasonal data, and are compared and illustrated using four real time series. We also briefly describe some of the other functionality available in the forecast package.",
"Catalogs of periodic variable stars contain large numbers of periodic light-curves (photometric time series data from the astrophysics domain). Separating anomalous objects from well-known classes is an important step towards the discovery of new classes of astronomical objects. Most anomaly detection methods for time series data assume either a single continuous time series or a set of time series whose periods are aligned. Light-curve data precludes the use of these methods as the periods of any given pair of light-curves may be out of sync. One may use an existing anomaly detection method if, prior to similarity calculation, one performs the costly act of aligning two light-curves, an operation that scales poorly to massive data sets. This paper presents PCAD, an unsupervised anomaly detection method for large sets of unsynchronized periodic time-series data, that outputs a ranked list of both global and local anomalies. It calculates its anomaly score for each light-curve in relation to a set of centroids produced by a modified k-means clustering algorithm. Our method is able to scale to large data sets through the use of sampling. We validate our method on both light-curve data and other time series data sets. We demonstrate its effectiveness at finding known anomalies, and discuss the effect of sample size and number of centroids on our results. We compare our method to naive solutions and existing time series anomaly detection methods for unphased data, and show that PCAD's reported anomalies are comparable to or better than all other methods. Finally, astrophysicists on our team have verified that PCAD finds true anomalies that might be indicative of novel astrophysical phenomena.",
"",
"Changepoints are abrupt variations in the generative parameters of a data sequence. Online detection of changepoints is useful in modelling and prediction of time series in application areas such as finance, biometrics, and robotics. While frequentist methods have yielded online filtering and prediction techniques, most Bayesian papers have focused on the retrospective segmentation problem. Here we examine the case where the model parameters before and after the changepoint are independent and we derive an online algorithm for exact inference of the most recent changepoint. We compute the probability distribution of the length of the current run,'' or time since the last changepoint, using a simple message-passing algorithm. Our implementation is highly modular so that the algorithm may be applied to a variety of types of data. We illustrate this modularity by demonstrating the algorithm on three different real-world data sets.",
"This paper considers outliers in multivariate time series analysis. It generalizes four types of disturbances commonly used in the univariate time series analysis to the multivariate case, and investigates dynamic effects of a multivariate outlier on individual components if marginal models are used. An innovational outlier of a vector series can introduce a patch of outliers for the marginal component models. The paper also proposes an iterative procedure to detect and handle multiple outliers. By comparing and contrasting results of univariate and multivariate outlier detections, one can gain insights into the characteristics of an outlier. An outlier in a component series mayor may not have significant impacts on the other components. We use real examples to demonstrate the proposed analysis.",
"A diagnostic procedure for detecting additive and innovation outliers as well as level shifts in a regression model with ARIMA errors is introduced. The procedure is based on a robust estimate of the model parameters and on innovation residuals computed by means of robust filtering. A Monte Carlo study shows that, when there is a large proportion of outliers, this procedure is more powerful than the classical methods based on maximum likelihood type estimates and Kalman filtering. Copyright © 2001 by John Wiley & Sons, Ltd."
]
} |
1607.02480 | 2507064736 | Much of the worlds data is streaming, time-series data, where anomalies give significant information in critical situations. Yet detecting anomalies in streaming data is a difficult task, requiring detectors to process data in real-time, and learn while simultaneously making predictions. We present a novel anomaly detection technique based on an on-line sequence memory algorithm called Hierarchical Temporal Memory (HTM). We show results from a live application that detects anomalies in financial metrics in real-time. We also test the algorithm on NAB, a published benchmark for real-time anomaly detection, where our algorithm achieves best-in-class results. | There have been a number of model-based approaches applied to specific domains. These tend to be extremely specific to the domain they are modeling. Examples include anomaly detection in aircraft engine measurements @cite_1 , cloud datacenter temperatures @cite_4 , and ATM fraud detection @cite_2 . While these approaches may have success in a specific domain, they are not suitable for general purpose applications. | {
"cite_N": [
"@cite_1",
"@cite_4",
"@cite_2"
],
"mid": [
"2101623967",
"2012054765",
"2040328877"
],
"abstract": [
"This paper presents a model-based anomaly detection architecture designed for analyzing streaming transient aircraft engine measurement data. The technique calculates and monitors residuals between sensed engine outputs and model predicted outputs for anomaly detection purposes. Pivotal to the performance of this technique is the ability to construct a model that accurately reflects the nominal operating performance of the engine. The dynamic model applied in the architecture is a piecewise linear design comprising steady-state trim points and dynamic state space matrices. A simple curve-fitting technique for updating the model trim point information based on steady-state information extracted from available nominal engine measurement data is presented. Results from the application of the model-based approach for processing actual engine test data are shown. These include both nominal fault-free test case data and seeded fault test case data. The results indicate that the updates applied to improve the model trim point information also improve anomaly detection performance. Recommendations for follow-on enhancements to the technique are also presented and discussed.© 2014 ASME",
"The growing importance, large scale, and high server density of high-performance computing datacenters make them prone to strategic attacks, misconfigurations, and failures (cooling as well as computing infrastructure). Such unexpected events lead to thermal anomalies - hotspots, fugues, and coldspots - which significantly impact the total cost of operation of datacenters. A model-based thermal anomaly detection mechanism, which compares expected (obtained using heat generation and extraction models) and observed thermal maps (obtained using thermal cameras) of datacenters is proposed. In addition, a Thermal Anomaly-aware Resource Allocation (TARA) scheme is designed to create time-varying thermal fingerprints of the datacenter so to maximize the accuracy and minimize the latency of the aforementioned model-based detection. TARA significantly improves the performance of model-based anomaly detection compared to state-of-the-art resource allocation schemes.",
"Model-based anomaly detection in technical systems is an important application field of artificial intelligence. We consider discrete event systems, which is a system class to which a wide range of relevant technical systems belong and for which no comprehensive model-based anomaly detection approach exists so far. The original contributions of this paper are threefold: First, we identify the types of anomalies that occur in discrete event systems and we propose a tailored behavior model that captures all anomaly types, called probabilistic deterministic timed-transition automata (PDTTA). Second, we present a new algorithm to learn a PDTTA from sample observations of a system. Third, we describe an approach to detect anomalies based on a learned PDTTA. An empirical evaluation in a practical application, namely ATM fraud detection, shows promising results."
]
} |
1607.02046 | 2467838519 | This paper addresses the problem of 3D human pose estimation in the wild. A significant challenge is the lack of training data, i.e., 2D images of humans annotated with 3D poses. Such data is necessary to train state-of-the-art CNN architectures. Here, we propose a solution to generate a large set of photorealistic synthetic images of humans with 3D pose annotations. We introduce an image-based synthesis engine that artificially augments a dataset of real images with 2D human pose annotations using 3D Motion Capture (MoCap) data. Given a candidate 3D pose our algorithm selects for each joint an image whose 2D pose locally matches the projected 3D pose. The selected images are then combined to generate a new synthetic image by stitching local image patches in a kinematically constrained manner. The resulting images are used to train an end-to-end CNN for full-body 3D pose estimation. We cluster the training data into a large number of pose classes and tackle pose estimation as a K-way classification problem. Such an approach is viable only with large training sets such as ours. Our method outperforms the state of the art in terms of 3D pose estimation in controlled environments (Human3.6M) and shows promising results for in-the-wild images (LSP). This demonstrates that CNNs trained on artificial images generalize well to real images. | 3D human pose estimation in monocular images. Recent approaches employ CNNs for 3D pose estimation in monocular images @cite_7 or in videos @cite_10 . Due to the lack of large scale training data, they are usually trained (and tested) on 3D MoCap data in constrained environments @cite_7 . Pose understanding in natural images is usually limited to 2D pose estimation @cite_36 @cite_20 @cite_42 . Recent work also tackles 3D pose understanding from 2D poses @cite_25 @cite_38 . Some approaches use as input the 2D joints automatically provided by a 2D pose detector @cite_21 @cite_1 , while others jointly solve the 2D and 3D pose estimation @cite_27 @cite_12 . Most similar to ours is the approach of @cite_13 who use a dual-source approach that combines 2D pose estimation with 3D pose retrieval. Our method uses the same two training sources, i.e., images with annotated 2D pose and 3D MoCap data. However, we combine both sources off-line to generate a large training set that is used to train an end-to-end CNN 3D pose classifier. This is shown to improve over @cite_13 , which can be explained by the fact that training is performed in an end-to-end fashion. | {
"cite_N": [
"@cite_13",
"@cite_38",
"@cite_7",
"@cite_36",
"@cite_42",
"@cite_21",
"@cite_1",
"@cite_27",
"@cite_12",
"@cite_10",
"@cite_25",
"@cite_20"
],
"mid": [
"2963013806",
"2178077220",
"2949812103",
"2155394491",
"2113325037",
"2088196373",
"2039262381",
"2111446867",
"",
"2285449971",
"1943191679",
"2952422028"
],
"abstract": [
"One major challenge for 3D pose estimation from a single RGB image is the acquisition of sufficient training data. In particular, collecting large amounts of training data that contain unconstrained images and are annotated with accurate 3D poses is infeasible. We therefore propose to use two independent training sources. The first source consists of images with annotated 2D poses and the second source consists of accurate 3D motion capture data. To integrate both sources, we propose a dual-source approach that combines 2D pose estimation with efficient and robust 3D pose retrieval. In our experiments, we show that our approach achieves state-of-the-art results and is even competitive when the skeleton structure of the two sources differ substantially.",
"Reconstructing 3D human poses from a single 2D image is an ill-posed problem without considering the human body model. Explicitly enforcing physiological constraints is known to be non-convex and usually leads to difficulty in finding an optimal solution. An attractive alternative is to learn a prior model of the human body from a set of human pose data. In this paper, we develop a new approach, namely pose locality constrained representation (PLCR), to model the 3D human body and use it to improve 3D human pose reconstruction. In this approach, the human pose space is first hierarchically divided into lower-dimensional pose subspaces by subspace clustering. After that, a block-structural pose dictionary is constructed by concatenating the basis poses from all the pose subspaces. Finally, PLCR utilizes the block-structural pose dictionary to explicitly encourage pose locality in human-body modeling – nonzero coefficients are only assigned to the basis poses from a small number of pose subspaces that are close to each other in the pose-subspace hierarchy. We combine PLCR into the matching-pursuit based 3D human-pose reconstruction algorithm and show that the proposed PLCR-based algorithm outperforms the state-of-the-art algorithm that uses the standard sparse representation and physiological regularity in reconstructing a variety of human poses from both synthetic data and real images.",
"This paper focuses on structured-output learning using deep neural networks for 3D human pose estimation from monocular images. Our network takes an image and 3D pose as inputs and outputs a score value, which is high when the image-pose pair matches and low otherwise. The network structure consists of a convolutional neural network for image feature extraction, followed by two sub-networks for transforming the image features and pose into a joint embedding. The score function is then the dot-product between the image and pose embeddings. The image-pose embedding and score function are jointly trained using a maximum-margin cost function. Our proposed framework can be interpreted as a special form of structured support vector machines where the joint feature space is discriminatively learned using deep neural networks. We test our framework on the Human3.6m dataset and obtain state-of-the-art results compared to other recent methods. Finally, we present visualizations of the image-pose embedding space, demonstrating the network has learned a high-level embedding of body-orientation and pose-configuration.",
"We present a method for estimating articulated human pose from a single static image based on a graphical model with novel pairwise relations that make adaptive use of local image measurements. More precisely, we specify a graphical model for human pose which exploits the fact the local image measurements can be used both to detect parts (or joints) and also to predict the spatial relationships between them (Image Dependent Pairwise Relations). These spatial relationships are represented by a mixture model. We use Deep Convolutional Neural Networks (DCNNs) to learn conditional probabilities for the presence of parts and their spatial relationships within image patches. Hence our model combines the representational flexibility of graphical models with the efficiency and statistical power of DCNNs. Our method significantly outperforms the state of the art methods on the LSP and FLIC datasets and also performs very well on the Buffy dataset without any training.",
"We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regres- sors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formula- tion which capitalizes on recent advances in Deep Learn- ing. We present a detailed empirical analysis with state-of- art or better performance on four academic benchmarks of diverse real-world images.",
