paper_id stringlengths 9 12 | venue stringclasses 139
values | year stringclasses 7
values | paper_title stringlengths 0 181 | paper_authors stringlengths 4 925 | paper_abstract stringlengths 1 5k | paper_keywords stringlengths 2 436 | paper_content stringlengths 0 100k | review_id stringlengths 9 12 | review_title stringlengths 0 500 | review_rating stringclasses 61
values | review_text stringlengths 2 28.3k | review_confidence stringclasses 13
values | text stringlengths 402 130k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
sWMF2rf9nQt | ICLR.cc/2021/Conference | 2021 | Misclassification Detection via Class Augmentation | ["Fei Zhu", "Xu-yao Zhang", "Chuang Wang", "Cheng-lin Liu"] | Despite the impressive performance in various pattern recognition tasks, deep neural networks (DNNs) are typically overconfident in their predictions, making it difficult to determine whether a test example is misclassified. In this paper, we propose a simple yet effective method of class augmentation (classAug) to add... | ["misclassification detection", "augmentation", "uncertainty", "confidence"] | ABSTRACTDespite the impressive performance in various pattern recognition tasks, deep neu-ral networks (DNNs) are typically overconfident in their predictions, making itdifficult to determine whether a test example is misclassified. In this paper, wepropose a simple yet effective method of class augmentation (classAug)... | LbjzxKtYAfL | Review for Misclassification Detection via Class Augmentation | 3: Clear rejection | This paper focuses on addressing the challenge of misclassification detection in DNNs. To this end, the authors propose a method of class augmentation (classAug) which is an extension of Mixup. They increased the number of classes during training by assigning new classes to the samples generated using between-class int... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Misclassification Detection via Class Augmentation
### Paper Abstract
Despite the impressive performance in various pattern recognition tasks, deep neural networks (DNNs... |
Bw7VC-DJUM | ICLR.cc/2021/Conference | 2021 | Learning Spatiotemporal Features via Video and Text Pair Discrimination | ["Tianhao Li", "Limin Wang"] | Current video representations heavily rely on learning from manually annotated video datasets which are time-consuming and expensive to acquire. We observe videos are naturally accompanied by abundant text information such as YouTube titles and Instagram captions. In this paper, we leverage this visual-textual connecti... | ["Spatiotemporal Feature Learning", "Video and Text Pair Discrimination", "Self-/Weakly Supervised Learning"] | ABSTRACTCurrent video representations heavily rely on learning from manually annotatedvideo datasets which are time-consuming and expensive to acquire. We observevideos are naturally accompanied by abundant text information such as YouTubetitles and Instagram captions. In this paper, we leverage this visual-textual con... | 45RjaN9deCh | Good work showcasing video representations can be trained with much fewer data samples (on the order of 100K) | 6: Marginally above acceptance threshold | *Summary:*
The paper proposes an approach to learn a video feature backbone in an unsupervised manner through the use of video titles (text modality) associated with user generated content from Youtube or Instagram. The key idea is to use a contrastive loss that increases the similarity score between a positive pair vs... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Spatiotemporal Features via Video and Text Pair Discrimination
### Paper Abstract
Current video representations heavily rely on learning from manually annotated... |
S1xRnxSYwS | ICLR.cc/2020/Conference | 2020 | Goten: GPU-Outsourcing Trusted Execution of Neural Network Training and Prediction | ["Lucien K.L. Ng", "Sherman S.M. Chow", "Anna P.Y. Woo", "Donald P. H. Wong", "Yongjun Zhao"] | Before we saw worldwide collaborative efforts in training machine-learning models or widespread deployments of prediction-as-a-service, we need to devise an efficient privacy-preserving mechanism which guarantees the privacy of all stakeholders (data contributors, model owner, and queriers). Slaom (ICLR ’19) preserves p... | ["machine learning", "security", "privacy", "TEE", "trusted processors", "Intel SGX", "GPU", "high-performance"] | ABSTRACTBefore we saw worldwide collaborative efforts in training machine-learning mod-els or widespread deployments of prediction-as-a-service, we need to devise anefficient privacy-preserving mechanism which guarantees the privacy of all stake-holders (data contributors, model owner, and queriers). Slalom (ICLR ’19) ... | HkguQA3BYB | Official Blind Review #3 | 6: Weak Accept | The paper builds a privacy-preserving training framework within a Trusted Execution Environment (TEE) such as Intel SGX. The work is heavily inspired from Slalom, which does privacy-preserving inference in TEEs. The main drawbacks of Slalom when extending to training are (1) weight quantization needs to be dynamics as ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Goten: GPU-Outsourcing Trusted Execution of Neural Network Training and Prediction
### Paper Abstract
Before we saw worldwide collaborative efforts in training machine-l... | |
low-53sFqn | robot-learning.org/CoRL/2023/Conference | 2023 | Fleet Active Learning: A Submodular Maximization Approach | ["Oguzhan Akcin", "Orhan Unuvar", "Onat Ure", "Sandeep P. Chinchali"] | In multi-robot systems, robots often gather data to improve the performance of their deep neural networks (DNNs) for perception and planning. Ideally, these robots should select the most informative samples from their local data distributions by employing active learning approaches. However, when the data collection is... | ["Active Learning", "Cloud Robotics", "Robotic Perception"] | Fleet Active Learning:A Submodular Maximization ApproachOguzhan Akcin1, Orhan Unuvar1, Onat Ure1and Sandeep P. Chinchali1Abstract: In multi-robot systems, robots often gather data to improve the perfor-mance of their deep neural networks (DNNs) for perception and planning. Ide-ally, these robots should select the most ... | UY1WOcGzK9 | Pros:
- the paper grounds itself in practical scenario of data gathering for autonomous cars. Perhaps in control literature, there has been similar set ups with distributed systems/sensing, e.g., sensor network placement, etc, the paper brings these ideas for practical robotic scenarios.
- the paper is generally well w... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Fleet Active Learning: A Submodular Maximization Approach
### Paper Abstract
In multi-robot systems, robots often gather data to improve the performance of their deep ne... | ||
S1eWbkSFPS | ICLR.cc/2020/Conference | 2020 | GRAPHS, ENTITIES, AND STEP MIXTURE | ["Kyuyong Shin", "Wonyoung Shin", "Jung-Woo Ha", "Sunyoung Kwon"] | Graph neural networks have shown promising results on representing and analyzing diverse graph-structured data such as social, citation, and protein interaction networks. Existing approaches commonly suffer from the oversmoothing issue, regardless of whether policies are edge-based or node-based for neighborhood aggreg... | ["Graph Neural Network", "Random Walk", "Attention"] | ABSTRACTGraph neural networks have shown promising results on representing and ana-lyzing diverse graph-structured data such as social, citation, and protein inter-action networks. Existing approaches commonly suffer from the oversmoothingissue, regardless of whether policies are edge-based or node-based for neighbor-h... | SJg4oGBstr | Official Blind Review #1 | 6: Weak Accept | The paper presents a graph neural network model that aims at improving the feature aggregation scheme to better handle distant nodes, therefore mitigating the “smoothing” problem of classic averaging.
I find the paper clearly motivated and easy to follow, although some sentences could be streamlined and some repetitio... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
GRAPHS, ENTITIES, AND STEP MIXTURE
### Paper Abstract
Graph neural networks have shown promising results on representing and analyzing diverse graph-structured data such... | |
rJxGLlBtwH | ICLR.cc/2020/Conference | 2020 | On the interaction between supervision and self-play in emergent communication | ["Ryan Lowe*", "Abhinav Gupta*", "Jakob Foerster", "Douwe Kiela", "Joelle Pineau"] | A promising approach for teaching artificial agents to use natural language involves using human-in-the-loop training. However, recent work suggests that current machine learning methods are too data inefficient to be trained in this way from scratch. In this paper, we investigate the relationship between two categorie... | ["multi-agent communication", "self-play", "emergent languages"] | ABSTRACTA promising approach for teaching artificial agents to use natural language involvesusing human-in-the-loop training. However, recent work suggests that currentmachine learning methods are too data inefficient to be trained in this way fromscratch. In this paper, we investigate the relationship between two cate... | BJxoajDCtS | Official Blind Review #1 | 6: Weak Accept | This paper investigated how two conflicting learning objectives; supervised and self-play updates could be combined with a focus on visual-grounded language tasks. With a different set of their combinations, the authors empirically found that alternating two learning updates may result in the best equilibrium state; co... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the interaction between supervision and self-play in emergent communication
### Paper Abstract
A promising approach for teaching artificial agents to use natural lang... | |
SkejkR4KDr | ICLR.cc/2020/Conference | 2020 | Layer Flexible Adaptive Computation Time for Recurrent Neural Networks | ["Lida Zhang", "Diego Klabjan"] | Deep recurrent neural networks perform well on sequence data and are the model of choice. However, it is a daunting task to decide the structure of the networks, i.e. the number of layers, especially considering different computational needs of a sequence. We propose a layer flexible recurrent neural network with adapt... | ["model", "sequence", "number", "layers", "recurrent neural networks", "sequence data", "choice"] | ABSTRACTDeep recurrent neural networks perform well on sequence data and are the modelof choice. However, it is a daunting task to decide the structure of the networks,i.e. the number of layers, especially considering different computational needsof a sequence. We propose a layer flexible recurrent neural network with ... | S1li9bECtB | Official Blind Review #3 | 3: Weak Reject | Summary:
The paper proposes to use adaptive computation time (ACT) for the number of layers (instead of the number of timesteps in the original paper) for a RNN. The intuition is that some inputs are more complex and may need more processing compare to other inputs.
Comments on the paper,
- I am slightly confused ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Layer Flexible Adaptive Computation Time for Recurrent Neural Networks
### Paper Abstract
Deep recurrent neural networks perform well on sequence data and are the model ... | |
6xhX4nnBhbJ | AKBC.ws/2021/Conference | 2021 | SHINRA2020-ML: Categorizing 30-language Wikipedia into fine-grained NE based on "``Resource by Collaborative Contribution" scheme | ["Satoshi Sekine", "Kouta Nakayama", "Maya Ando", "Yu Usami", "Masako Nomoto", "Koji Matsuda"] | This paper describes a Knowledge Base construction project, SHINRA and particularly the SHINRA2020-ML task. The SHINRA2020-ML task is to categorize 30-language Wikipedia pages into fine-grained named entity categories, called "Extended Named Entity (ENE)". It is one of the three tasks conducted on SHINRA since 2018. S... | ["Structured Knowledge", "Wikipedia", "Shared-task", "Resource by Collaborative Contribution"] | Automated Knowledge Base Construction (2020) Conference paperSHINRA2020-ML: Categorizing 30-language Wikipediainto ne-grained NE based on\Resource by Collaborative Contribution" schemeSatoshi Sekine satoshi.sekine@riken.jpKouta Nakayama kouta.nakayama@riken.jpMaya Ando maya@kzd.biglobe.ne.jpYu Usami yu@usami.llcMasako ... | -8zbCGZFyWM | Report of collaborative effort to classify Wikipedia pages in 30 new languages. | 7: Good paper, accept | This paper presents the design and results of a shared task that fosed on categorizing Wikipedia entity pages in 30 languages. The task attracted 10 participants, working on overlapping subsets of the target languages, and the output predictions will be shared with the community as a knowledge resource.
The task requi... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
SHINRA2020-ML: Categorizing 30-language Wikipedia into fine-grained NE based on "``Resource by Collaborative Contribution" scheme
### Paper Abstract
This paper describes... |
rJNwDjAqYX | ICLR.cc/2019/Conference | 2019 | Large-Scale Study of Curiosity-Driven Learning | ["Yuri Burda", "Harri Edwards", "Deepak Pathak", "Amos Storkey", "Trevor Darrell", "Alexei A. Efros"] | Reinforcement learning algorithms rely on carefully engineered rewards from the environment that are extrinsic to the agent. However, annotating each environment with hand-designed, dense rewards is difficult and not scalable, motivating the need for developing reward functions that are intrinsic to the agent.
Curiosi... | ["exploration", "curiosity", "intrinsic reward", "no extrinsic reward", "unsupervised", "no-reward", "skills"] | ABSTRACTReinforcement learning algorithms rely on carefully engineered rewards from theenvironment that are extrinsic to the agent. However, annotating each environ-ment with hand-designed, dense rewards is difficult and not scalable, motivatingthe need for developing reward functions that are intrinsic to the agent. C... | Ske_-TWah7 | Using curiosity-based reward exclusively works for game environments; not clear that this would be the case for more practical settings and findings regarding varying effectiveness of observation representation are largely incomplete. However, the core finding should influence additional research in game environments. | 6: Marginally above acceptance threshold | The authors consider the setting of a RL agent that exclusively receives intrinsic reward during training that is intended to model curiosity; technically, ‘curiosity’ is quantified by the ability of the agent to predict its own forward dynamics [Pathak, et al., ICML17]. This study primarily centers around an initially... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Large-Scale Study of Curiosity-Driven Learning
### Paper Abstract
Reinforcement learning algorithms rely on carefully engineered rewards from the environment that are ex... |
B1GRtj05t7 | ICLR.cc/2019/Conference | 2019 | NA | ["NA"] | NA | [] | Hyg9GolSnQ | The spatial value decomposition is just successor representation? | 4: Ok but not good enough - rejection | The paper is quite clearly written and the authors has explained the ideas nicely. There are also some convincing results that that the proposed algorithm, spatial monte-carlo method improved agent's performance in several continuous control tasks.
However, a critical problem I found is that the so-called 'spatial val... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NA
### Paper Abstract
NA
### Paper Keywords
[]
### Paper Content
<|im_end|>
<|im_start|>assistant
### Review Title
The spatial value decomposition is just successor re... | |
H1lQJ1HYwS | ICLR.cc/2020/Conference | 2020 | Deep amortized clustering | ["Juho Lee", "Yoonho Lee", "Yee Whye Teh"] | We propose a \textit{deep amortized clustering} (DAC), a neural architecture which learns to cluster datasets efficiently using a few forward passes. DAC implicitly learns what makes a cluster, how to group data points into clusters, and how to count the number of clusters in datasets. DAC is meta-learned using labelle... | ["clustering", "amortized inference", "meta learning", "deep learning"] | ABSTRACTWe propose a deep amortized clustering (DAC), a neural architecture which learnsto cluster datasets efficiently using a few forward passes. DAC implicitly learnswhat makes a cluster, how to group data points into clusters, and how to countthe number of clusters in datasets. DAC is meta-learned using labelled da... | r1gwMlFp_S | Official Blind Review #3 | 3: Weak Reject | This paper proposed a deep amortized clustering framework which learns to cluster data efficiently based on the combination of set transformer and amortized clustering. The main motivation is to learn clustering rules from labeled data sets and generalize to new data sets, so as to avoid manually defining clusterin... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deep amortized clustering
### Paper Abstract
We propose a \textit{deep amortized clustering} (DAC), a neural architecture which learns to cluster datasets efficiently us... | |
0gLzHrE_t3z | aclweb.org/ACL/2020/Workshop/NLP-COVID | 2020 | CORD-19: The COVID-19 Open Research Dataset | ["Lucy Lu Wang", "Kyle Lo", "Yoganand Chandrasekhar", "Russell Reas", "Jiangjiang Yang", "Doug Burdick", "Darrin Eide", "Kathryn Funk", "Yannis Katsis", "Rodney Kinney", "Yunyao Li", "Ziyang Liu", "William Merrill", "Paul Mooney", "Dewey Murdick", "Devvret Rishi", "Jerry Sheehan", "Zhihong Shen", "Brandon Stilson", "Al... | The COVID-19 Open Research Dataset (CORD-19) is a growing resource of scientific papers on COVID-19 and related historical coronavirus research. CORD-19 is designed to facilitate the development of text mining and information retrieval systems over its rich collection of metadata and structured full text papers. Since ... | ["COVID-19", "Coronavirus", "text mining", "NLP corpus", "open access", "shared task", "information extraction", "information retrieval"] | CORD-19: The C OVID -19 Open Research DatasetLucy Lu Wang1;Kyle Lo1;Yoganand Chandrasekhar1Russell Reas1Jiangjiang Yang1Douglas Burdick2Darrin Eide3Kathryn Funk4Yannis Katsis2Rodney Kinney1Yunyao Li2Ziyang Liu6William Merrill1Paul Mooney5Dewey Murdick7Devvret Rishi5Jerry Sheehan4Zhihong Shen3Brandon Stilson1Alex D. Wad... | QSKmsC8qzKu | CORD-19 is an excellent resource with an impressive integration work for the research community to fight COVID-19. | 9: Top 15% of accepted papers, strong accept | The authors present the CORD-19 data set and describe how it has been developed and continues to be developed. The CORD-19 data set is a valuable resource that provides access to the latest literature about COVID-19 and coronaviruses and it is updated daily with over 200k downloads. The generation of the CORD-19 requir... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
CORD-19: The COVID-19 Open Research Dataset
### Paper Abstract
The COVID-19 Open Research Dataset (CORD-19) is a growing resource of scientific papers on COVID-19 and re... |
Hkg4W2AcFm | ICLR.cc/2019/Conference | 2019 | Overcoming the Disentanglement vs Reconstruction Trade-off via Jacobian Supervision | ["Jos\u00e9 Lezama"] | A major challenge in learning image representations is the disentangling of the factors of variation underlying the image formation. This is typically achieved with an autoencoder architecture where a subset of the latent variables is constrained to correspond to specific factors, and the rest of them are considered n... | ["disentangling", "autoencoders", "jacobian", "face manipulation"] | ABSTRACTA major challenge in learning image representations is the disentangling of thefactors of variation underlying the image formation. This is typically achievedwith an autoencoder architecture where a subset of the latent variables is con-strained to correspond to specific factors, and the rest of them are consid... | BJxo72kq3X | Nice results on image manipulation | 7: Good paper, accept | The paper aims to learn an autoencoder that can be used to effectively encode the known attributes/ generative factors and this allows easy and controlled manipulation of the images while producing realistic images.
