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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
BkUDW_lCb | ICLR.cc/2018/Conference | 2018 | Pointing Out SQL Queries From Text | ["Chenglong Wang", "Marc Brockschmidt", "Rishabh Singh"] | The digitization of data has resulted in making datasets available to millions of users in the form of relational databases and spreadsheet tables. However, a majority of these users come from diverse backgrounds and lack the programming expertise to query and analyze such tables. We present a system that allows for qu... | ["Program Synthesis", "Semantic Parsing", "WikiTable", "SQL", "Pointer Network"] | ABSTRACTThe digitization of data has resulted in making datasets available to millions ofusers in the form of relational databases and spreadsheet tables. However, a ma-jority of these users come from diverse backgrounds and lack the programmingexpertise to query and analyze such tables. We present a system that allows... | rk5LF4OeM | The contributions have already been done in the past | 3: Clear rejection | This paper proposes a model for solving the WikiSQL dataset that was released recently.
The main issues with the paper is that its contributions are not new.
* The first claimed contribution is to use typing at decoding time (they don't say why but this helps search and learning). Restricting the type of the decoded ... | 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
Pointing Out SQL Queries From Text
### Paper Abstract
The digitization of data has resulted in making datasets available to millions of users in the form of relational d... |
tL89RnzIiCd | ICLR.cc/2021/Conference | 2021 | Hopfield Networks is All You Need | ["Hubert Ramsauer", "Bernhard Sch\u00e4fl", "Johannes Lehner", "Philipp Seidl", "Michael Widrich", "Lukas Gruber", "Markus Holzleitner", "Thomas Adler", "David Kreil", "Michael K Kopp", "G\u00fcnter Klambauer", "Johannes Brandstetter", "Sepp Hochreiter"] | We introduce a modern Hopfield network with continuous states and a corresponding update rule. The new Hopfield network can store exponentially (with the dimension of the associative space) many patterns, retrieves the pattern with one update, and has exponentially small retrieval errors. It has three types of energy m... | ["Modern Hopfield Network", "Energy", "Attention", "Convergence", "Storage Capacity", "Hopfield layer", "Associative Memory"] | ABSTRACTWe introduce a modern Hopfield network with continuous states and a correspond-ing update rule. The new Hopfield network can store exponentially (with thedimension of the associative space) many patterns, retrieves the pattern with oneupdate, and has exponentially small retrieval errors. It has three types of e... | d0DwejhNgq | Very interesting but missing critical content | 7: Good paper, accept | This work extends the binary Hopfield network (Demircigil et al., 2017) to continuous patterns and states. Connections are drawn between the result model to the attention layers of the transformers, the pooling operation of LSTM, similarity search, and fully connected layers. Experimental results are briefly described ... | 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
Hopfield Networks is All You Need
### Paper Abstract
We introduce a modern Hopfield network with continuous states and a corresponding update rule. The new Hopfield netw... |
lH6TlQO_4P | eswc-conferences.org/ESWC/2021/Conference/Research_Track | 2021 | Expressibility of OWL Axioms with Patterns | ["Aaron Eberhart", "Cogan M Shimizu", "Sulogna Chowdhury", "Md Kamruzzaman Sarker", "Pascal Hitzler"] | The high expressivity of the Web Ontology Language (OWL) makes it possible to describe complex relationships between classes, roles, and individuals in an ontology. However, this high expressivity can be an obstacle to correct usage and wide adoption. Past attempts to ameliorate this have included the development of sp... | ["Ontology", "OWL Axioms", "Semantic Expressibility", "Simple Axioms", "Axiom Patterns"] | Expressibility of OWL Axioms with PatternsAaron Eberhart , Cogan Shimizu, Sulogna Chowdhury, Md. KamruzzamanSarker, and Pascal HitzlerData Semantics Lab, Kansas State University, USAaaroneberhart@ksu.eduAbstract. The high expressivity of the Web Ontology Language (OWL)makes it possible to describe complex relationships... | 1cp1rMYZJVt | new metrics about ontology axiomatization | 1: Weak Accept | This paper presents two metrics to evaluate the complexity of ontology.
The description logics identified is SROIQ(D).
The methods is based on a definition of simple axiom patterns.
The authors want to test the hypothesis that ontology used simple patterns.
Simple pattern should have at most 3 elements.
A list of 17... | 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
Expressibility of OWL Axioms with Patterns
### Paper Abstract
The high expressivity of the Web Ontology Language (OWL) makes it possible to describe complex relationship... |
VYfotZsQV5S | ICLR.cc/2021/Conference | 2021 | MISSO: Minimization by Incremental Stochastic Surrogate Optimization for Large Scale Nonconvex and Nonsmooth Problems | ["Belhal Karimi", "Hoi To Wai", "Eric Moulines", "Ping Li"] | Many constrained, nonconvex and nonsmooth optimization problems can be tackled using the majorization-minimization (MM) method which alternates between constructing a surrogate function which upper bounds the objective function, and then minimizing this surrogate. For problems which minimize a finite sum of functions, ... | ["nonconvex", "optimization", "stochastic", "sampling", "MCMC", "majorization-minimization"] | ABSTRACTMany constrained, nonconvex and nonsmooth optimization problems can be tack-led using the majorization-minimization (MM) method which alternates betweenconstructing a surrogate function which upper bounds the objective function, andthen minimizing this surrogate. For problems which minimize a finite sum offunct... | 7WMCspilEAL | Reviewer 5's Report | 3: Clear rejection | This paper develops a stochastic MM-type algorithm to minimize a finite sum. Essentially, the stochastic method draws one sample at each iteration, and find a majorization surrogate for the corresponding loss, and find the minimizer for the updated total loss.
Overall, I don't find the paper well-developed and doesn't... | 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
MISSO: Minimization by Incremental Stochastic Surrogate Optimization for Large Scale Nonconvex and Nonsmooth Problems
### Paper Abstract
Many constrained, nonconvex and ... |
r17_wzJPM | ICLR.cc/2018/Workshop | 2018 | SGD on Random Mixtures: Private Machine Learning under Data Breach Threats | ["Kangwook Lee", "Kyungmin Lee", "Hoon Kim", "Changho Suh", "Kannan Ramchandran"] | We propose Stochastic Gradient Descent on Random Mixtures (SGDRM) as a simple way of protecting data under data breach threats. We show that SGDRM converges to the globally optimal point for deep neural networks with linear activations while being differentially private. We also train nonlinear neural networks with pri... | ["SGD on random mixtures", "SGDRM", "differential privacy"] | ABSTRACTWe propose Stochastic Gradient Descent on Random Mixtures (SGDRM) as asimple way of protecting data under data breach threats. We show that SGDRMconverges to the globally optimal point for deep neural networks with linear acti-vations while being differentially private. We also train nonlinear neural networkswi... | SyHHPz-tG | The result seems to be theoretically sound, but the setting is not clear | 7: Good paper, accept | The motivation is clear and well understood.
The result seems to be theoretically sound, but the setting is not clear.
The authors mention that the convergence result is specific to linear neural network models. If so, the resulting model is represented by a linear model. If this is correct, the analytic solution is o... | 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
SGD on Random Mixtures: Private Machine Learning under Data Breach Threats
### Paper Abstract
We propose Stochastic Gradient Descent on Random Mixtures (SGDRM) as a simp... |
xHP1W_Y0jF | thecvf.com/ECCV/2020/Workshop/BIC | 2020 | Registration of multi-modal volumetric images of embryos by establishing correspondences between cells | ["Manan Lalit", "Florian Jug", "Pavel Tomancak"] | Early development of an animal from an egg involves a rapid increase in cell number and several cell fate specification events which are accompanied by dynamic morphogenetic changes.
In order to correlate the morphological changes with the underlying genetic events, one typically needs to monitor the living system with... | ["Cell Correspondence", "Image Registration", "In-Situ Hybridisation", "\\textit{Platynereis dumerilii}", "Iterative Closest Point", "Shape Context"] | 000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044ECCV#100ECCV#100Registration of multi-modal volume... | ph8NKdPC_fW | Excellent work | 10: Top 5% of accepted papers, seminal paper | The paper presents a methodology for registration of high-resolution gene expression data and live microscopy images of Platynereis dumerilii embrios in 3D. The method detects nuclei, computes descriptors, finds correspondences between cells and aligns the data. One of the core strategies used in this methodology is 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
Registration of multi-modal volumetric images of embryos by establishing correspondences between cells
### Paper Abstract
Early development of an animal from an egg invo... |
S1eL4kBYwr | ICLR.cc/2020/Conference | 2020 | UNITER: Learning UNiversal Image-TExt Representations | ["Yen-Chun Chen", "Linjie Li", "Licheng Yu", "Ahmed El Kholy", "Faisal Ahmed", "Zhe Gan", "Yu Cheng", "Jingjing Liu"] | Joint image-text embedding is the bedrock for most Vision-and-Language (V+L) tasks, where multimodality inputs are jointly processed for visual and textual understanding. In this paper, we introduce UNITER, a UNiversal Image-TExt Representation, learned through large-scale pre-training over four image-text datasets (CO... | ["Self-supervised Representation Learning", "Large-scale Pre-training", "Vision and Language"] | ABSTRACTJoint image-text embedding is the bedrock for most Vision-and-Language (V+L)tasks, where multimodality inputs are jointly processed for visual and textual un-derstanding. In this paper, we introduce UNITER, a UNiversal Image-TExt Rep-resentation, learned through large-scale pre-training over four image-text dat... | BkeMz3WRFB | Official Blind Review #3 | 6: Weak Accept | This is an impressive paper. LIke BERT, it proposes a tranformer based approach to derive a pre-trained network for representing images and texts. The resulting pre-trained network, used in 9 different tasks, advances the SOTA on all the tasks.
The major limitation of this paper is why. Why does it happen? How this re... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
UNITER: Learning UNiversal Image-TExt Representations
### Paper Abstract
Joint image-text embedding is the bedrock for most Vision-and-Language (V+L) tasks, where multim... | |
Skj8Kag0Z | ICLR.cc/2018/Conference | 2018 | Stabilizing Adversarial Nets with Prediction Methods | ["Abhay Yadav", "Sohil Shah", "Zheng Xu", "David Jacobs", "Tom Goldstein"] | Adversarial neural networks solve many important problems in data science, but are notoriously difficult to train. These difficulties come from the fact that optimal weights for adversarial nets correspond to saddle points, and not minimizers, of the loss function. The alternating stochastic gradient methods typically ... | ["adversarial networks", "optimization"] | ABSTRACTAdversarial neural networks solve many important problems in data science, butare notoriously difficult to train. These difficulties come from the fact that optimalweights for adversarial nets correspond to saddle points, and not minimizers, ofthe loss function. The alternating stochastic gradient methods typic... | ry-q9ZOlf | A simple modification to alternating stochastic gradient for GAN training, which stabilizes training, essentially for free. Clever and useful idea, solid and insightful analysis, good presentation. | 9: Top 15% of accepted papers, strong accept | This paper proposes a simple modification to the standard alternating stochastic gradient method for GAN training, which stabilizes training, by adding a prediction step.