"Markerless 3D human pose detection from a single image is a severely underconstrained problem because different 3D poses can have similar image projections. In order to handle this ambiguity, current approaches rely on prior shape models that can only be correctly adjusted if 2D image features are accurately detected. Unfortunately, although current 2D part detector algorithms have shown promising results, they are not yet accurate enough to guarantee a complete disambiguation of the 3D inferred shape. In this paper, we introduce a novel approach for estimating 3D human pose even when observations are noisy. We propose a stochastic sampling strategy to propagate the noise from the image plane to the shape space. This provides a set of ambiguous 3D shapes, which are virtually undistinguishable from their image projections. Disambiguation is then achieved by imposing kinematic constraints that guarantee the resulting pose resembles a 3D human shape. We validate the method on a variety of situations in which state-of-the-art 2D detectors yield either inaccurate estimations or partly miss some of the body parts.",
"Human pose estimation is a key step to action recognition. We propose a method of estimating 3D human poses from a single image, which works in conjunction with an existing 2D pose joint detector. 3D pose estimation is challenging because multiple 3D poses may correspond to the same 2D pose after projection due to the lack of depth information. Moreover, current 2D pose estimators are usually inaccurate which may cause errors in the 3D estimation. We address the challenges in three ways: (i) We represent a 3D pose as a linear combination of a sparse set of bases learned from 3D human skeletons. (ii) We enforce limb length constraints to eliminate anthropomorphically implausible skeletons. (iii) We estimate a 3D pose by minimizing the 1-norm error between the projection of the 3D pose and the corresponding 2D detection. The 1-norm loss term is robust to inaccurate 2D joint estimations. We use the alternating direction method (ADM) to solve the optimization problem efficiently. Our approach outperforms the state-of-the-arts on three benchmark datasets.",
"We introduce a novel approach to automatically recover 3D human pose from a single image. Most previous work follows a pipelined approach: initially, a set of 2D features such as edges, joints or silhouettes are detected in the image, and then these observations are used to infer the 3D pose. Solving these two problems separately may lead to erroneous 3D poses when the feature detector has performed poorly. In this paper, we address this issue by jointly solving both the 2D detection and the 3D inference problems. For this purpose, we propose a Bayesian framework that integrates a generative model based on latent variables and discriminative 2D part detectors based on HOGs, and perform inference using evolutionary algorithms. Real experimentation demonstrates competitive results, and the ability of our methodology to provide accurate 2D and 3D pose estimations even when the 2D detectors are inaccurate.",
"",
"This paper addresses the challenge of 3D full-body human pose estimation from a monocular image sequence. Here, two cases are considered: (i) the image locations of the human joints are provided and (ii) the image locations of joints are unknown. In the former case, a novel approach is introduced that integrates a sparsity-driven 3D geometric prior and temporal smoothness. In the latter case, the former case is extended by treating the image locations of the joints as latent variables. A deep fully convolutional network is trained to predict the uncertainty maps of the 2D joint locations. The 3D pose estimates are realized via an Expectation-Maximization algorithm over the entire sequence, where it is shown that the 2D joint location uncertainties can be conveniently marginalized out during inference. Empirical evaluation on the Human3.6M dataset shows that the proposed approaches achieve greater 3D pose estimation accuracy over state-of-the-art baselines. Further, the proposed approach outperforms a publicly available 2D pose estimation baseline on the challenging PennAction dataset.",
"Estimating 3D human pose from 2D joint locations is central to the analysis of people in images and video. To address the fact that the problem is inherently ill posed, many methods impose a prior over human poses. Unfortunately these priors admit invalid poses because they do not model how joint-limits vary with pose. Here we make two key contributions. First, we collect a motion capture dataset that explores a wide range of human poses. From this we learn a pose-dependent model of joint limits that forms our prior. Both dataset and prior are available for research purposes. Second, we define a general parametrization of body pose and a new, multi-stage, method to estimate 3D pose from 2D joint locations using an over-complete dictionary of poses. Our method shows good generalization while avoiding impossible poses. We quantitatively compare our method with recent work and show state-of-the-art results on 2D to 3D pose estimation using the CMU mocap dataset. We also show superior results using manual annotations on real images and automatic detections on the Leeds sports pose dataset.",
"This paper proposes a new hybrid architecture that consists of a deep Convolutional Network and a Markov Random Field. We show how this architecture is successfully applied to the challenging problem of articulated human pose estimation in monocular images. The architecture can exploit structural domain constraints such as geometric relationships between body joint locations. We show that joint training of these two model paradigms improves performance and allows us to significantly outperform existing state-of-the-art techniques."
]
} |
1607.02046 | 2467838519 | This paper addresses the problem of 3D human pose estimation in the wild. A significant challenge is the lack of training data, i.e., 2D images of humans annotated with 3D poses. Such data is necessary to train state-of-the-art CNN architectures. Here, we propose a solution to generate a large set of photorealistic synthetic images of humans with 3D pose annotations. We introduce an image-based synthesis engine that artificially augments a dataset of real images with 2D human pose annotations using 3D Motion Capture (MoCap) data. Given a candidate 3D pose our algorithm selects for each joint an image whose 2D pose locally matches the projected 3D pose. The selected images are then combined to generate a new synthetic image by stitching local image patches in a kinematically constrained manner. The resulting images are used to train an end-to-end CNN for full-body 3D pose estimation. We cluster the training data into a large number of pose classes and tackle pose estimation as a K-way classification problem. Such an approach is viable only with large training sets such as ours. Our method outperforms the state of the art in terms of 3D pose estimation in controlled environments (Human3.6M) and shows promising results for in-the-wild images (LSP). This demonstrates that CNNs trained on artificial images generalize well to real images. | Synthetic pose data. A number of works have considered the use of synthetic data for human pose estimation. Synthetic data have been used for upper body @cite_18 , full-body silhouettes @cite_33 , hand-object interactions @cite_15 , full-body pose from depth @cite_22 or egocentric RGB-D scenes @cite_4 . Recently, Zuffi and Black @cite_2 used a 3D mesh-model to sample synthetic exemplars and fit 3D scans. @cite_3 , a scene-specific pedestrian detectors was learned without real data while @cite_16 synthesized virtual samples with a generative model to enhance the classification performance of a discriminative model. @cite_19 , pictures of 2D characters were animated by fitting and deforming a 3D mesh model. Later, @cite_6 augmented labelled training images with small perturbations in a similar way. These methods require a perfect segmentation of the humans in the images. Park and Ramanan @cite_44 synthesized hypothetical poses for tracking purposes by applying geometric transformations to the first frame of a video sequence. We also use image-based synthesis to generate images but our rendering engine combines image regions from several images to create images with associated 3D poses. | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_22",
"@cite_33",
"@cite_3",
"@cite_6",
"@cite_44",
"@cite_19",
"@cite_2",
"@cite_15",
"@cite_16"
],
"mid": [
"2152926413",
"1906662973",
"2060280062",
"2123503110",
"",
"2073246097",
"1950149599",
"2010073476",
"1901654600",
"2165272793",
""
],
"abstract": [
"Example-based methods are effective for parameter estimation problems when the underlying system is simple or the dimensionality of the input is low. For complex and high-dimensional problems such as pose estimation, the number of required examples and the computational complexity rapidly become prohibitively high. We introduce a new algorithm that learns a set of hashing functions that efficiently index examples relevant to a particular estimation task. Our algorithm extends locality-sensitive hashing, a recently developed method to find approximate neighbors in time sublinear in the number of examples. This method depends critically on the choice of hash functions that are optimally relevant to a particular estimation problem. Experiments demonstrate that the resulting algorithm, which we call parameter-sensitive hashing, can rapidly and accurately estimate the articulated pose of human figures from a large database of example images.",
"We tackle the problem of estimating the 3D pose of an individual's upper limbs (arms+hands) from a chest mounted depth-camera. Importantly, we consider pose estimation during everyday interactions with objects. Past work shows that strong pose+viewpoint priors and depth-based features are crucial for robust performance. In egocentric views, hands and arms are observable within a well defined volume in front of the camera. We call this volume an egocentric workspace. A notable property is that hand appearance correlates with workspace location. To exploit this correlation, we classify arm+hand configurations in a global egocentric coordinate frame, rather than a local scanning window. This greatly simplify the architecture and improves performance. We propose an efficient pipeline which 1) generates synthetic workspace exemplars for training using a virtual chest-mounted camera whose intrinsic parameters match our physical camera, 2) computes perspective-aware depth features on this entire volume and 3) recognizes discrete arm+hand pose classes through a sparse multi-class SVM. We achieve state-of-the-art hand pose recognition performance from egocentric RGB-D images in real-time.",
"We propose a new method to quickly and accurately predict human pose---the 3D positions of body joints---from a single depth image, without depending on information from preceding frames. Our approach is strongly rooted in current object recognition strategies. By designing an intermediate representation in terms of body parts, the difficult pose estimation problem is transformed into a simpler per-pixel classification problem, for which efficient machine learning techniques exist. By using computer graphics to synthesize a very large dataset of training image pairs, one can train a classifier that estimates body part labels from test images invariant to pose, body shape, clothing, and other irrelevances. Finally, we generate confidence-scored 3D proposals of several body joints by reprojecting the classification result and finding local modes. The system runs in under 5ms on the Xbox 360. Our evaluation shows high accuracy on both synthetic and real test sets, and investigates the effect of several training parameters. We achieve state-of-the-art accuracy in our comparison with related work and demonstrate improved generalization over exact whole-skeleton nearest neighbor matching.",
"We describe a learning-based method for recovering 3D human body pose from single images and monocular image sequences. Our approach requires neither an explicit body model nor prior labeling of body parts in the image. Instead, it recovers pose by direct nonlinear regression against shape descriptor vectors extracted automatically from image silhouettes. For robustness against local silhouette segmentation errors, silhouette shape is encoded by histogram-of-shape-contexts descriptors. We evaluate several different regression methods: ridge regression, relevance vector machine (RVM) regression, and support vector machine (SVM) regression over both linear and kernel bases. The RVMs provide much sparser regressors without compromising performance, and kernel bases give a small but worthwhile improvement in performance. The loss of depth and limb labeling information often makes the recovery of 3D pose from single silhouettes ambiguous. To handle this, the method is embedded in a novel regressive tracking framework, using dynamics from the previous state estimate together with a learned regression value to disambiguate the pose. We show that the resulting system tracks long sequences stably. For realism and good generalization over a wide range of viewpoints, we train the regressors on images resynthesized from real human motion capture data. The method is demonstrated for several representations of full body pose, both quantitatively on independent but similar test data and qualitatively on real image sequences. Mean angular errors of 4-6 spl deg are obtained for a variety of walking motions.",
"",
"State-of-the-art methods for human detection and pose estimation require many training samples for best performance. While large, manually collected datasets exist, the captured variations w.r.t. appearance, shape and pose are often uncontrolled thus limiting the overall performance. In order to overcome this limitation we propose a new technique to extend an existing training set that allows to explicitly control pose and shape variations. For this we build on recent advances in computer graphics to generate samples with realistic appearance and background while modifying body shape and pose. We validate the effectiveness of our approach on the task of articulated human detection and articulated pose estimation. We report close to state of the art results on the popular Image Parsing [25] human pose estimation benchmark and demonstrate superior performance for articulated human detection. In addition we define a new challenge of combined articulated human detection and pose estimation in real-world scenes.",
"We address the task of articulated pose estimation from video sequences. We consider an interactive setting where the initial pose is annotated in the first frame. Our system synthesizes a large number of hypothetical scenes with different poses and camera positions by applying geometric deformations to the first frame. We use these synthetic images to generate a custom labeled training set for the video in question. This training data is then used to learn a regressor (for future frames) that predicts joint locations from image data. Notably, our training set is so accurate that nearest-neighbor (NN) matching on low-resolution pixel features works well. As such, we name our underlying representation “tiny synthetic videos”. We present quantitative results the Friends benchmark dataset that suggests our simple approach matches or exceed state-of-the-art.",
"This article presents a new method to animate photos of 2D characters using 3D motion capture data. Given a single image of a person or essentially human-like subject, our method transfers the motion of a 3D skeleton onto the subject's 2D shape in image space, generating the impression of a realistic movement. We present robust solutions to reconstruct a projective camera model and a 3D model pose which matches best to the given 2D image. Depending on the reconstructed view, a 2D shape template is selected which enables the proper handling of occlusions. After fitting the template to the character in the input image, it is deformed as-rigid-as-possible by taking the projected 3D motion data into account. Unlike previous work, our method thereby correctly handles projective shape distortion. It works for images from arbitrary views and requires only a small amount of user interaction. We present animations of a diverse set of human (and nonhuman) characters with different types of motions, such as walking, jumping, or dancing.",