To achieve this, ordinarily, the encoder produces latent code with two components y and z where y are c... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Overcoming the Disentanglement vs Reconstruction Trade-off via Jacobian Supervision
### Paper Abstract
A major challenge in learning image representations is the disenta... |
HJlWXhC5Km | ICLR.cc/2019/Conference | 2019 | Learning to Control Visual Abstractions for Structured Exploration in Deep Reinforcement Learning | ["catalin ionescu", "tejas kulkarni", "aaron van de oord", "andriy mnih", "vlad mnih"] | Exploration in environments with sparse rewards is a key challenge for reinforcement learning. How do we design agents with generic inductive biases so that they can explore in a consistent manner instead of just using local exploration schemes like epsilon-greedy? We propose an unsupervised reinforcement learning agen... | ["exploration", "deep reinforcement learning", "intrinsic motivation", "unsupervised learning"] | ABSTRACTIONS FORSTRUCTURED EXPLORATION IN DEEP REINFORCE -MENT LEARNINGAnonymous authorsPaper under double-blind reviewABSTRACTExploration in environments with sparse rewards is a key challenge for reinforce-ment learning. How do we design agents with generic inductive biases so that theycan explore in a consistent man... | rkg7LLJ6h7 | The paper is unfortunately written quite confusingly such that it is hard to evaluate the contribution of the potentially interesting ideas. | 4: Ok but not good enough - rejection | The appproach introduces visual abstractions that are used for reinforcement learning. The abstractions are learned using a lower bound on the mutual information and options are created to generate different measurements for each abstraction. The algorithm hence learns to "control" each abstraction as well as to select... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Control Visual Abstractions for Structured Exploration in Deep Reinforcement Learning
### Paper Abstract
Exploration in environments with sparse rewards is a... |
tIjRAiFmU3y | ICLR.cc/2021/Conference | 2021 | An Unsupervised Deep Learning Approach for Real-World Image Denoising | ["Dihan Zheng", "Sia Huat Tan", "Xiaowen Zhang", "Zuoqiang Shi", "Kaisheng Ma", "Chenglong Bao"] | Designing an unsupervised image denoising approach in practical applications is a challenging task due to the complicated data acquisition process. In the real-world case, the noise distribution is so complex that the simplified additive white Gaussian (AWGN) assumption rarely holds, which significantly deteriorates th... | ["Real-world image denoising", "unsupervised image denoising"] | ABSTRACTDesigning an unsupervised image denoising approach in practical applications isa challenging task due to the complicated data acquisition process. In the real-world case, the noise distribution is so complex that the simplified additive whiteGaussian (AWGN) assumption rarely holds, which significantly deteriora... | z7QvpJZ8Qw- | Official Review | 6: Marginally above acceptance threshold | The paper proposes an unsupervised approach for denoising for which works on a single noisy image. The central idea is to use a neural network to map the noisy image to a latent image space where the noisy distribution follows Gaussian (like Variance Stabliziation Transform) and then use a off-the-shelf Gaussian denois... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
An Unsupervised Deep Learning Approach for Real-World Image Denoising
### Paper Abstract
Designing an unsupervised image denoising approach in practical applications is ... |
oPQBinIjBt | FAST.org.ar/2022/Workshop | 2022 | On Possible Extensions to the Smalltalk Syntax | ["Leandro Caniglia", "Javier Burroni"] | Even though changing it is fairly easy, the Smalltalk syntax has remained essentially stable for many decades. The reason is that people are cautious about side effects that could mess up something provably good. This doesn't mean that there aren't variations one might propose, but that the real task is to identify a c... | ["possible extensions", "smalltalk syntax", "smalltalk", "easy", "stable", "many decades", "reason", "people", "cautious", "side effects"] | Published in FAST Workshop on Smalltalk Related Technologies (11/2022)On Possible Extensions to the Smalltalk SyntaxLeandro Caniglia leandro.caniglia@fast.org.arFundación Argentina de SmalltalkJavier Burroni javier.burroni@gmail.comUniversity of Massachusetts, AmherstReviewed on OpenReview: https: // openreview. net/ ... | YvsXdUbZaa | Interesting syntax ideas, motivating examples, but maybe does too much at once | 8: Top 50% of accepted papers, clear accept | The paper presents possible syntax extensions for the Smalltalk syntax
- In 3-6 we have some smaller "tricks" to extend ST80 syntax
- Sections 7-10 explain the idea of using pragma syntax to add sections of foreign languages
- 11-15 presents smaller examples for this feature
- 17 presents a larger use case that use... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On Possible Extensions to the Smalltalk Syntax
### Paper Abstract
Even though changing it is fairly easy, the Smalltalk syntax has remained essentially stable for many d... |
HJxbbY_7PV | ICLR.cc/2019/Workshop/LLD | 2019 | ONLY SPARSITY BASED LOSS FUNCTION FOR LEARNING REPRESENTATIONS | ["Vivek Bakaraju", "Kishore Reddy Konda"] | We study the emergence of sparse representations in neural networks. We show that in unsupervised
models with regularization, the emergence of sparsity is the result of the input data samples being
distributed along highly non-linear or discontinuous manifold. We also derive a similar argument
for discriminatively trai... | ["Sparsity", "Unsupervised Learning", "Single Layer Models"] | ABSTRACTWe study the emergence of sparse representations in neural networks. We show that in unsupervisedmodels with regularization, the emergence of sparsity is the result of the input data samples beingdistributed along highly non-linear or discontinuous manifold. We also derive a similar argumentfor discriminatively... | Skl3LnM_KE | The paper proposes an interesting loss ruction that relies on independence of datapoints, unfortunately, it also depends on the data to work. | 3: Marginally above acceptance threshold | The authors propose a loss function that encourages different samples to be mapped to different feature vectors. The loss is analyzed both in the context of an autoencoder and a single layer supervised neural network.
One issue the authors would consider is introducing more clarity in the exposition of the work. For in... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ONLY SPARSITY BASED LOSS FUNCTION FOR LEARNING REPRESENTATIONS
### Paper Abstract
We study the emergence of sparse representations in neural networks. We show that i... |
HJ1kmv9xx | ICLR.cc/2017/conference | 2017 | LR-GAN: Layered Recursive Generative Adversarial Networks for Image Generation | ["Jianwei Yang", "Anitha Kannan", "Dhruv Batra", "Devi Parikh"] | We present LR-GAN: an adversarial image generation model which takes scene structure and context into account. Unlike previous generative adversarial networks (GANs), the proposed GAN learns to generate image background and foregrounds separately and recursively, and stitch the foregrounds on the background in a contex... | ["Computer vision", "Deep learning", "Unsupervised Learning"] | ABSTRACTWe present LR-GAN: an adversarial image generation model which takes scenestructure and context into account. Unlike previous generative adversarial net-works (GANs), the proposed GAN learns to generate image background and fore-grounds separately and recursively, and stitch the foregrounds on the backgroundin ... | BJM4eQUEg | a figure-ground shape aware GAN mode for image generation | 6: Marginally above acceptance threshold | The paper proposes a model for image generation where the back-ground is generated first and then the foreground is pasted in by generating first a foregound mask and corresponding appearance, curving the appearance image using the mask and transforming the mask using predicted affine transform to paste it on top of th... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
LR-GAN: Layered Recursive Generative Adversarial Networks for Image Generation
### Paper Abstract
We present LR-GAN: an adversarial image generation model which takes sc... |
YEv8xafoAQ | ICLR.cc/2021/Conference | 2021 | Text as Neural Operator: Image Manipulation by Text Instruction | ["Tianhao Zhang", "Hung-Yu Tseng", "Lu Jiang", "Honglak Lee", "Irfan Essa", "Weilong Yang"] | In recent years, text-guided image manipulation has gained increasing attention in the image generation research field. Recent works have proposed to deal with a simplified setting where the input image only has a single object and the text modification is acquired by swapping image captions or labels. In this paper, w... | ["Image Manipulation", "Generative Adversarial Networks"] | ABSTRACTIn recent years, text-guided image manipulation has gained increasing attentionin the image generation research field. Recent works have proposed to deal witha simplified setting where the input image only has a single object and the textmodification is acquired by swapping image captions or labels. In this pap... | tTX-ZXSMTwJ | Official Review | 6: Marginally above acceptance threshold | ### Summary
Authors proposed an architecture to edit image content given text operators. I think the paper is well organized, very clear, easy-to-follow, and experimental results are sound. Some details regarding baseline architecture should be clearer (i.e., differences in image and text encoders). One of the contrib... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Text as Neural Operator: Image Manipulation by Text Instruction
### Paper Abstract
In recent years, text-guided image manipulation has gained increasing attention in the... |
SkxnuTNtDS | ICLR.cc/2020/Conference | 2020 | Quadratic GCN for graph classification | ["Omer Nagar", "Yoram Louzoun"] | Graph Convolutional Networks (GCN) have been extensively used to classify nodes in graphs and have been shown to outperform most other node classification methods. However, there are currently few GCN based formalism for graph classification tasks. In this task, graphs of different sizes (e.g. graphs representing diffe... | ["GCN", "Quadratic activation", "graph classification"] | ABSTRACTGraph Convolutional Networks (GCN) have been extensively used to classifynodes in graphs and have been shown to outperform most other node classific-ation methods. However, there are currently few GCN based formalism for graphclassification tasks. In this task, graphs of different sizes (e.g. graphs representin... | rJxWvaFGsr | Official Blind Review #2 | 1: Reject | The paper proposes a simple modification to existing graph NN methods for classifying the entire graph of possibly variable size. The problem is of interest. The proposed method is on the other hand fairly simple and straightforward, as shown in Fig.1 and eq(4), as a simple function of the adjacency matrix A that incor... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Quadratic GCN for graph classification
### Paper Abstract
Graph Convolutional Networks (GCN) have been extensively used to classify nodes in graphs and have been shown t... | |
HygrdpVKvr | ICLR.cc/2020/Conference | 2020 | NAS evaluation is frustratingly hard | ["Antoine Yang", "Pedro M. Esperan\u00e7a", "Fabio M. Carlucci"] | Neural Architecture Search (NAS) is an exciting new field which promises to be as much as a game-changer as Convolutional Neural Networks were in 2012. Despite many great works leading to substantial improvements on a variety of tasks, comparison between different methods is still very much an open issue. While most al... | ["neural architecture search", "nas", "benchmark", "reproducibility", "harking"] | ABSTRACTNeural Architecture Search (NAS) is an exciting new field which promises to beas much as a game-changer as Convolutional Neural Networks were in 2012. De-spite many great works leading to substantial improvements on a variety of tasks,comparison between different methods is still very much an open issue. Whilem... | HygAVxO3OH | Official Blind Review #2 | 1: Reject | In this submission, the authors conduct a series of experiments on five image classification datasets to compare several existing NAS methods. Based on these experimental results, they point out: 1) how a network is trained (i.e., training protocols/tricks such as DropPath, Cutout) plays an important role for the final... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NAS evaluation is frustratingly hard
### Paper Abstract
Neural Architecture Search (NAS) is an exciting new field which promises to be as much as a game-changer as Convo... | |
LNtTXJ9XXr | ICLR.cc/2021/Conference | 2021 | Adversarial Masking: Towards Understanding Robustness Trade-off for Generalization | ["Minhao Cheng", "Zhe Gan", "Yu Cheng", "Shuohang Wang", "Cho-Jui Hsieh", "Jingjing Liu"] | Adversarial training is a commonly used technique to improve model robustness against adversarial examples. Despite its success as a defense mechanism, adversarial training often fails to generalize well to unperturbed test data. While previous work assumes it is caused by the discrepancy between robust and non-robust ... | ["Adversarial Machine Learning", "Adversarial Robustness", "Adversarial Training", "Generalization"] | ABSTRACTAdversarial training is a commonly used technique to improve model robustnessagainst adversarial examples. Despite its success as a defense mechanism, ad-versarial training often fails to generalize well to unperturbed test data. Whileprevious work assumes it is caused by the discrepancy between robust and non-... | KuEnOsFHYdP | Nice empirical observation on the trade-off between generalization and robustness | 7: Good paper, accept | The paper observes that the rescaling operation in the batch normalization layer and the ReLU activation learn to select different features for standard and adversarial trainings. The authors call this effect "Adversarial Masking." Based on this observation, the authors then propose Rob-Mask that achieves good standard... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adversarial Masking: Towards Understanding Robustness Trade-off for Generalization
### Paper Abstract
Adversarial training is a commonly used technique to improve model ... |
Hpxrls8yAn | ICLR.cc/2021/Conference | 2021 | Robust Memory Augmentation by Constrained Latent Imagination | ["Yao Mu", "Yuzheng Zhuang", "Bin Wang", "Wulong Liu", "Shengbo Eben Li", "Jianye HAO"] | The latent dynamics model summarizes an agent’s high dimensional experiences in a compact way. While learning from imagined trajectories by the latent model is confirmed to has great potential to facilitate behavior learning, the lack of memory diversity limits generalization capability. Inspired by a neuroscience expe... | ["Memory Augmentation", "Model-based reinforcement learning", "Latent imagination"] | ABSTRACTThe latent dynamics model summarizes an agent’s high dimensional experiencesin a compact way. While learning from imagined trajectories by the latent model isconfirmed to have great potential to facilitate behavior learning, the lack of memorydiversity limits generalization capability. Inspired by a neuroscienc... | OQ2BfRq0SIL | Review | 5: Marginally below acceptance threshold | Summary:
The paper proposes a new method CLION - aimed at improving the performance of model-based RL from pixels on the DeepMind Control (DMControl) benchmarks. A leading model-based RL Dreamer is used as a baseline. The paper proposes an improvement over Dreamer by trying to use augmented latent states and rollouts ... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust Memory Augmentation by Constrained Latent Imagination
### Paper Abstract
The latent dynamics model summarizes an agent’s high dimensional experiences in a compact... |
BJl9ZTVKwB | ICLR.cc/2020/Conference | 2020 | MIM: Mutual Information Machine | ["Micha Livne", "Kevin Swersky", "David J. Fleet"] | We introduce the Mutual Information Machine (MIM), an autoencoder framework
for learning joint distributions over observations and latent states.