This is a clever and useful idea, and the paper is very well written. The proposed method is very clearly motivated, both intuitively and mathemati... | 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
Stabilizing Adversarial Nets with Prediction Methods
### Paper Abstract
Adversarial neural networks solve many important problems in data science, but are notoriously di... |
_A15qsPswaK | robot-learning.org/CoRL/2023/Conference | 2023 | HYDRA: Hybrid Robot Actions for Imitation Learning | ["Suneel Belkhale", "Yuchen Cui", "Dorsa Sadigh"] | Imitation Learning (IL) is a sample efficient paradigm for robot learning using expert demonstrations. However, policies learned through IL suffer from state distribution shift at test time, due to compounding errors in action prediction which lead to previously unseen states. Choosing an action representation for the ... | ["Imitation Learning", "Robotics", "Manipulation"] | HYDRA: Hybrid Robot Actionsfor Imitation LearningSuneel Belkhale, Yuchen Cui, Dorsa SadighStanford University, Stanford CA{belkhale, yuchenc, dorsa }@stanford.eduAbstract: Imitation Learning (IL) is a sample efficient paradigm for robot learn-ing using expert demonstrations. However, policies learned through IL suffer ... | HgWW3_YmuvA | Remarks and Questions:
- It seems like the data labeling method performs linear interpolation between cartesian end-effector locations:
- The authors describe this low-level action relabeling as the application of a “linear controller” to “imagine” new actions. This description is unnecessary, and also confusing (the ... | 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
HYDRA: Hybrid Robot Actions for Imitation Learning
### Paper Abstract
Imitation Learning (IL) is a sample efficient paradigm for robot learning using expert demonstratio... | ||
nLYMajjctMh | ICLR.cc/2021/Conference | 2021 | Federated Learning of a Mixture of Global and Local Models | ["Filip Hanzely", "Peter Richtarik"] | We propose a new optimization formulation for training federated learning models. The standard formulation has the form of an empirical risk minimization problem constructed to find a single global model trained from the private data stored across all participating devices. In contrast, our formulation seeks an explici... | ["optimization", "federated learning", "personalization", "local SGD"] | ABSTRACTWe propose a new optimization formulation for training federated learning mod-els. The standard formulation has the form of an empirical risk minimizationproblem constructed to find a single global model trained from the private datastored across all participating devices. In contrast, our formulation seeks an ... | sDg6GZHRktx | Interesting insights into federated learning, possibly limited by the focus on strong convexity | 6: Marginally above acceptance threshold | Interesting insights into federated learning, possibly limited by the focus on strong convexity
This paper considers distributed training problems arising in the context of federated learning. It proposes a novel framing of the problem as a compromise between fitting a model locally to the data available at a device, ... | 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
Federated Learning of a Mixture of Global and Local Models
### Paper Abstract
We propose a new optimization formulation for training federated learning models. The stand... |
BJPlJVywz | ICLR.cc/2018/Workshop | 2018 | NAM - Unsupervised Cross-Domain Image Mapping without Cycles or GANs | ["Yedid Hoshen", "Lior Wolf"] | Several methods were recently proposed for Unsupervised Domain Mapping, which is the task of translating images between domains without prior knowledge of correspondences. Current approaches suffer from an instability in training due to relying on GANs which are powerful but highly sensitive to hyper-parameters and suf... | ["Generative Models", "GAN", "Cross Domain Mapping", "Unsupervised Image Translation"] | ABSTRACTSeveral methods were recently proposed for Unsupervised Domain Mapping,which is the task of translating images between domains without prior knowledgeof correspondences. Current approaches suffer from an instability in training dueto relying on GANs which are powerful but highly sensitive to hyper-parametersand... | HJkTa91KM | Accept | 7: Good paper, accept | This workshop paper introdcues a new domain mapping approach which relies on generative models pretrained on the source domain, and aligns with target images with an image sampled from the source domain. The joint optimization carried out during this mapping on the mapping function underlines the contribution.
The 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
NAM - Unsupervised Cross-Domain Image Mapping without Cycles or GANs
### Paper Abstract
Several methods were recently proposed for Unsupervised Domain Mapping, which is ... |
E-PcUeaDbzv | graphicsinterface.org/Graphics_Interface/2022/Conference | 2022 | Automatic Asymmetric Weight Distribution Detection and Correction Utilizing Electrical Muscle Stimulation | ["Ravi Kiran Kattoju", "Eugene M Taranta", "Ryan Ghamandi", "Joseph LaViola"] | Postural control is a constant re-establishment process for the maintenance of balance and stability. Asymmetric weight distribution (AWD), characterized by uneven leg loading, leads to increased instability, injury, and progressive deterioration of posture and gait. Postural self-correction is automatically affected b... | ["Wearable computing", "Preventive Healthcare", "Posture correction", "Asymmetric weight distribution", "Electrical Muscle Stimulation"] | ABSTRACTPostural control is a constant re-establishment process for the main-tenance of balance and stability. Asymmetric weight distribution(AWD), characterized by uneven leg loading, leads to increasedinstability, injury, and progressive deterioration of posture and gait.Postural self-correction is automatically affe... | u8p0z0L1FLR | The study is important and well designed and conducted. The results are clear as well and show promise in helping to correct AWD. | 8: Top 50% of accepted papers, clear accept | This paper reports the results of an experimental study that compared the response time and usability of three different techniques for correcting asymmetric weight distribution (AWD). The authors of the paper developed a technique that used EMS feedback with auto AWD correction, which was compared against two self-cor... | 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
Automatic Asymmetric Weight Distribution Detection and Correction Utilizing Electrical Muscle Stimulation
### Paper Abstract
Postural control is a constant re-establishm... |
SJxJtYkPG | ICLR.cc/2018/Workshop | 2018 | A Language and Compiler View on Differentiable Programming | ["Fei Wang", "Tiark Rompf"] | Current and emerging deep learning architectures call for an expressive high-level programming style with end-to-end differentiation and for a high-performance implementation at the same time. But the current generation of deep learning
frameworks either limits expressiveness and ease of use for increased performance (... | ["multi-stage programming", "delimited continuations", "computation graph", "PyTorch", "TensorFlow", "compiler"] | ABSTRACTCurrent and emerging deep learning architectures call for an expressive high-levelprogramming style with end-to-end differentiation and for a high-performanceimplementation at the same time. But the current generation of deep learningframeworks either limits expressiveness and ease of use for increased performa... | r1uXjW-Yf | Interesting perspective about bridging NNs and FP, few evidences in real case | 6: Marginally above acceptance threshold | The paper describes the construction method of the neural network with reverse-mode automatic differentiation using functional programming, especially reset/shift operators, and the combination of those results and the multi-stage programming paradigms.
The perspective of using delimited continuation shown in the paper... | 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
A Language and Compiler View on Differentiable Programming
### Paper Abstract
Current and emerging deep learning architectures call for an expressive high-level programm... |
SJlh2jR9FX | ICLR.cc/2019/Conference | 2019 | Learning with Reflective Likelihoods | ["Adji B. Dieng", "Kyunghyun Cho", "David M. Blei", "Yann LeCun"] | Models parameterized by deep neural networks have achieved state-of-the-art results in many domains. These models are usually trained using the maximum likelihood principle with a finite set of observations. However, training deep probabilistic models with maximum likelihood can lead to the issue we refer to... | ["new learning criterion", "penalized maximum likelihood", "posterior inference in deep generative models", "input forgetting issue", "latent variable collapse issue"] | ABSTRACTModels parameterized by deep neural networks have achieved state-of-the-art re-sults in many domains. These models are usually trained using the maximumlikelihood principle with a finite set of observations. However, training deep prob-abilistic models with maximum likelihood can lead to the issue we refer to a... | rJlp3F0d2Q | Interesting ideas that need further refinement | 4: Ok but not good enough - rejection | Summary:
This paper proposes maximizing the “reflective likelihood,” which the authors define as: E_x E_y [log q(y|x) - \alpha log q(y)] where the expectations are taken over the data, q is the classifier, and \alpha is a weight on the log q(y) term. The paper derives the reflective likelihood for classification mode... | 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 with Reflective Likelihoods
### Paper Abstract
Models parameterized by deep neural networks have achieved state-of-the-art results in many domains. These m... |
HU50NPRFZC | MICCAI.org/2022/Challenge/FLARE | 2022 | A semi-supervised segmentation network based on noisy student learning for MICCAI FLARE22 Challenge | ["Nuo Tong", "Kunru Wang", "Rui Wu", "Huiying Yang", "Hu Huang"] | Abdominal organ segmentation is very important for clinical applications. However, manually annotating organs from CT scans is time-consuming and labor-intensive. Therefore, it is hard to get access to a large amount of annotated data. Semi-supervised learning is an effective method to use unlabeled data to reduce data... | ["multi-organ segmentation", "semi-supervised learning", "noisy-student", "pseudo-labels"] | Asemi-supervisedsegmentationnetworkbasedonnoisystudentlearningforMICCAIFLARE22ChallengeNuoTong1,KunruWang1,RuiWu1,HuiyingYang1,HuHuang11XidianUniversity,Xi’an,Shaanxi710071,Chinakunruwang@stu.xidian.edu.cnAbstract.Abdominalorgansegmentationisveryimportantforclinicalapplications.However,manuallyannotatingorgansfromCTsca... | 3UlSbl3VZET | A Poorly-Written Paper with Insufficient Experiments | 5: Marginally below acceptance threshold | This paper uses a teacher network to provide pseudo-labels for training a noisy nnU-Net student. Some comparisons are made to show the effectiveness of the proposed method results.
Pros:
+ This paper adopts a nnU-Net-based teacher model to train a noisy student for better performance, which is a good choice.