"We propose a new 3D model of the human body that is both realistic and part-based. The body is represented by a graphical model in which nodes of the graph correspond to body parts that can independently translate and rotate in 3D and deform to represent different body shapes and to capture pose-dependent shape variations. Pairwise potentials define a “stitching cost” for pulling the limbs apart, giving rise to the stitched puppet (SP) model. Unlike existing realistic 3D body models, the distributed representation facilitates inference by allowing the model to more effectively explore the space of poses, much like existing 2D pictorial structures models. We infer pose and body shape using a form of particle-based max-product belief propagation. This gives SP the realism of recent 3D body models with the computational advantages of part-based models. We apply SP to two challenging problems involving estimating human shape and pose from 3D data. The first is the FAUST mesh alignment challenge, where ours is the first method to successfully align all 3D meshes with no pose prior. The second involves estimating pose and shape from crude visual hull representations of complex body movements.",
"This paper presents a method for vision based estimation of the pose of human hands in interaction with objects. Despite the fact that most robotics applications of human hand tracking involve grasping and manipulation of objects, the majority of methods in the literature assume a free hand, isolated from the surrounding environment. Our hand tracking method is non-parametric, performing a nearest neighbor search in a large database (100000 entries) of hand poses with and without grasped objects. The system operates in real time, it is robust to self occlusions, object occlusions and segmentation errors, and provides full hand pose reconstruction from markerless video. Temporal consistency in hand pose is taken into account, without explicitly tracking the hand in the high dimensional pose space.",
""
]
} |
1607.02061 | 2516768980 | Several studies on sentence processing suggest that the mental lexicon keeps track of the mutual expectations between words. Current DSMs, however, represent context words as separate features, thereby loosing important information for word expectations, such as word interrelations. In this paper, we present a DSM that addresses this issue by defining verb contexts as joint syntactic dependencies. We test our representation in a verb similarity task on two datasets, showing that joint contexts achieve performances comparable to single dependencies or even better. Moreover, they are able to overcome the data sparsity problem of joint feature spaces, in spite of the limited size of our training corpus. | A number of studies in sentence processing suggests that verbs activate expectations on their typical argument nouns and vice versa @cite_20 @cite_18 and nouns do the same with other nouns occurring as co-arguments in the same events @cite_17 @cite_7 . Experimental subjects seem to exploit a rich event knowledge to activate or inhibit dynamically the representations of the potential arguments. This phenomenon, generally referred to as @cite_20 , supports the idea of a mental lexicon arranged as a web of mutual expectations. | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_20",
"@cite_17"
],
"mid": [
"2094931100",
"2009342863",
"",
"2161895009"
],
"abstract": [
"We explore the implications of an event-based expectancy generation approach to language understanding, suggesting that one useful strategy employed by comprehenders is to generate expectations about upcoming words. We focus on two questions: (1) What role is played by elements other than verbs in generating expectancies? (2) What connection exists between expectancy generation and event-based knowledge? Because verbs follow their arguments in many constructions (particularly in verb-final languages), deferring expectations until the verb seems inefficient. Both human data and computational modeling suggest that other sentential elements may also play a role in predictive processing and that these constraints often reflect knowledge regarding typical events. We investigated these predictions, using both short and long stimulus onset asynchrony priming. Robust priming obtained when verbs were named aloud following typical agents, patients, instruments, and locations, suggesting that event memory is organized so that nouns denoting entities and objects activate the classes of events in which they typically play a role. These computations are assumed to be an important component of expectancy generation in sentence processing.",
"This research tests whether comprehenders use their knowledge of typical events in real time to process verbal arguments. In self-paced reading and event-related brain potential (ERP) experiments, we used materials in which the likelihood of a specific patient noun (brakes or spelling) depended on the combination of an agent and verb (mechanic checked vs. journalist checked). Reading times were shorter at the word directly following the patient for the congruent than the incongruent items. Differential N400s were found earlier, immediately at the patient. Norming studies ruled out any account of these results based on direct relations between the agent and patient. Thus, comprehenders dynamically combine information about real-world events based on intrasentential agents and verbs, and this combination then rapidly influences online sentence interpretation.",
"",
"An increasing number of results in sentence and discourse processing demonstrate that comprehension relies on rich pragmatic knowledge about real-world events, and that incoming words incrementally activate such knowledge. If so, then even outside of any larger context, nouns should activate knowledge of the generalized events that they denote or typically play a role in. We used short stimulus onset asynchrony priming to demonstrate that (1) event nouns prime people (sale–shopper) and objects (trip–luggage) commonly found at those events; (2) location nouns prime people animals (hospital–doctor) and objects (barn–hay) commonly found at those locations; and (3) instrument nouns prime things on which those instruments are commonly used (key–door), but not the types of people who tend to use them (hose–gardener). The priming effects are not due to normative word association. On our account, facilitation results from event knowledge relating primes and targets. This has much in common with computational models like LSA or BEAGLE in which one word primes another if they frequently occur in similar contexts. LSA predicts priming for all six experiments, whereas BEAGLE correctly predicted that priming should not occur for the instrument–people relation but should occur for the other five. We conclude that event-based relations are encoded in semantic memory and computed as part of word meaning, and have a strong influence on language comprehension."
]
} |
1607.02061 | 2516768980 | Several studies on sentence processing suggest that the mental lexicon keeps track of the mutual expectations between words. Current DSMs, however, represent context words as separate features, thereby loosing important information for word expectations, such as word interrelations. In this paper, we present a DSM that addresses this issue by defining verb contexts as joint syntactic dependencies. We test our representation in a verb similarity task on two datasets, showing that joint contexts achieve performances comparable to single dependencies or even better. Moreover, they are able to overcome the data sparsity problem of joint feature spaces, in spite of the limited size of our training corpus. | Some past works in computational linguistics @cite_21 @cite_6 @cite_12 @cite_14 modeled thematic fit estimations by means of dependency-based or of thematic roles-based DSMs. However, these semantic spaces are built similarly to traditional DSMs as they split verb arguments into separate vector dimensions. By using syntactic-semantic links, they encode the relation between an event and each of its participants, but they do not encode directly the relation between participants co-occurring in the same event. | {
"cite_N": [
"@cite_14",
"@cite_21",
"@cite_12",
"@cite_6"
],
"mid": [
"2950577311",
"2128870637",
"1983578042",
"2120660105"
],
"abstract": [
"We propose two novel model architectures for computing continuous vector representations of words from very large data sets. The quality of these representations is measured in a word similarity task, and the results are compared to the previously best performing techniques based on different types of neural networks. We observe large improvements in accuracy at much lower computational cost, i.e. it takes less than a day to learn high quality word vectors from a 1.6 billion words data set. Furthermore, we show that these vectors provide state-of-the-art performance on our test set for measuring syntactic and semantic word similarities.",
"Research into corpus-based semantics has focused on the development of ad hoc models that treat single tasks, or sets of closely related tasks, as unrelated challenges to be tackled by extracting different kinds of distributional information from the corpus. As an alternative to this \"one task, one model\" approach, the Distributional Memory framework extracts distributional information once and for all from the corpus, in the form of a set of weighted word-link-word tuples arranged into a third-order tensor. Different matrices are then generated from the tensor, and their rows and columns constitute natural spaces to deal with different semantic problems. In this way, the same distributional information can be shared across tasks such as modeling word similarity judgments, discovering synonyms, concept categorization, predicting selectional preferences of verbs, solving analogy problems, classifying relations between word pairs, harvesting qualia structures with patterns or example pairs, predicting the typical properties of concepts, and classifying verbs into alternation classes. Extensive empirical testing in all these domains shows that a Distributional Memory implementation performs competitively against task-specific algorithms recently reported in the literature for the same tasks, and against our implementations of several state-of-the-art methods. The Distributional Memory approach is thus shown to be tenable despite the constraints imposed by its multi-purpose nature.",
"How do people know as much as they do with as little information as they get? The problem takes many forms; learning vocabulary from text is an especially dramatic and convenient case for research. A new general theory of acquired similarity and knowledge representation, latent semantic analysis (LSA), is presented and used to successfully simulate such learning and several other psycholinguistic phenomena. By inducing global knowledge indirectly from local co-occurrence data in a large body of representative text, LSA acquired knowledge about the full vocabulary of English at a comparable rate to schoolchildren. LSA uses no prior linguistic or perceptual similarity knowledge; it is based solely on a general mathematical learning method that achieves powerful inductive effects by extracting the right number of dimensions (e.g., 300) to represent objects and contexts. Relations to other theories, phenomena, and problems are sketched.",
"The aim of this paper is to present a computational model of the dynamic composition and update of verb argument expectations using Distributional Memory, a state-of-the-art framework for distributional semantics. The experimental results conducted on psycholinguistic data sets show that the model is able to successfully predict the changes on the patient argument thematic fit produced by different types of verb agents."
]
} |
1607.02061 | 2516768980 | Several studies on sentence processing suggest that the mental lexicon keeps track of the mutual expectations between words. Current DSMs, however, represent context words as separate features, thereby loosing important information for word expectations, such as word interrelations. In this paper, we present a DSM that addresses this issue by defining verb contexts as joint syntactic dependencies. We test our representation in a verb similarity task on two datasets, showing that joint contexts achieve performances comparable to single dependencies or even better. Moreover, they are able to overcome the data sparsity problem of joint feature spaces, in spite of the limited size of our training corpus. | adopted an analogous approach, relying on a huge learning corpus (1.6 Teraword) to build composite-feature vectors. Their model outperformed a traditional DSM on the similarity subset of the WordSim-353 test set @cite_5 . | {
"cite_N": [
"@cite_5"
],
"mid": [
"2067438047"
],
"abstract": [
"Keyword-based search engines are in widespread use today as a popular means for Web-based information retrieval. Although such systems seem deceptively simple, a considerable amount of skill is required in order to satisfy non-trivial information needs. This paper presents a new conceptual paradigm for performing search in context, that largely automates the search process, providing even non-professional users with highly relevant results. This paradigm is implemented in practice in the IntelliZap system, where search is initiated from a text query marked by the user in a document she views, and is guided by the text surrounding the marked query in that document (“the context”). The context-driven information retrieval process involves semantic keyword extraction and clustering to automatically generate new, augmented queries. The latter are submitted to a host of general and domain-specific search engines. Search results are then semantically reranked, using context. Experimental results testify that using context to guide search, effectively offers even inexperienced users an advanced search tool on the Web."
]
} |
1607.02061 | 2516768980 | Several studies on sentence processing suggest that the mental lexicon keeps track of the mutual expectations between words. Current DSMs, however, represent context words as separate features, thereby loosing important information for word expectations, such as word interrelations. In this paper, we present a DSM that addresses this issue by defining verb contexts as joint syntactic dependencies. We test our representation in a verb similarity task on two datasets, showing that joint contexts achieve performances comparable to single dependencies or even better. Moreover, they are able to overcome the data sparsity problem of joint feature spaces, in spite of the limited size of our training corpus. | introduced a probabilistic similarity scheme for modeling the so-called joint context. By making use of the Kneser-Ney language model @cite_16 and of a probabilistic distributional measure, they were able to overcome data sparsity, outperforming a wide variety of DSMs on two similarity tasks, evaluated on VerbSim @cite_11 and on a set of 1,000 verbs extracted from WordNet @cite_19 . On the basis of their results, the authors claimed that composite-feature models are particularly advantageous for measuring verb similarity. | {
"cite_N": [
"@cite_19",
"@cite_16",
"@cite_11"
],
"mid": [
"",
"1934041838",
"2170682101"
],
"abstract": [
"",
"In stochastic language modeling, backing-off is a widely used method to cope with the sparse data problem. In case of unseen events this method backs off to a less specific distribution. In this paper we propose to use distributions which are especially optimized for the task of backing-off. Two different theoretical derivations lead to distributions which are quite different from the probability distributions that are usually used for backing-off. Experiments show an improvement of about 10 in terms of perplexity and 5 in terms of word error rate.",
"This paper presents and compares WordNet-based and distributional similarity approaches. The strengths and weaknesses of each approach regarding similarity and relatedness tasks are discussed, and a combination is presented. Each of our methods independently provide the best results in their class on the RG and WordSim353 datasets, and a supervised combination of them yields the best published results on all datasets. Finally, we pioneer cross-lingual similarity, showing that our methods are easily adapted for a cross-lingual task with minor losses."