The model formulation reflects two key design principles: 1) symmetry, to encourage
the encoder and decoder to learn different factorizations of the sam... | ["Mutual Information", "Representation Learning", "Generative Models", "Probability Density Estimator"] | ABSTRACTWe introduce the Mutual Information Machine (MIM), an autoencoder frameworkfor learning joint distributions over observations and latent states. The model for-mulation reflects two key design principles: 1) symmetry, to encourage the encoderand decoder to learn different factorizations of the same underlying di... | rkxjPJvb5B | Official Blind Review #1 | 6: Weak Accept | This paper defines a new learning objective of an autoencoder framework for learning joint distributions over observations and latent states, where the objective is the joint entropy of M_s, an equally weighted mixture of the encoding and decoding distributions. Since it involves the marginal distribution which is int... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
MIM: Mutual Information Machine
### Paper Abstract
We introduce the Mutual Information Machine (MIM), an autoencoder framework for learning joint distributions o... | |
w_BtePbtmx4 | ICLR.cc/2021/Conference | 2021 | Accelerating DNN Training through Selective Localized Learning | ["Sarada Krithivasan", "Sanchari Sen", "Swagath Venkataramani", "Anand Raghunathan"] | Training Deep Neural Networks (DNNs) places immense compute requirements on the underlying hardware platforms, expending large amounts of time and energy. We proposeLoCal+SGD, a new algorithmic approach to accelerate DNN train-ing by selectively combining localized or Hebbian learning within a StochasticGradient Descen... | ["Efficient DNN Training"] | ABSTRACTTraining Deep Neural Networks (DNNs) places immense compute requirementson the underlying hardware platforms, expending large amounts of time and energy.We propose LoCal+SGD , a new algorithmic approach to accelerate DNN train-ing by selectively combining localized or Hebbian learning within a StochasticGradien... | ZaXIt7ZDPhl | A possible useful training practice, but some important experiments are missing | 6: Marginally above acceptance threshold | This paper try to leverage the benefit of Hebb learning to reduce CNN training time cost. In order to achieve this, a learning mode selection algorithm is proposed to progressively increase number of layers using Hebb learning. The writing of this paper is good and the idea is also interesting, however, the experimen... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Accelerating DNN Training through Selective Localized Learning
### Paper Abstract
Training Deep Neural Networks (DNNs) places immense compute requirements on the underl... |
Bkel6ertwS | ICLR.cc/2020/Conference | 2020 | Learning DNA folding patterns with Recurrent Neural Networks | ["Michal Rozenwald", "Aleksandra Galitsyna", "Ekaterina Khrameeva", "Grigory Sapunov", "Mikhail S. Gelfand"] |
The recent expansion of machine learning applications to molecular biology proved to have a significant contribution to our understanding of biological systems, and genome functioning in particular. Technological advances enabled the collection of large epigenetic datasets, including information about various DNA bind... | ["Machine Learning", "Recurrent Neural Networks", "3D chromatin structure", "topologically associating domains", "computational biology."] | ABSTRACTThe recent expansion of Machine Learning applications to molecular biologyproved to have a significant contribution to our understanding of biological sys-tems, and genome functioning in particular. Technological advances enabled thecollection of large epigenetic datasets, including information about various DN... | HylmAQEc_B | Official Blind Review #2 | 1: Reject | The authors compared a bidirectional LSTM to traditional machine learning models for genomic contact prediction. Although the paper presents an interesting application of ML, I vote for rejection since i) the paper is similar to previously published works and lacks methodological novelty, ii) the description about the... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning DNA folding patterns with Recurrent Neural Networks
### Paper Abstract
The recent expansion of machine learning applications to molecular biology proved to ha... | |
To4Wy2NEM2 | ICLR.cc/2021/Conference | 2021 | Adaptive Optimizers with Sparse Group Lasso | ["Yun Yue", "Suo Tong", "Zhen Zhang", "Yongchao Liu", "Chunyang Wen", "Huanjun Bao", "Jinjie Gu", "Yixiang Mu"] | We develop a novel framework that adds the regularizers to a family of adaptive optimizers in deep learning, such as MOMENTUM, ADAGRAD, ADAM, AMSGRAD, ADAHESSIAN, and create a new class of optimizers, which are named GROUP MOMENTUM, GROUP ADAGRAD, GROUP ADAM, GROUP AMSGRAD and GROUP ADAHESSIAN, etc., accordingly. We es... | ["adaptive optimizers", "sparse group lasso", "DNN models", "online optimization"] | ABSTRACTWe develop a novel framework that adds the regularizers to a family of adaptiveoptimizers in deep learning, such as M OMENTUM , ADAGRAD , ADAM , AMS-GRAD, ADAHESSIAN , and create a new class of optimizers, which are namedGROUP MOMENTUM , GROUP ADAGRAD , GROUP ADAM , GROUP AMSG RADand G ROUP ADAHESSIAN , etc., a... | IgExjU-Knjk | The paper proposes a novel framework to add sparsity regularizers to adaptive optimizers in deep learning such as Adam, Adagrad... A generic algorithm is proposed and the theoretical convergence and regret guarantees are provided. Empirical evaluations on ad click logs datasets are provided to show the effectiveness of... | 5: Marginally below acceptance threshold | Summary:
The paper proposes a novel framework to add sparsity regularizers to adaptive optimizers in deep learning such as Adam, Adagrad... A generic algorithm is proposed and the theoretical convergence and regret guarantees are provided. Empirical evaluations on ad click logs datasets are provided to show the effect... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adaptive Optimizers with Sparse Group Lasso
### Paper Abstract
We develop a novel framework that adds the regularizers to a family of adaptive optimizers in deep learnin... |
KD0M-BbuCYx | MICCAI.org/2022/Challenge/FLARE | 2022 | Semi-Supervised Detection, Identification and Segmentation for Abdominal Organs | ["Mingze Sun", "Yankai Jiang", "Heng Guo"] | Abdominal organ segmentation is an important prerequisite in many medical image analysis applications. Methods based on U-Net have demonstrated their scalability and achieved great success in different organ segmentation tasks. However, the limited number of data and labels hinders the training process of these methods... | ["Semi-supervised", "Organ detection", "Organ segmentation"] | Semi-Supervised Detection, Identification andSegmentation for Abdominal OrgansMingze Sun, Yankai Jiang, and Heng GuoAlibaba DAMO Academygh205191@alibaba-inc.comAbstract. Abdominal organ segmentation is an important prerequi-site in many medical image analysis applications. Methods based onU-Net have demonstrated their ... | 0Y2bCybzA3 | nnU-Net as the teacher model and three-stage abdominal organ segmentation frame | 9: Top 15% of accepted papers, strong accept | The authors use nnU-Net as the teacher model and design a consistency measuring strategy to generate high quality pseudo labels and propose a three-stage abdominal organ segmentation frame: (1) identify the abdomen region using a lightweight segmentation network. (2) detect different abdominal organs using DETR. (3) se... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Semi-Supervised Detection, Identification and Segmentation for Abdominal Organs
### Paper Abstract
Abdominal organ segmentation is an important prerequisite in many medi... |
rk6cfpRjZ | ICLR.cc/2018/Conference | 2018 | Learning Intrinsic Sparse Structures within Long Short-Term Memory | ["Wei Wen", "Yuxiong He", "Samyam Rajbhandari", "Minjia Zhang", "Wenhan Wang", "Fang Liu", "Bin Hu", "Yiran Chen", "Hai Li"] | Model compression is significant for the wide adoption of Recurrent Neural Networks (RNNs) in both user devices possessing limited resources and business clusters requiring quick responses to large-scale service requests. This work aims to learn structurally-sparse Long Short-Term Memory (LSTM) by reducing the sizes of... | ["Sparsity", "Model Compression", "Acceleration", "LSTMs", "Recurrent Neural Networks", "Structural Learning"] | ABSTRACTModel compression is significant for the wide adoption of Recurrent Neural Net-works (RNNs) in both user devices possessing limited resources and business clus-ters requiring quick responses to large-scale service requests. This work aims tolearn structurally-sparse Long Short-Term Memory (LSTM) by reducing the... | r14eWGtez | Review | 7: Good paper, accept | The authors propose a technique to compress LSTMs in RNNs by using a group Lasso regularizer which results in structured sparsity, by eliminating individual hidden layer inputs at a particular layer. The authors present experiments on unidirectional and bidirectional LSTM models which demonstrate the effectiveness of t... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Intrinsic Sparse Structures within Long Short-Term Memory
### Paper Abstract
Model compression is significant for the wide adoption of Recurrent Neural Networks... |
Xb8xvrtB8Ce | ICLR.cc/2021/Conference | 2021 | Bag of Tricks for Adversarial Training | ["Tianyu Pang", "Xiao Yang", "Yinpeng Dong", "Hang Su", "Jun Zhu"] | Adversarial training (AT) is one of the most effective strategies for promoting model robustness. However, recent benchmarks show that most of the proposed improvements on AT are less effective than simply early stopping the training procedure. This counter-intuitive fact motivates us to investigate the implementation ... | ["Adversarial Training", "Robustness", "Adversarial Examples"] | ABSTRACTAdversarial training (AT) is one of the most effective strategies for promotingmodel robustness. However, recent benchmarks show that most of the proposedimprovements on AT are less effective than simply early stopping the training pro-cedure. This counter-intuitive fact motivates us to investigate the implemen... | 2q69CZ2nh1C | Great systematic survey - but with limitations that should be addressed or pointed out | 7: Good paper, accept | The paper systematically reviews different hyperparameter settings and training strategies used for PGD adversarial training on CIFAR-10. Based on that, it derives practical takeaways (“bag of tricks”) and puts forward a standard baseline setting for future work.
I think this work is a valuable contribution to the res... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Bag of Tricks for Adversarial Training
### Paper Abstract
Adversarial training (AT) is one of the most effective strategies for promoting model robustness. However, rece... |
rkeJRhNYDH | ICLR.cc/2020/Conference | 2020 | TabFact: A Large-scale Dataset for Table-based Fact Verification | ["Wenhu Chen", "Hongmin Wang", "Jianshu Chen", "Yunkai Zhang", "Hong Wang", "Shiyang Li", "Xiyou Zhou", "William Yang Wang"] | The problem of verifying whether a textual hypothesis holds based on the given evidence, also known as fact verification, plays an important role in the study of natural language understanding and semantic representation. However, existing studies are mainly restricted to dealing with unstructured evidence (e.g., natur... | ["Fact Verification", "Tabular Data", "Symbolic Reasoning"] | ABSTRACTThe problem of verifying whether a textual hypothesis holds based on the givenevidence, also known as fact verification, plays an important role in the study ofnatural language understanding and semantic representation. However, existingstudies are mainly restricted to dealing with unstructured evidence (e.g., ... | r1lp3ZLT5S | Official Blind Review #4 | 8: Accept | Updated review:
Thank you for addressing the comments and making relevant edits. Additionally, the HIT section provides a lot more insights into the data collection process. I've updated my score based on the responses/edits made.
------
This paper is about a dataset (TABFACT) aimed at promoting research for fact-ver... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
TabFact: A Large-scale Dataset for Table-based Fact Verification
### Paper Abstract
The problem of verifying whether a textual hypothesis holds based on the given eviden... | |
4emQEegFhSy | ICLR.cc/2021/Conference | 2021 | Adaptive Multi-model Fusion Learning for Sparse-Reward Reinforcement Learning | ["Giseung Park", "Whiyoung Jung", "Sungho Choi", "Youngchul Sung"] | In this paper, we consider intrinsic reward generation for sparse-reward reinforcement learning based on model prediction errors. In typical model-prediction-error-based intrinsic reward generation, an agent has a learning model for the underlying environment. Then intrinsic reward is designed as the error between the ... | ["sparse-reward RL", "intrinsic reward generation", "adaptive fusion", "information geometry", "scale-free property"] | ABSTRACTIn this paper, we consider intrinsic reward generation for sparse-reward reinforce-ment learning based on model prediction errors. In typical model-prediction-error-based intrinsic reward generation, an agent has a learning model for the underlyingenvironment. Then, intrinsic reward is designed as the error bet... | R4rxPzX52-V | Contribution of the paper might be insufficient | 5: Marginally below acceptance threshold | Summary
The papers looks at the problem of using intrinsic rewards to help agent explore in sparse reward settings. The paper proposes combining multiple intrinsic rewards and proposes a meta-gradient based method to learning the fusion of these intrinsic rewards.
Strengths
1. Learning in a sparse rewards setting is ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adaptive Multi-model Fusion Learning for Sparse-Reward Reinforcement Learning
### Paper Abstract
In this paper, we consider intrinsic reward generation for sparse-reward... |
gIJfALc6Gz | EMNLP/2020/Workshop/NLP-COVID | 2020 | Expressive Interviewing: A Conversational System for Coping with COVID-19 | ["Charles Welch", "Allison Lahnala", "Veronica Perez-Rosas", "Siqi Shen", "Sarah Seraj", "Larry An", "Kenneth Resnicow", "James Pennebaker", "Rada Mihalcea"] | The ongoing COVID-19 pandemic has raised concerns for many regarding personal and public health implications, financial security and economic stability. Alongside many other unprecedented challenges, there are increasing concerns over social isolation and mental health. We introduce Expressive Interviewing -- an interv... | ["expressive interviewing", "conversational system", "motivational interviewing", "expressive writing"] | Expressive Interviewing:A Conversational System for Coping with COVID-19Charles Welch, Allison Lahnala, Ver ́onica P ́erez-Rosas, Siqi Shen, Sarah Seraj,Larry An, Kenneth Resnicowy, James Pennebaker, Rada MihalceaComputer Science & Engineering, University of MichiganDepartment of Psychology, University of TexasMedical ... | uLxABArJYcC | Interesting but needs more detail, proper analyses and method | 4: Ok but not good enough - rejection | Overall - I like the idea and the paper is well-written. The online system is great - I did it and I much appreciate the authors' effort on making it available.
I feel that some comments need addressing;
- intro paragraph: references are missing. It would also be good to look at some of thee other work on well-being,... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Expressive Interviewing: A Conversational System for Coping with COVID-19
### Paper Abstract
The ongoing COVID-19 pandemic has raised concerns for many regarding persona... |
ryE98iR5tm | ICLR.cc/2019/Conference | 2019 | Practical lossless compression with latent variables using bits back coding | ["James Townsend", "Thomas Bird", "David Barber"] | Deep latent variable models have seen recent success in many data domains. Lossless compression is an application of these models which, despite having the potential to be highly useful, has yet to be implemented in a practical manner. We present '`Bits Back with ANS' (BB-ANS), a scheme to perform lossless compression ... | ["compression", "variational auto-encoders", "deep latent gaussian models", "lossless compression", "latent variables", "approximate inference", "variational inference"] | ABSTRACTDeep latent variable models have seen recent success in many data domains.Lossless compression is an application of these models which, despite havingthe potential to be highly useful, has yet to be implemented in a practical man-ner. We present ‘Bits Back with ANS’ (BB-ANS), a scheme to perform losslesscompres... | ryeexzmp37 | Very nice paper, but with practical limitations | 8: Top 50% of accepted papers, clear accept | This paper is built on a simple but profound observation: Frey's bits-back coding algorithm can be implemented much more elegantly when replacing arithmetic coding (AC) with asymmetric numerical systems (ANS), a much more recent development not known at the time, simply due to the fact that it encodes symbols in a stac... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Practical lossless compression with latent variables using bits back coding
### Paper Abstract
Deep latent variable models have seen recent success in many data domains.... |
lWaz5a9lcFU | ICLR.cc/2021/Conference | 2021 | EEC: Learning to Encode and Regenerate Images for Continual Learning | ["Ali Ayub", "Alan Wagner"] | The two main impediments to continual learning are catastrophic forgetting and memory limitations on the storage of data. To cope with these challenges, we propose a novel, cognitively-inspired approach which trains autoencoders with Neural Style Transfer to encode and store images. Reconstructed images from encoded ep... | ["Continual Learning", "Catastrophic Forgetting", "Cognitively-inspired Learning"] | ABSTRACTThe two main impediments to continual learning are catastrophic forgetting andmemory limitations on the storage of data. To cope with these challenges, wepropose a novel, cognitively-inspired approach which trains autoencoders withNeural Style Transfer to encode and store images. During training on a new task,r... | stv-HRT9-s | Good Paper | 6: Marginally above acceptance threshold | ##########################################################################
Summary:
The paper proposes an approach which trains autoencoders with Neural Style Transfer to encode and store images. The method is applied to the problem of continual learning to overcome the catastrophic forgetting and memory limitation... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
EEC: Learning to Encode and Regenerate Images for Continual Learning
### Paper Abstract
The two main impediments to continual learning are catastrophic forgetting and me... |
2Ey_1FeNtOC | ICLR.cc/2021/Conference | 2021 | Minimum Description Length Recurrent Neural Networks | ["Nur Lan", "Emmanuel Chemla", "Roni Katzir"] | Recurrent neural networks (RNNs) face two well-known challenges: (a) the difficulty of such networks to generalize appropriately as opposed to memorizing, especially from very short input sequences (generalization); and (b) the difficulty for us to understand the knowledge that the network has attained (transparency). ... | ["recurrent neural network", "neural network", "language modeling", "minimum description length", "genetic algorithm", "semantics", "syntax"] | ABSTRACTRecurrent neural networks (RNNs) face two well-known challenges: (a) the dif-ficulty of such networks to generalize appropriately as opposed to memorizing,especially from very short input sequences (generalization); and (b) the difficultyfor us to understand the knowledge that the network has attained (transpar... | _vUn7TRwMpQ | Interesting direction, but the experimental setup needs to be corrected | 3: Clear rejection | This paper presents an approach to training recurrent neural networks with the mix objective of minimizing cross entropy (or something similar, which is not clearly defined) and minimizing minimum description length (MDL) by using a binary representation of the network which is optimized with a genetic algorithm. The m... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Minimum Description Length Recurrent Neural Networks
### Paper Abstract
Recurrent neural networks (RNNs) face two well-known challenges: (a) the difficulty of such netwo... |
S1JHhv6TW | ICLR.cc/2018/Conference | 2018 | Boosting Dilated Convolutional Networks with Mixed Tensor Decompositions | ["Nadav Cohen", "Ronen Tamari", "Amnon Shashua"] | The driving force behind deep networks is their ability to compactly represent rich classes of functions. The primary notion for formally reasoning about this phenomenon is expressive efficiency, which refers to a situation where one network must grow unfeasibly large in order to replicate functions of another. To date... | ["Deep Learning", "Expressive Efficiency", "Dilated Convolutions", "Tensor Decompositions"] | ABSTRACTThe driving force behind deep networks is their ability to compactly represent richclasses of functions. The primary notion for formally reasoning about this phe-nomenon is expressive efficiency, which refers to a situation where one networkmust grow unfeasibly large in order to replicate functions of another. ... | SyEtTPclG | Very interesting and thorough paper. | 9: Top 15% of accepted papers, strong accept | To date the theoretical advantage of deep learning has focused on the concept of "expressive efficiency" where one network must grow much larger to replicate functions that another "more efficient" network can produce. This has focused so far on depth (i.e. shallow networks have to grow much larger than deeper networks... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Boosting Dilated Convolutional Networks with Mixed Tensor Decompositions
### Paper Abstract
The driving force behind deep networks is their ability to compactly represen... |
r1lq1hRqYQ | ICLR.cc/2019/Conference | 2019 | From Language to Goals: Inverse Reinforcement Learning for Vision-Based Instruction Following | ["Justin Fu", "Anoop Korattikara", "Sergey Levine", "Sergio Guadarrama"] | Reinforcement learning is a promising framework for solving control problems, but its use in practical situations is hampered by the fact that reward functions are often difficult to engineer. Specifying goals and tasks for autonomous machines, such as robots, is a significant challenge: conventionally, reward function... | ["inverse reinforcement learning", "language grounding", "instruction following", "language-based learning"] | ABSTRACTReinforcement learning is a promising framework for solving control problems,but its use in practical situations is hampered by the fact that reward functions areoften difficult to engineer. Specifying goals and tasks for autonomous machines,such as robots, is a significant challenge: conventionally, reward fun... | SkgNkygonm | Limited technical merit and significance | 5: Marginally below acceptance threshold | This paper applies IRL to the cases of multiple tasks/environments and multimodal input features involving natural language (text) and vision (images). It is interesting to see the better performance of their proposed approaches with language-conditioned rewards over language-conditioned policies. The paper is written ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
From Language to Goals: Inverse Reinforcement Learning for Vision-Based Instruction Following
### Paper Abstract
Reinforcement learning is a promising framework for solv... |
3Wp8HM2CNdR | ICLR.cc/2021/Conference | 2021 | Whitening for Self-Supervised Representation Learning | ["Aleksandr Ermolov", "Aliaksandr Siarohin", "Enver Sangineto", "Nicu Sebe"] | Most of the self-supervised representation learning methods are based on the contrastive loss and the instance-discrimination task, where augmented versions of the same image instance ("positives") are contrasted with instances extracted from other images ("negatives"). For the learning to be effective, a lot of negat... | ["self-supervised learning", "unsupervised learning", "contrastive loss", "triplet loss", "whitening"] | ABSTRACTMost of the self-supervised representation learning methods are based on the con-trastive loss and the instance-discrimination task, where augmented versions ofthe same image instance (“positives”) are contrasted with instances extracted fromother images (“negatives”). For the learning to be effective, a lot of... | 1h488jkojc | Interesting method | 6: Marginally above acceptance threshold | ##########################################################################
Summary:
The paper provides a simple self-supervision loss function which is computed using only positive samples. In particular, the proposed loss function is based on the whitening of a latent space-features. Results show competitive but n... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Whitening for Self-Supervised Representation Learning
### Paper Abstract
Most of the self-supervised representation learning methods are based on the contrastive loss an... |
1flmvXGGJaa | ICLR.cc/2021/Conference | 2021 | NAS-Bench-301 and the Case for Surrogate Benchmarks for Neural Architecture Search | ["Julien Niklas Siems", "Lucas Zimmer", "Arber Zela", "Jovita Lukasik", "Margret Keuper", "Frank Hutter"] | The most significant barrier to the advancement of Neural Architecture Search (NAS) is its demand for large computational resources, which hinders scientifically sound empirical evaluations. As a remedy, several tabular NAS benchmarks were proposed to simulate runs of NAS methods in seconds. However, all existing tabu... | ["Neural Architecture Search", "Benchmarking", "Performance Prediction", "Deep Learning"] | ABSTRACTThe most significant barrier to the advancement of Neural Architecture Search(NAS) is its demand for large computational resources, which hinders scientifi-cally sound empirical evaluations. As a remedy, several tabular NAS benchmarkswere proposed to simulate runs of NAS methods in seconds. However, all exist-i... | kPw3Od9dRf3 | Great effort! Some remaining questions. | 8: Top 50% of accepted papers, clear accept | ****Update after rebuttal ****
I am increasing my rating for the paper as they did all the experiments I had asked for and updated the paper accordingly.