+... | 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
A semi-supervised segmentation network based on noisy student learning for MICCAI FLARE22 Challenge
### Paper Abstract
Abdominal organ segmentation is very important for... |
Sk36NgFeg | ICLR.cc/2017/conference | 2017 | Filling in the details: Perceiving from low fidelity visual input | ["Farahnaz A. Wick", "Michael L. Wick", "Marc Pomplun"] | Humans perceive their surroundings in great detail even though most of our visual field is reduced to low-fidelity color-deprived (e.g., dichromatic) input by the retina. In contrast, most deep learning architectures deploy computational resources homogeneously to every part of the visual input. Is such a prodigal depl... | ["Deep learning", "Computer vision", "Semi-Supervised Learning"] | ABSTRACTHumans perceive their surroundings in great detail even though most of our vi-sual field is reduced to low-fidelity color-deprived (e.g., dichromatic) input bythe retina. In contrast, most deep learning architectures deploy computationalresources homogeneously to every part of the visual input. Is such a prodig... | SyiMEjbEg | Interesting idea, evaluation could be improved | 6: Marginally above acceptance threshold | I like the idea the paper is exploring. Nevertheless I see some issues with the analysis:
- To get a better understanding of the quality of the results, I think at least some state-of-the-art comparisons should be included (e.g. by setting d times d pixel patches too their average and applying a denoising autoencoder)... | 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
Filling in the details: Perceiving from low fidelity visual input
### Paper Abstract
Humans perceive their surroundings in great detail even though most of our visual fi... |
oweBPxtma_i | ICLR.cc/2021/Conference | 2021 | A self-explanatory method for the black box problem on discrimination part of CNN | ["Jinwei Zhao", "Qizhou Wang", "Wanli Qiu", "Guo Xie", "Wei Wang", "Xinhong Hei", "Deyu Meng"] | Recently, for finding inherent causality implied in CNN, the black box problem of its discrimination part, which is composed of all fully connected layers of the CNN, has been studied by different scientific communities. Many methods were proposed, which can extract various interpretable models from the optimal discrim... | ["Convolution neural network", "Interpretability performance", "Markov random field"] | ABSTRACTRecently, for finding inherent causality implied in CNN, the black box problemof its discrimination part, which is composed of all fully connected layers of theCNN, has been studied by different scientific communities. Many methods wereproposed, which can extract various interpretable models from the optimal di... | BAVYJ09zazC | Significant issues in the language severally limit this work, though there are content issues as well. | 3: Clear rejection | I found this paper very difficult to follow as it has many grammatical and syntactic errors. I believe it needs a significant amount of editing in order for the paper to be published in english. In particular careful attention should be made to omitted particles and pluralization. This alone is a barrier to publicatio... | 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
A self-explanatory method for the black box problem on discrimination part of CNN
### Paper Abstract
Recently, for finding inherent causality implied in CNN, the black b... |
QcQ7fsx8QI | NoDaLiDa/2023/Conference | 2023 | Parsing the Old Using the New and Vice Versa: Fine-tuning Pre-Trained Language Models for Icelandic | ["Mathias Hans Erik Stenlund", "Be\u00e1ta Megyesi"] | In this study, we present experiments on parsing historical Icelandic by using a pre-trained ConvBERT language model for modern Icelandic which is then fine-tuned on modern Icelandic, historical Icelandic, and on a combination of both. Using the dependency parser DiaParser, the models are evaluated on both modern and h... | ["parsing", "icelandic", "ConvBERT", "historical", "modern", "fine-tune", "pre-trained", "LAS"] | 10000010020030040050060070080090100110120130140150160170180190200210220230240250260270280290300310320330340350360370380390400410420430440450460470480490500510520530540550560570580590600610620630640650660670680690700710720730740750760770780790800810820830840850860870880890900910920930940950960970980991001011021031041051... | x9syz0iOVX | Good evaluation paper on parsing modern Icelandic training on historical material and vice versa. | 8: Top 50% of accepted papers, clear accept | The authors look at training a dependency parser for Icelandic by finetuning a BERT-like model. In particular, they look at domain effects, where "domain" here is defined by temporal difference: training on historical material and applying to modern material and the other way around. The paper is very clear, well-writt... | 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
Parsing the Old Using the New and Vice Versa: Fine-tuning Pre-Trained Language Models for Icelandic
### Paper Abstract
In this study, we present experiments on parsing h... |
uL5eaPThII2 | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | Extraction of Union and Intersection Axioms from Biomedical Text | ["Nikhil Sachdeva", "Monika Jain", "Raghava Mutharaju"] | Many ontology, especially the ones created automatically by the ontology learning systems, have only shallow relationships between
the concepts, i.e., simple subclass relations. Expressive axioms such as the class union and intersection are not part of the ontology. These expressive axioms make the ontology rich and pl... | ["Ontology Learning", "Ontology Enrichment", "Axiom Extraction"] | Extraction of Union and Intersection Axiomsfrom Biomedical TextNikhil Sachdeva [0000000175444552], Monika Jain[0000000176977772], andRaghava Mutharaju[0000000324213935]Knowledgeable Computing and Reasoning Lab, IIIT-Delhi, India.fnikhil16061, monikaja, raghava.mutharaju g@iiitd.ac.inAbstract. Many ontology, especially ... | o4kv9_Iwx0T | The presented approach appears as useful to make an existing ontology more meaningful, but is is not clearly introduced. | 6: Marginally above acceptance threshold | Summary
The paper proposes a mechanism to extract union and intersection axioms from a text, looking for concepts of a reference ontology. As a result, an ontology is enriched with the extracted axioms. As most of ontology learning systems have the capability of extracting classes and the hierarchical relations among ... | 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
Extraction of Union and Intersection Axioms from Biomedical Text
### Paper Abstract
Many ontology, especially the ones created automatically by the ontology learning sys... |
BJelsDvo84 | ICLR.cc/2019/Workshop/LLD | 2019 | EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks | ["Jason Wei", "Kai Zou"] | We present EDA: easy data augmentation techniques for boosting performance on text classification tasks. EDA consists of four simple but powerful operations: synonym replacement, random insertion, random swap, and random deletion. On five text classification tasks, we show that EDA improves performance for both convolu... | ["Data Augmentation", "Text Classification", "Natural Language Processing"] | ABSTRACTWe present EDA :easydataaugmentation techniques for boosting performanceon text classification tasks. EDA consists of four simple but powerful opera-tions: synonym replacement, random insertion, random swap, and random dele-tion. On five text classification tasks, we show that EDA improves performancefor both c... | SJeisCOTOE | Simple data augmentation techniques for text that work well especially on small datasets | 4: Top 50% of accepted papers, clear accept | Summary: The paper proposes 4 data augmentation techniques for text: synonym replacement, random insertion, random swap, and random deletion. These techniques are validated on 5 text classification datasets, showing increased performance of up to 3% when the dataset size is small. Ablation studies show that all four te... | 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
EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks
### Paper Abstract
We present EDA: easy data augmentation techniques for boo... |
HJeIU0VYwB | ICLR.cc/2020/Conference | 2020 | ADA+: A GENERIC FRAMEWORK WITH MORE ADAPTIVE EXPLICIT ADJUSTMENT FOR LEARNING RATE | ["Yue Zhao", "Xiangsheng Huang", "Ludan Kou"] | Although adaptive algorithms have achieved significant success in training deep neural networks with faster training speed, they tend to have poor generalization performance compared to SGD with Momentum(SGDM). One of the state-of-the-art algorithms, PADAM, is proposed to close the generalization gap of adaptive method... | ["Optimization", "Adaptive Methods", "Convergence", "Convolutional Neural Network"] | ABSTRACTAlthough adaptive algorithms have achieved significant success in training deepneural networks with faster training speed, they tend to have poor generaliza-tion performance compared to S GD WITH MOMENTUM (SGDM). One of thestate-of-the-art algorithms, P ADAM , is proposed to close the generalization gapof adapt... | H1g4Pa9zcS | Official Blind Review #4 | 1: Reject | This paper proposes a new variation on adaptive learning rate algorithm that builds on a prior work Padam, which is also a concurrent submission to this conference. The method is empirically validated through through various domains and neural networks architecture. Though the empirical results are extensive, I am lean... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ADA+: A GENERIC FRAMEWORK WITH MORE ADAPTIVE EXPLICIT ADJUSTMENT FOR LEARNING RATE
### Paper Abstract
Although adaptive algorithms have achieved significant success in t... | |
Hyl5V0EYvB | ICLR.cc/2020/Conference | 2020 | Testing Robustness Against Unforeseen Adversaries | ["Daniel Kang*", "Yi Sun*", "Dan Hendrycks", "Tom Brown", "Jacob Steinhardt"] | Most existing defenses against adversarial attacks only consider robustness to L_p-bounded distortions. In reality, the specific attack is rarely known in advance and adversaries are free to modify images in ways which lie outside any fixed distortion model; for example, adversarial rotations lie outside the set of L_p... | ["adversarial examples", "adversarial training", "adversarial attacks"] | ABSTRACTMost existing defenses against adversarial attacks only consider robustness to Lp-bounded distortions. In reality, the specific attack is rarely known in advance andadversaries are free to modify images in ways which lie outside any fixed distor-tion model; for example, image rotations lie outside the set of Lp... | BJgnvSjpFr | Official Blind Review #2 | 3: Weak Reject | This paper considers the robustness against adversarial attacks in learning of deep neural networks. Other than the usual Lp-distortions, the authors introduce novel adversarial distortions, JPEG, Fog, Gabor and Snow. Then, the authors propose the summary metric, unforeseen attack robustness (UAR), to measure the robus... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Testing Robustness Against Unforeseen Adversaries
### Paper Abstract
Most existing defenses against adversarial attacks only consider robustness to L_p-bounded distortio... | |
SJgmR0NKPr | ICLR.cc/2020/Conference | 2020 | Training Recurrent Neural Networks Online by Learning Explicit State Variables | ["Somjit Nath", "Vincent Liu", "Alan Chan", "Xin Li", "Adam White", "Martha White"] | Recurrent neural networks (RNNs) allow an agent to construct a state-representation from a stream of experience, which is essential in partially observable problems. However, there are two primary issues one must overcome when training an RNN: the sensitivity of the learning algorithm's performance to truncation length... | ["Recurrent Neural Network", "Partial Observability", "Online Prediction", "Incremental Learning"] | ABSTRACTRecurrent neural networks (RNNs) allow an agent to construct a state-representation from a stream of experience, which is essential in partially ob-servable problems. However, there are two primary issues one must overcomewhen training an RNN: the sensitivity of the learning algorithm’s performance totruncation... | SkeYFd2EqS | Official Blind Review #2 | 3: Weak Reject | In this paper, the authors reformulate the RNN training objective to explicitly learn the state vectors, and propose an algorithm called Fixed Point Propagation (FPP Algorithm 1). The authors motivate the derivation of FPP in Section 3, provide some theoretical convergence results in Section 4, and demonstrate experime... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Training Recurrent Neural Networks Online by Learning Explicit State Variables
### Paper Abstract
Recurrent neural networks (RNNs) allow an agent to construct a state-re... | |
rJlf_RVKwr | ICLR.cc/2020/Conference | 2020 | Sensible adversarial learning | ["Jungeum Kim", "Xiao Wang"] | The trade-off between robustness and standard accuracy has been consistently reported in the machine learning literature. Although the problem has been widely studied to understand and explain this trade-off, no studies have shown the possibility of a no trade-off solution. In this paper, motivated by the fact that the... | ["adversarial learning", "deep neural networks", "trade-off", "margins", "sensible reversion", "sensible robustness"] | ABSTRACTThe trade-off between robustness and standard accuracy has been consistently re-ported in the machine learning literature. Although the problem has been widelystudied to understand and explain this trade-off, the problem seems to remain asan open problem. In this paper, motivated by the fact that the high dimen... | BygQh5fMoH | Official Blind Review #3 | 3: Weak Reject | Summary:
The paper studies the phenomenon of trade-off between robust and standard accuracies that is usually observed in adversarial training. Many existing studies try to understand this trade-off and show that it is unavoidable. In contrast, this work shows that under a sensible definition of adversarial risk, there... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Sensible adversarial learning
### Paper Abstract
The trade-off between robustness and standard accuracy has been consistently reported in the machine learning literature... | |
SyeYiyHFDH | ICLR.cc/2020/Conference | 2020 | Convergence Analysis of a Momentum Algorithm with Adaptive Step Size for Nonconvex Optimization | ["Anas Barakat", "Pascal Bianchi"] | Although Adam is a very popular algorithm for optimizing the weights of neural networks, it has been recently shown that it can diverge even in simple convex optimization examples. Therefore, several variants of Adam have been proposed to circumvent this convergence issue. In this work, we study the algorithm for smoot... | ["nonconvex optimization", "adaptive methods"] | ABSTRACTAlthough A DAM is a very popular algorithm for optimizing the weights of neuralnetworks, it has been recently shown that it can diverge even in simple convex op-timization examples. Several variants of A DAM have been proposed to circumventthis convergence issue. In this work, we study the A DAM algorithm for s... | H1exRRr0FH | Official Blind Review #3 | 3: Weak Reject | This work analyzes the performance of ADAM algorithm under a bounded step size assumption. it proves convergence rate for the deterministic setting and prove a convergence result for the stochastic setting.
However, I have a few concerns as follows. The first two comments are the major reasons of my rating.
1. the 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
Convergence Analysis of a Momentum Algorithm with Adaptive Step Size for Nonconvex Optimization
### Paper Abstract
Although Adam is a very popular algorithm for optimizi... | |
HCiIL2hbU-0 | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | Reasoning over Bayesian Networks using Semantic Artificial Neural Networks | ["Sotirios Batsakis", "Grigoris Antoniou"] | Representation of application domains, related concepts and their dependencies is often achieved using Bayesian Networks. In Bayesian Networks nodes represent random variables and arcs represent their dependencies. Since inference over Bayesian Networks is a complex task in this work a novel approach for representing a... | ["Graph based knowledge representation", "Beyesian Nets", "Semantic Neural Networks"] | Reasoning over Bayesian Networks using SemanticArtificial Neural NetworksSotirios Batsakis1;2( )and Grigoris Antoniou11University of Huddersfield, UK2Technical University of Crete, Greecefs.batsakis,g.antoniou g@hud.ac.ukAbstract. Representation of application domains, related concepts and their de-pendencies is often ... | 3JVOr3B3TNi | self-contained paper | 7: Good paper, accept | This paper further elaborates on the idea of Semantic Artificial Neural Networks, which is conceptually very needed line of research.