]
} |
1607.02078 | 2949844559 | Motivation: Histone modifications are among the most important factors that control gene regulation. Computational methods that predict gene expression from histone modification signals are highly desirable for understanding their combinatorial effects in gene regulation. This knowledge can help in developing 'epigenetic drugs' for diseases like cancer. Previous studies for quantifying the relationship between histone modifications and gene expression levels either failed to capture combinatorial effects or relied on multiple methods that separate predictions and combinatorial analysis. This paper develops a unified discriminative framework using a deep convolutional neural network to classify gene expression using histone modification data as input. Our system, called DeepChrome, allows automatic extraction of complex interactions among important features. To simultaneously visualize the combinatorial interactions among histone modifications, we propose a novel optimization-based technique that generates feature pattern maps from the learnt deep model. This provides an intuitive description of underlying epigenetic mechanisms that regulate genes. Results: We show that DeepChrome outperforms state-of-the-art models like Support Vector Machines and Random Forests for gene expression classification task on 56 different cell-types from REMC database. The output of our visualization technique not only validates the previous observations but also allows novel insights about combinatorial interactions among histone modification marks, some of which have recently been observed by experimental studies. | The combinatorial effect of histone modifications in regulating gene expression has been studied throughout literature ( @cite_2 ). To better understand this relationship, scientists have generated experimental datasets quantifying gene expression and histone modification signals across different cell-types. These datasets have been made available through large-scale repositories, the latest being the Roadmap Epigenome Project (REMC) ( @cite_27 ). * -12pt | {
"cite_N": [
"@cite_27",
"@cite_2"
],
"mid": [
"2076154138",
"2099035207"
],
"abstract": [
"The reference human genome sequence set the stage for studies of genetic variation and its association with human disease, but epigenomic studies lack a similar reference. To address this need, the NIH Roadmap Epigenomics Consortium generated the largest collection so far of human epigenomes for primary cells and tissues. Here we describe the integrative analysis of 111 reference human epigenomes generated as part of the programme, profiled for histone modification patterns, DNA accessibility, DNA methylation and RNA expression. We establish global maps of regulatory elements, define regulatory modules of coordinated activity, and their likely activators and repressors. We show that disease- and trait-associated genetic variants are enriched in tissue-specific epigenomic marks, revealing biologically relevant cell types for diverse human traits, and providing a resource for interpreting the molecular basis of human disease. Our results demonstrate the central role of epigenomic information for understanding gene regulation, cellular differentiation and human disease.",
"In the nuclei of eukaryotic cells, DNA wraps around the octamer of histone proteins to form the nucleosome, in a structure like ‘beads on a string’, which makes up the basic unit of chromatin. Chromatin further folds into higherlevel structures, loosely or tightly, which helps to determine the accessibility of the DNA. For instance, actively transcribed regions tend to be in looser chromatin structures so that transcription factors and RNA polymerases can access the genes. Chromatin structure can be altered by various post-translational modifications of the N-terminal tail residues of histone proteins. For example, acetylation of a lysine residue can neutralize its positive charge and weaken the binding between the histone and the negatively charged DNA, which exposes the DNA to regulatory proteins. Methylation is another common type of histone modification; for example, the lysine at the fourth position of the H3 histone can be mono-, dior tri-methylated (denoted as H3K4me1, H3K4me2 and H3K4me3, respectively). By examining histone modification patterns at highly conserved noncoding regions in mouse embryonic stem cells, found ‘bivalent domains’ of histone modifications (i.e., harboring both the repressive mark H3K27me3 and the active mark H3K4me3) near genes with poised transcription [1]. When embryonic stem cells differentiate into more specialized cells (e.g., neural precursor cells), a subset of the bivalent domains are resolved (i.e., H3K27me3 becomes weaker, while H3K4me3 becomes stronger, and these loci coincide with genes that are actively transcribed in neural precursor cells). Thus, combinations of histone marks are indicative of transcriptional states. mapped 20 histone methylations of lysine and arginine residues in human CD4 T cells using chromatin immunoprecipitation followed by sequencing (ChIP-seq) [2]. They found that monomethylated H3K27, H3K9, H4K20, H3K79 and H2BK5 were linked to gene activation, while trimethylated H3K27, H3K9 and H3K79 were linked to gene repression. In a later study, the group profiled 39 additional histone modifications in human CD4 T cells [3]. They identified more than 3000 genes that were highly expressed in these cells and the promoters of these genes showed high levels of 17 histone modifications (called a histone modification module). Other studies also investigated the correlation between individual histone marks and gene expression, although not in a quantitative way [4,5]."
]
} |
1607.02078 | 2949844559 | Motivation: Histone modifications are among the most important factors that control gene regulation. Computational methods that predict gene expression from histone modification signals are highly desirable for understanding their combinatorial effects in gene regulation. This knowledge can help in developing 'epigenetic drugs' for diseases like cancer. Previous studies for quantifying the relationship between histone modifications and gene expression levels either failed to capture combinatorial effects or relied on multiple methods that separate predictions and combinatorial analysis. This paper develops a unified discriminative framework using a deep convolutional neural network to classify gene expression using histone modification data as input. Our system, called DeepChrome, allows automatic extraction of complex interactions among important features. To simultaneously visualize the combinatorial interactions among histone modifications, we propose a novel optimization-based technique that generates feature pattern maps from the learnt deep model. This provides an intuitive description of underlying epigenetic mechanisms that regulate genes. Results: We show that DeepChrome outperforms state-of-the-art models like Support Vector Machines and Random Forests for gene expression classification task on 56 different cell-types from REMC database. The output of our visualization technique not only validates the previous observations but also allows novel insights about combinatorial interactions among histone modification marks, some of which have recently been observed by experimental studies. | @cite_20 applied Support Vector Machine (SVM) models on worm datasets ( @cite_23 ) and reformulated the task as gene expression classification and prediction. The authors divided regions flanking transcription start site (TSS) and transcription termination site (TTS) into 100 base-pair (bp) bins and used the histone modification signal in each bin as a feature for the SVM. To incorporate information from all positions or bins, they trained different models for different bins that resulted in 160 models for 160 bins. They validated the existence of the quantitive relationship between histone modifications and gene expression by such bin-specific modeling. Furthermore, using a separate linear regression model, the paper inferred pair-wise interactions between different histone modifications using binary combinatorial terms. Since it is infeasible to consider all possible higher order interaction terms through polynomial regression, Bayesian networks were then used for modeling such relationships. However, Bayesian networks do not take into consideration local neighboring bin information and their highly connected output network is difficult to interpret. | {
"cite_N": [
"@cite_23",
"@cite_20"
],
"mid": [
"2150904790",
"2164116506"
],
"abstract": [
"Despite the successes of genomics, little is known about how genetic information produces complex organisms. A look at the crucial functional elements of fly and worm genomes could change that.",
"We develop a statistical framework to study the relationship between chromatin features and gene expression. This can be used to predict gene expression of protein coding genes, as well as microRNAs. We demonstrate the prediction in a variety of contexts, focusing particularly on the modENCODE worm datasets. Moreover, our framework reveals the positional contribution around genes (upstream or downstream) of distinct chromatin features to the overall prediction of expression levels."
]
} |
1607.02078 | 2949844559 | Motivation: Histone modifications are among the most important factors that control gene regulation. Computational methods that predict gene expression from histone modification signals are highly desirable for understanding their combinatorial effects in gene regulation. This knowledge can help in developing 'epigenetic drugs' for diseases like cancer. Previous studies for quantifying the relationship between histone modifications and gene expression levels either failed to capture combinatorial effects or relied on multiple methods that separate predictions and combinatorial analysis. This paper develops a unified discriminative framework using a deep convolutional neural network to classify gene expression using histone modification data as input. Our system, called DeepChrome, allows automatic extraction of complex interactions among important features. To simultaneously visualize the combinatorial interactions among histone modifications, we propose a novel optimization-based technique that generates feature pattern maps from the learnt deep model. This provides an intuitive description of underlying epigenetic mechanisms that regulate genes. Results: We show that DeepChrome outperforms state-of-the-art models like Support Vector Machines and Random Forests for gene expression classification task on 56 different cell-types from REMC database. The output of our visualization technique not only validates the previous observations but also allows novel insights about combinatorial interactions among histone modification marks, some of which have recently been observed by experimental studies. | In order to to elucidate the possible combinatorial roles of histone modifications in gene regulation, @cite_32 applied rule learning on the T-cells datasets ( @cite_5 ) and produced 83 valid rules for gene expression (high) and repression (low). The authors selected the 20 most discriminative histone modifications as input into a rule learning system. They used several heuristics to filter out unexpected rules that were obtained by the learning system after scanning the entire search space. However, this study does not consider detailed feature patterns across local bins and does not perform prediction of gene expression. | {
"cite_N": [
"@cite_5",
"@cite_32"
],
"mid": [
"2083245183",
"630889805"
],
"abstract": [
"Keji Zhao and colleagues report genome-wide maps of 18 histone lysine acetylations in human CD4+ T cells as detected by ChIP-sequencing. Analysis of the data along with genome-wide maps of histone lysine methylations revealed a common module of 17 modifications associated with 25 of genes.",
"Gene regulation, despite being investigated in a large number of works, is still yet to be well understood. The mechanisms that control gene expression is one of the open problems. Epigenetic factors, among others, are assumed to have a role to play. In this work, we focus on DNA methylation and post-translational histone modifications (PTMs). These individually have been shown to contribute to controlling of gene expression. However, neither can totally account for the expression levels, i.e. low or high. Therefore, the hypothesis of their combinatorial role, as two of the most influencing factors, has been established and discussed in literature. Taking a computational approach based on rule induction, we derived (83 ) rules and identified some key PTMs that have considerable effects such as H2BK5ac, H3K79me123, H4K91ac, and H3K4me3. Also, some interesting patterns of DNA methylation and PTMs that can explain the low expression of genes in CD4 (+ ) T cell. The results include previously reported patterns as well as some new valid ones which could give some new insights to the process in question."
]
} |
1607.02078 | 2949844559 | Motivation: Histone modifications are among the most important factors that control gene regulation. Computational methods that predict gene expression from histone modification signals are highly desirable for understanding their combinatorial effects in gene regulation. This knowledge can help in developing 'epigenetic drugs' for diseases like cancer. Previous studies for quantifying the relationship between histone modifications and gene expression levels either failed to capture combinatorial effects or relied on multiple methods that separate predictions and combinatorial analysis. This paper develops a unified discriminative framework using a deep convolutional neural network to classify gene expression using histone modification data as input. Our system, called DeepChrome, allows automatic extraction of complex interactions among important features. To simultaneously visualize the combinatorial interactions among histone modifications, we propose a novel optimization-based technique that generates feature pattern maps from the learnt deep model. This provides an intuitive description of underlying epigenetic mechanisms that regulate genes. Results: We show that DeepChrome outperforms state-of-the-art models like Support Vector Machines and Random Forests for gene expression classification task on 56 different cell-types from REMC database. The output of our visualization technique not only validates the previous observations but also allows novel insights about combinatorial interactions among histone modification marks, some of which have recently been observed by experimental studies. | @cite_18 leveraged the correlated nature of epigenetic signals in the REMC database, including histone modifications. Their tool, ChromImpute, imputed signals for a particular new sample using an ensemble of regression trees on all the other signals and samples. EFilter ( @cite_28 ), a linear estimation algorithm, predicted gene expression in a new sample by using imputed expression levels from similar samples. Unlike the studies discussed above, these works focus on imputing or predicting signals for new samples. | {
"cite_N": [
"@cite_28",
"@cite_18"
],
"mid": [
"2049684926",
"2019251848"
],
"abstract": [
"Optimized algorithms from the field of electrical-signal processing improve the identification of genomic signals from diverse high-throughput sequencing experiments, such as ChIP-seq, DNase-seq and FAIRE-seq.",
"Large-scale epigenomic profiles are predicted from experimental data using multiple regression tree models."
]
} |
1607.01490 | 2466952995 | Ontologies are one of the core foundations of the Semantic Web. To participate in Semantic Web projects, domain experts need to be able to understand the ontologies involved. Visual notations can provide an overview of the ontology and help users to understand the connections among entities. However, the users first need to learn the visual notation before they can interpret it correctly. Controlled natural language representation would be readable right away and might be preferred in case of complex axioms, however, the structure of the ontology would remain less apparent. We propose to combine ontology visualizations with contextual ontology verbalizations of selected ontology (diagram) elements, displaying controlled natural language (CNL) explanations of OWL axioms corresponding to the selected visual notation elements. Thus, the domain experts will benefit from both the high-level overview provided by the graphical notation and the detailed textual explanations of particular elements in the diagram. | In the field of textual ontology verbalizations there has been some exploration of how to make verbalizations more convenient for users. One approach that has been tried is grouping verbalizations by entities. It produces a kind of a dictionary, where records are entities (class, property, individual), and every record contains verbalizations of axioms that refer to this entity. The resulting document is significantly larger than a plain, non-grouped verbalization because many axioms may refer to multiple entities and thus will be repeated in each entity. Nevertheless, the grouped presentation was preferred by users @cite_15 . Our approach can be considered a generalization of this approach, where a dictionary is replaced by an ontology visualization that serves as a map of the ontology. | {
"cite_N": [
"@cite_15"
],
"mid": [
"2133999934"
],
"abstract": [
"The SWAT Tools ontology verbaliser generates a hierarchically organised hypertext designed for easy comprehension and navigation. The document structure, inspired by encyclopedias and glossaries, is organised at a number of levels. At the top level, a heading is generated for every concept in the ontology; at the next level, each entry is subdivided into logically-based headings like 'Definition' and 'Examples'; at the next, sentences are aggregated when they have parts in common; at the lowest level, phrases are hyperlinked to concept headings. One consequence of this organisation is that some statements are repeated because they are relevant to more than one entry; this means that the text is longer than one in which statements are simply listed. This trade-off between organisation and brevity is investigated in a user study."