**************************
Summary:
While NAS has made tremendous advances in recent past, benchmarking algorithms with respect to each other still remains a challe... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NAS-Bench-301 and the Case for Surrogate Benchmarks for Neural Architecture Search
### Paper Abstract
The most significant barrier to the advancement of Neural Architect... |
dK8vOIBENa3 | logconference.io/LOG/2022/Conference | 2022 | Transductive Linear Probing: A Novel Framework for Few-Shot Node Classification | ["Zhen Tan", "Song Wang", "Kaize Ding", "Jundong Li", "huan liu"] | Few-shot node classification is tasked to provide accurate predictions for nodes from novel classes with only few representative labeled nodes. This problem has drawn tremendous attention for its projection to prevailing real-world applications, such as product categorization for newly added commodity categories on an ... | ["node classification", "few-shot learning", "graph neural networks", "graph contrastive learning"] | Transductive Linear Probing: A Novel Frameworkfor Few-Shot Node ClassificationZhen Tan∗Arizona State Universityztan36@asu.eduSong Wang∗University of Virginiasw3wv@virginia.eduKaize Ding∗Arizona State Universitykding9@asu.eduJundong LiUniversity of Virginiajundong@virginia.eduHuan LiuArizona State Universityhuanliu@asu.... | VEdJ1SqVdI | **Summary:**
The paper proposes a comprehensive empirical study that compares meta-learning methods to graph contrastive learning methods in the context of few-shot node classification. In particular, the authors demonstrate and argue for the superiority of representation learning methods in this context.
**Strengths:... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Transductive Linear Probing: A Novel Framework for Few-Shot Node Classification
### Paper Abstract
Few-shot node classification is tasked to provide accurate predictions... | |||
OVInxmrJ_u | NoDaLiDa/2023/Conference | 2023 | Translated Benchmarks Can Be Misleading: the Case of Estonian Question Answering | ["Hele-Andra Kuulmets", "Mark Fishel"] | Translated test datasets are a popular and cheaper alternative to native test datasets. However, one of the properties of translated data is the existence of cultural knowledge unfamiliar to the target language speakers. This can make translated test datasets differ significantly from native target datasets. As a resul... | ["benchmarks", "question-answering", "translated datasets"] | Translated Benchmarks Can Be Misleading: the Case of EstonianQuestion AnsweringHele-Andra Kuulmets Mark FishelInstitute of Computer ScienceUniversity of Tartufhele-andra.kuulmets, mark.fisel g@ut.eeAbstractTranslated test datasets are a popular andcheaper alternative to native test datasets.However, one of the properti... | _uoNNf9Tsp | The paper addresses the differences in using native vs. translated test sets in extractive Question Answering | 7: Good paper, accept | The paper discusses the effects of using a native test set or a translated test set when generating benchmarks for extractive Question Answering.
The subject can be positioned in the global context in NLP of how low resource languages can benefit from large data sets, models and methods developed for and with English ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Translated Benchmarks Can Be Misleading: the Case of Estonian Question Answering
### Paper Abstract
Translated test datasets are a popular and cheaper alternative to nat... |
SkeRTsAcYm | ICLR.cc/2019/Conference | 2019 | Phase-Aware Speech Enhancement with Deep Complex U-Net | ["Hyeong-Seok Choi", "Jang-Hyun Kim", "Jaesung Huh", "Adrian Kim", "Jung-Woo Ha", "Kyogu Lee"] | Most deep learning-based models for speech enhancement have mainly focused on estimating the magnitude of spectrogram while reusing the phase from noisy speech for reconstruction. This is due to the difficulty of estimating the phase of clean speech. To improve speech enhancement performance, we tackle the phase estima... | ["speech enhancement", "deep learning", "complex neural networks", "phase estimation"] | ABSTRACTMost deep learning-based models for speech enhancement have mainly focusedon estimating the magnitude of spectrogram while reusing the phase from noisyspeech for reconstruction. This is due to the difficulty of estimating the phaseof clean speech. To improve speech enhancement performance, we tackle thephase es... | SyxRGSbR37 | well written & rather experimental paper -- for the experts mostly | 6: Marginally above acceptance threshold | The paper is written, provides good description of the state-of-the-art and comprehensive experimental results.
The methological contribution is mild, essentially changing a buiding block in a state-of-the-art neural architecture.
The paper is for the expert audience mostly and is difficult to grasp without a good back... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Phase-Aware Speech Enhancement with Deep Complex U-Net
### Paper Abstract
Most deep learning-based models for speech enhancement have mainly focused on estimating the ma... |
cfqJY583gim | ICLR.cc/2021/Conference | 2021 | On the Benefits of Early Fusion in Multimodal Representation Learning | ["Sabera J Talukder", "George Barnum", "Yisong Yue"] | Intelligently reasoning about the world often requires integrating data from multiple modalities, as any individual modality may contain unreliable or incomplete information. Prior work in multimodal learning fuses input modalities only after significant independent processing. On the other hand, the brain performs mul... | ["Multimodal Learning", "Representations", "Noise", "Audio", "Visual", "Fusion", "Biologically Inspired"] | ABSTRACTIntelligently reasoning about the world often requires integrating data from mul-tiple modalities, as any individual modality may contain unreliable or incompleteinformation. Prior work in multimodal learning fuses input modalities only af-ter significant independent processing. On the other hand, the brain per... | R1pyPPprd_ | Official Blind Review -- AnonReviewer3 | 4: Ok but not good enough - rejection | **Summary**
In this work, the authors systematically study the role of early fusion in multi-modal learning across visual and audio perceptual tasks. The authors investigate this topic with a convolutional LSTM architecture using the MNIST image classification dataset and the Free Spoken Digit datasets. The authors fi... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Benefits of Early Fusion in Multimodal Representation Learning
### Paper Abstract
Intelligently reasoning about the world often requires integrating data from mul... |
rJ0JwFcex | ICLR.cc/2017/conference | 2017 | Neuro-Symbolic Program Synthesis | ["Emilio Parisotto", "Abdel-rahman Mohamed", "Rishabh Singh", "Lihong Li", "Dengyong Zhou", "Pushmeet Kohli"] | Recent years have seen the proposal of a number of neural architectures for the problem of Program Induction. Given a set of input-output examples, these architectures are able to learn mappings that generalize to new test inputs. While achieving impressive results, these approaches have a number of important limitatio... | ["Deep learning", "Structured prediction"] | ABSTRACTRecent years have seen the proposal of a number of neural architectures for theproblem of Program Induction. Given a set of input-output examples, these ar-chitectures are able to learn mappings that generalize to new test inputs. Whileachieving impressive results, these approaches have a number of important li... | rkpg7VPNl | Strong ideas for an important problem | 7: Good paper, accept | This paper sets out to tackle the program synthesis problem: given a set of input/output pairs discover the program that generated them. The authors propose a bipartite model, with one component that is a generative model of tree-structured programs and the other component an input/output pair encoder for conditioning.... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neuro-Symbolic Program Synthesis
### Paper Abstract
Recent years have seen the proposal of a number of neural architectures for the problem of Program Induction. Given a... |
r1gx60NKPS | ICLR.cc/2020/Conference | 2020 | JAUNE: Justified And Unified Neural language Evaluation | ["Hassan Kan\u00e9", "Yusuf Kocyigit", "Ali Abdalla", "Pelkins Ajanoh", "Mohamed Coulibali"] | We review the limitations of BLEU and ROUGE -- the most popular metrics used to assess reference summaries against hypothesis summaries, and introduce JAUNE: a set of criteria for what a good metric should behave like and propose concrete ways to use recent Transformers-based Language Models to assess reference summar... | ["NLP", "Evaluation Metrics", "Summarization", "Translation", "BLEU", "ROUGE", "Transformers"] | ABSTRACTWe review the limitations of BLEU and ROUGE – the most popular metricsused to assess reference summaries against hypothesis summaries, and introduceJAUNE: a set of criteria for what a good metric should behave like and proposeconcrete ways to use recent Transformers-based Language Models to assess ref-erence su... | HyeKi-nlcH | Official Blind Review #3 | 1: Reject |
=== Summary ===
The authors motivate the development of new (automatic) metrics to evaluate language generation by using similarity with a given reference: standard metrics like BLEU, ROUGE or METEOR have been shown to have poor correlation with human judgment on a number of tasks and are vulnerable to changes in wor... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
JAUNE: Justified And Unified Neural language Evaluation
### Paper Abstract
We review the limitations of BLEU and ROUGE -- the most popular metrics used to assess referen... | |
rJeXS04FPH | ICLR.cc/2020/Conference | 2020 | DeFINE: Deep Factorized Input Token Embeddings for Neural Sequence Modeling | ["Sachin Mehta", "Rik Koncel-Kedziorski", "Mohammad Rastegari", "Hannaneh Hajishirzi"] | For sequence models with large vocabularies, a majority of network parameters lie in the input and output layers. In this work, we describe a new method, DeFINE, for learning deep token representations efficiently. Our architecture uses a hierarchical structure with novel skip-connections which allows for the use of lo... | ["sequence modeling", "input representations", "language modeling", "word embedding"] | ABSTRACTFor sequence models with large vocabularies, a majority of network parameterslie in the input and output layers. In this work, we describe a new method,DeFINE , for learning deep token representations efficiently. Our architectureuses a hierarchical structure with novel skip-connections which allows for theuse ... | SJeIxKeYqr | Official Blind Review #4 | 6: Weak Accept | This paper describes an approach to learn word embedding functions more efficiently and with fewer parameters. This is done by replacing the embedding lookup function which is typical in NLP tasks such as language modeling and machine translation with a hierarchical embedding model. This allows for a low dimensional em... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
DeFINE: Deep Factorized Input Token Embeddings for Neural Sequence Modeling
### Paper Abstract
For sequence models with large vocabularies, a majority of network paramet... | |
BkgYPREtPr | ICLR.cc/2020/Conference | 2020 | Symplectic Recurrent Neural Networks | ["Zhengdao Chen", "Jianyu Zhang", "Martin Arjovsky", "L\u00e9on Bottou"] | We propose Symplectic Recurrent Neural Networks (SRNNs) as learning algorithms that capture the dynamics of physical systems from observed trajectories. SRNNs model the Hamiltonian function of the system by a neural networks, and leverage symplectic integration, multiple-step training and initial state optimization to ... | ["Hamiltonian systems", "learning physical laws", "symplectic integrators", "recurrent neural networks", "inverse problems"] | Published as a conference paper at ICLR 2020Symplectic Recurrent Neural NetworksZhengdao Chena;c, Jianyu Zhangb;c,Martin Arjovskya, Léon Bottouc;aaNew York University, New York, USAbTianjin University, Tianjin, ChinacFacebook AI Research, New York, USAAbstractWe propose Symplectic Recurrent Neural Networks (SRNNs) as ... | Hyg1X9VTqS | Official Blind Review #3 | 8: Accept | Summary:
The paper improves upon Hamiltonian Neural Networks to model physical systems from observed trajectories. Specifically, the authors propose to use (i) better integrator, (ii) multi-step learning, and (iii) initial state optimization. Authors experimentally show that all these improvements are beneficial in mo... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Symplectic Recurrent Neural Networks
### Paper Abstract
We propose Symplectic Recurrent Neural Networks (SRNNs) as learning algorithms that capture the dynamics of physi... | |
KrHA1GyDF4n | icaps-conference.org/ICAPS/2022/Workshop/XAIP | 2022 | Repair Suggestions for Planning Domains with Missing Actions Effects | ["Alba Gragera", "Raquel Fuentetaja", "\u00c1ngel Garc\u00eda-Olaya", "Fernando Fern\u00e1ndez"] | In Automated Planning (AP) a proper definition of the domain and problem files is assumed. However, producing complete model descriptions is a time consuming and challenging task, especially for non-experts. It is easy to make mistakes when creating formal models, turning the planning task unsolvable for the planners. ... | ["Classical Planning", "Incomplete Domains"] | Repair Suggestions for Planning Domains with Missing Actions EffectsAlba Gragera, Raquel Fuentetaja, ́Angel Garc ́ıa-Olaya, Fernando Fern ́andezUniversidad Carlos III de Madrid, Legan ́es, 28911, Madrid, Spainagragera@pa.uc3m.es, {rfuentet, agolaya, ffernand }@inf.uc3m.esAbstractIn Automated Planning (AP) a proper defi... | W-Z2OqCU3bP | An acceptable paper | 7: Good paper, accept | The paper intends to turn an unsolvable planning problem into a solvable one by adding state variables to actions' positive effects, and such changes can serve as an explanation about why the problem is unsolvable. The author solved the problem of finding such changes by reducing it to another planning problem. The red... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Repair Suggestions for Planning Domains with Missing Actions Effects
### Paper Abstract
In Automated Planning (AP) a proper definition of the domain and problem files is... |
TOzBiasXxyX | eswc-conferences.org/ESWC/2021/Conference/Resources_Track | 2021 | CSKG: The CommonSense Knowledge Graph | ["Filip Ilievski", "Pedro Szekely", "Bin Zhang"] | Sources of commonsense knowledge aim to support applications in natural language understanding, computer vision, and knowledge graphs. These sources contain complementary knowledge to each other, which makes their integration desired. Yet, such integration is not trivial because of their different foci, modeling approa... | ["Commonsense Knowledge", "Knowledge graphs", "Embeddings"] | CSKG: The CommonSense Knowledge GraphFilip Ilievski, Pedro Szekely, and Bin ZhangInformation Sciences Institute, University of Southern Californiafilievski,pszekely,binzhang g@isi.eduAbstract. Sources of commonsense knowledge support applications innatural language understanding, computer vision, and knowledge graphs.G... | Sad5w-RZtSU | Interesting paper, novel and clear, but lacking in evaluation. | 1: Weak Accept | The paper presents a "Commonsense KG" based on the merging 7 knowledge sources using 5 integration/mapping methods into a single graph. The paper is easy to read, and the topic is important and relevant to ESWC. The approach/result is novel, however a better description/comparison versus existing work would have been a... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
CSKG: The CommonSense Knowledge Graph
### Paper Abstract
Sources of commonsense knowledge aim to support applications in natural language understanding, computer vision,... |
ryQz_ZfHz | ICLR.cc/2018/Conference | 2018 | Neural Variational Sparse Topic Model | ["Anonymous"] | Effectively inferring discriminative and coherent latent topics of short texts is a critical task for many real world applications. Nevertheless, the task has been proven to be a great challenge for traditional topic models due to the data sparsity problem induced by the characteristics of short texts. Moreover, the co... | ["Variational Autoencoder", "Sparse Topical Coding", "Neural Variational Inference"] | ABSTRACTEffectively inferring discriminative and coherent latent topics of short texts is acritical task for many real world applications. Nevertheless, the task has beenproven to be a great challenge for traditional topic models due to the data spar-sity problem induced by the characteristics of short texts. Moreover,... | Bk1eUSLHG | The manuscript needs to be clarified more precisely. | 3: Clear rejection | This paper proposes a neural variational sparse topic model to model topics of short text via variational auto encoding and sparse coding.