The paper, although necessarily brief, contains all elements of a research paper. The authors have even managed to present pseudocode, example and a mini-benchmark on three datasets.
P... | 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
Reasoning over Bayesian Networks using Semantic Artificial Neural Networks
### Paper Abstract
Representation of application domains, related concepts and their dependenc... |
BJInEZsTb | ICLR.cc/2018/Conference | 2018 | Learning Representations and Generative Models for 3D Point Clouds | ["Panos Achlioptas", "Olga Diamanti", "Ioannis Mitliagkas", "Leonidas Guibas"] | Three-dimensional geometric data offer an excellent domain for studying representation learning and generative modeling. In this paper, we look at geometric data represented as point clouds. We introduce a deep autoencoder (AE) network with excellent reconstruction quality and generalization ability. The learned repres... | ["representation learning", "auto-encoders", "3D point clouds", "generative models", "GANs", "Gaussian Mixture Models"] | ABSTRACTThree-dimensional geometric data offer an excellent domain for studying repre-sentation learning and generative modeling. In this paper, we look at geometricdata represented as point clouds. We introduce a deep autoencoder (AE) networkwith state-of-the-art reconstruction quality and generalization ability. The ... | B1Mvg-qlM | Well written paper with new important results | 8: Top 50% of accepted papers, clear accept | 3D data processing is very important topic nowadays, since it has a lot of applications: robotics, AR/VR, etc.
Current approaches to 2D image processing based on Deep Neural Networks provide very accurate results and a wide variety of different architectures for image modelling, generation, classification, retrieval.
... | 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
Learning Representations and Generative Models for 3D Point Clouds
### Paper Abstract
Three-dimensional geometric data offer an excellent domain for studying representat... |
RayUtcIlGz | ICLR.cc/2021/Conference | 2021 | Training Invertible Linear Layers through Rank-One Perturbations | ["Andreas Kr\u00e4mer", "Jonas K\u00f6hler", "Frank Noe"] | Many types of neural network layers rely on matrix properties such as invertibility or orthogonality.
Retaining such properties during optimization with gradient-based stochastic optimizers is a challenging task, which is usually addressed by either reparameterization of the affected parameters or by directly optimizin... | ["Parameter Perturbation", "Reparameterization", "Invertible Neural Networks", "Normalizing Flows", "Rank-one update"] | ABSTRACTMany types of neural network layers rely on matrix properties such as invertibil-ity or orthogonality. Retaining such properties during optimization with gradient-based stochastic optimizers is a challenging task, which is usually addressed byeither reparameterization of the affected parameters or by directly o... | FOwNrGzxJ4z | Parameter perturbations for building invertible matrices applied to density estimation | 6: Marginally above acceptance threshold | # Summary
This work proposes parameter perturbations for the training of neural networks while preserving some properties. It particularly focuses on the case of invertible layers and demonstrates improvements over classical architectures for density estimation with normalizing flows. It is empirically shown that param... | 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
Training Invertible Linear Layers through Rank-One Perturbations
### Paper Abstract
Many types of neural network layers rely on matrix properties such as invertibility o... |
DILxQP08O3B | ICLR.cc/2021/Conference | 2021 | VTNet: Visual Transformer Network for Object Goal Navigation | ["Heming Du", "Xin Yu", "Liang Zheng"] | Object goal navigation aims to steer an agent towards a target object based on observations of the agent. It is of pivotal importance to design effective visual representations of the observed scene in determining navigation actions. In this paper, we introduce a Visual Transformer Network (VTNet) for learning informa... | ["vtnet", "visual transformer network", "agent", "visual representations", "navigation actions", "navigation", "target object", "observations"] | ABSTRACTObject goal navigation aims to steer an agent towards a target object based onobservations of the agent. It is of pivotal importance to design effective visualrepresentations of the observed scene in determining navigation actions. In thispaper, we introduce a Visual Transformer Network (VTNet) for learning inf... | OOSgzqW2IDO | Effective method, good results, but writing requires improvement | 6: Marginally above acceptance threshold | **Paper summary**
The paper addresses the problem of navigation towards objects (ObjectNav) in a virtual environment. The idea of the paper is to incorporate spatial information of objects using a transformer-based framework called Visual Transformer Network. The paper compares the results with a number of state-of-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
VTNet: Visual Transformer Network for Object Goal Navigation
### Paper Abstract
Object goal navigation aims to steer an agent towards a target object based on observatio... |
SkHDoG-Cb | ICLR.cc/2018/Conference | 2018 | Simulated+Unsupervised Learning With Adaptive Data Generation and Bidirectional Mappings | ["Kangwook Lee", "Hoon Kim", "Changho Suh"] | Collecting a large dataset with high quality annotations is expensive and time-consuming. Recently, Shrivastava et al. (2017) propose Simulated+Unsupervised (S+U) learning: It first learns a mapping from synthetic data to real data, translates a large amount of labeled synthetic data to the ones that resemble real data... | ["bidirectional mappings", "mapping", "synthetic data", "adaptive data generation", "real data", "learning model", "algorithms", "performances", "various tasks", "room"] | ABSTRACTCollecting a large dataset with high quality annotations is expensive and time-consuming. Recently, Shrivastava et al. (2017) propose Simulated+Unsupervised(S+U) learning : It first learns a mapping from synthetic data to real data, trans-lates a large amount of labeled synthetic data to the ones that resemble ... | S1uLIj8lG | This paper considers a generative approach in semi-supervised setting, which is basically a combination of cycleGAN and the Apple's S+U leaning. | 6: Marginally above acceptance threshold | * sec.2.2 is about label-preserving translation and many notations are introduced. However, it is not clear what label here refers to, and it does not shown in the notation so far at all. Only until the end of sec.2.2, the function F(.) is introduced and its revelation - Google Search as label function is discussed onl... | 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
Simulated+Unsupervised Learning With Adaptive Data Generation and Bidirectional Mappings
### Paper Abstract
Collecting a large dataset with high quality annotations is e... |
rkLyJl-0- | ICLR.cc/2018/Conference | 2018 | Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks | ["Shankar Krishnan", "Ying Xiao", "Rif. A. Saurous"] | Progress in deep learning is slowed by the days or weeks it takes to train large models. The natural solution of using more hardware is limited by diminishing returns, and leads to inefficient use of additional resources. In this paper, we present a large batch, stochastic optimization algorithm that is both faster tha... | ["Deep Learning", "Optimization"] | ABSTRACTProgress in deep learning is slowed by the days or weeks it takes to train largemodels. The natural solution of using more hardware is limited by diminishing re-turns, and leads to inefficient use of additional resources. In this paper, we presenta large batch, stochastic optimization algorithm that is both fas... | ByPYAMtgG | [UPDATED] Interesting idea, nice experiments, good motivation but lacks theoretical understanding | 6: Marginally above acceptance threshold | The paper proposes a new algorithm, where they claim to use Hessian implicitly and are using a motivation from power-series. In general, I like the paper.
To me, Algorithm 1 looks like some kind of proximal-point type algorithm. Algorithm 2 is more heuristic approach, with a couple of parameters to tune it. Given the... | 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
Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks
### Paper Abstract
Progress in deep learning is slowed by the days or weeks it takes to tr... |
JbAqsfbYsJy | ICLR.cc/2021/Conference | 2021 | Action and Perception as Divergence Minimization | ["Danijar Hafner", "Pedro A Ortega", "Jimmy Ba", "Thomas Parr", "Karl Friston", "Nicolas Heess"] | We introduce a unified objective for action and perception of intelligent agents. Extending representation learning and control, we minimize the joint divergence between the combined system of agent and environment and a target distribution. Intuitively, such agents use perception to align their beliefs with the world,... | ["objective functions", "reinforcement learning", "information theory", "probabilistic modeling", "control as inference", "exploration", "intrinsic motivation", "world models"] | ABSTRACTWe introduce a unified objective for action and perception of intelligent agents.Extending representation learning and control, we minimize the joint divergencebetween the combined system of agent and environment and a target distribution.Intuitively, such agents use perception to align their beliefs with the w... | nDlpy1miu1R | This paper discusses about the general training objective for perception, action and beliefs. The authors formulate a joint KL minimization objective and derived modern methods like variational inference, control, MaxEnt RL to show the correspondance to their unified framework. This work is a theorectical-based paper ... | 6: Marginally above acceptance threshold | The authors proposed to use the joint KL divergence between the generative joint distribution and the target distribution (containing latent variables which could correspond to latent parts we wanted to model (e.g. beliefs). It was illustrative to discuss decomposing the joint KL into different ways and thus forming in... | 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
Action and Perception as Divergence Minimization
### Paper Abstract
We introduce a unified objective for action and perception of intelligent agents. Extending represent... |
rylUOn4Yvr | ICLR.cc/2020/Conference | 2020 | ROBUST DISCRIMINATIVE REPRESENTATION LEARNING VIA GRADIENT RESCALING: AN EMPHASIS REGULARISATION PERSPECTIVE | ["Xinshao Wang", "Yang Hua", "Elyor Kodirov", "Neil M. Robertson"] | It is fundamental and challenging to train robust and accurate Deep Neural Networks (DNNs) when semantically abnormal examples exist. Although great progress has been made, there is still one crucial research question which is not thoroughly explored yet: What training examples should be focused and how much more shoul... | ["examples weighting", "emphasis regularisation", "gradient scaling", "abnormal training examples"] | ABSTRACTIt is fundamental and challenging to train robust and accurate Deep Neural Net-works (DNNs) when semantically abnormal examples exist. Although greatprogress has been made, there is still one crucial research question which is notthoroughly explored yet: What training examples should be focused on and howmuch m... | rJgz97JRYB | Official Blind Review #1 | 6: Weak Accept | Summary:
The paper proposes a method for noise robustness based on scaling gradients of examples. By choosing the proper scaling parameters (alpha and beta), the method recovers standard losses such as CCE, MAE, and GCE, while also recovering other losses. The method is strongly related to reweighting training examples... | <|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 DISCRIMINATIVE REPRESENTATION LEARNING VIA GRADIENT RESCALING: AN EMPHASIS REGULARISATION PERSPECTIVE
### Paper Abstract
It is fundamental and challenging to trai... | |
rkgOuJBtPS | ICLR.cc/2020/Conference | 2020 | Task-Agnostic Robust Encodings for Combating Adversarial Typos | ["Erik Jones", "Robin Jia", "Aditi Raghunathan", "Percy Liang"] | Despite achieving excellent benchmark performance, state-of-the-art NLP models can still be easily fooled by adversarial perturbations such as typos. Previous heuristic defenses cannot guard against the exponentially large number of possible perturbations, and previous certified defenses only work with limited model si... | ["Natural language processing", "adversarial examples", "robustness"] | ABSTRACTDespite achieving excellent benchmark performance, state-of-the-art NLP modelscan still be easily fooled by adversarial perturbations such as typos. Previousheuristic defenses cannot guard against the exponentially large number of possibleperturbations, and previous certified defenses only work with limited mod... | Skx73bFAYr | Official Blind Review #1 | 3: Weak Reject | This paper explores adversarially robust neural encodings for text that ensure robustness to text typos. The key idea is to balance two criteria, namely (i) the fidelity of a model, i.e., accuracy for a task, and (ii) stability, that captures the robustness of the model to perturbations. The paper proposes two methods ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Task-Agnostic Robust Encodings for Combating Adversarial Typos
### Paper Abstract
Despite achieving excellent benchmark performance, state-of-the-art NLP models can stil... | |
Vzp2aRidnh | NoDaLiDa/2023/Conference | 2023 | ASR Language Resources for Faroese | ["Carlos Daniel Hern\u00e1ndez Mena", "Annika Simonsen", "Jon Gudnason"] | The aim of this work is to present a set of novel language resources in Faroese suitable for the field of Automatic Speech Recognition including: an ASR corpus comprised of 109 hours of transcribed speech data, acoustic models in systems such as WAV2VEC2, NVIDIA-NeMo, Kaldi and PocketSphinx; a set of n-gram language m... | ["Faroese", "Acoustic Model", "Faroese ASR", "Language Model", "Pronunciation Dictionary", "Faroese Language Resources"] | ASR Language Resources for FaroeseCarlos MenaLanguage and V oice Lab,Reykjavík University,102 Reykjavík, Icelandcarlosm@ru.isAnnika SimonsenUniversity of Iceland,107 Reykjavík, Icelandannika@hi.isJón GuðnasonLanguage and V oice Lab,Reykjavík University,102 Reykjavík, Icelandjg@ru.isAbstractThe aim of this work is... | nnmzj2wCDX | Excellent work | 8: Top 50% of accepted papers, clear accept | This is a very good paper that describes the latest resources for Faroese ASR. It is well-written, clear, and definitely helpful for anybody who would like to work with Faroese ASR. I fully agree with the claim that authors make in conclusions - Faroese ASR is no longer under-developed due to this work. (line 746)
Stil... | 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
ASR Language Resources for Faroese
### Paper Abstract
The aim of this work is to present a set of novel language resources in Faroese suitable for the field of Automati... |
SJlEs1HKDr | ICLR.cc/2020/Conference | 2020 | Attentive Sequential Neural Processes | ["Jaesik Yoon", "Gautam Singh", "Sungjin Ahn"] | Sequential Neural Processes (SNP) is a new class of models that can meta-learn a temporal stochastic process of stochastic processes by modeling temporal transition between Neural Processes. As Neural Processes (NP) suffers from underfitting, SNP is also prone to the same problem, even more severely due to its temporal... | ["meta-learning", "neural processes", "attention", "sequential modeling"] | ABSTRACTSequential Neural Processes (SNP) is a new class of models that temporally ex-tends Neural Processes (NP) and can meta-learn a sequence of stochastic pro-cesses. This learned function however under-fits the provided contexts as is alsothe case in NP. Applying attention to the contexts resolves this in NP but si... | ryeKOFZW5S | Official Blind Review #3 | 3: Weak Reject | This paper deals with the underfitting problem happening in neural process and sequential neural process (SNP). The idea is to incorporate the attention scheme in SNP and carry out the so-called attentive sequential neural process (ASNP) for sequence learning.