]
} |
1607.01443 | 2460123873 | We present Breakout, a group interaction platform for online courses that enables the creation and measurement of face-to-face peer learning groups in online settings. Breakout is designed to help students easily engage in synchronous, video breakout session based peer learning in settings that otherwise force students to rely on asynchronous text-based communication. The platform also offers data collection and intervention tools for studying the communication patterns inherent in online learning environments. The goals of the system are twofold: to enhance student engagement in online learning settings and to create a platform for research into the relationship between distributed group interaction patterns and learning outcomes. | In the Meeting Mediator project, showed that real-time visualizations have the potential to increase interaction balance, decrease dominant behavior, and increase collaboration in physical meetings @cite_12 . Later, the same system was shown to increase the effectiveness of distributed collaboration @cite_1 . While some of these works address distributed communication and real-time feedback, ours is the first to offer an open, accessible research tool for measuring group communication at scale and in real-world environments rather than laboratory experiments. | {
"cite_N": [
"@cite_1",
"@cite_12"
],
"mid": [
"2023275508",
"2090252063"
],
"abstract": [
"Sociometric feedback visualizes social signals among group members to increase their awareness of their communication patterns. We deployed the Meeting Mediator, a real-time sociometric feedback system to groups participating in two rounds of a social dilemma task: in one round, all members were co-located and in the other round, the members were geographically distributed. Laboratory results show that the sociometric feedback successfully increases the speaking time and the frequency of turn transitions of groups that are initially distributed and later co-located, and also leads to a higher cooperation rate, increasing the overall earnings of these groups. In addition, the sociometric feedback helps groups have a more consistent pattern of behavior before and after a change in their geographic distribution. Therefore, the sociometric feedback influences the communication patterns of distributed groups and makes them more cooperative. Furthermore, the sociometric feedback helps groups sustain those patterns of communication even after a change in geographic distribution.",
"We present the Meeting Mediator (MM), a real-time portable system that detects social interactions and provides persuasive feedback to enhance group collaboration. Social interactions is captured using Sociometric badges [17] and are visualized on mobile phones to promote behavioral change. Particularly in distributed collaborations, MM attempts to bridge the gap among the distributed groups by detecting and communicating social signals. In a study on brainstorming and problem solving meetings, MM had a significant effect on overlapping speaking time and interactivity level without distracting the subjects. The Sociometric badges were also able to detect dominant players in the group and measure their influence on other participants. Most interestingly, in groups with one or more dominant people, MM effectively reduced the dynamical difference between co-located and distributed collaboration as well as the behavioral difference between dominant and non-dominant people. Our system encourages change in group dynamics that may lead to higher performance and satisfaction. We envision that MM will be deployed in real-world organizations to improve interactions across various group collaboration contexts."
]
} |
1607.01639 | 2470356888 | The use of TLS by malware poses new challenges to network threat detection because traditional pattern-matching techniques can no longer be applied to its messages. However, TLS also introduces a complex set of observable data features that allow many inferences to be made about both the client and the server. We show that these features can be used to detect and understand malware communication, while at the same time preserving the privacy of benign uses of encryption. These data features also allow for accurate malware family attribution of network communication, even when restricted to a single, encrypted flow. To demonstrate this, we performed a detailed study of how TLS is used by malware and enterprise applications. We provide a general analysis on millions of TLS encrypted flows, and a targeted study on 18 malware families composed of thousands of unique malware samples and ten-of-thousands of malicious TLS flows. Importantly, we identify and accommodate the bias introduced by the use of a malware sandbox. The performance of a malware classifier is correlated with a malware family's use of TLS, i.e., malware families that actively evolve their use of cryptography are more difficult to classify. We conclude that malware's usage of TLS is distinct from benign usage in an enterprise setting, and that these differences can be effectively used in rules and machine learning classifiers. | Identifying threats in encryption poses significant challenges. Nevertheless, the security community has put forth two solutions to solve this problem. The first involves decrypting all traffic that flows through a security appliance: Man-in-the-Middle (MITM) @cite_30 . Once the traffic has been decrypted, traditional signature-based methods, such as Snort @cite_6 , can be applied. While this approach can be successful at finding threats, there are several important shortcomings. First, this method does not respect the privacy of the users on the network. Second, this method is computationally expensive and difficult to deploy and maintain. Third, this method relies on malware clients and servers to not change their behavior when a MITM interposes itself. | {
"cite_N": [
"@cite_30",
"@cite_6"
],
"mid": [
"2076014973",
"1674877186"
],
"abstract": [
"Web-based applications rely on the HTTPS protocol to guarantee privacy and security in transactions ranging from home banking, e-commerce, and e-procurement to those that deal with sensitive data such as career and identity information. Users trust this protocol to prevent unauthorized viewing of their personal, financial, and confidential information over the Web.",
"Network intrusion detection systems (NIDS) are an important part of any network security architecture. They provide a layer of defense which monitors network traffic for predefined suspicious activity or patterns, and alert system administrators when potential hostile traffic is detected. Commercial NIDS have many differences, but Information Systems departments must face the commonalities that they share such as significant system footprint, complex deployment and high monetary cost. Snort was designed to address these issues."
]
} |
1607.01639 | 2470356888 | The use of TLS by malware poses new challenges to network threat detection because traditional pattern-matching techniques can no longer be applied to its messages. However, TLS also introduces a complex set of observable data features that allow many inferences to be made about both the client and the server. We show that these features can be used to detect and understand malware communication, while at the same time preserving the privacy of benign uses of encryption. These data features also allow for accurate malware family attribution of network communication, even when restricted to a single, encrypted flow. To demonstrate this, we performed a detailed study of how TLS is used by malware and enterprise applications. We provide a general analysis on millions of TLS encrypted flows, and a targeted study on 18 malware families composed of thousands of unique malware samples and ten-of-thousands of malicious TLS flows. Importantly, we identify and accommodate the bias introduced by the use of a malware sandbox. The performance of a malware classifier is correlated with a malware family's use of TLS, i.e., malware families that actively evolve their use of cryptography are more difficult to classify. We conclude that malware's usage of TLS is distinct from benign usage in an enterprise setting, and that these differences can be effectively used in rules and machine learning classifiers. | There has been previous work that uses active probing @cite_7 and passive monitoring to gain visibility into how TLS is used in the wild @cite_13 . Unlike @cite_13 , our results specifically highlight malware's use of the TLS protocol, and show how data features from TLS can be used in rules and classifiers. | {
"cite_N": [
"@cite_13",
"@cite_7"
],
"mid": [
"1910624757",
"1984097153"
],
"abstract": [
"The majority of electronic communication today happens either via email or chat. Thanks to the use of standardised protocols electronic mail (SMTP, IMAP, POP3) and instant chat (XMPP, IRC) servers can be deployed in a decentralised but interoperable fashion. These protocols can be secured by providing encryption with the use of TLS---directly or via the STARTTLS extension---and leverage X.509 PKIs or ad hoc methods to authenticate communication peers. However, many combination of these mechanisms lead to insecure deployments. We present the largest study to date that investigates the security of the email and chat infrastructures. We used active Internet-wide scans to determine the amount of secure service deployments, and passive monitoring to investigate if user agents actually use this opportunity to secure their communications. We addressed both the client-to-server interactions as well as server-to-server forwarding mechanisms that these protocols offer, and the use of encryption and authentication methods in the process. Our findings shed light on an insofar unexplored area of the Internet. The truly frightening result is that most of our communication is poorly secured in transit.",
"Transport Layer Security is the standard, widely deployed protocol for securing client-server communications over the Internet. TLS is designed to prevent eavesdropping, tampering, and message forgery for client-server applications. Here, the author looks at the collection of standards that make up TLS, including its history, protocol, and future."
]
} |
1607.01657 | 2463954311 | We consider the task of graph exploration. An @math -node graph has unlabeled nodes, and all ports at any node of degree @math are arbitrarily numbered @math . A mobile agent has to visit all nodes and stop. The exploration time is the number of edge traversals. We consider the problem of how much knowledge the agent has to have a priori, in order to explore the graph in a given time, using a deterministic algorithm. This a priori information (advice) is provided to the agent by an oracle, in the form of a binary string, whose length is called the size of advice. We consider two types of oracles. The instance oracle knows the entire instance of the exploration problem, i.e., the port-numbered map of the graph and the starting node of the agent in this map. The map oracle knows the port-numbered map of the graph but does not know the starting node of the agent. We first consider exploration in polynomial time, and determine the exact minimum size of advice to achieve it. This size is @math , for both types of oracles. When advice is large, there are two natural time thresholds: @math for a map oracle, and @math for an instance oracle, that can be achieved with sufficiently large advice. We show that, with a map oracle, time @math cannot be improved in general, regardless of the size of advice. We also show that the smallest size of advice to achieve this time is larger than @math , for any @math . For an instance oracle, advice of size @math is enough to achieve time @math . We show that, with any advice of size @math , the time of exploration must be at least @math , for any @math , and with any advice of size @math , the time must be @math . We also investigate minimum advice sufficient for fast exploration of hamiltonian graphs. | Another direction of research concerns exploration of anonymous graphs. In this case it is impossible to explore arbitrary graphs and stop after exploration, if no marking of nodes is allowed, and if nothing is known about the graph. Hence some authors @cite_19 @cite_24 allow pebbles which the agent can drop on nodes to recognize already visited ones, and then remove them and drop them in other places. A more restrictive scenario assumes a stationary token that is fixed at the starting node of the agent @cite_16 @cite_34 . Exploring anonymous graphs without the possibility of marking nodes (and thus possibly without stopping) is investigated, e.g., in @cite_22 @cite_0 . The authors concentrate attention not on the cost of exploration but on the minimum amount of memory sufficient to carry out this task. In the absence of marking nodes, in order to guarantee stopping after exploration, some knowledge about the graph is required, e.g., an upper bound on its size @cite_16 @cite_32 . | {
"cite_N": [
"@cite_22",
"@cite_32",
"@cite_34",
"@cite_24",
"@cite_19",
"@cite_0",
"@cite_16"
],
"mid": [
"2017617294",
"2049516232",
"",
"2154255883",
"2149888497",
"",
"1578829633"
],
"abstract": [
"A robot with k-bit memory has to explore a tree whose nodes are unlabeled and edge ports are locally labeled at each node. The robot has no a priori knowledge of the topology of the tree or of its size, and its aim is to traverse all the edges. While O(log Δ) bits of memory suffice to explore any tree of maximum degree Δ if stopping is not required, we show that bounded memory is not sufficient to explore with stop all trees of bounded degree (indeed Ω (log log log n) bits of memory are needed for some such trees of size n). For the more demanding task requiring to stop at the starting node after completing exploration, we show a sharper lower bound Ω (log n) on required memory size, and present an algorithm to accomplish this task with O(log2 n)-bit memory, for all n-node trees.",
"We present a deterministic, log-space algorithm that solves st-connectivity in undirected graphs. The previous bound on the space complexity of undirected st-connectivity was log4 3(ṡ) obtained by Armoni, Ta-Shma, Wigderson and Zhou (JACM 2000). As undirected st-connectivity is complete for the class of problems solvable by symmetric, nondeterministic, log-space computations (the class SL), this algorithm implies that SL e L (where L is the class of problems solvable by deterministic log-space computations). Independent of our work (and using different techniques), Trifonov (STOC 2005) has presented an O(log n log log n)-space, deterministic algorithm for undirected st-connectivity. Our algorithm also implies a way to construct in log-space a fixed sequence of directions that guides a deterministic walk through all of the vertices of any connected graph. Specifically, we give log-space constructible universal-traversal sequences for graphs with restricted labeling and log-space constructible universal-exploration sequences for general graphs.",
"",
"We show that two cooperating robots can learn exactly any strongly-connected directed graph with n indistinguishable nodes in expected time polynomial in n. We introduce a new type of homing sequence for two robots which helps the robots recognize certain previously-seen nodes. We then present an algorithm in which the robots learn the graph and the homing sequence simultaneously by wandering actively through the graph. Unlike most previous learning results using homing sequences, our algorithm does not require a teacher to provide counterexamples. Furthermore, the algorithm can use efficiently any additional information available that distinguishes nodes. We also present an algorithm in which the robots learn by taking random walks. The rate at which a random walk converges to the stationary distribution is characterized by the conductance of the graph. Our random-walk algorithm learns in expected time polynomial in n and in the inverse of the conductance and is more efficient than the homing-sequence algorithm for high-conductance graphs. >",
"Exploring and mapping an unknown environment is a fundamental problem that is studied in a variety of contexts. Many results have focused on finding efficient solutions to restricted versions of the problem. In this paper, we consider a model that makes very limited assumptions about the environment and solve the mapping problem in this general setting. We model the environment by an unknown directed graph G, and consider the problem of a robot exploring and mapping G. The edges emanating from each vertex are numbered from ‘1’ to ‘d’, but we do not assume that the vertices of G are labeled. Since the robot has no way of distinguishing between vertices, it has no hope of succeeding unless it is given some means of distinguishing between vertices. For this reason we provide the robot with a “pebble”—a device that it can place on a vertex and use to identify the vertex later. In this paper we show: (1) If the robot knows an upper bound on the number of vertices then it can learn the graph efficiently with only one pebble. (2) If the robot does not know an upper bound on the number of vertices n, then (log log n) pebbles are both necessary and sufficient. In both cases our algorithms are deterministic. C © 2002 Elsevier Science (USA)",
"",
"We study the problem of mapping an unknown environment represented as an unlabelled undirected graph. A robot (or automaton) starting at a single vertex of the graph G has to traverse the graph and return to its starting point building a map of the graph in the process. We are interested in the cost of achieving this task (whenever possible) in terms of the number of edge traversal made by the robot. Another optimization criteria is to minimize the amount of information that the robot has to carry when moving from node to node in the graph. We present efficient algorithms for solving map construction using a robot that is not allowed to mark any vertex of the graph, assuming the knowledge of only an upper bound on the size of the graph. We also give universal algorithms (independent of the size of the graph) for map construction when only the starting location of the robot is marked. Our solutions apply the technique of universal exploration sequences to solve the map construction problem under various constraints. We also show how the solution can be adapted to solve other problems such as the gathering of two identical robots dispersed in an unknown graph."