Overall, many parts of the paper need to be further clarified or justified. I tried to list some major and minor concerns/comments below:
* 'Collapsed document code \theta': In s... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Variational Sparse Topic Model
### Paper Abstract
Effectively inferring discriminative and coherent latent topics of short texts is a critical task for many real ... |
hpZXQ6VVFxo | CUHK.edu.hk/2021/Course/IERG5350 | 2021 | Deploying Reinforcement Learning in Water Transport | ["Donald P. H. Wong", "Tsz Kui Chow"] | In this project, we deployed various types of reinforcement learning algorithms to resolves the rewards maximization problem in water sailing by reaching the destination with the highest priority using the smallest number of steps. We have implemented our own environment for this project and trained different agents us... | ["steps", "reinforcement learning", "water transport", "project", "destination", "environment", "policy iteration", "value iteration", "dqn", "various types"] | Deploying Reinforcement Learning in WaterTransportPak Hei WongDepartment of Information EngineeringThe Chinese University of Hong KongShatin, Hong Kongwph019@ie.cuhk.edu.hkTsz Kui ChowDepartment of Information EngineeringThe Chinese University of Hong KongShatin, Hong Kong1155093841@link.cuhk.edu.hkAbstractIn this proj... | l3twecQSRWS | A decent extension to the FrozenLake-v0 environment for water transport | 6: Marginally above acceptance threshold | __General__
The project simulates a worldwide water transport problem with reinforcement learning with an environment extending FrozenLake-v0. To obtain an optimal route, the authors implemented vanilla policy iteration, value iteration, and DQN algorithms. All 3 algorithms are evaluated with different combinations of... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deploying Reinforcement Learning in Water Transport
### Paper Abstract
In this project, we deployed various types of reinforcement learning algorithms to resolves the re... |
B1gJ1L2aW | ICLR.cc/2018/Conference | 2018 | Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionality | ["Xingjun Ma", "Bo Li", "Yisen Wang", "Sarah M. Erfani", "Sudanthi Wijewickrema", "Grant Schoenebeck", "Dawn Song", "Michael E. Houle", "James Bailey"] | Deep Neural Networks (DNNs) have recently been shown to be vulnerable against adversarial examples, which are carefully crafted instances that can mislead DNNs to make errors during prediction. To better understand such attacks, a characterization is needed of the properties of regions (the so-called `adversarial subsp... | ["Adversarial Subspace", "Local Intrinsic Dimensionality", "Deep Neural Networks"] | ABSTRACTDeep Neural Networks (DNNs) have recently been shown to be vulnerable againstadversarial examples, which are carefully crafted instances that can mislead DNNsto make errors during prediction. To better understand such attacks, a characteri-zation is needed of the properties of regions (the so-called ‘adversaria... | S1tVnWqxM | Intuitive solution to important problem; well written | 7: Good paper, accept | The authors clearly describe the problem being addressed in the manuscript and motivate their solution very clearly. The proposed solution seems very intuitive and the empirical evaluations demonstrates its utility. My main concern is the underlying assumption (if I understand correctly) that the adversarial attack tec... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionality
### Paper Abstract
Deep Neural Networks (DNNs) have recently been shown to be vulnerable agains... |
HJe88xBKPr | ICLR.cc/2020/Conference | 2020 | Mixed Precision Training With 8-bit Floating Point | ["Naveen Mellempudi", "Sudarshan Srinivasan", "Dipankar Das", "Bharat Kaul"] | Reduced precision computation is one of the key areas addressing the widening’compute gap’, driven by an exponential growth in deep learning applications. In recent years, deep neural network training has largely migrated to 16-bit precision,with significant gains in performance and energy efficiency. However, attempts... | ["8-bit training", "8-bit floating point", "low precision training", "deep learning"] | ABSTRACTReduced precision computation is one of the key areas addressing the widening’compute gap’, driven by an exponential growth in deep learning applications. Inrecent years, deep neural network training has largely migrated to 16-bit precision,with significant gains in performance and energy efficiency. However, a... | SyxwlXOjcH | Official Blind Review #1 | 6: Weak Accept | Originality: The paper proposed a new scaling loss strategy for mixed-precision (8-bit mainly) training and verified the importance of rounding (quantization) error issue for low-precision training.
Quality: The authors clearly illustrated the benefit of their proposed loss strategy and the importance of quantization... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Mixed Precision Training With 8-bit Floating Point
### Paper Abstract
Reduced precision computation is one of the key areas addressing the widening’compute gap’, driven ... | |
HJg3Rp4FwH | ICLR.cc/2020/Conference | 2020 | Policy Optimization In the Face of Uncertainty | ["Tung-Long Vuong", "Han Nguyen", "Hai Pham", "Kenneth Tran"] | Model-based reinforcement learning has the potential to be more sample efficient than model-free approaches. However, existing model-based methods are vulnerable to model bias, which leads to poor generalization and asymptotic performance compared to model-free counterparts. In this paper, we propose a novel policy opt... | ["Reinforcement Learning", "Model-based Reinforcement Learning"] | ABSTRACTModel-based reinforcement learning has the potential to be more sample efficientthan model-free approaches. However, existing model-based methods are vulnera-ble to model bias, which leads to poor generalization and asymptotic performancecompared to model-free counterparts. In this paper, we propose a novel pol... | Byg_BlmItS | Official Blind Review #2 | 3: Weak Reject | I enjoyed this paper overall, and I think the idea is a good one. However there remain significant issues with the paper that preclude me giving a good score. Firstly, there is almost no discussion of the environment model. Anyone who has worked on Model Based RL will tell you that the details here are crucial. This de... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Policy Optimization In the Face of Uncertainty
### Paper Abstract
Model-based reinforcement learning has the potential to be more sample efficient than model-free approa... | |
B7v4QMR6Z9w | ICLR.cc/2021/Conference | 2021 | Federated Learning Based on Dynamic Regularization | ["Durmus Alp Emre Acar", "Yue Zhao", "Ramon Matas", "Matthew Mattina", "Paul Whatmough", "Venkatesh Saligrama"] | We propose a novel federated learning method for distributively training neural network models, where the server orchestrates cooperation between a subset of randomly chosen devices in each round. We view Federated Learning problem primarily from a communication perspective and allow more device level computations to s... | ["Federated Learning", "Deep Neural Networks", "Distributed Optimization"] | ABSTRACTWe propose a novel federated learning method for distributively training neuralnetwork models, where the server orchestrates cooperation between a subset ofrandomly chosen devices in each round. We view Federated Learning problemprimarily from a communication perspective and allow more device level computa-tion... | 4J9CEtjWwj | Interesting new algorithm for distributed optimization / federated learning | 7: Good paper, accept | * Context: the authors propose a new distributed optimization algorithm, called FedDyn, to minimize a sum of smooth functions. Their motivation is the context of federated learning, in which each function is the loss of one user corresponding to its data stored locally. The goal is to reduce the communication burden to... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Federated Learning Based on Dynamic Regularization
### Paper Abstract
We propose a novel federated learning method for distributively training neural network models, whe... |
Sk_P2Q9sG | MIDL.amsterdam/2018/Conference | 2018 | Uncertainty quantification using Bayesian neural networks in classification: Application to ischemic stroke lesion segmentation | ["Yongchan Kwon", "Joong-Ho Won", "Beom Joon Kim", "Myunghee Cho Paik"] | Most recent research of neural networks in the field of computer vision has focused on improving accuracy of point predictions by developing various network architectures or learning algorithms. Uncertainty quantification accompanied by point estimation can lead to a more informed decision, and the quality of predictio... | ["Uncertainty quantification", "Bayesian neural network", "Ischemic stroke lesion segmentation", "Probabilistic interpretation"] | Uncertainty quantification using Bayesian neuralnetworks in classification: Application to ischemicstroke lesion segmentationYongchan KwonDepartment of StatisticsSeoul National Universityykwon0407@snu.ac.krJoong-Ho WonDepartment of StatisticsSeoul National Universitywonj@stats.snu.ac.krBeom Joon KimDepartment of Neurol... | SyB1Eg-Cz | In this paper the authors build on the work of Kendall and Gal for quantifying the aleatoric and epistemic uncertainty for each pixel, in segmentation problems. They improve upon that work by avoiding to learn extra components to model the variance by exploiting that the variance of a multinomial is a function of its m... | 3: Marginally above acceptance threshold | The paper is well written, the literature related to the paper is thoroughly described, considering the limited amount of space available in the paper. The experiments give some empirical evidences favoring the proposed model, although it is difficult to compare.
It would have been nice to see a couple of plots for al... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Uncertainty quantification using Bayesian neural networks in classification: Application to ischemic stroke lesion segmentation
### Paper Abstract
Most recent research o... |
pAbm1qfheGk | ICLR.cc/2021/Conference | 2021 | Learning Neural Generative Dynamics for Molecular Conformation Generation | ["Minkai Xu", "Shitong Luo", "Yoshua Bengio", "Jian Peng", "Jian Tang"] | We study how to generate molecule conformations (i.e., 3D structures) from a molecular graph. Traditional methods, such as molecular dynamics, sample conformations via computationally expensive simulations. Recently, machine learning methods have shown great potential by training on a large collection of conformation d... | ["Molecular conformation generation", "deep generative models", "continuous normalizing flow", "energy-based models"] | ABSTRACTWe study how to generate molecule conformations ( i.e., 3D structures) from amolecular graph. Traditional methods, such as molecular dynamics, sample con-formations via computationally expensive simulations. Recently, machine learn-ing methods have shown great potential by training on a large collection of con-... | LnQoFGQTj5y | Interesting new generative model for molecular conformations with strong results; Some questions on evaluation | 6: Marginally above acceptance threshold | This paper presents an approach to generate diverse small molecule conformations given its graph by combining a conditional flow-based model with an energy-based model. Sampling is performed in two separate stages: 1) a normalizing flow produces a distribution over interatomic distances (which is then postprocessed int... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Neural Generative Dynamics for Molecular Conformation Generation
### Paper Abstract
We study how to generate molecule conformations (i.e., 3D structures) from a... |
0ciyfd4SvbI | ICML.cc/2021/Workshop/AutoML | 2021 | Latency-Aware Neural Architecture Search with Multi-Objective Bayesian Optimization | ["David Eriksson", "Pierce I-Jen Chuang", "Samuel Daulton", "Peng Xia", "Akshat Shrivastava", "Arun Babu", "Shicong Zhao", "Ahmed A Aly", "Ganesh Venkatesh", "Maximilian Balandat"] | When tuning the architecture and hyperparameters of large machine learning models for on-device deployment, it is desirable to understand the optimal trade-offs between on-device latency and model accuracy. In this work, we leverage recent methodological advances in Bayesian optimization over high-dimensional search s... | ["Bayesian Optimization", "Gaussian Process", "AutoML", "Natural Language Understanding"] | 8th ICML Workshop on Automated Machine Learning (2021)Latency-Aware Neural Architecture Search withMulti-Objective Bayesian OptimizationDavid Erikssonderiksson@fb.comFacebookPierce I-Jen Chuangpichuang@fb.comFacebookSamuel Daultonsdaulton@fb.comFacebookPeng Xia pengxia@fb.comFacebookAkshat Shrivastava akshats@fb.comFac... | Xk4DQofO9-x | Nice Application, Incremental Technical Contribution, Limited Experiment | 6: Marginally above acceptance threshold | The paper presents a Bayesian optimization approach to optimize the architecture and hyperparameters of a ML model for natural language understanding. A focus of this paper is to perform multi-objective optimization to have high accuracy and high latency property for on-device deployment.
Strengths:
Interesting applic... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Latency-Aware Neural Architecture Search with Multi-Objective Bayesian Optimization
### Paper Abstract
When tuning the architecture and hyperparameters of large machine ... |
ryeoxnRqKQ | ICLR.cc/2019/Conference | 2019 | NATTACK: A STRONG AND UNIVERSAL GAUSSIAN BLACK-BOX ADVERSARIAL ATTACK | ["Yandong Li", "Lijun Li", "Liqiang Wang", "Tong Zhang", "Boqing Gong"] | Recent works find that DNNs are vulnerable to adversarial examples, whose changes from the benign ones are imperceptible and yet lead DNNs to make wrong predictions. One can find various adversarial examples for the same input to a DNN using different attack methods. In other words, there is a population of adversaria... | ["adversarial attack", "black-box", "evolutional strategy", "policy gradient"] | ABSTRACTRecent works find that DNNs are vulnerable to adversarial examples, whosechanges from the benign ones are imperceptible and yet lead DNNs to makewrong predictions. One can find various adversarial examples for the same in-put to a DNN using different attack methods. In other words, there is a populationof adver... | S1g3Hhj_hm | NATTACK: A STRONG AND UNIVERSAL GAUSSIAN BLACK-BOX ADVERSARIAL ATTACK | 4: Ok but not good enough - rejection | Summary: In this paper the authors discuss a black-box method to learn
adversarial inputs to DNNs which are "close" to some nominal example
but nevertheless get misclassified. The algorithm essentially tries to
learn the mean of a joint Gaussian distribution over image
perturbations so that the perturbed image has high... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NATTACK: A STRONG AND UNIVERSAL GAUSSIAN BLACK-BOX ADVERSARIAL ATTACK
### Paper Abstract
Recent works find that DNNs are vulnerable to adversarial examples, whose chang... |
B1GOWV5eg | ICLR.cc/2017/conference | 2017 | Learning to Repeat: Fine Grained Action Repetition for Deep Reinforcement Learning | ["Sahil Sharma", "Aravind S. Lakshminarayanan", "Balaraman Ravindran"] | Reinforcement Learning algorithms can learn complex behavioral patterns for sequential decision making tasks wherein an agent interacts with an environment and acquires feedback in the form of rewards sampled from it. Traditionally, such algorithms make decisions, i.e., select actions to execute, at every single time s... | ["Deep learning", "Reinforcement Learning"] | ABSTRACTReinforcement Learning algorithms can learn complex behavioral patterns for se-quential decision making tasks wherein an agent interacts with an environmentand acquires feedback in the form of rewards sampled from it. Traditionally, suchalgorithms make decisions, i.e., select actions to execute, at every single... | SkazJMfVg | Review | 7: Good paper, accept | This paper provides a simple method to handle action repetitions. They make the action a tuple (a,x), where a is the action chosen, and x the number of repetitions. Overall they report some improvements over A3C/DDPG, dramatic in some games, moderate in other. The idea seems natural and there is a wealth of experiment ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Repeat: Fine Grained Action Repetition for Deep Reinforcement Learning
### Paper Abstract
Reinforcement Learning algorithms can learn complex behavioral patt... |
B1lOzpVtDB | ICLR.cc/2020/Conference | 2020 | Conversation Generation with Concept Flow | ["Houyu Zhang", "Zhenghao Liu", "Chenyan Xiong", "Zhiyuan Liu"] | Human conversations naturally evolve around related entities and connected concepts, while may also shift from topic to topic. This paper presents ConceptFlow, which leverages commonsense knowledge graphs to explicitly model such conversation flows for better conversation response generation. ConceptFlow grounds the co... | ["Grounded Natural Language Generation", "Conversation Response Generation", "ConceptNet", "ConceptFlow", "Conversation Flow"] | ABSTRACTHuman conversations naturally evolve around related entities and connected con-cepts, while may also shift from topic to topic. This paper presents ConceptFlow,which leverages commonsense knowledge graphs to explicitly model such conver-sation flows for better conversation response generation. ConceptFlow groun... | rygX5gdiFr | Official Blind Review #1 | 6: Weak Accept | The paper proposes a system for generating a single-turn response to a posted utterance in an open-domain dialog setting. It follows but improves on a previous work that leverages commonsense in generating the responses. The specific proposal is to extract keywords from the incoming utterance, match them to concepts ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Conversation Generation with Concept Flow
### Paper Abstract
Human conversations naturally evolve around related entities and connected concepts, while may also shift fr... | |
SJiFvr9el | ICLR.cc/2017/conference | 2017 | Linear Time Complexity Deep Fourier Scattering Network and Extension to Nonlinear Invariants | ["Randall Balestriero", "Herve Glotin"] | In this paper we propose a scalable version of a state-of-the-art deterministic time-
invariant feature extraction approach based on consecutive changes of basis and
nonlinearities, namely, the scattering network. The first focus of the paper is to
extend the scattering network to allow the use of higher order nonlinea... | ["Unsupervised Learning", "Applications", "Deep learning"] | ABSTRACTIn this paper we propose a scalable version of a state-of-the-art deterministic time-invariant feature extraction approach based on consecutive changes of basis andnonlinearities, namely, the scattering network. The first focus of the paper is toextend the scattering network to allow the use of higher order non... | B1wW-ZQEx | 5: Marginally below acceptance threshold | This work proposes 3 improvements to scattering networks: (1) a non-linearity that allows Fourier-domain computation, (2) compact-supported (in the Fourier domain) representations, and (3) computing additional variance features.