Strength:
1. A combination of attention into SNP.
2. Some ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Attentive Sequential Neural Processes
### Paper Abstract
Sequential Neural Processes (SNP) is a new class of models that can meta-learn a temporal stochastic process of ... | |
BJewlyStDr | ICLR.cc/2020/Conference | 2020 | On Bonus Based Exploration Methods In The Arcade Learning Environment | ["Adrien Ali Taiga", "William Fedus", "Marlos C. Machado", "Aaron Courville", "Marc G. Bellemare"] | Research on exploration in reinforcement learning, as applied to Atari 2600 game-playing, has emphasized tackling difficult exploration problems such as Montezuma's Revenge (Bellemare et al., 2016). Recently, bonus-based exploration methods, which explore by augmenting the environment reward, have reached above-human a... | ["exploration", "arcade learning environment", "bonus-based methods"] | ABSTRACTResearch on exploration in reinforcement learning, as applied to Atari 2600 game-playing, has emphasized tackling difficult exploration problems such as MON-TEZUMA ’SREVENGE (Bellemare et al., 2016). Recently, bonus-based explorationmethods, which explore by augmenting the environment reward, have reachedabove-... | SklK7x6-cS | Official Blind Review #3 | 6: Weak Accept | #rebuttal responses
I change the score to be weak accept as the authors do not provide any comparison result on Rainbow without the prioritized replay buffer during the rebuttal phase. I also agree with Reviewer 1's opinion that the authors do not provide some fixing method, such as combining the noisy networks and 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
On Bonus Based Exploration Methods In The Arcade Learning Environment
### Paper Abstract
Research on exploration in reinforcement learning, as applied to Atari 2600 game... | |
h1MwZpU6Iv | MIDL.io/2020/Conference | 2020 | The Performance of Deep U-Net Pre-Clinical Organ-wise Segmentation in the Presence of Low Counting Statistics. | ["R L Smith", "Noah Evans", "Victor Florence", "S Paisey", "C Marshall"] | Micro-PET-CT allows non-invasive monitoring of biological processes, disease progression and therapy response. Morphological information provided by the CT allows organ / tissue delineation for subsequent quantification of the physiological information depicted by the PET. Deep learning with convolutional neural networ... | ["deep learning", "pre-clinical imaging", "PET-CT", "image segmentation", "artificial intelligence"] | h25PR6dVQy | Interesting research question on the influence of CT dosage on segmentation accuracy of DL models, with several essential missing details. | 1: Strong reject | Summary:
A pre-trained organ segmentation network based on CNN model is evaluated with different noise and CT dose settings. These variations are simulated by adding structured noise. It is reported that the DICE accuracy is reasonable even when CT dosage is reduced by 30%
Strengths:
+ The question of reducing CT d... | 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
The Performance of Deep U-Net Pre-Clinical Organ-wise Segmentation in the Presence of Low Counting Statistics.
### Paper Abstract
Micro-PET-CT allows non-invasive monito... | |
B4SHgqe1kvX | ICLR.cc/2021/Conference | 2021 | Accurate Word Representations with Universal Visual Guidance | ["Haojie Yu", "Zhuosheng Zhang", "hai zhao", "Rui Wang", "Masao Utiyama"] | Word representation is a fundamental component in neural language understanding models. Recently, pre-trained language models (PrLMs) offer a new performant method of contextualized word representations by leveraging the sequence-level context for modeling. Although the PrLMs generally give more accurate contextualized... | ["Natural Language Processing", "Visual Representation", "Multimodal Language Representation", "Natural Language Understanding"] | ABSTRACTWord representation is a fundamental component in neural language understand-ing models. Recently, pre-trained language models (PrLMs) offer a new perfor-mant method of contextualized word representations by leveraging the sequence-level context for modeling. Although the PrLMs generally give more accurateconte... | ihRVb9C7EN | The idea is interesting, but there are flaws in experiments and model design. | 4: Ok but not good enough - rejection |
1.Objective of the work: This work proposes to integrate external visual knowledge into the contextualized word representations (CWP). The goal is to equip CWP with the ability to disambiguate and make CWP "more contextualized."
2.Metrits of the paper: Most multimodal researches focus on solving multimodal tasks, li... | 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
Accurate Word Representations with Universal Visual Guidance
### Paper Abstract
Word representation is a fundamental component in neural language understanding models. R... |
HyyHX4gZM | ICLR.cc/2018/Conference | 2018 | Towards Quantum Inspired Convolution Networks | ["Davi Geiger", "Zvi Kedem"] | Deep Convolution Neural Networks (CNNs), rooted by the pioneer work of \cite{Hinton1986,LeCun1985,Alex2012}, and summarized in \cite{LeCunBengioHinton2015}, have been shown to be very useful in a variety of fields. The state-of-the art CNN machines such as image rest net \cite{He_2016_CVPR} are described by real v... | ["quantum technique", "convolution networks", "shape detection"] | ABSTRACTDeep Convolution Neural Networks (CNNs), rooted by the pioneer work ofRumelhart et al. (1986); LeCun (1985); Krizhevsky et al. (2012), and summa-rized in LeCun et al. (2015), have been shown to be very useful in a variety offields. The state-of-the art CNN machines such as image rest net He et al. (2016)are des... | ry_O471GG | Really related to CNNs? | 5: Marginally below acceptance threshold | This paper presented the study on complex kernel for shape detection, which is inspired by quantum theory. This paper started from the description of the shape detection problem, then introduced the traditional Bayesian and Hough transform approaches. In Section 4, the proposed quantum inspired complex kernel was descr... | 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
Towards Quantum Inspired Convolution Networks
### Paper Abstract
Deep Convolution Neural Networks (CNNs), rooted by the pioneer work of \cite{Hinton1986,LeCun1985,Alex2... |
Bys4ob-Rb | ICLR.cc/2018/Conference | 2018 | Certified Defenses against Adversarial Examples | ["Aditi Raghunathan", "Jacob Steinhardt", "Percy Liang"] | While neural networks have achieved high accuracy on standard image classification benchmarks, their accuracy drops to nearly zero in the presence of small adversarial perturbations to test inputs. Defenses based on regularization and adversarial training have been proposed, but often followed by new, stronger attacks ... | ["adversarial examples", "certificate of robustness", "convex relaxations"] | ABSTRACTWhile neural networks have achieved high accuracy on standard image classifi-cation benchmarks, their accuracy drops to nearly zero in the presence of smalladversarial perturbations to test inputs. Defenses based on regularization and ad-versarial training have been proposed, but often followed by new, stronger... | SkwJQwogM | The main idea of using upper bound (as opposed to lower bound) is reasonable. However, I find there are some limitations/weakness of the proposed method: | 5: Marginally below acceptance threshold | This paper derived an upper bound on adversarial perturbation for neural networks with one hidden layer. The upper bound is derived via (1) theorem of middle value; (2) replace the middle value by the maximum (eq 4); (3) replace the maximum of the gradient value (locally) by the global maximal value (eq 5); (4) this le... | 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
Certified Defenses against Adversarial Examples
### Paper Abstract
While neural networks have achieved high accuracy on standard image classification benchmarks, their ... |
vsxYOZoPvne | eswc-conferences.org/ESWC/2021/Conference/Research_Track | 2021 | Do Embeddings Actually Capture Knowledge Graph Semantics? | ["Nitisha Jain", "Jan-Christoph Kalo", "Wolf-Tilo Balke", "Ralf Krestel"] | Knowledge graph embeddings that generate vector space representations of knowledge graph triples, have gained considerable popularity in past years. Several embedding models have been proposed that achieve state-of-the-art performance for the task of triple completion in knowledge graphs. Relying on the presumed semant... | ["knowledge graph embeddings", "semantic representation", "entity similarity"] | Do Embeddings Actually Capture KnowledgeGraph Semantics?Nitisha Jain1, Jan-Christoph Kalo2, Wolf-Tilo Balke2, Ralf Krestel11Hasso-Plattner-Institut, University of Potsdam, Germany2Technische Universit at Braunschweig, Germanyfnitisha.jain,ralf.krestel g@hpi.de,fkalo,balkeg@ifis.cs.tu-bs.deAbstract. Knowledge graph embe... | 4b0rBEG7Jh0 | The paper presents interesting criticism of widespread Knowledge Graph embedding models. The authors designed several experiments to show that KG embedding models did not capture semantics in the case of realistic datasets. While the paper is clearly written and I think it presents original ideas with a widespread exte... | 1: Weak Accept | I acknowledge that I have read the rebuttal. I would like to thank you to the authors for their answers. I modified the review accordingly.
==
Nowadays embeddings is a hot topic for many fields with regard to a representation. Knowledge Graph Embedding Models are widely used in Semantic Web. Systems based on embeddin... | 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
Do Embeddings Actually Capture Knowledge Graph Semantics?