]
} |
1607.01657 | 2463954311 | We consider the task of graph exploration. An @math -node graph has unlabeled nodes, and all ports at any node of degree @math are arbitrarily numbered @math . A mobile agent has to visit all nodes and stop. The exploration time is the number of edge traversals. We consider the problem of how much knowledge the agent has to have a priori, in order to explore the graph in a given time, using a deterministic algorithm. This a priori information (advice) is provided to the agent by an oracle, in the form of a binary string, whose length is called the size of advice. We consider two types of oracles. The instance oracle knows the entire instance of the exploration problem, i.e., the port-numbered map of the graph and the starting node of the agent in this map. The map oracle knows the port-numbered map of the graph but does not know the starting node of the agent. We first consider exploration in polynomial time, and determine the exact minimum size of advice to achieve it. This size is @math , for both types of oracles. When advice is large, there are two natural time thresholds: @math for a map oracle, and @math for an instance oracle, that can be achieved with sufficiently large advice. We show that, with a map oracle, time @math cannot be improved in general, regardless of the size of advice. We also show that the smallest size of advice to achieve this time is larger than @math , for any @math . For an instance oracle, advice of size @math is enough to achieve time @math . We show that, with any advice of size @math , the time of exploration must be at least @math , for any @math , and with any advice of size @math , the time must be @math . We also investigate minimum advice sufficient for fast exploration of hamiltonian graphs. | Providing nodes or agents with arbitrary kinds of information that can be used to perform network tasks more efficiently has been previously proposed in @cite_33 @cite_4 @cite_21 @cite_6 @cite_1 @cite_18 @cite_20 @cite_25 @cite_9 @cite_23 @cite_30 @cite_27 @cite_8 in contexts ranging from graph coloring to broadcasting and leader election. This approach was referred to as algorithms with advice . The advice is given either to nodes of the network or to mobile agents performing some network task. In the first case, instead of advice, the term informative labeling schemes is sometimes used, if different nodes can get different information. | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_4",
"@cite_33",
"@cite_8",
"@cite_9",
"@cite_21",
"@cite_1",
"@cite_6",
"@cite_27",
"@cite_23",
"@cite_25",
"@cite_20"
],
"mid": [
"1983693678",
"2038319432",
"2046334554",
"2025590344",
"2174013141",
"2181598850",
"2109659895",
"1971694274",
"1975011672",
"2056295140",
"2034501275",
"",
"1975595616"
],
"abstract": [
"We study deterministic broadcasting in radio networks in the recently introduced framework of network algorithms with advice. We concentrate on the problem of trade-offs between the number of bits of information (size of advice) available to nodes and the time in which broadcasting can be accomplished. In particular, we ask what is the minimum number of bits of information that must be available to nodes of the network, in order to broadcast very fast. For networks in which constant time broadcast is possible under a complete knowledge of the network we give a tight answer to the above question: O(n) bits of advice are sufficient but o(n) bits are not, in order to achieve constant broadcasting time in all these networks. This is in sharp contrast with geometric radio networks of constant broadcasting time: we show that in these networks a constant number of bits suffices to broadcast in constant time. For arbitrary radio networks we present a broadcasting algorithm whose time is inverse-proportional to the size of the advice.",
"We study the amount of knowledge about the network that is required in order to efficiently solve a task concerning this network. The impact of available information on the efficiency of solving network problems, such as communication or exploration, has been investigated before but assumptions concerned availability of particular items of information about the network, such as the size, the diameter, or a map of the network. In contrast, our approach is quantitative: we investigate the minimum number of bits of information (bits of advice) that has to be given to an algorithm in order to perform a task with given efficiency. We illustrate this quantitative approach to available knowledge by the task of tree exploration. A mobile entity (robot) has to traverse all edges of an unknown tree, using as few edge traversals as possible. The quality of an exploration algorithm A is measured by its competitive ratio, i.e., by comparing its cost (number of edge traversals) to the length of the shortest path containing all edges of the tree. Depth-First-Search has competitive ratio 2 and, in the absence of any information about the tree, no algorithm can beat this value. We determine the minimum number of bits of advice that has to be given to an exploration algorithm in order to achieve competitive ratio strictly smaller than 2. Our main result establishes an exact threshold number of bits of advice that turns out to be roughly loglogD, where D is the diameter of the tree. More precisely, for any constant c, we construct an exploration algorithm with competitive ratio smaller than 2, using at most loglogD-c bits of advice, and we show that every algorithm using loglogD-g(D) bits of advice, for any function g unbounded from above, has competitive ratio at least 2.",
"We study the problem of the amount of information required to draw a complete or a partial map of a graph with unlabeled nodes and arbitrarily labeled ports. A mobile agent, starting at any node of an unknown connected graph and walking in it, has to accomplish one of the following tasks: draw a complete map of the graph, i.e., find an isomorphic copy of it including port numbering, or draw a partial map, i.e., a spanning tree, again with port numbering. The agent executes a deterministic algorithm and cannot mark visited nodes in any way. None of these map drawing tasks is feasible without any additional information, unless the graph is a tree. Hence we investigate the minimum number of bits of information (minimum size of advice) that has to be given to the agent to complete these tasks. It turns out that this minimum size of advice depends on the number n of nodes or the number m of edges of the graph, and on a crucial parameter @m, called the multiplicity of the graph, which measures the number of nodes that have an identical view of the graph. We give bounds on the minimum size of advice for both above tasks. For @m=1 our bounds are asymptotically tight for both tasks and show that the minimum size of advice is very small. For @m>1 the minimum size of advice increases abruptly. In this case our bounds are asymptotically tight for topology recognition and asymptotically almost tight for spanning tree construction.",
"We consider the following problem. Given a rooted tree @math , label the nodes of @math in the most compact way such that, given the labels of two nodes @math and @math , one can determine in constant time, by looking only at the labels, whether @math is ancestor of @math . The best known labeling scheme is rather straightforward and uses labels of length at most @math bits each, where @math is the number of nodes in the tree. Our main result in this paper is a labeling scheme with maximum label length @math . Our motivation for studying this problem is enhancing the performance of web search engines. In the context of this application each indexed document is a tree, and the labels of all trees are maintained in main memory. Therefore even small improvements in the maximum label length are important.",
"[L. Blin, P. Fraigniaud, N. Nisse, S. Vial, Distributing chasing of network intruders, in: 13th Colloquium on Structural Information and Communication Complexity, SIROCCO, in: LNCS, vol. 4056, Springer-Verlag, 2006, pp. 70-84] introduced a new measure of difficulty for a distributed task in a network. The smallest number of bits of advice of a distributed problem is the smallest number of bits of information that has to be available to nodes in order to accomplish the task efficiently. Our paper deals with the number of bits of advice required to perform efficiently the graph searching problem in a distributed setting. In this variant of the problem, all searchers are initially placed at a particular node of the network. The aim of the team of searchers is to clear a contaminated graph in a monotone connected way, i.e., the cleared part of the graph is permanently connected, and never decreases while the search strategy is executed. Moreover, the clearing of the graph must be performed using the optimal number of searchers, i.e. the minimum number of searchers sufficient to clear the graph in a monotone connected way in a centralized setting. We show that the minimum number of bits of advice permitting the monotone connected and optimal clearing of a network in a distributed setting is @Q(nlogn), where n is the number of nodes of the network. More precisely, we first provide a labelling of the vertices of any graph G, using a total of O(nlogn) bits, and a protocol using this labelling that enables the optimal number of searchers to clear G in a monotone connected distributed way. Then, we show that this number of bits of advice is optimal: any distributed protocol requires @W(nlogn) bits of advice to clear a network in a monotone connected way, using an optimal number of searchers.",
"Topology recognition is one of the fundamental distributed tasks in networks. Each node of an anonymous network has to deterministically produce an isomorphic copy of the underlying graph, with all ports correctly marked. This task is usually unfeasible without any a priori information. Such information can be provided to nodes as advice. An oracle knowing the network can give a possibly different string of bits to each node, and all nodes must reconstruct the network using this advice, after a given number of rounds of communication. During each round each node can exchange arbitrary messages with all its neighbors and perform arbitrary local computations. The time of completing topology recognition is the number of rounds it takes, and the size of advice is the maximum length of a string given to nodes. We investigate tradeoffs between the time in which topology recognition is accomplished and the minimum size of advice that has to be given to nodes. We provide upper and lower bounds on the minimum size of advice that is sufficient to perform topology recognition in a given time, in the class of all graphs of size n and diameter D≤αn, for any constant α<1. In most cases, our bounds are asymptotically tight. More precisely, if the allotted time is D-k, where 0<k≤D, then the optimal size of advice is i¾?n 2 logn D-k+1. If the allotted time is D, then this optimal size is i¾?n logn. If the allotted time is D+k, where 0<k≤D 2, then the optimal size of advice is i¾?1+logn k. The only remaining gap between our bounds is for time D+k, where D 2<k≤D. In this time interval our upper bound remains O1+logn k, while the lower bound that holds for any time is 1. This leaves a gap if D∈ologn. Finally, we show that for time 2D+1, one bit of advice is both necessary and sufficient. Our results show how sensitive is the minimum size of advice to the time allowed for topology recognition: allowing just one round more, from D to D+1, decreases exponentially the advice needed to accomplish this task.",
"We consider a model for online computation in which the online algorithm receives, together with each request, some information regarding the future, referred to as advice. The advice is a function, defined by the online algorithm, of the whole request sequence. The advice provided to the online algorithm may allow an improvement in its performance, compared to the classical model of complete lack of information regarding the future. We are interested in the impact of such advice on the competitive ratio, and in particular, in the relation between the size b of the advice, measured in terms of bits of information per request, and the (improved) competitive ratio. Since b=0 corresponds to the classical online model, and b=@?log|A|@?, where A is the algorithm's action space, corresponds to the optimal (offline) one, our model spans a spectrum of settings ranging from classical online algorithms to offline ones. In this paper we propose the above model and illustrate its applicability by considering two of the most extensively studied online problems, namely, metrical task systems (MTS) and the k-server problem. For MTS we establish tight (up to constant factors) upper and lower bounds on the competitive ratio of deterministic and randomized online algorithms with advice for any choice of 1@?b@?@Q(logn), where n is the number of states in the system: we prove that any randomized online algorithm for MTS has competitive ratio @W(log(n) b) and we present a deterministic online algorithm for MTS with competitive ratio O(log(n) b). For the k-server problem we construct a deterministic online algorithm for general metric spaces with competitive ratio k^O^(^1^ ^b^) for any choice of @Q(1)@?b@?logk.",
"We study the amount of knowledge about a communication network that must be given to its nodes in order to efficiently disseminate information. Our approach is quantitative: we investigate the minimum total number of bits of information (minimum size of advice) that has to be available to nodes, regardless of the type of information provided. We compare the size of advice needed to perform broadcast and wakeup (the latter is a broadcast in which nodes can transmit only after getting the source information), both using a linear number of messages (which is optimal). We show that the minimum size of advice permitting the wakeup with a linear number of messages in an n-node network, is @Q(nlogn), while the broadcast with a linear number of messages can be achieved with advice of size O(n). We also show that the latter size of advice is almost optimal: no advice of size o(n) can permit to broadcast with a linear number of messages. Thus an efficient wakeup requires strictly more information about the network than an efficient broadcast.",
"We study the problem of the amount of information (advice) about a graph that must be given to its nodes in order to achieve fast distributed computations. The required size of the advice enables to measure the information sensitivity of a network problem. A problem is information sensitive if little advice is enough to solve the problem rapidly (i.e., much faster than in the absence of any advice), whereas it is information insensitive if it requires giving a lot of information to the nodes in order to ensure fast computation of the solution. In this paper, we study the information sensitivity of distributed graph coloring.",