The technical contributions seem worthwhile, since #1 and #2 may result in better speed, w... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Linear Time Complexity Deep Fourier Scattering Network and Extension to Nonlinear Invariants
### Paper Abstract
In this paper we propose a scalable version of a state-of... | |
BJlA6eBtvH | ICLR.cc/2020/Conference | 2020 | Differentiable Hebbian Consolidation for Continual Learning | ["Vithursan Thangarasa", "Thomas Miconi", "Graham W. Taylor"] | Continual learning is the problem of sequentially learning new tasks or knowledge while protecting previously acquired knowledge. However, catastrophic forgetting poses a grand challenge for neural networks performing such learning process. Thus, neural networks that are deployed in the real world often struggle in sce... | ["continual learning", "catastrophic forgetting", "Hebbian learning", "synaptic plasticity", "neural networks"] | ABSTRACTContinual learning is the problem of sequentially learning new tasks or knowledgewhile protecting previously acquired knowledge. However, catastrophic forgettingposes a grand challenge for neural networks performing such learning process.Thus, neural networks that are deployed in the real world often struggle i... | r1xvrhmUdr | Official Blind Review #1 | 3: Weak Reject | This paper addresses the continual learning setting, and aims to mitigate catastrophic forgetting, with results on Permuted MNIST, Split MNIST, Vision Datasets Mixture, and their own class-imbalanced version of the Permuted MNIST dataset. The authors propose to augment differentiable plastic weights - a general neural ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Differentiable Hebbian Consolidation for Continual Learning
### Paper Abstract
Continual learning is the problem of sequentially learning new tasks or knowledge while pr... | |
r87xPSd89gq | ICLR.cc/2022/Workshop/OSC | 2022 | Invariant Causal Representation Learning for Generalization in Imitation and Reinforcement Learning | ["Chaochao Lu", "Jos\u00e9 Miguel Hern\u00e1ndez-Lobato", "Bernhard Sch\u00f6lkopf"] | A fundamental challenge in imitation and reinforcement learning is to learn policies, representations, or dynamics that do not build on spurious correlations and generalize beyond the specific environments that they were trained on. We investigate these generalization problems from a unified view. For this, we propose ... | ["imitation", "representations", "dynamics", "generalization", "policies", "invariant causal representation", "reinforcement", "fundamental challenge", "reinforcement learning"] | ABSTRACTA fundamental challenge in imitation and reinforcement learning is to learn poli-cies, representations, or dynamics that do not build on spurious correlations andgeneralize beyond the specific environments that they were trained on. We inves-tigate these generalization problems from a unified view. For this, we... | SpzxRdXMlMc | Review | 2: Good workshop paper, accept | This paper proposes an approach to learn an invariant casual representation to improve generalization in both imitation and reinforcement learning. The paper is well positioned for this workshop, and provides extensive theoretical support for results
I unfortunately do not have much background on casual representation... | 1: The reviewer's evaluation is an educated guess | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Invariant Causal Representation Learning for Generalization in Imitation and Reinforcement Learning
### Paper Abstract
A fundamental challenge in imitation and reinforce... |
CZ8Y3NzuVzO | ICLR.cc/2021/Conference | 2021 | What Should Not Be Contrastive in Contrastive Learning | ["Tete Xiao", "Xiaolong Wang", "Alexei A Efros", "Trevor Darrell"] | Recent self-supervised contrastive methods have been able to produce impressive transferable visual representations by learning to be invariant to different data augmentations. However, these methods implicitly assume a particular set of representational invariances (e.g., invariance to color), and can perform poorly w... | ["Self-supervised learning", "Contrastive learning", "Representation learning"] | ABSTRACTRecent self-supervised contrastive methods have been able to produce impres-sive transferable visual representations by learning to be invariant to different dataaugmentations. However, these methods implicitly assume a particular set of rep-resentational invariances (e.g., invariance to color), and can perform... | Y-rfIIEWV6Y | interesting findings regarding augmentation vs embedding subspaces. | 6: Marginally above acceptance threshold | The current mainstream way of doing contrastive learning in Instance discrimination is to train the network to associate two independently augmented versions of the same image. The augmentation process often consists of several stages such as cropping, blurring, etc.
The authors argue that some augmentation (such as ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
What Should Not Be Contrastive in Contrastive Learning
### Paper Abstract
Recent self-supervised contrastive methods have been able to produce impressive transferable vi... |
yxafu6ZtUux | ICLR.cc/2021/Conference | 2021 | AN ONLINE SEQUENTIAL TEST FOR QUALITATIVE TREATMENT EFFECTS | ["Chengchun Shi", "Shikai Luo", "Rui Song", "Hongtu Zhu"] | Tech companies (e.g., Google or Facebook) often use randomized online experiments and/or A/B testing primarily based on the average treatment effects to compare their new product with an old one. However, it is also critically important to detect qualitative treatment effects such that the new one may significantly out... | ["sequential testing", "A/B testing", "qualitative treatment effects", "bootstrap"] | ABSTRACTTech companies (e.g., Google or Facebook) often use randomized online experi-ments and/or A/B testing primarily based on the average treatment effects to com-pare their new product with an old one. However, it is also critically importantto detect qualitative treatment effects such that the new one may signific... | fJZDfhxUN2 | Official Blind Review #1 | 6: Marginally above acceptance threshold | This paper proposed a powerful online sequential test which can efficiently detect qualitative treatment effects (QTE). The test algorithm involves adaptive randomization, sequential monitoring and online updating. Theoretical guarantee on the Type-I error is presented.
Overall, the paper is well-written, with a clea... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
AN ONLINE SEQUENTIAL TEST FOR QUALITATIVE TREATMENT EFFECTS
### Paper Abstract
Tech companies (e.g., Google or Facebook) often use randomized online experiments and/or A... |
rkxdexBYPB | ICLR.cc/2020/Conference | 2020 | Group-Transformer: Towards A Lightweight Character-level Language Model | ["Sungrae Park", "Geewook Kim", "Junyeop Lee", "Junbum Cha", "Ji-Hoon Kim Hwalsuk Lee"] | Character-level language modeling is an essential but challenging task in Natural Language Processing.
Prior works have focused on identifying long-term dependencies between characters and have built deeper and wider networks for better performance. However, their models require substantial computational resources, wh... | ["Transformer", "Lightweight model", "Language Modeling", "Character-level language modeling"] | ABSTRACTCharacter-level language modeling based on Transformer has brought greatsuccess by alleviating limitation of recursive operation. However, existingTransformer-based models require substantial computational resources, whichhinders the usability of character-level language models in applications with lim-ited res... | B1e_weqbKH | Official Blind Review #1 | 6: Weak Accept | Summary: This paper proposes a lightweight alternative to the design of self-attention based Transformers on character-level language modeling (LM). The approach was motivated by the similar technique that has been applied on group convolutions, but with a few notable differences too, such as inter-group mixing and low... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Group-Transformer: Towards A Lightweight Character-level Language Model
### Paper Abstract
Character-level language modeling is an essential but challenging task in Natu... | |
sCZbhBvqQaU | ICLR.cc/2021/Conference | 2021 | Robust Reinforcement Learning on State Observations with Learned Optimal Adversary | ["Huan Zhang", "Hongge Chen", "Duane S Boning", "Cho-Jui Hsieh"] | We study the robustness of reinforcement learning (RL) with adversarially perturbed state observations, which aligns with the setting of many adversarial attacks to deep reinforcement learning (DRL) and is also important for rolling out real-world RL agent under unpredictable sensing noise. With a fixed agent policy, w... | ["reinforcement learning", "robustness", "adversarial attacks", "adversarial defense"] | ABSTRACTWe study the robustness of reinforcement learning (RL) with adversarially per-turbed state observations, which aligns with the setting of many adversarial at-tacks to deep reinforcement learning (DRL) and is also important for rolling outreal-world RL agent under unpredictable sensing noise. With a fixed agent ... | oaVa5eMhyic | Official Blind Review #2 | 7: Good paper, accept | Summary of the paper: The paper studies adversarial attacks in RL, focusing both on the design of optimal attack strategies on RL agents, as well as robust training RL procedures for mitigating attacks. Building on the results of (Zhang et al., 2020), the paper proposes a new learning framework (ATLA), that simultaneo... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust Reinforcement Learning on State Observations with Learned Optimal Adversary
### Paper Abstract
We study the robustness of reinforcement learning (RL) with adversa... |
RJhC_uIwdsj | NeurIPS.cc/2022/Workshop/SyntheticData4ML | 2022 | Visual Pre-training for Navigation: What Can We Learn from Noise? | ["Yanwei Wang", "Ching-Yun Ko", "Pulkit Agrawal"] | In visual navigation, one powerful paradigm is to predict actions from observations directly. Training such an end-to-end system allows representations that are useful for downstream tasks to emerge automatically. However, the lack of inductive bias makes this system data-hungry. We hypothesize a sufficient representat... | ["Synthetic Noise Data", "Self-Supervised Learning", "Embodied AI", "Visual Navigation"] | Visual Pre-training for Navigation:What Can We Learn from Noise?Yanwei WangMIT CSAILyanwei@mit.eduChing-Yun KoMIT RLEcyko@mit.eduPulkit AgrawalMIT CSAILpulkitag@mit.eduFigure 1: A rollout of our navigation policy pre-trained with random crop prediction onsynthetic noise images . Given a goal image and the current view,... | LYNqIFveErN | Nice contribution to visual navigation | 7: Good paper, accept | This paper proposes to train one aspect (the image encoder) of a visual navigation system using artificial data. The system learns to move the modeled camera to zoom in on a given crop from the image.
PROS:
The method is simple but works well.
The curriculum experiments are insightful
Many ablation experiments off... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Visual Pre-training for Navigation: What Can We Learn from Noise?
### Paper Abstract
In visual navigation, one powerful paradigm is to predict actions from observations ... |
S1NnlZnjG | MIDL.amsterdam/2018/Conference | 2018 | Iterative fully convolutional neural networks for automatic vertebra segmentation | ["Nikolas Lessmann", "Bram van Ginneken", "Pim A. de Jong", "Ivana Isgum"] | Precise segmentation of the vertebrae is often required for automatic detection of vertebral abnormalities. This especially enables incidental detection of abnormalities such as compression fractures in images that were acquired for other diagnostic purposes. While many CT and MR scans of the chest and abdomen cover a ... | ["convolutional neural networks", "vertebrae", "first", "instance memory", "vertebra", "iterative", "automatic detection"] | Iterative fully convolutional neural networks forautomatic vertebra segmentationNikolas LessmannImage Sciences InstituteUniversity Medical Center UtrechtBram van GinnekenDiagnostic Image Analysis GroupRadboud University Medical CenterPim A. de JongDepartment of RadiologyUniversity Medical Center UtrechtIvana IšgumImag... | r1MnKmApf | The authors propose an interesting iterative instance segmentation framework based on FCN for the successive (bottom-up/top-down) segmentation of vertebrae in CT scans. A memory module is integrated into the original U-Net architecture to remember the vertebras (i.e., instances) have been segmented and move forward to ... | 3: Marginally above acceptance threshold | The authors extend the original U-Net architecture by including a memory module for successive instance segmentation as well as an additional contracting path for the identification of the visibility (i.e., completely vs. partially visible) of each instance.
However, the practical significance of the latter classificat... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Iterative fully convolutional neural networks for automatic vertebra segmentation
### Paper Abstract
Precise segmentation of the vertebrae is often required for automati... |
r1xI-gHFDH | ICLR.cc/2020/Conference | 2020 | How can we generalise learning distributed representations of graphs? | ["Paul M Scherer", "Pietro Lio"] | We propose a general framework to construct unsupervised models capable of learning distributed representations of discrete structures such as graphs based on R-Convolution kernels and distributed semantics research. Our framework combines the insights and observations of Deep Graph Kernels and Graph2Vec towards a unif... | ["graphs", "distributed representations", "similarity learning"] | ABSTRACTThis paper consists of two parts. Firstly, we specify a general framework forconstructing systems capable of learning distributed representations of graphsin an unsupervised manner. We show how systems such as Deep Graph Ker-nels, Graph2Vec and Anonymous Walk Embeddings can be formulated under thisframework. In... | BkxnW6eCFS | Official Blind Review #3 | 3: Weak Reject | The paper presents an unsupervised method for graph embedding.
Despite having good experimental results, the paper is not of the quality to be accepted to the conference yet. The approach is rather a mix of previous works and hence not novel.
In particular, the algorithm for WL decomposition is almost fully taken f... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
How can we generalise learning distributed representations of graphs?
### Paper Abstract
We propose a general framework to construct unsupervised models capable of learn... | |
ry80wMW0W | ICLR.cc/2018/Conference | 2018 | Hierarchical Subtask Discovery with Non-Negative Matrix Factorization | ["Adam C. Earle", "Andrew M. Saxe", "Benjamin Rosman"] | Hierarchical reinforcement learning methods offer a powerful means of planning flexible behavior in complicated domains. However, learning an appropriate hierarchical decomposition of a domain into subtasks remains a substantial challenge. We present a novel algorithm for subtask discovery, based on the recently introd... | ["Reinforcement Learning", "Hierarchy", "Subtask Discovery", "Linear Markov Decision Process"] | ABSTRACTHierarchical reinforcement learning methods offer a powerful means of planningflexible behavior in complicated domains. However, learning an appropriate hierar-chical decomposition of a domain into subtasks remains a substantial challenge. Wepresent a novel algorithm for subtask discovery, based on the recently... | BkITkWpZM | Extends previous work that considered multitask learning in RL and proposed a hierarchical learner based on concurrent execution of many actions in parallel. Current work proposes an autonomous way to build the hierarchy and discover subtasks. | 7: Good paper, accept | The present paper extends a previous work by Saxe et al (2017) that considered multitask learning in RL and proposed a hierarchical learner based on concurrent execution of many actions in parallel. That framework made heavy use of the framework of linearly solvable Markov decision process (LMDP) proposed by Todorov, w... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Hierarchical Subtask Discovery with Non-Negative Matrix Factorization
### Paper Abstract
Hierarchical reinforcement learning methods offer a powerful means of planning f... |
aa0705s2Qc | ICLR.cc/2021/Conference | 2021 | Measuring Visual Generalization in Continuous Control from Pixels | ["Jake Grigsby", "Yanjun Jane Qi"] | Self-supervised learning and data augmentation have significantly reduced the performance gap between state and image-based reinforcement learning agents in continuous control tasks. However, it is still unclear whether current techniques can face the variety of visual conditions required by real-world environments. We... | ["Continuous Control", "Reinforcement Learning"] | Under review as a conference paper at ICLR 2021Measuring Visual Generalization inContinuous Control from PixelsAnonymous authorsPaper under double-blind reviewAbstractSelf-supervised learning and data augmentation have signicantly reducedthe performance gap between state and image-based reinforcement learn-ing agents i... | nv3IM_jTTO | Official Blind Review #2 | 6: Marginally above acceptance threshold |
This paper extends the DeepMind Control Suite benchmark by adding a series of visual variations across different tasks (e.g. lighting, color, background, textures, camera angles etc.). They also contrast a few recent self-supervised learning and data augmentation RL methods and measure these agents’ ability to general... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Measuring Visual Generalization in Continuous Control from Pixels
### Paper Abstract
Self-supervised learning and data augmentation have significantly reduced the perfor... |
r1e30AEKPr | ICLR.cc/2020/Conference | 2020 | A Group-Theoretic Framework for Knowledge Graph Embedding | ["Tong Yang", "Long Sha", "Pengyu Hong"] | We have rigorously proved the existence of a group algebraic structure hidden in relational knowledge embedding problems, which suggests that a group-based embedding framework is essential for model design. Our theoretical analysis explores merely the intrinsic property of the embedding problem itself without introduci... | ["group theory", "knowledge graph embedding", "representation learning"] | ABSTRACTWe have rigorously proved the existence of a group algebraic structure hiddenin relational knowledge embedding problems, which suggests that a group-basedembedding framework is essential for designing embedding models. Our theoreti-cal analysis explores merely the intrinsic property of the embedding problem its... | SylHMCB0FH | Official Blind Review #1 | 3: Weak Reject | This paper start merely by studying the graph reconstruction problem and prove that the intrinsic structure of this task itself automatically produces the complete definition of groups. it seems to be a novel result. Based on this result, one could construct embedding models that naturally accommodate all possible loca... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Group-Theoretic Framework for Knowledge Graph Embedding
### Paper Abstract
We have rigorously proved the existence of a group algebraic structure hidden in relational ... | |
dFBRrTMjlyL | ICLR.cc/2021/Conference | 2021 | Bidirectionally Self-Normalizing Neural Networks | ["Yao Lu", "Stephen Gould", "Thalaiyasingam Ajanthan"] | The problem of exploding and vanishing gradients has been a long-standing obstacle that hinders the effective training of neural networks. Despite various tricks and techniques that have been employed to alleviate the problem in practice, there still lacks satisfactory theories or provable solutions. In this paper, we ... | ["neural networks", "problem", "practice", "gradients", "obstacle", "effective training", "various tricks", "techniques", "satisfactory theories", "provable solutions"] | ABSTRACTThe problem of exploding and vanishing gradients has been a long-standing ob-stacle that hinders the effective training of neural networks. Despite various tricksand techniques that have been employed to alleviate the problem in practice, therestill lacks satisfactory theories or provable solutions. In this pap... | 6YbssJTKZiT | Good idea and theory; substandard experimental part. | 4: Ok but not good enough - rejection | Summary: The authors introduce a variant
of self-normalizing networks (SNNs) that also maintain
the norm of the errors during backpropagation. The presented
Bidirectional self-normalizing networks (BSNN)
have interesting theoretical properties, but are hardly
assessed empirically. There are severe flaws in
the method... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Bidirectionally Self-Normalizing Neural Networks
### Paper Abstract
The problem of exploding and vanishing gradients has been a long-standing obstacle that hinders the e... |
w6p7UMtf-0S | ICLR.cc/2021/Conference | 2021 | Improving Few-Shot Visual Classification with Unlabelled Examples | ["Peyman Bateni", "Jarred Barber", "Jan-Willem van de Meent", "Frank Wood"] | We propose a transductive meta-learning method that uses unlabelled instances to improve few-shot image classification performance. Our approach combines a regularized Mahalanobis-distance-based soft k-means clustering procedure with a modified state of the art neural adaptive feature extractor to achieve improved test... | ["Meta-learning", "Few-shot image classification", "Transductive few-shot learning"] | ABSTRACTWe propose a transductive meta-learning method that uses unlabelled instancesto improve few-shot image classification performance. Our approach combines aregularized Mahalanobis-distance-based soft k-means clustering procedure witha modified state of the art neural adaptive feature extractor to achieve improved... | kdgRWoJTnoV | This paper proposes a transductive few-shot classification method on the basis of the simple Conditional Neural Adaptive Processes (CNAPS) introduced by Bateni et al. The proposed method, called transductive CNAPS, extends the simple CNAPS by exploiting the query set both in the feature adaptation stage and classificat... | 6: Marginally above acceptance threshold | Pros:
1) The paper is well-motivated, clearly written and neatly organized. I like the writing style of this paper: explaining very clearly what the improvements are and how they perform with respect to the counterpart methods.