### Paper Abstract
Knowledge graph embeddings that generate vector space representations of knowledge graph tri... |
zgMPc_48Zb | ICLR.cc/2021/Conference | 2021 | Differentially Private Generative Models Through Optimal Transport | ["Tianshi Cao", "Alex Bie", "Karsten Kreis", "Sanja Fidler"] | Although machine learning models trained on massive data have led to breakthroughs in several areas, their deployment in privacy-sensitive domains remains limited due to restricted access to data. Generative models trained with privacy constraints on private data can sidestep this challenge and provide indirect access ... | ["Differential Privacy", "Generative Learning", "GAN", "Optimal Transport"] | ABSTRACTAlthough machine learning models trained on massive data have led to break-throughs in several areas, their deployment in privacy-sensitive domains remainslimited due to restricted access to data. Generative models trained with privacyconstraints on private data can sidestep this challenge and provide indirect ... | pf6MMEK8z2h | Proposed method is not well motivated and is not novel | 4: Ok but not good enough - rejection | This paper presents a differentially private method for training a generative model. The proposed method takes advantage of the Sinkhorn divergence to achieve robustness against the hyperparameters' choices. The authors also introduce a cost function enabling the generative model to generate images associated with a sp... | 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
Differentially Private Generative Models Through Optimal Transport
### Paper Abstract
Although machine learning models trained on massive data have led to breakthroughs ... |
H1WgVz-AZ | ICLR.cc/2018/Conference | 2018 | Learning Approximate Inference Networks for Structured Prediction | ["Lifu Tu", "Kevin Gimpel"] | Structured prediction energy networks (SPENs; Belanger & McCallum 2016) use neural network architectures to define energy functions that can capture arbitrary dependencies among parts of structured outputs. Prior work used gradient descent for inference, relaxing the structured output to a set of continuous variables a... | ["Approximate Inference Networks", "Structured Prediction", "Multi-Label Classification", "Sequence Labeling"] | ABSTRACTStructured prediction energy networks (SPENs; Belanger & McCallum 2016) useneural network architectures to define energy functions that can capture arbitrarydependencies among parts of structured outputs. Prior work used gradient descentfor inference, relaxing the structured output to a set of continuous variab... | Sk0CEftxG | New idea for training structured predictors, but unclear motivation and evaluation | 5: Marginally below acceptance threshold | The paper proposes training ``inference networks,'' which are neural network structured predictors. The setup is analogous to generative adversarial networks, where the role of the discriminator is played by a structured prediction energy network (SPEN) and the generator is played by an inference network.
The idea is ... | 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 Approximate Inference Networks for Structured Prediction
### Paper Abstract
Structured prediction energy networks (SPENs; Belanger & McCallum 2016) use neural n... |
_O9YLet0wvN | ICLR.cc/2021/Conference | 2021 | Closing the Generalization Gap in One-Shot Object Detection | ["Claudio Michaelis", "Matthias Bethge", "Alexander S Ecker"] | Despite substantial progress in object detection and few-shot learning, detecting objects based on a single example - one-shot object detection - remains a challenge. A central problem is the generalization gap: Object categories used during training are detected much more reliably than novel ones. We here show that th... | ["One-Shot Learning", "Few-Shot Learning", "Object Detection", "One-Shot Object Detection", "Generalization"] | ABSTRACTDespite substantial progress in object detection and few-shot learning, detectingobjects based on a single example – one-shot object detection – remains a chal-lenge. A central problem is the generalization gap: Object categories used duringtraining are detected much more reliably than novel ones. We here show ... | uUPMDaNplpb | Further study is needed to verify the generalization ability of the main claims. | 6: Marginally above acceptance threshold | This paper provides a variety of studies to understand the generalization gap between known and novel classes in one-shot object detection. The studies are carried out by using siamese Faster R-CNN framework on four benchmark datasets. The most notable observation was that it was more important to increase the number o... | 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
Closing the Generalization Gap in One-Shot Object Detection
### Paper Abstract
Despite substantial progress in object detection and few-shot learning, detecting objects ... |
pAj7zLJK05U | ICLR.cc/2021/Conference | 2021 | AttackDist: Characterizing Zero-day Adversarial Samples by Counter Attack | ["Simin Chen", "Zihe Song", "Lei Ma", "Cong Liu", "Wei Yang"] | Deep Neural Networks (DNNs) have been shown vulnerable to adversarial attacks, which could produce adversarial samples that easily fool the state-of-the-art DNNs. The harmfulness of adversarial attacks calls for the defense mechanisms under fire. However, the relationship between adversarial attacks and defenses is lik... | ["adversarial samples", "attackdist", "adversarial attacks", "dnns", "defense", "perturbation", "counter attack attackdist", "vulnerable", "harmfulness"] | ABSTRACTDeep Neural Networks (DNNs) have been shown vulnerable to adversarial at-tacks, which could produce adversarial samples that easily fool the state-of-the-artDNNs.The harmfulness of adversarial attacks calls for the defense mechanismsunder fire. However, the relationship between adversarial attacks and defensesi... | nv8uxq46l1A | NA | 5: Marginally below acceptance threshold | Summary: this paper is about adversarial detection based on counter-attack. The main intuition is that the adversarial sample lies closer to the decision boundary and hence if we do counter-attack to the data, the perturbation is expected to be much smaller than the clean data.
Although the idea sounds interesting, I ... | 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
AttackDist: Characterizing Zero-day Adversarial Samples by Counter Attack
### Paper Abstract
Deep Neural Networks (DNNs) have been shown vulnerable to adversarial attack... |
b9g0vxzYa_ | logconference.io/LOG/2022/Conference | 2022 | Influence-Based Mini-Batching for Graph Neural Networks | ["Johannes Gasteiger", "Chendi Qian", "Stephan G\u00fcnnemann"] | Using graph neural networks for large graphs is challenging since there is no clear way of constructing mini-batches. To solve this, previous methods have relied on sampling or graph clustering. While these approaches often lead to good training convergence, they introduce significant overhead due to expensive random d... | ["GNN", "graph neural network", "graph", "scalability", "batching", "influence", "local clustering"] | Influence-Based Mini-Batching for Graph Neural NetworksJohannes Gasteiger∗Technical University of Munich†johannes.gasteiger@tum.deChendi Qian∗Technical University of Munichchendi.qian@tum.deStephan GünnemannTechnical University of Munichstephan.guennemann@tum.deAbstractUsing graph neural networks for large graphs is c... | EceCkd18sV | This paper proposes a mini-batching approach for training GNNs with a focus on model behavior during inference. The proposed mini-batching is based on the influence score of nodes on the outputs. The influence scores between nodes are approximated with Personalized PageRank.
The experimental results show the proposed... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Influence-Based Mini-Batching for Graph Neural Networks
### Paper Abstract
Using graph neural networks for large graphs is challenging since there is no clear way of con... | |||
C70cp4Cn32 | ICLR.cc/2021/Conference | 2021 | Multi-Level Local SGD: Distributed SGD for Heterogeneous Hierarchical Networks | ["Timothy Castiglia", "Anirban Das", "Stacy Patterson"] | We propose Multi-Level Local SGD, a distributed stochastic gradient method for learning a smooth, non-convex objective in a multi-level communication network with heterogeneous workers. Our network model consists of a set of disjoint sub-networks, with a single hub and multiple workers; further, workers may have differ... | ["Machine Learning", "Stochastic Gradient Descent", "Federated Learning", "Hierarchical Networks", "Distributed", "Heterogeneous", "Convergence Analysis"] | ABSTRACTWe propose Multi-Level Local SGD, a distributed stochastic gradient method forlearning a smooth, non-convex objective in a multi-level communication networkwith heterogeneous workers. Our network model consists of a set of disjoint sub-networks, with a single hub and multiple workers; further, workers may have ... | h2kaDsr4b0y | Unsure about how this compares to grouping SGD | 6: Marginally above acceptance threshold | The paper proposed a multi level SGD algorithm, where workers are assigned to different groups, and each group averages local workers' model.
The idea is a natural extension on existing SGD algorithms (and should be beneficial in heterogeneous networks). The proof should not be hard based on existing asynchronous/dec... | 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
Multi-Level Local SGD: Distributed SGD for Heterogeneous Hierarchical Networks
### Paper Abstract
We propose Multi-Level Local SGD, a distributed stochastic gradient met... |
y2I4gyAGlCB | ICLR.cc/2021/Conference | 2021 | Imagine That! Leveraging Emergent Affordances for 3D Tool Synthesis | ["Yizhe Wu", "Sudhanshu Kasewa", "Oliver Groth", "Sasha Salter", "Kevin Li Sun", "Oiwi Parker Jones", "Ingmar Posner"] | In this paper we explore the richness of information captured by the latent space of a vision-based generative model. The model combines unsupervised generative learning with a task-based performance predictor to learn and to exploit task-relevant object affordances given visual observations from a reaching task, invo... | ["Affordance Learning", "Imagination", "Generative Models", "Activation Maximisation"] | ABSTRACTIn this paper we explore the richness of information captured by the latent spaceof a vision-based generative model. The model combines unsupervised genera-tive learning with a task-based performance predictor to learn and to exploit task-relevant objectaffordancesgiven visual observations from a reaching task,... | QnqwFETJPw | Initial review | 5: Marginally below acceptance threshold | Summary
-------
By combining a task-success classifier with the latent space of a tool-shape generative model, this paper shows that an activation-maximization approach can generate tool shapes which can succeed at particular tasks.
Positives
---------
The paper addresses the interesting topic of affordances, and ho... | 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
Imagine That! Leveraging Emergent Affordances for 3D Tool Synthesis
### Paper Abstract
In this paper we explore the richness of information captured by the latent space ... |
BkghKgStPH | ICLR.cc/2020/Conference | 2020 | Continual Learning using the SHDL Framework with Skewed Replay Distributions | ["Amarjot Singh", "Jay McClelland"] | Human and animals continuously acquire, adapt as well as transfer knowledge throughout their lifespan. The ability to learn continuously is crucial for the effective functioning of agents interacting with the real world and processing continuous streams of information. Continuous learning has been a long-standing chall... | ["Continual Learning", "Catastrophic Forgetting", "SHDL", "CIFAR-100"] | ABSTRACTHuman and animals continuously acquire, adapt as well as transfer knowledgethroughout their lifespan. The ability to learn continuously is crucial for the ef-fective functioning of agents interacting with the real world and processing con-tinuous streams of information. Continuous learning has been a long-stand... | HJgg5GG59B | Official Blind Review #4 | 1: Reject | I think there might be some interesting ideas in the work, but I think the authors somehow did not manage to position themselves well within the *recent* works on the topic or even with respect to what continual learning (CL) is understood to be in these recent works.