"This paper addresses the problem of locally verifying global properties. Several natural questions are studied, such as “how expensive is local verification?” and more specifically, “how expensive is local verification compared to computation?” A suitable model is introduced in which these questions are studied in terms of the number of bits a vertex needs to communicate. The model includes the definition of a proof labeling scheme (a pair of algorithms- one to assign the labels, and one to use them to verify that the global property holds). In addition, approaches are presented for the efficient construction of schemes, and upper and lower bounds are established on the bit complexity of schemes for multiple basic problems. The paper also studies the role and cost of unique identities in terms of impossibility and complexity, in the context of proof labeling schemes. Previous studies on related questions deal with distributed algorithms that simultaneously compute a configuration and verify that this configuration has a certain desired property. It turns out that this combined approach enables the verification to be less costly sometimes, since the configuration is typically generated so as to be easily verifiable. In contrast, our approach separates the configuration design from the verification. That is, it first generates the desired configuration without bothering with the need to verify it, and then handles the task of constructing a suitable verification scheme. Our approach thus allows for a more modular design of algorithms, and has the potential to aid in verifying properties even when the original design of the structures for maintaining them was done without verification in mind.",
"We consider the problem of labeling the nodes of a graph in a way that will allow one to compute the distance between any two nodes directly from their labels (without using any additional information). Our main interest is in the minimal length of labels needed in different cases. We obtain upper and lower bounds for several interesting families of graphs. In particular, our main results are the following. For general graphs, we show that the length needed is Θ(n). For trees, we show that the length needed is Θ(log2 n). For planar graphs, we show an upper bound of O(√nlogn) and a lower bound of Ω(n1 3). For bounded degree graphs, we show a lower bound of Ω(√n). The upper bounds for planar graphs and for trees follow by a more general upper bound for graphs with a r(n)-separator. The two lower bounds, however, are obtained by two different arguments that may be interesting in their own right. We also show some lower bounds on the length of the labels, even if it is only required that distances be approximated to a multiplicative factor s. For example, we show that for general graphs the required length is Ω(n) for every s < 3. We also consider the problem of the time complexity of the distance function once the labels are computed. We show that there are graphs with optimal labels of length 3 log n, such that if we use any labels with fewer than n bits per label, computing the distance function requires exponential time. A similar result is obtained for planar and bounded degree graphs.",
"",
"We use the recently introduced advising scheme framework for measuring the difficulty of locally distributively computing a Minimum Spanning Tree (MST). An (m,t)-advising scheme for a distributed problem P is a way, for every possible input I of P, to provide an \"advice\" (i.e., a bit string) about I to each node so that: (1) the maximum size of the advices is at most m bits, and (2) the problem P can be solved distributively in at most t rounds using the advices as inputs. In case of MST, the output returned by each node of a weighted graph G is the edge leading to its parent in some rooted MST T of G. Clearly, there is a trivial (log n,0)-advising scheme for MST (each node is given the local port number of the edge leading to the root of some MST T), and it is known that any (0,t)-advising scheme satisfies t ≥ Ω (√n). Our main result is the construction of an (O(1),O(log n))-advising scheme for MST. That is, by only giving a constant number of bits of advice to each node, one can decrease exponentially the distributed computation time of MST in arbitrary graph, compared to algorithms dealing with the problem in absence of any a priori information. We also consider the average size of the advices. On the one hand, we show that any (m,0)-advising scheme for MST gives advices of average size Ω(log n). On the other hand we design an (m,1)-advising scheme for MST with advices of constant average size, that is one round is enough to decrease the average size of the advices from log(n) to constant."
]
} |
1607.01059 | 2462710658 | Sparse representation-based classification (SRC), proposed by , seeks the sparsest decomposition of a test sample over the dictionary of training samples, with classification to the most-contributing class. Because it assumes test samples can be written as linear combinations of their same-class training samples, the success of SRC depends on the size and representativeness of the training set. Our proposed classification algorithm enlarges the training set by using local principal component analysis to approximate the basis vectors of the tangent hyperplane of the class manifold at each training sample. The dictionary in SRC is replaced by a local dictionary that adapts to the test sample and includes training samples and their corresponding tangent basis vectors. We use a synthetic data set and three face databases to demonstrate that this method can achieve higher classification accuracy than SRC in cases of sparse sampling, nonlinear class manifolds, and stringent dimension reduction. | The approach of using tangent hyperplanes for pattern recognition is not new. When the data is assumed to lie on a low-dimensional manifold, local tangent hyperplanes are a simple and intuitive approach to enhancing the data set and gaining insight into the manifold structure. Our proposed method is very much related to (TDC) @cite_16 @cite_12 @cite_24 , which constructs local tangent hyperplanes of the class manifolds, computes the distances between these hyperplanes and the given test sample, and then classifies the test sample to the class with the closest hyperplane. We show in Section that our proposed method's integration of tangent hyperplane basis vectors into the sparse representation framework generally outperforms TDC. | {
"cite_N": [
"@cite_24",
"@cite_16",
"@cite_12"
],
"mid": [
"2038481941",
"1537355730",
""
],
"abstract": [
"Distance measure is quite important for pattern recognition. Utilizing invariance in image data, tangent distance is very powerful in classifying handwritten digits. For this measure a set of invariant transformations must be known a priori. But in many practical problems, it is very difficult to know these transformations. In this paper, an algorithm is proposed to approximate the invariant tangent distance exclusively from the data. By virtue of ideas arising from manifold learning, the algorithm needs no prior transformations and can be applied to more classification problems. k-nearest neighbor rule based on the new distance are implemented for classification problems. Experimental results on synthetic and real datasets illustrate its validity.",
"In order to achieve good generalization with neural networks overfitting must be controlled. Weight penalty factors are one common method of providing this control. However, using weight penalties creates the additional search problem of finding the optimal penalty factors. MacKay [5] proposed an approximate Bayesian framework for training neural networks, in which penalty factors are treated as hyperparameters and found in an iterative search. However, for classification networks trained with cross-entropy error, this search is slow and unstable, and it is not obvious how to improve it. This paper describes and compares several strategies for controlling this search. Some of these strategies greatly improve the speed and stability of the search. Test runs on a range of tasks are described.",
""
]
} |
1607.01059 | 2462710658 | Sparse representation-based classification (SRC), proposed by , seeks the sparsest decomposition of a test sample over the dictionary of training samples, with classification to the most-contributing class. Because it assumes test samples can be written as linear combinations of their same-class training samples, the success of SRC depends on the size and representativeness of the training set. Our proposed classification algorithm enlarges the training set by using local principal component analysis to approximate the basis vectors of the tangent hyperplane of the class manifold at each training sample. The dictionary in SRC is replaced by a local dictionary that adapts to the test sample and includes training samples and their corresponding tangent basis vectors. We use a synthetic data set and three face databases to demonstrate that this method can achieve higher classification accuracy than SRC in cases of sparse sampling, nonlinear class manifolds, and stringent dimension reduction. | On the other hand, approaches to address the limiting linear subspace assumption (i.e., the assumption that class manifolds are linear subspaces) in SRC have been proposed. For example, extended sparse coding and dictionary learning to general Riemannian manifolds @cite_11 . Admittedly only a first step in meeting their ultimate objective, 's work requires explicit knowledge of the class manifolds. This is an unsatisfiable condition in many real-world classification problems and is not a requirement of our proposed algorithm. Alternatively, have been effective in overcoming SRC's linearity assumption, as nonlinear relationships in the original space may be linear in kernel space given an appropriate choice of kernel @cite_8 . Several local'' modifications of SRC implicitly ameliorate the linearity assumption; in @cite_9 and (LSDL-SRC) @cite_15 , for instance, coefficients of the representation are constrained by their corresponding training samples' distances to the test sample, and so these algorithms need only assume linearity at the local level. Our proposed method is designed to improve not only the locality but also the accuracy of the approximation of the test sample in terms of its ground truth class. Section contains an experimental comparison between our proposed method and LSDL-SRC, as well as a discussion thereof. | {
"cite_N": [
"@cite_8",
"@cite_9",
"@cite_15",
"@cite_11"
],
"mid": [
"1964749215",
"1989243451",
"2108024382",
"2147942061"
],
"abstract": [
"Sparse representation has attracted great attention in the past few years. Sparse representation based classification (SRC) algorithm was developed and successfully used for classification. In this paper, a kernel sparse representation based classification (KSRC) algorithm is proposed. Samples are mapped into a high dimensional feature space first and then SRC is performed in this new feature space by utilizing kernel trick. Since samples in the high dimensional feature space are unknown, we cannot perform KSRC directly. In order to overcome this difficulty, we give the method to solve the problem of sparse representation in the high dimensional feature space. If an appropriate kernel is selected, in the high dimensional feature space, a test sample is probably represented as the linear combination of training samples of the same class more accurately. Therefore, KSRC has more powerful classification ability than SRC. Experiments of face recognition, palmprint recognition and finger-knuckle-print recognition demonstrate the effectiveness of KSRC.",
"Applications of sparse signal representation in image processing and pattern recognition have attracted a great deal of attention. Sparse representation based classification (SRC) methods emphasizes on sparse representation computed by l\"1-minimization to exploit the underlying sparsity in the problem domain, and argued the importance of sparse representation that improved the discrimination to achieve robust and accurate classification results. Recently, many studies have shown the role of collaborative representation (CR) in SRC, which actually improved the classification accuracy. In this paper, we proposed a novel collaborative neighbor representation method for multi-class classification based on l\"2-minimization approach with the assumption of locally linear embedding (LLE). The proposed method represents a test sample over the dictionary by automatically choosing optimal nearest basis spanned in the same linear subspace as of test sample. The proposed representation method achieves competitive classification accuracy via optimal neighbor representation having discriminative learning power. Extensive experiments on real-world face and digit databases are performed to analyze the performance of the proposed method against SRC methods. Result clearly shows that the proposed method achieves competitive results for face recognition and pattern classification, and is significantly much faster and comparably accurate than SRC based classification methods.",
"Motivated by image reconstruction, sparse representation based classification (SRC) has been shown to be an effective method for applications like face recognition. In this paper, we propose a locality-sensitive dictionary learning algorithm for SRC, in which the designed dictionary is able to preserve local data structure, resulting in improved image classification. During the dictionary update and sparse coding stages in the proposed algorithm, we provide closed-form solutions and enforce the data locality constraint throughout the learning process. In contrast to previous dictionary learning approaches utilizing sparse representation techniques, which did not (or only partially) take data locality into consideration, our algorithm is able to produce a more representative dictionary and thus achieves better performance. We conduct experiments on databases designed for face and handwritten digit recognition. For such reconstruction-based classification problems, we will confirm that our proposed method results in better or comparable performance as state-of-the-art SRC methods do, while less training time for dictionary learning can be achieved.",
"Existing dictionary learning algorithms are based on the assumption that the data are vectors in an Euclidean vector space Rd, and the dictionary is learned from the training data using the vector space structure of Rd and its Euclidean L2-metric. However, in many applications, features and data often originated from a Riemannian manifold that does not support a global linear (vector space) structure. Furthermore, the extrinsic viewpoint of existing dictionary learning algorithms becomes inappropriate for modeling and incorporating the intrinsic geometry of the manifold that is potentially important and critical to the application. This paper proposes a novel framework for sparse coding and dictionary learning for data on a Riemannian manifold, and it shows that the existing sparse coding and dictionary learning methods can be considered as special (Euclidean) cases of the more general framework proposed here. We show that both the dictionary and sparse coding can be effectively computed for several important classes of Riemannian manifolds, and we validate the proposed method using two well-known classification problems in computer vision and medical imaging analysis."