2) The ablation study validates the effectiveness of the improved adaptation of the featu... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Improving Few-Shot Visual Classification with Unlabelled Examples
### Paper Abstract
We propose a transductive meta-learning method that uses unlabelled instances to imp... |
rkeeoeHYvr | ICLR.cc/2020/Conference | 2020 | AdvCodec: Towards A Unified Framework for Adversarial Text Generation | ["Boxin Wang", "Hengzhi Pei", "Han Liu", "Bo Li"] | Machine learning (ML) especially deep neural networks (DNNs) have been widely applied to real-world applications. However, recent studies show that DNNs are vulnerable to carefully crafted \emph{adversarial examples} which only deviate from the original data by a small magnitude of perturbation.
While there has been ... | ["adversarial text generation", "tree-autoencoder", "human evaluation"] | ABSTRACTWhile there has been great interest on generating imperceptible adversarial ex-amples in continuous data domain (e.g. image and audio) to explore the modelvulnerabilities, generating adversarial text in the discrete domain is still challeng-ing. The main contribution of this paper is to propose a general target... | Byg5OXlgcr | Official Blind Review #1 | 3: Weak Reject | Motivated by recent development of attack/defense methods addressing the vulnerability of deep CNN classifiers for images, this paper proposes an attack framework for adversarial text generation, in which an autoencoder is employed to map discrete text to a high-dimensional continuous latent space, standard iterative o... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
AdvCodec: Towards A Unified Framework for Adversarial Text Generation
### Paper Abstract
Machine learning (ML) especially deep neural networks (DNNs) have been widely ap... | |
Sygt9yBtPS | ICLR.cc/2020/Conference | 2020 | Text Embedding Bank Module for Detailed Image Paragraph Caption | ["Zengming Shen", "Arjun Gupta", "Thomas S. Huang"] | Image paragraph captioning is the task of automatically generating multiple sentences for describing images in grain-fined and coherent text. Existing typical deep learning-based models for image captioning consist of an image encoder to extract visual features and a language model decoder, which has shown promising re... | ["caption", "text embedding"] | ABSTRACTImage paragraph captioning is the task of automatically generating multiple sen-tences for describing images in grain-fined and coherent text. Existing typicaldeep learning-based models for image captioning consist of an image encoder toextract visual features and a language model decoder, which has shown promi... | S1lootKTKS | Official Blind Review #2 | 1: Reject | # 1. Summary
The paper tackles the problem of generating long, paragraph-like, fine-grained captions from describing images. It proposed a text embedding bank (TEB) model to encode a paragraph into a fixed-length representation.
The strengths of the paper are:
* The tackled problem is relevant to the computer visio... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Text Embedding Bank Module for Detailed Image Paragraph Caption
### Paper Abstract
Image paragraph captioning is the task of automatically generating multiple sentences ... | |
Byx9RUONcE | MIDL.io/2019/Conference/Abstract | 2019 | Uncertainty handling in intra-operative multispectral imaging with invertible neural networks | ["Tim J. Adler", "Lynton Ardizzone", "Leonardo Ayala", "Janek Gr\u00f6hl", "Anant Vemuri", "Sebastian J. Wirkert", "Beat P. M\u00fcller-Stich", "Carsten Rother", "Ullrich K\u00f6the", "Lena Maier-Hein"] | Replacing traditional open surgery with minimally-invasive techniques for complicated interventions such as partial tumor resection or anastomosis is one of the most important challenges in modern healthcare. In these and many other procedures, characterization of the tissue remains challenging by means of visual inspe... | ["surgical data science", "multispectral imaging", "ambiguity", "uncertainty estimation", "deep learning", "invertible neural networks", "out of distribution detection"] | Proceedings of Machine Learning Research { Under Review:1{4, 2019 Extended Abstract { MIDL 2019 submissionUncertainty handling in intra-operative multispectralimaging with invertible neural networksTim J. Adler1;2t.adler@dkfz.deLynton Ardizzone3, Leonardo Ayala1, Janek Gr ohl1;4, Anant Vemuri1, SebastianJ. Wirkert1, Be... | BkllKuqIsV | Unclear but interesting | 3: Accept | The motivation of the application seems unclear and should be written more directly.
The abstract discusses invertible neural networks and their use in out of distribution detection. It is unclear if the input is an image patch or a distribution of frequencies (I lean to distribution of frequencies given the figure ... | 3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Uncertainty handling in intra-operative multispectral imaging with invertible neural networks
### Paper Abstract
Replacing traditional open surgery with minimally-invasi... |
nhIsVl2UoMt | ICLR.cc/2021/Conference | 2021 | Additive Poisson Process: Learning Intensity of Higher-Order Interaction in Stochastic Processes | ["Simon Luo", "Feng Zhou", "Lamiae Azizi", "Mahito Sugiyama"] | We present the Additive Poisson Process (APP), a novel framework that can model the higher-order interaction effects of the intensity functions in stochastic processes using lower dimensional projections. Our model combines the techniques in information geometry to model higher-order interactions on a statistical manif... | ["Poisson Process", "Log-Linear Model", "Energy-Based Model", "Generalized Additive Models", "Information Geometry"] | ABSTRACTWe present the Additive Poisson Process (APP), a novel framework that can modelthe higher-order interaction effects of the intensity functions in point processesusing lower dimensional projections. Our model combines the techniques in in-formation geometry to model higher-order interactions on a statistical man... | o3BsNZZ568E | Paper too unclear to understand & assess contributions | 4: Ok but not good enough - rejection | ## Summary
This work claims to address the problem of learning the structure of interactions between multiple point processes. This can, e.g., be used to predict co-occurrences between different types of events. The authors develop a method that (a) is theoretically principled and (b) scales to large datasets.
## Rea... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Additive Poisson Process: Learning Intensity of Higher-Order Interaction in Stochastic Processes
### Paper Abstract
We present the Additive Poisson Process (APP), a nove... |
Mwuc0Plt_x2 | ICLR.cc/2021/Conference | 2021 | RG-Flow: A hierarchical and explainable flow model based on renormalization group and sparse prior | ["Hong-Ye Hu", "Dian Wu", "Yi-Zhuang You", "Bruno Olshausen", "Yubei Chen"] | Flow-based generative models have become an important class of unsupervised learning approaches. In this work, we incorporate the key idea of renormalization group (RG) and sparse prior distribution to design a hierarchical flow-based generative model, called RG-Flow, which can separate different scale information of i... | ["Unsupervised learning", "representation learning", "flow-based generative model", "renormalization group", "sparse encoding"] | ABSTRACTFlow-based generative models have become an important class of unsupervisedlearning approaches. In this work, we incorporate the key idea of renormaliza-tion group (RG) and sparse prior distribution to design a hierarchical flow-basedgenerative model, called RG-Flow, which can separate information at differents... | N6B0VtZR3oR | Official Review 2 | 5: Marginally below acceptance threshold | Summary:
The submission suggests a generative model with a normalizing flow based on a renormalization group idea. This gives rise to a multi-scale latent representation by arranging bijective transformations hierarchically with depth scaling logarithmically with the input dimension. Extensive experiments on CelebA are... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
RG-Flow: A hierarchical and explainable flow model based on renormalization group and sparse prior
### Paper Abstract
Flow-based generative models have become an importa... |
Byg0DsCqYQ | ICLR.cc/2019/Conference | 2019 | Robust Conditional Generative Adversarial Networks | ["Grigorios G. Chrysos", "Jean Kossaifi", "Stefanos Zafeiriou"] | Conditional generative adversarial networks (cGAN) have led to large improvements in the task of conditional image generation, which lies at the heart of computer vision. The major focus so far has been on performance improvement, while there has been little effort in making cGAN more robust to noise. The regression (o... | ["conditional GAN", "unsupervised pathway", "autoencoder", "robustness"] | ABSTRACTConditional generative adversarial networks (cGAN) have led to large improve-ments in the task of conditional image generation, which lies at the heart ofcomputer vision. The major focus so far has been on performance improvement,while there has been little effort in making cGAN more robust to noise. The regres... | B1x_dKfthX | Explanation is not clear and experiments are weak | 6: Marginally above acceptance threshold | Authors propose to augment a conditional GAN model with an unsupervised branch for spanning target manifold and show better performance than the conditional GAN in natural scene generation and face generation.
However the novelty is limited and not well explained.
1.Similar idea of using an autoencoder as another bran... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust Conditional Generative Adversarial Networks
### Paper Abstract
Conditional generative adversarial networks (cGAN) have led to large improvements in the task of co... |
H1lMogrKDH | ICLR.cc/2020/Conference | 2020 | LEARNING DIFFICULT PERCEPTUAL TASKS WITH HODGKIN-HUXLEY NETWORKS | ["Alan Lockett", "Ankit Patel", "Paul Pfaffinger"] | This paper demonstrates that a computational neural network model using ion channel-based conductances to transmit information can solve standard computer vision datasets at near state-of-the-art performance. Although not fully biologically accurate, this model incorporates fundamental biophysical principles underlying... | ["conductance-weighted averaging", "neural modeling", "normalization methods"] | ABSTRACTThis paper demonstrates that a computational neural network model using ionchannel-based conductances to transmit information can solve standard computervision datasets at near state-of-the-art performance. Although not fully biologicallyaccurate, this model incorporates fundamental biophysical principles under... | HJgyLhgI9r | Official Blind Review #1 | 3: Weak Reject | I wanted to first thank the authors for their work on connecting biologically inspired CWA to deep learning. The paper overall reads nicely.
The authors make the assumption that reviewers are fully aware of biological terminology related to CWA, however, this may not be the case. It took a few hours of searching to b... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
LEARNING DIFFICULT PERCEPTUAL TASKS WITH HODGKIN-HUXLEY NETWORKS
### Paper Abstract
This paper demonstrates that a computational neural network model using ion channel-b... | |
B42UJTVdDZ5 | ICLR.cc/2022/Workshop/DGM4HSD | 2022 | MMVAE+: Enhancing the Generative Quality of Multimodal VAEs without Compromises | ["Emanuele Palumbo", "Imant Daunhawer", "Julia E Vogt"] | Multimodal VAEs have recently gained attention as efficient models for weakly-supervised generative learning with a large number of modalities. However, all existing variants of multimodal VAEs are affected by a non-trivial trade-off between generative quality and generative coherence. We focus on the mixture-of-expert... | ["multimodal learning", "variational autoencoder", "deep generative models"] | ABSTRACTMultimodal V AEs have recently gained attention as efficient models for weakly-supervised generative learning with a large number of modalities. However, allexisting variants of multimodal V AEs are affected by a non-trivial trade-off be-tween generative quality and generative coherence. We focus on the mixture... | BMMeCfHi5fc | Official Review | 6: Marginally above acceptance threshold | The paper studies an improvement of mixture-of-experts multimodal VAE.
The main contribution of the training objective is revising the $\beta$ regularizer for the MMVAE training. The authors also provide some theoretical justification.
Overall, the results are good, but some other non-VAE based generative modeling b... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
MMVAE+: Enhancing the Generative Quality of Multimodal VAEs without Compromises
### Paper Abstract
Multimodal VAEs have recently gained attention as efficient models for... |
vjSwY0mftz | ICLR.cc/2023/Workshop/RRL | 2023 | Action Inference by Maximising Evidence: Zero-Shot Imitation from Observation with World Models | ["Xingyuan Zhang", "Philip Becker-Ehmck", "Patrick van der Smagt", "Maximilian Karl"] | Unlike most reinforcement learning agents which require an unrealistic amount of environment interactions to learn a new behaviour, humans excel at learning quickly by merely observing and imitating others. This ability highly depends on the fact that humans have a model of their own embodiment that allows them to infe... | ["Imitation Learning", "Transfer Learning", "World Models", "Variational Inference"] | ABSTRACTUnlike most reinforcement learning agents which require an unrealistic amountof environment interactions to learn a new behaviour, humans excel at learningquickly by merely observing and imitating others. This ability highly dependson the fact that humans have a model of their own embodiment that allows themto ... | wdhNqmDEu2S | "Transfer learning" in agents by imitating experts | 4: Good paper, strong accept | This paper proposes the AIME framework - Action Inference by Maximizing Evidence - a 2 step approach to train reinforcement learning agents where the agent first learns a world model from actions and observations, and then imitates the behavior of an expert based on a set of observations.