E.g. CL is a generic learning problem, and most a... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Continual Learning using the SHDL Framework with Skewed Replay Distributions
### Paper Abstract
Human and animals continuously acquire, adapt as well as transfer knowled... | |
AwPGPgExiYA | ICLR.cc/2021/Conference | 2021 | Differentiable Learning of Graph-like Logical Rules from Knowledge Graphs | ["Hongzhi Shi", "quanming yao", "Yong Li"] | Logical rules inside a knowledge graph (KG) are essential for reasoning, logical inference, and rule mining. However, existing works can only handle simple, i.e., chain-like and tree-like, rules and cannot capture KG's complex semantics, which can be better captured by graph-like rules. Besides, learning graph-like rul... | ["knowledge graph", "logical rules", "logical query"] | ABSTRACTLogical rules inside a knowledge graph (KG) are essential for reasoning, logicalinference, and rule mining. However, existing works can only handle simple, i.e.,chain-like and tree-like, rules and cannot capture KG’s complex semantics, whichcan be better captured by graph-like rules. Besides, learning graph-lik... | pHLDv1Et_yo | Addressing a difficult problem with moderate success | 5: Marginally below acceptance threshold | This paper addresses the problem of logic structure learning, AKA, finding rules that generate a noisy knowledge graph. The innovation here is that the learning algorithm can address a much broader spectrum of rules than current methods. The key ingredient of the proposed method is a differentiable scoring function tha... | 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
Differentiable Learning of Graph-like Logical Rules from Knowledge Graphs
### Paper Abstract
Logical rules inside a knowledge graph (KG) are essential for reasoning, log... |
HJlRFlHFPS | ICLR.cc/2020/Conference | 2020 | Unsupervised Distillation of Syntactic Information from Contextualized Word Representations | ["Shauli Ravfogel", "Yanai Elazar", "Jacob Goldberger", "Yoav Goldberg"] | Contextualized word representations, such as ELMo and BERT, were shown to perform well on a various of semantic and structural (syntactic) task. In this work, we tackle the task of unsupervised disentanglement between semantics and structure in neural language representations: we aim to learn a transformation of the co... | ["dismantlement", "contextualized word representations", "language models", "representation learning"] | ABSTRACTContextualized word representations, such as ELMo and BERT, were shown toperform well on various semantic and structural (syntactic) task. In this work,we tackle the task of unsupervised disentanglement between semantics and struc-ture in neural language representations: we aim to learn a transformation of thec... | rJxPTT-IYH | Official Blind Review #2 | 6: Weak Accept | Summary:
=========
This paper aims to disentangle semantics and syntax in contextualized word representations. The main idea is to learn a transformation of the contexualized representations that will make two word representations to be more similar if they appear in the same syntactic context, but less similar if they... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Unsupervised Distillation of Syntactic Information from Contextualized Word Representations
### Paper Abstract
Contextualized word representations, such as ELMo and BERT... | |
rklXaoAcFX | ICLR.cc/2019/Conference | 2019 | Geomstats: a Python Package for Riemannian Geometry in Machine Learning | ["Nina Miolane", "Johan Mathe", "Claire Donnat", "Mikael Jorda", "Xavier Pennec"] | We introduce geomstats, a Python package for Riemannian modelization and optimization over manifolds such as hyperspheres, hyperbolic spaces, SPD matrices or Lie groups of transformations. Our contribution is threefold. First, geomstats allows the flexible modeling of many a machine learning problem through an efficien... | ["Riemannian geometry", "Python package", "machine learning", "deep learning"] | ABSTRACTWe introduce geomstats , a Python package for Riemannian modelization andoptimization over manifolds such as hyperspheres, hyperbolic spaces, spaces ofSPD matrices or Lie groups of transformations. Our contribution is threefold. First,geomstats allows the flexible modeling of many a machine learning problemthro... | SkefgeUtnX | geomstats has a potential to generate large impact to the community | 8: Top 50% of accepted papers, clear accept | This white paper presents the geomstats package. The package provides tools for Riemannian modelization and
optimization over manifolds. Especially, the package supports several important manifolds: hyperspheres, hyperbolic spaces, spaces of SPD matrices or Lie groups of transformations.
Pros:
1. the paper shows eve... | 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
Geomstats: a Python Package for Riemannian Geometry in Machine Learning
### Paper Abstract
We introduce geomstats, a Python package for Riemannian modelization and optim... |
ByxAOoR5K7 | ICLR.cc/2019/Conference | 2019 | Policy Generalization In Capacity-Limited Reinforcement Learning | ["Rachel A. Lerch", "Chris R. Sims"] | Motivated by the study of generalization in biological intelligence, we examine reinforcement learning (RL) in settings where there are information-theoretic constraints placed on the learner's ability to represent a behavioral policy. We first show that the problem of optimizing expected utility within capacity-limite... | ["reinforcement learning", "generalization", "capacity constraints", "information theory"] | ABSTRACTMotivated by the study of generalization in biological intelligence, we examinereinforcement learning (RL) in settings where there are information-theoreticconstraints placed on the learner’s ability to represent a behavioral policy. Wefirst show that the problem of optimizing expected utility within capacity-l... | HyeiEgwjn7 | Inciteful and of general interest. | 7: Good paper, accept | ## Summary
The authors identify a synergy between the rate distortion (RD) and reinforcement learning (RL) literature. RD work shows how to optimise resources when capacity is limited and the authors transfer this idea to RL and posit a novel algorithm based on the Actor critic algorithm. In experiments this is shown ... | 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
Policy Generalization In Capacity-Limited Reinforcement Learning
### Paper Abstract
Motivated by the study of generalization in biological intelligence, we examine reinf... |
XPZIaotutsD | ICLR.cc/2021/Conference | 2021 | DEBERTA: DECODING-ENHANCED BERT WITH DISENTANGLED ATTENTION | ["Pengcheng He", "Xiaodong Liu", "Jianfeng Gao", "Weizhu Chen"] | Recent progress in pre-trained neural language models has significantly improved the performance of many natural language processing (NLP) tasks. In this paper we propose a new model architecture DeBERTa (Decoding-enhanced BERT with disentangled attention) that improves the BERT and RoBERTa models using two novel techn... | ["Transformer", "Attention", "Natural Language Processing", "Language Model Pre-training", "Position Encoding"] | ABSTRACTRecent progress in pre-trained neural language models has significantly improvedthe performance of many natural language processing (NLP) tasks. In this pa-per we propose a new model architecture DeBERTa (Decoding- enhanced BERTwith disentangled attention) that improves the BERT and RoBERTa models usingtwo nove... | WYwTmQDzGb3 | DeBERTa: Decoding-Enhanced BERT with Disentangle Attention | 6: Marginally above acceptance threshold | Summary and Contributions
The authors proposed an extension to the word representation transformer architecture that takes into account disentangle features for position and content. The disentangle of attention is based on the composition of a content and position parameter matrices, in addition with combinations of... | 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
DEBERTA: DECODING-ENHANCED BERT WITH DISENTANGLED ATTENTION
### Paper Abstract
Recent progress in pre-trained neural language models has significantly improved the perfo... |
QjINdYOfq0b | ICLR.cc/2021/Conference | 2021 | ABS: Automatic Bit Sharing for Model Compression | ["Jing Liu", "Bohan Zhuang", "Peng Chen", "Yong Guo", "Chunhua Shen", "Jianfei Cai", "Mingkui Tan"] | We present Automatic Bit Sharing (ABS) to automatically search for optimal model compression configurations (e.g., pruning ratio and bitwidth). Unlike previous works that consider model pruning and quantization separately, we seek to optimize them jointly. To deal with the resultant large designing space, we propose a ... | ["Quantization", "Pruning", "Model Compression", "AutoML"] | ABSTRACTWe present Automatic Bit Sharing (ABS) to automatically search for optimalmodel compression configurations (e.g., pruning ratio and bitwidth). Unlike pre-vious works that consider model pruning and quantization separately, we seekto optimize them jointly. To deal with the resultant large designing space, weprop... | zmSsNI1hbK | Good work, but the novelty might be questionable | 6: Marginally above acceptance threshold | The paper describes the method to determine optimal quantization bit-width and pruning configuration for the neural network compression. Different from other approaches, the proposed method integrates multiple bit configurations (including pruning) into a single architecture, which is named “Super-bit”. The architectur... | 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
ABS: Automatic Bit Sharing for Model Compression
### Paper Abstract
We present Automatic Bit Sharing (ABS) to automatically search for optimal model compression configur... |
OZgVHzdKicb | ICLR.cc/2021/Conference | 2021 | Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples | ["Kevin Li", "Abhishek Gupta", "Vitchyr H. Pong", "Ashwin Reddy", "Aurick Zhou", "Justin Yu", "Sergey Levine"] | Exploration in reinforcement learning is, in general, a challenging problem. In this work, we study a more tractable class of reinforcement learning problems defined by data that provides examples of successful outcome states. In this case, the reward function can be obtained automatically by training a classifier to c... | ["Reinforcement Learning", "Goal Reaching", "Bayesian Classification", "Reward Inference"] | ABSTRACTExploration in reinforcement learning is, in general, a challenging problem. Inthis work, we study a more tractable class of reinforcement learning problemsdefined by data that provides examples of successful outcome states. In this case,the reward function can be obtained automatically by training a classifier... | NutPG60XnTO | some interesting ideas but require deeper analysis | 5: Marginally below acceptance threshold | This paper considers the problem of learning a policy for an MDP with unspecified reward, given user-provided goal states. To this end, a reward model and a policy are jointly learned: the reward model is the conditional normalized maximum likelihood (CNML) learned from a training set consisting of the example goal sta... | 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
Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples
### Paper Abstract
Exploration in reinforcement learning is, in general,... |
SJlRWC4FDB | ICLR.cc/2020/Conference | 2020 | Adversarial Attacks on Copyright Detection Systems | ["Parsa Saadatpanah", "Ali Shafahi", "Tom Goldstein"] | It is well-known that many machine learning models are susceptible to adversarial attacks, in which an attacker evades a classifier by making small perturbations to inputs. This paper discusses how industrial copyright detection tools, which serve a central role on the web, are susceptible to adversarial attacks. We di... | ["adversarial attacks", "copyright detection systems", "susceptible", "attacks", "system", "attacker", "classifier", "small perturbations"] | ABSTRACTIt is well-known that many machine learning models are susceptible to adversarialattacks, in which an attacker evades a classifier by making small perturbationsto inputs. This paper discusses how industrial copyright detection tools, whichserve a central role on the web, are susceptible to adversarial attacks. ... | ryxuDm2AKr | Official Blind Review #1 | 3: Weak Reject | This paper points out that the industrial copyright detection systems are susceptible to adversarial attacks and discuss the reason why these systems, especially the neural-network-based systems, are vulnerable to attacks. The authors describe a widely used music identification method and propose a simple gradient meth... | <|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 Attacks on Copyright Detection Systems
### Paper Abstract
It is well-known that many machine learning models are susceptible to adversarial attacks, in which... | |
B1ckMDqlg | ICLR.cc/2017/conference | 2017 | Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer | ["Noam Shazeer", "*Azalia Mirhoseini", "*Krzysztof Maziarz", "Andy Davis", "Quoc Le", "Geoffrey Hinton", "Jeff Dean"] | The capacity of a neural network to absorb information is limited by its number of parameters. Conditional computation, where parts of the network are active on a per-example basis, has been proposed in theory as a way of dramatically increasing model capacity without a proportional increase in computation. In practi... | ["Deep learning"] | ABSTRACTThe capacity of a neural network to absorb information is limited by its number ofparameters. Conditional computation, where parts of the network are active on aper-example basis, has been proposed in theory as a way of dramatically increas-ing model capacity without a proportional increase in computation. In p... | r10X7Es4g | Elegant use of MoE for expanding model capacity, but it would be very nice to discuss MoE alternatives in terms of computational efficiency and other factors. | 6: Marginally above acceptance threshold | Paper Strengths:
-- Elegant use of MoE for expanding model capacity and enabling training large models necessary for exploiting very large datasets in a computationally feasible manner
-- The effective batch size for training the MoE drastically increased also
-- Interesting experimental results on the effects of i... | 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
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
### Paper Abstract
The capacity of a neural network to absorb information is limited by ... |