]
} |
1607.01059 | 2462710658 | Sparse representation-based classification (SRC), proposed by , seeks the sparsest decomposition of a test sample over the dictionary of training samples, with classification to the most-contributing class. Because it assumes test samples can be written as linear combinations of their same-class training samples, the success of SRC depends on the size and representativeness of the training set. Our proposed classification algorithm enlarges the training set by using local principal component analysis to approximate the basis vectors of the tangent hyperplane of the class manifold at each training sample. The dictionary in SRC is replaced by a local dictionary that adapts to the test sample and includes training samples and their corresponding tangent basis vectors. We use a synthetic data set and three face databases to demonstrate that this method can achieve higher classification accuracy than SRC in cases of sparse sampling, nonlinear class manifolds, and stringent dimension reduction. | Other classification algorithms have been proposed that are similar to ours in that they aim to enlarge or otherwise enhance the training set in SRC. Such methods for face recognition, for example, include the use of virtual images that exploit the symmetry of the human face, as in both the method of @cite_18 and @cite_34 . Though visual comparison of these virtual images and our recovered tangent vectors (see Section ) could be informative, our proposed method can be used for general classification. | {
"cite_N": [
"@cite_18",
"@cite_34"
],
"mid": [
"2117286860",
"1954340199"
],
"abstract": [
"The face almost always has an axis-symmetrical structure. However, as the face usually does not have an absolutely frontal pose when it is imaged, the majority of face images are not symmetrical images. These facts inspire us that the mirror image of the face image might be viewed as a representation of the face with a possible pose opposite to that of the original face image. In this paper we propose a scheme to produce the mirror image of the face and integrate the original face image and its mirror image for representation-based face recognition. This scheme is simple and computationally efficient. Almost all the representation-based classification methods can be improved by this scheme. The underlying rationales of the scheme are as follows: first, the use of the mirror image can somewhat overcome the misalignment problem of the face image in face recognition. Second, it is able to somewhat eliminate the side-effect of the variation of the pose and illumination of the original face image. The experiments show that the proposed scheme can greatly improve the accuracy of the representation-based classification methods. The proposed scheme might be also helpful for improving other face recognition methods.",
"We proposed a method to use virtual available facial images for face recognition.We improved the linear regression classification method for face recognition.The classification accuracy on sample pairs are better than that on original samples.This method achieves lower classification error rates than many other methods.This method performs well even when there are few training samples of each class. Sparse representation classification, as one of the state-of-the-art classification methods, has been widely studied and successfully applied in face recognition since it was proposed by In this study, we proposed a method to generate virtual available facial images and modified the well-known linear regression classification (LRC) and collaborative representation based classification (CRC) for face recognition. The new method integrates the original and virtual symmetry facial images to form a training sample set of large size. Experimental results show that the proposed method can achieve better performance than most of the competitive face recognition methods, e.g. LRC, CRC, INNC, SRC, RCR, RRC and the method in (2014). This promising performance is mainly attributed to the fact that the sample combination scheme used in the new method can exploit limited original training samples to produce a large number of available training samples and to convey sufficient variations of the original training samples."
]
} |
1607.01059 | 2462710658 | Sparse representation-based classification (SRC), proposed by , seeks the sparsest decomposition of a test sample over the dictionary of training samples, with classification to the most-contributing class. Because it assumes test samples can be written as linear combinations of their same-class training samples, the success of SRC depends on the size and representativeness of the training set. Our proposed classification algorithm enlarges the training set by using local principal component analysis to approximate the basis vectors of the tangent hyperplane of the class manifold at each training sample. The dictionary in SRC is replaced by a local dictionary that adapts to the test sample and includes training samples and their corresponding tangent basis vectors. We use a synthetic data set and three face databases to demonstrate that this method can achieve higher classification accuracy than SRC in cases of sparse sampling, nonlinear class manifolds, and stringent dimension reduction. | Additionally, there have been many local modifications to the sparse representation framework with objectives other than classification. For example, 's (RSSL) @cite_20 uses the @math -norm for sparse feature extraction, combining high-level semantics with low-level, locality-preserving features. In the feature selection algorithm (CGSSL) by @cite_3 , features are jointly selected using sparse regularization (via the @math -norm) and a non-negative spectral clustering objective. Not only are the selected features sparse; they also are the most discriminative features in terms of predicting the cluster indicators in both the original space and a lower-dimensional subspace on which the data is assumed to lie. | {
"cite_N": [
"@cite_3",
"@cite_20"
],
"mid": [
"2084028080",
"2056935845"
],
"abstract": [
"Many pattern analysis and data mining problems have witnessed high-dimensional data represented by a large number of features, which are often redundant and noisy. Feature selection is one main technique for dimensionality reduction that involves identifying a subset of the most useful features. In this paper, a novel unsupervised feature selection algorithm, named clustering-guided sparse structural learning (CGSSL), is proposed by integrating cluster analysis and sparse structural analysis into a joint framework and experimentally evaluated. Nonnegative spectral clustering is developed to learn more accurate cluster labels of the input samples, which guide feature selection simultaneously. Meanwhile, the cluster labels are also predicted by exploiting the hidden structure shared by different features, which can uncover feature correlations to make the results more reliable. Row-wise sparse models are leveraged to make the proposed model suitable for feature selection. To optimize the proposed formulation, we propose an efficient iterative algorithm. Finally, extensive experiments are conducted on 12 diverse benchmarks, including face data, handwritten digit data, document data, and biomedical data. The encouraging experimental results in comparison with several representative algorithms and the theoretical analysis demonstrate the efficiency and effectiveness of the proposed algorithm for feature selection.",
"To uncover an appropriate latent subspace for data representation, in this paper we propose a novel Robust Structured Subspace Learning (RSSL) algorithm by integrating image understanding and feature learning into a joint learning framework. The learned subspace is adopted as an intermediate space to reduce the semantic gap between the low-level visual features and the high-level semantics. To guarantee the subspace to be compact and discriminative, the intrinsic geometric structure of data, and the local and global structural consistencies over labels are exploited simultaneously in the proposed algorithm. Besides, we adopt the @math -norm for the formulations of loss function and regularization respectively to make our algorithm robust to the outliers and noise. An efficient algorithm is designed to solve the proposed optimization problem. It is noted that the proposed framework is a general one which can leverage several well-known algorithms as special cases and elucidate their intrinsic relationships. To validate the effectiveness of the proposed method, extensive experiments are conducted on diversity datasets for different image understanding tasks, i.e., image tagging, clustering, and classification, and the more encouraging results are achieved compared with some state-of-the-art approaches."
]
} |
1607.01299 | 2471395359 | We study the problem of planning Pareto-optimal journeys in public transit networks. Most existing algorithms and speed-up techniques work by computing subjourneys to intermediary stops until the destination is reached. In contrast, the trip-based model focuses on trips and transfers between them, constructing journeys as a sequence of trips. In this paper, we develop a speed-up technique for this model inspired by principles behind existing state-of-the-art speed-up techniques, Transfer Pattern and Hub Labelling. The resulting algorithm allows us to compute Pareto-optimal (with respect to arrival time and number of transfers) 24-hour profiles on very large real-world networks in less than half a millisecond. Compared to the current state of the art for bicriteria queries on public transit networks, this is up to two orders of magnitude faster, while increasing preprocessing overhead by at most one order of magnitude. | RAPTOR @cite_19 foregoes modelling the data as a graph and instead operates directly on the timetable data. In addition to travel time and number of transfers, they also consider price as a criteria. The Connection Scan Algorithm (CSA) @cite_12 also eschews graphs and instead works on an ordered array of connections to find Pareto-optimal journeys with respect to travel time and number of transfers. Accelerated CSA @cite_7 is a speed-up technique for CSA that works via partitioning of the network. Unlike the original CSA, it was only evaluated as a single-criterion algorithm, using the number of transfers as a tiebreaker between journeys with identical arrival time. | {
"cite_N": [
"@cite_19",
"@cite_7",
"@cite_12"
],
"mid": [
"",
"2259825555",
"115854788"
],
"abstract": [
"",
"We study the problem of efficiently computing journeys in timetable networks. Our algorithm optimally answers profile queries, computing all journeys given a time interval. Our study demonstrates that queries can be answered optimally on large country-scale timetable networks within several milliseconds and fast delay integration is possible. Previous work either had to drop optimality or only considered comparatively small timetable networks. Our technique is a combination of the Connection Scan Algorithm and multilevel overlay graphs.",
"This paper studies the problem of computing optimal journeys in dynamic public transit networks. We introduce a novel algorithmic framework, called Connection Scan Algorithm (CSA), to compute journeys. It organizes data as a single array of connections, which it scans once per query. Despite its simplicity, our algorithm is very versatile. We use it to solve earliest arrival and multi-criteria profile queries. Moreover, we extend it to handle the minimum expected arrival time (MEAT) problem, which incorporates stochastic delays on the vehicles and asks for a set of (alternative) journeys that in its entirety minimizes the user’s expected arrival time at the destination. Our experiments on the dense metropolitan network of London show that CSA computes MEAT queries, our most complex scenario, in 272 ms on average."
]
} |
1607.01299 | 2471395359 | We study the problem of planning Pareto-optimal journeys in public transit networks. Most existing algorithms and speed-up techniques work by computing subjourneys to intermediary stops until the destination is reached. In contrast, the trip-based model focuses on trips and transfers between them, constructing journeys as a sequence of trips. In this paper, we develop a speed-up technique for this model inspired by principles behind existing state-of-the-art speed-up techniques, Transfer Pattern and Hub Labelling. The resulting algorithm allows us to compute Pareto-optimal (with respect to arrival time and number of transfers) 24-hour profiles on very large real-world networks in less than half a millisecond. Compared to the current state of the art for bicriteria queries on public transit networks, this is up to two orders of magnitude faster, while increasing preprocessing overhead by at most one order of magnitude. | Public Transit Labelling (PTL) @cite_13 uses, as the name implies, a hub labelling approach. It requires extensive preprocessing and produces a very large amount of auxiliary data, but leads to very low query times, even for multi-criteria queries. Timetable Labelling (TTL) @cite_0 is another labelling-based approach, which has been extended in the context of databases by Efentakis @cite_24 . However, TTL only performs single-criterion queries regarding arrival time. | {
"cite_N": [
"@cite_0",
"@cite_13",
"@cite_24"
],
"mid": [
"2037403365",
"2951056795",
"2397633064"
],
"abstract": [
"A public transportation network can often be modeled as a timetable graph where (i) each node represents a station; and (ii) each directed edge (u,v) is associated with a timetable that records the departure (resp. arrival) time of each vehicle at station u (resp. v). Several techniques have been proposed for various types of route planning on timetable graphs, e.g., retrieving the route from a node to another with the shortest travel time. These techniques, however, either provide insufficient query efficiency or incur significant space overheads. This paper presents Timetable Labelling (TTL), an efficient indexing technique for route planning on timetable graphs. The basic idea of TTL is to associate each node @math with a set of labels, each of which records the shortest travel time from u to some other node v given a certain departure time from u; such labels would then be used during query processing to improve efficiency. In addition, we propose query algorithms that enable TTL to support three popular types of route planning queries, and investigate how we reduce the space consumption of TTL with advanced preprocessing and label compression methods. By conducting an extensive set of experiments on real world datasets, we demonstrate that TTL significantly outperforms the states of the art in terms of query efficiency, while incurring moderate preprocessing and space overheads.",
"We study the journey planning problem in public transit networks. Developing efficient preprocessing-based speedup techniques for this problem has been challenging: current approaches either require massive preprocessing effort or provide limited speedups. Leveraging recent advances in Hub Labeling, the fastest algorithm for road networks, we revisit the well-known time-expanded model for public transit. Exploiting domain-specific properties, we provide simple and efficient algorithms for the earliest arrival, profile, and multicriteria problems, with queries that are orders of magnitude faster than the state of the art.",
"Recent scientic literature focuses on answering Earliest Arrival (EA), Latest Departure (LD) and Shortest Duration (SD) queries in (schedule-based) public transportation networks. Unfortunately, most of the existing solutions operate in main memory, making the proposed methods hard to scale for larger instances and dicult to integrate in a multi-user environment. This work proposes PTLDB (Public Transporation Labels on the DataBase), a novel, scalable, pure-SQL framework for answering EA, LD and SD queries, implemented entirely on an open-source database system. Moreover, we formulate four new types of queries targeting public transportation networks, namely the Earliest Arrival and Latest Departure k-Nearest Neighbor (kNN) and Oneto-many queries and propose novel ways to eciently answer them within PTLDB. Our experimentation will show that the proposed solution is fast, scalable and easy to use, making our PTLDB framework a serious contender for use in real-world applications."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.