The theoretical and implement... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Action Inference by Maximising Evidence: Zero-Shot Imitation from Observation with World Models
### Paper Abstract
Unlike most reinforcement learning agents which requir... |
R-hlehDAkOu | ICML.cc/2020/Workshop/Artemiss | 2020 | The Dynamic Latent Block Model for Sparse and Evolving Count Matrices | ["Giulia Marchello", "Marco Corneli", "Charles Bouveyron"] | We consider here the problem of co-clustering count matrices with a high level of missing values that may evolve along the time. We introduce a generative model, named dynamic latent block model (dLBM), to handle this situation and which extends the classical binary latent block model (LBM) to the dynamic case. The mod... | ["latent block model", "dynamic count data", "missing values", "sparsity"] | The Dynamic Latent Block Model for Sparse and Evolving Count MatricesGiulia Marchello* 1 2Marco Corneli* 2 3Charles Bouveyron* 2AbstractWe consider here the problem of co-clusteringcount matrices with a high level of missing valuesthat may evolve along the time. We introduce agenerative model, named dynamic latent bloc... | qdgOEOYcT4 | The paper has some shortcomings, but it is a good workshop paper. | 7: Good paper, accept | The authors propose a latent block model for sparse count matrices that evolve temporally. In this context, a 0-entry in the count matrix is considered a missing value. Similar to Corneli et al. (2016), the entries in the matrix is assumed to follow a non-homogeneous Poisson process, where the rate parameter is a tenso... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The Dynamic Latent Block Model for Sparse and Evolving Count Matrices
### Paper Abstract
We consider here the problem of co-clustering count matrices with a high level o... |
HkgCdUaMq4 | MIDL.io/2019/Conference/Abstract | 2019 | Screening Mammogram Classification with Prior Exams | ["Jungkyu Park", "Jason Phang", "Yiqiu Shen", "Nan Wu", "S. Gene Kim", "Linda Moy", "Kyunghyun Cho", "Krzysztof J. Geras"] | Radiologists typically compare a patient's most recent breast cancer screening exam to their previous ones in making informed diagnoses. To reflect this practice, we propose new neural network models that compare pairs of screening mammograms from the same patient. We train and evaluate our proposed models on over 665,... | ["breast cancer screening", "mammography", "convolutional neural networks", "breast cancer screening exam classification", "utilizing prior exams"] | Medical Imaging with Deep Learning 2019 MIDL 2019 – Extended Abstract TrackScreening Mammogram Classification with Prior ExamsJungkyu Park1, Jason Phang1, Yiqiu Shen1, Nan Wu1, S. Gene Kim2, LindaMoy2, Kyunghyun Cho1, Krzysztof J. Geras2;11Center for Data Science, New York University2Department of Radiology, New York U... | B1l5YrZLo4 | A novel approach with through experiments | 4: Strong accept | The abstract presents a model to classify mammography images as benign or malignant from multiple time steps. Images from different screening times are aligned by finding the parameters of the affine transformation. The paper presents two different models trained with and without aligned pairs to perform the classifica... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Screening Mammogram Classification with Prior Exams
### Paper Abstract
Radiologists typically compare a patient's most recent breast cancer screening exam to their previ... |
SyxC9TEtPH | ICLR.cc/2020/Conference | 2020 | Conditional Invertible Neural Networks for Guided Image Generation | ["Lynton Ardizzone", "Carsten L\u00fcth", "Jakob Kruse", "Carsten Rother", "Ullrich K\u00f6the"] | In this work, we address the task of natural image generation guided by a conditioning input. We introduce a new architecture called conditional invertible neural network (cINN). It combines the purely generative INN model with an unconstrained feed-forward network, which efficiently pre-processes the conditioning inpu... | ["Invertible neural networks", "generative models", "conditional generation"] | ABSTRACTIn this work, we address the task of natural image generation guided by a condi-tioning input. We introduce a new architecture called conditional invertible neuralnetwork (cINN). It combines the purely generative INN model with an uncon-strained feed-forward network, which efficiently preprocesses the condition... | H1xjIDIp5B | Official Blind Review #2 | 3: Weak Reject | Authors provide an extension to the invertible generative models, by fashioning them into conditional invertible generative networks. The networks are conditioned at the input with a feed-forward network initialized with a VGG based classification network. Conditioning is implemented within the coupling layers (Dinh et... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Conditional Invertible Neural Networks for Guided Image Generation
### Paper Abstract
In this work, we address the task of natural image generation guided by a condition... | |
VcCQar2ny4G | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | Ontology Reasoning for Explanatory Feedback Generation to Teach how Algorithms Work | ["Anton Anikin", "Oleg Aleksandrovich Sychev", "Mikhail Denisov"] | Developing algorithms using control structures and understanding their building blocks are essential skills in mastering programming. In this work, we applied a formal model and reasoning rules for Jena reasoner to build execution trace for the given algorithm and find fault reasons if the student provides an incorrect... | ["Ontology reasoning", "Constraint-based tutoring systems", "Program execution trace", "Error detection"] | Ontology Reasoning for Explanatory FeedbackGeneration to Teach how Algorithms Work?Anton Anikin1[0000000306614284], Oleg Sychev1[0000000272962538], andMikhail Denisov1[000000021216610X]Volgograd State Technical University, Lenin Ave, 28, Volgograd, 400005, Russiaanton@anikin.name, oasychev@gmail.comAbstract. Developing... | GD8gdeFg2xe | the demonstrated tool should be more thoroughly documented | 4: Ok but not good enough - rejection | The demo should present a tool for learning programming whose novelty is its use of an ontology for classifying errors. The application presented in the paper is relevant but too few details are given either on the added value of Semantic Web technologies or on the effectiveness and reusability of the tool itself. I th... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Ontology Reasoning for Explanatory Feedback Generation to Teach how Algorithms Work
### Paper Abstract
Developing algorithms using control structures and understanding t... |
B1eibJrtwr | ICLR.cc/2020/Conference | 2020 | Abstractive Dialog Summarization with Semantic Scaffolds | ["Lin Yuan", "Zhou Yu"] | The demand for abstractive dialog summary is growing in real-world applications. For example, customer service center or hospitals would like to summarize customer service interaction and doctor-patient interaction. However, few researchers explored abstractive summarization on dialogs due to the lack of suitable data... | ["Abstractive Summarization", "Dialog", "Multi-task Learning"] | ABSTRACTIVE DIALOG SUMMARIZATIONWITH SEMANTIC SCAFFOLDSAnonymous authorsPaper under double-blind reviewABSTRACTThe demand for abstractive dialog summary is growing in real-world applications.For example, customer service center or hospitals would like to summarize cus-tomer service interaction and doctor-patient intera... | BkeK_KTlcH | Official Blind Review #2 | 3: Weak Reject | === Summary ===
The authors propose a new abstractive dialog summarization dataset and task based on the MultiWOZ dataset. Unlike previous work which targets very short descriptions of dialog transcripts (e.g. 'industrial designer presentation'), this paper looks to generate long descriptions of the entire dialog usin... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Abstractive Dialog Summarization with Semantic Scaffolds
### Paper Abstract
The demand for abstractive dialog summary is growing in real-world applications. For example,... | |
SJxpsxrYPS | ICLR.cc/2020/Conference | 2020 | PROGRESSIVE LEARNING AND DISENTANGLEMENT OF HIERARCHICAL REPRESENTATIONS | ["Zhiyuan Li", "Jaideep Vitthal Murkute", "Prashnna Kumar Gyawali", "Linwei Wang"] | Learning rich representation from data is an important task for deep generative models such as variational auto-encoder (VAE). However, by extracting high-level abstractions in the bottom-up inference process, the goal of preserving all factors of variations for top-down generation is compromised. Motivated by the conc... | ["generative model", "disentanglement", "progressive learning", "VAE"] | ABSTRACTLearning rich representation from data is an important task for deep generativemodels such as variational auto-encoder (V AE). However, by extracting high-levelabstractions in the bottom-up inference process, the goal of preserving all factorsof variations for top-down generation is compromised. Motivated by th... | Bkxu1H8y9B | Official Blind Review #1 | 6: Weak Accept | This paper proposed a method for training Variational Ladder Autoencoder (VLAE) using a progressive learning strategy. In comparison to the generative model using a progressive learning strategy, the proposed method focuses not only on the image generation but also on extracting and disentangling hierarchical represent... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
PROGRESSIVE LEARNING AND DISENTANGLEMENT OF HIERARCHICAL REPRESENTATIONS
### Paper Abstract
Learning rich representation from data is an important task for deep generati... | |
rkzUYjCcFm | ICLR.cc/2019/Conference | 2019 | FAST OBJECT LOCALIZATION VIA SENSITIVITY ANALYSIS | ["Mohammad K. Ebrahimpour", "David C. Noelle"] | Deep Convolutional Neural Networks (CNNs) have been repeatedly shown to perform well on image classification tasks, successfully recognizing a broad array of objects when given sufficient training data. Methods for object localization, however, are still in need of substantial improvement. Common approaches to this pro... | ["Internal Representations", "Sensitivity Analysis", "Object Detection"] | ABSTRACTDeep Convolutional Neural Networks (CNNs) have been repeatedly shown to per-form well on image classification tasks, successfully recognizing a broad array ofobjects when given sufficient training data. Methods for object localization, how-ever, are still in need of substantial improvement. In this paper, we of... | ryxzjQsrh7 | More oriented towards object localization rather than model interpretation | 6: Marginally above acceptance threshold |
===================
SUMMARY
===================
The paper proposes a method to extend the output of a network trained for visual object recognition (i.e. image classification) with bounding box coordinates that can serve to localize the recognized object.
This process is referred to as "object localization", and it... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
FAST OBJECT LOCALIZATION VIA SENSITIVITY ANALYSIS
### Paper Abstract
Deep Convolutional Neural Networks (CNNs) have been repeatedly shown to perform well on image classi... |
BkeStsCcKQ | ICLR.cc/2019/Conference | 2019 | Critical Learning Periods in Deep Networks | ["Alessandro Achille", "Matteo Rovere", "Stefano Soatto"] | Similar to humans and animals, deep artificial neural networks exhibit critical periods during which a temporary stimulus deficit can impair the development of a skill. The extent of the impairment depends on the onset and length of the deficit window, as in animal models, and on the size of the neural network. Deficit... | ["Critical Period", "Deep Learning", "Information Theory", "Artificial Neuroscience", "Information Plasticity"] | ABSTRACTSimilar to humans and animals, deep artificial neural networks exhibit critical periods dur-ing which a temporary stimulus deficit can impair the development of a skill. The extent ofthe impairment depends on the onset and length of the deficit window, as in animal mod-els, and on the size of the neural network... | SJg8cOOka7 | Very good paper identifying a novel phenomenology in training of deep neural networks: the presence of "critical periods" (reminiscent of the same phenomenon in many biological brain circuits) where perturbations in training can permanently affect the final performance of the model. | 8: Top 50% of accepted papers, clear accept | The authors analyze the learning dynamics in deep neural networks and identify an intriguing phenomenon that reflects what in biological learning is known as critical period: a relatively short time window early in post-natal development where organisms become particularly sensitive to particular changes in experience.... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Critical Learning Periods in Deep Networks
### Paper Abstract
Similar to humans and animals, deep artificial neural networks exhibit critical periods during which a temp... |
zGLniPvGsWT | auai.org/UAI/2022/Workshop/CRL | 2022 | Towards Computing an Optimal Abstraction for Structural Causal Models | ["Fabio Massimo Zennaro", "Paolo Turrini", "Theo Damoulas"] | Working with causal model at different levels of abstraction is an important feature of science. Existing work has already considered the problem of expressing formally the relation of abstraction between causal models. In this paper, we focus on the problem of learning abstractions. We start by defining the learning p... | ["structural causal models", "SCM", "abstraction", "levels of abstraction", "learning"] | ABSTRACTIONDefinition 3 (Abstraction [Rischel, 2020]) LetM =⟨M[X],M[φX]⟩andM′=⟨M′[X′],M′[φX′]⟩be twoSCMs. An abstraction αfromMtoM′is a tuple ⟨R, a, α i⟩where:•R⊆ X defines a subset of relevant variables in M;•a:R→ X′is a surjective function mapping relevantvariables RinMto variables in M′;•αi:M[a−1(X′i)]→ M′[X′i]is a ... | 7nsMvC4_d21 | Review | 6: Marginally above acceptance threshold | ### Summary of the paper
This work introduces an optimization problem meant to identify interesting causal abstraction. The objective to minimize contains two terms, one measuring interventional inconsistency, and another measuring informational loss. The latter term prevents the learner from finding degenerate solutio... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Computing an Optimal Abstraction for Structural Causal Models
### Paper Abstract
Working with causal model at different levels of abstraction is an important fea... |
rJeBJJBYDB | ICLR.cc/2020/Conference | 2020 | Chart Auto-Encoders for Manifold Structured Data | ["Stephan Schonsheck", "Jie Chen", "Rongjie Lai"] | Auto-encoding and generative models have made tremendous successes in image and signal representation learning and generation. These models, however, generally employ the full Euclidean space or a bounded subset (such as $[0,1]^l$) as the latent space, whose trivial geometry is often too simplistic to meaningfully ref... | ["Auto-encoder", "differential manifolds", "multi-charted latent space"] | ABSTRACTAuto-encoding and generative models have made tremendous successes in imageand signal representation learning and generation. These models, however, gen-erally employ the full Euclidean space or a bounded subset (such as [0;1]l) as thelatent space, whose flat geometry is often too simplistic to meaningfully ref... | HJeMrVc3Kr | Official Blind Review #3 | 3: Weak Reject | Summary:
The authors provide a model that is based on multiple Auto-Encoders, and it is claimed that each Auto-Encoder learns a local chart map of the data manifold.
In general the paper is ok written and tries to present the idea and the theoretical background in a nice way. However, there are few things that in my... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Chart Auto-Encoders for Manifold Structured Data
### Paper Abstract
Auto-encoding and generative models have made tremendous successes in image and signal representati... | |
HJlfAo09KX | ICLR.cc/2019/Conference | 2019 | Guaranteed Recovery of One-Hidden-Layer Neural Networks via Cross Entropy | ["Haoyu Fu", "Yuejie Chi", "Yingbin Liang"] | We study model recovery for data classification, where the training labels are generated from a one-hidden-layer fully -connected neural network with sigmoid activations, and the goal is to recover the weight vectors of the neural network. We prove that under Gaussian inputs, the empirical risk function using cross ent... | ["cross entropy", "neural networks", "parameter recovery"] | ABSTRACTWe study model recovery for data classification, where the training labels aregenerated from a one-hidden-layer fully-connected neural network with sigmoidactivations, and the goal is to recover the weight vectors of the neural network. Weprove that under Gaussian inputs, the empirical risk function using cross... | HylHS_0L37 | Review | 5: Marginally below acceptance threshold | Paper Summary:
This paper studies the problem of recovering a true underlying neural network (assuming there exists one) with cross-entropy loss. This paper shows if the input is standard Gaussian, within a small ball around the ground truth, the objective function is strongly convex and smooth if there is a sufficient... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Guaranteed Recovery of One-Hidden-Layer Neural Networks via Cross Entropy
### Paper Abstract
We study model recovery for data classification, where the training labels a... |
EKV158tSfwv | ICLR.cc/2021/Conference | 2021 | Efficient Continual Learning with Modular Networks and Task-Driven Priors | ["Tom Veniat", "Ludovic Denoyer", "MarcAurelio Ranzato"] | Existing literature in Continual Learning (CL) has focused on overcoming catastrophic forgetting, the inability of the learner to recall how to perform tasks observed in the past.
There are however other desirable properties of a CL system, such as the ability to transfer knowledge from previous tasks and to scale mem... | ["Continual learning", "Lifelong learning", "Benchmark", "Modular network", "Neural Network"] | ABSTRACTExisting literature in Continual Learning (CL) has focused on overcoming catas-trophic forgetting, the inability of the learner to recall how to perform tasks ob-served in the past. There are however other desirable properties of a CL system,such as the ability to transfer knowledge from previous tasks and to s... | aFqI0FKFN1- | AnonReviewer2 Review | 6: Marginally above acceptance threshold | **Summary of paper**
This paper has two main contributions. Firstly, it introduces a new set of benchmarks designed to proble different qualities that may be important for continual learning methods. Secondly, it introduces a new modular method for continual learning, which learns to add modules when learning on a new... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Efficient Continual Learning with Modular Networks and Task-Driven Priors
### Paper Abstract
Existing literature in Continual Learning (CL) has focused on overcoming cat... |
IpPQmzj4T_ | ICLR.cc/2021/Conference | 2021 | Teleport Graph Convolutional Networks | ["Hongyang Gao", "Shuiwang Ji"] | We consider the limitations in message-passing graph neural networks. In message-passing operations, each node aggregates information from its neighboring nodes. To enlarge the receptive field, graph neural networks need to stack multiple message-passing graph convolution layers, which leads to the over-fitting issue a... | ["over-smoothing"] | ABSTRACTWe consider the limitations in message-passing graph neural networks. Inmessage-passing operations, each node aggregates information from its neigh-boring nodes. To enlarge the receptive field, graph neural networks need to stackmultiple message-passing graph convolution layers, which leads to the over-fittingi... | 2cHEyWBv6l5 | # Official Review of Teleport Graph Convolution Net | 5: Marginally below acceptance threshold | A new architecture for graph neural networks, which the authors name as Teleport Graph Convolutional Networks (TGL), is proposed in this paper. Teleport graph convolution layer is proposed to address the limitations in message-passing operations of graph neural networks: 1. over-smoothing and 2. over-fitting. The ar... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Teleport Graph Convolutional Networks
### Paper Abstract
We consider the limitations in message-passing graph neural networks. In message-passing operations, each node a... |
SkHl6MWC- | ICLR.cc/2018/Conference | 2018 | Regularization Neural Networks via Constrained Virtual Movement Field | ["Zhendong Zhang", "Cheolkon Jung"] | We provide a novel thinking of regularization neural networks. We smooth the objective of neural networks w.r.t small adversarial perturbations of the inputs. Different from previous works, we assume the adversarial perturbations are caused by the movement field. When the magnitude of movement field approaches 0, we ca... | ["movement field", "problem", "regularization neural networks", "virtual movement field", "adversarial perturbations", "smoothness", "novel thinking", "objective", "neural networks"] | ABSTRACTWe provide a novel thinking of regularization neural networks. We smooth the ob-jective of neural networks w.r.t small adversarial perturbations of the inputs. Dif-ferent from previous works, we assume the adversarial perturbations are caused bythe movement field. When the magnitude of movement field approaches... | Sy_2ES9lG | Review | 5: Marginally below acceptance threshold | This paper proposes to regularize neural networks by the invariance to certain types of transforms. This is framed into a minimax problem, which yields a closed form regularization when constrained to simple types of transforms.
The basic idea of using derivative to measure sensitivity has been widely known, and is r... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Regularization Neural Networks via Constrained Virtual Movement Field
### Paper Abstract
We provide a novel thinking of regularization neural networks. We smooth the ob... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.