BJ_UL-k0b | ICLR.cc/2018/Conference | 2018 | Recasting Gradient-Based Meta-Learning as Hierarchical Bayes | ["Erin Grant", "Chelsea Finn", "Sergey Levine", "Trevor Darrell", "Thomas Griffiths"] | Meta-learning allows an intelligent agent to leverage prior learning episodes as a basis for quickly improving performance on a novel task. Bayesian hierarchical modeling provides a theoretical framework for formalizing meta-learning as inference for a set of parameters that are shared across tasks. Here, we reformulat... | ["meta-learning", "learning to learn", "hierarchical Bayes", "approximate Bayesian methods"] | ABSTRACTMeta-learning allows an intelligent agent to leverage prior learning episodes as abasis for quickly improving performance on a novel task. Bayesian hierarchicalmodeling provides a theoretical framework for formalizing meta-learning as infer-ence for a set of parameters that are shared across tasks. Here, we ref... | Hkv54AYeM | Novel view on MAML, well presented. | 7: Good paper, accept | The paper reformulates the model-agnostic meta-learning algorithm (MAML) in terms of inference for parameters of a prior distribution in a hierarchical Bayesian model. This provides an interesting and, as far as I can tell, novel view on MAML. The paper uses this view to improve the MAML algorithm. The writing of the p... | 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
Recasting Gradient-Based Meta-Learning as Hierarchical Bayes
### Paper Abstract
Meta-learning allows an intelligent agent to leverage prior learning episodes as a basis ... |
HLqzzQWA7Z9 | ICLR.cc/2022/Workshop/EmeCom | 2022 | Situated Communication: A Solution to Over-communication between Artificial Agents | ["Aleksandra Kalinowska", "Elnaz Davoodi", "Florian Strub", "Kory Mathewson", "Todd Murphey", "Patrick Pilarski"] | Most research on communication emergence between reinforcement learning (RL) agents explores unsituated communication in one-step referential tasks. The tasks are not temporally interactive and lack time pressures typically present in natural communication and language learning. In these settings, agents can successful... | ["Emergent communication", "Multi-agent communication", "Multi-step interactions", "Artificial agents"] | ABSTRACTMost research on communication emergence between reinforcement learning(RL) agents explores unsituated communication in one-step referential tasks. Thetasks are not temporally interactive and lack time pressures typically present innatural communication and language learning. In these settings, agents can suc-c... | r9UeXuy6wMq | Review | Accept | ### Summary:
This paper studies emergent communication in dynamic, temporally-extended environments and presents situated communication as a mechanism for incentivizing efficient and concise emergent communication protocols. Situated communication introduces an opportunity cost to communication -- by choosing a communi... | 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
Situated Communication: A Solution to Over-communication between Artificial Agents
### Paper Abstract
Most research on communication emergence between reinforcement lear... |
HyetFnEFDS | ICLR.cc/2020/Conference | 2020 | Diving into Optimization of Topology in Neural Networks | ["Kun Yuan", "Quanquan Li", "Yucong Zhou", "Jing Shao", "Junjie Yan"] | Seeking effective networks has become one of the most crucial and practical areas in deep learning. The architecture of a neural network can be represented as a directed acyclic graph, whose nodes denote transformation of layers and edges represent information flow. Despite the selection of \textit{micro} node operatio... | ["optimization", "topology", "neural networks", "neural network", "connections", "edges", "experiments", "effective networks", "crucial", "practical areas"] | ABSTRACTSeeking effective networks has become one of the most crucial and practical areasin deep learning. The architecture of a neural network can be represented as a di-rected acyclic graph, whose nodes denote the transformation of layers and edgesrepresent information flow. Despite the selection of micro node operat... | SJlSSBqYcr | Official Blind Review #3 | 6: Weak Accept | *UPDATE* I have read the other reviews, authors' comments and the revised version of the manuscript. I have modified my rating to accept. The updated version, with variance across runs reported, comparison to randomly wired networks, and clearer writing, is substantially better. The core idea is simple to understand in... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Diving into Optimization of Topology in Neural Networks
### Paper Abstract
Seeking effective networks has become one of the most crucial and practical areas in deep lear... | |
r1MxciCcKm | ICLR.cc/2019/Conference | 2019 | Connecting the Dots Between MLE and RL for Sequence Generation | ["Bowen Tan*", "Zhiting Hu*", "Zichao Yang", "Ruslan Salakhutdinov", "Eric P. Xing"] | Sequence generation models such as recurrent networks can be trained with a diverse set of learning algorithms. For example, maximum likelihood learning is simple and efficient, yet suffers from the exposure bias problem. Reinforcement learning like policy gradient addresses the problem but can have prohibitively poor ... | ["sequence generation", "maximum likelihood learning", "reinforcement learning", "policy optimization", "text generation", "reward augmented maximum likelihood", "exposure bias"] | ABSTRACTSequence generation models such as recurrent networks can be trained with adiverse set of learning algorithms. For example, maximum likelihood learning issimple and efficient, yet suffers from the exposure bias problem. Reinforcementlearning like policy gradient addresses the problem but can have prohibitively ... | B1xuDmP93Q | AR3 Review: Connecting the Dots Between MLE and RL for Sequence Generation | 5: Marginally below acceptance threshold | The authors propose a more unified view of disparate methods for training sequence models. Specifically, a multi-term objective L(q,theta) consisting of:
1) The standard reward maximization objective of policy gradient, E_{p_\theta}[R],
2) A weighted (weight alpha) reverse KL divergence of the parametric policy and a ... | 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
Connecting the Dots Between MLE and RL for Sequence Generation
### Paper Abstract
Sequence generation models such as recurrent networks can be trained with a diverse set... |
GJnpCsLQThe | ICLR.cc/2021/Conference | 2021 | Gradient Descent Ascent for Min-Max Problems on Riemannian Manifolds | ["Feihu Huang", "Shangqian Gao", "Heng Huang"] | In the paper, we study a class of useful non-convex minimax optimization problems on Riemanian manifolds and propose a class of Riemanian gradient descent ascent algorithms to solve these minimax problems. Specifically, we propose a new
Riemannian gradient descent ascent (RGDA) algorithm for the deterministic minimax... | ["Min-Max Optimization", "Riemannian Manifold", "Robust Training"] | ABSTRACTIn the paper, we study a class of useful non-convex minimax optimization prob-lems on Riemanian manifolds and propose a class of Riemanian gradient descentascent algorithms to solve these minimax problems. Specifically, we propose anew Riemannian gradient descent ascent (RGDA) algorithm for the deterministicmin... | sF0jCQ7ifA9 | The paper focuses on an important class of problems | 7: Good paper, accept | The paper proposes Riemannian algorithms for min-max problems where the min problem is over a manifold and the max problem is a strongly convex problem. It presents a rigorous convergence analysis of the algorithms proposed. Finally, experiments show the good performance of the algorithms for robust training of DNNs ov... | 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
Gradient Descent Ascent for Min-Max Problems on Riemannian Manifolds
### Paper Abstract
In the paper, we study a class of useful non-convex minimax optimization problems... |
QkQCcFsUtk | ICLR.cc/2021/Conference | 2021 | Unsupervised Word Translation Pairing using Refinement based Point Set Registration | ["Silviu Oprea", "Sourav Dutta", "Haytham Assem"] | Cross-lingual alignment of word embeddings play an important role in knowledge transfer across languages, for improving machine translation and other multi-lingual applications. Current unsupervised approaches rely on similarities in geometric structure of word embedding spaces across languages, to learn structure-pres... | ["point", "refinement", "word embeddings", "languages", "framework", "unsupervised word translation", "registration", "alignment", "important role"] | ABSTRACTCross-lingual alignment of word embeddings play an important role in knowl-edge transfer across languages, for improving machine translation and othermulti-lingual applications. Current unsupervised approaches rely on similari-ties in geometric structure of word embedding spaces across languages, to learnstruct... | Cnj2UzKvuG | Unsupervised Word Translation Pairing using Refinement based Point Set Registration | 4: Ok but not good enough - rejection | This paper proposes BioSpere, a framework for unsupervised mapping of bilingual word embeddings onto the same vector space, using adversarial initialization and point set registration algorithm.
Pros:
1. The align module designs an adversarial training paradigm together with cycle-consistent training to distinguish sy... | 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
Unsupervised Word Translation Pairing using Refinement based Point Set Registration
### Paper Abstract
Cross-lingual alignment of word embeddings play an important role ... |
IqtonxWI0V3 | ICLR.cc/2021/Conference | 2021 | TropEx: An Algorithm for Extracting Linear Terms in Deep Neural Networks | ["Martin Trimmel", "Henning Petzka", "Cristian Sminchisescu"] | Deep neural networks with rectified linear (ReLU) activations are piecewise linear functions, where hyperplanes partition the input space into an astronomically high number of linear regions. Previous work focused on counting linear regions to measure the network's expressive power and on analyzing geometric properties... | ["linear regions", "linear terms", "deep learning theory", "deep neural networks", "rectified linear unit", "relu network", "piecewise linear function", "tropical function"] | ABSTRACTDeep neural networks with rectified linear (ReLU) activations are piecewise linearfunctions, where hyperplanes partition the input space into an astronomically highnumber of linear regions. Previous work focused on counting linear regions tomeasure the network’s expressive power and on analyzing geometric prope... | q4Jo0_S2jR | TropEx: An Algorithm for Extracting Linear Terms in Deep Neural Networks | 6: Marginally above acceptance threshold | Summary: This paper studies the role of linear terms in the network performance using nontrivial tropical algebra inspired algorithms. In particular, the paper extracts linear terms associated with the linear regions of only the training points, and uses this to generate an extracted network function. The paper propose... | 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
TropEx: An Algorithm for Extracting Linear Terms in Deep Neural Networks
### Paper Abstract
Deep neural networks with rectified linear (ReLU) activations are piecewise l... |
HyNA5iRcFQ | ICLR.cc/2019/Conference | 2019 | Detecting Egregious Responses in Neural Sequence-to-sequence Models | ["Tianxing He", "James Glass"] | In this work, we attempt to answer a critical question: whether there exists some input sequence that will cause a well-trained discrete-space neural network sequence-to-sequence (seq2seq) model to generate egregious outputs (aggressive, malicious, attacking, etc.). And if such inputs exist, how to find them efficient... | ["Deep Learning", "Natural Language Processing", "Adversarial Attacks", "Dialogue Response Generation"] | ABSTRACTIn this work, we attempt to answer a critical question: whether there existssome input sequence that will cause a well-trained discrete-space neural networksequence-to-sequence (seq2seq) model to generate egregious outputs (aggressive,malicious, attacking, etc.). And if such inputs exist, how to find them effic... | SklZWlF9nm | An interesting paper | 7: Good paper, accept | Main contribution: devising and evaluating an algorithm to find inputs that trigger arbitrary "egregious" outputs ("I will kill you") in vanilla sequence-to-sequence models, as a white-box attack on NLG models.
Clarity:
The paper is overall clear. I found some of the appendices (esp. B and C) to be important for under... | 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
Detecting Egregious Responses in Neural Sequence-to-sequence Models
### Paper Abstract
In this work, we attempt to answer a critical question: whether there exists some ... |
BylyV1BtDB | ICLR.cc/2020/Conference | 2020 | FR-GAN: Fair and Robust Training | ["Yuji Roh", "Kangwook Lee", "Gyeong Jo Hwang", "Steven Euijong Whang", "Changho Suh"] | We consider the problem of fair and robust model training in the presence of data poisoning. Ensuring fairness usually involves a tradeoff against accuracy, so if the data poisoning is mistakenly viewed as additional bias to be fixed, the accuracy will be sacrificed even more. We demonstrate that this phenomenon indeed... | ["generative adversarial networks", "model fairness", "model robustness"] | ABSTRACTWe consider the problem of fairandrobust model training in the presence of datapoisoning. Ensuring fairness usually involves a tradeoff against accuracy, so if thedata poisoning is mistakenly viewed as additional bias to be fixed, the accuracywill be sacrificed even more. We demonstrate that this phenomenon ind... | ryg38nhhtS | Official Blind Review #3 | 3: Weak Reject | This paper introduces a new method for training a classifier that simultaneously optimizes for a fairness criterion and robustness to data poisoning. The method is shown to increase measures of fairness and reduce inaccuracy on poisoned data relative to classifiers that only consider accuracy or fairness. Extensive res... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
FR-GAN: Fair and Robust Training
### Paper Abstract
We consider the problem of fair and robust model training in the presence of data poisoning. Ensuring fairness usuall... | |
HJeEP04KDH | ICLR.cc/2020/Conference | 2020 | Quantized Reinforcement Learning (QuaRL) | ["Srivatsan Krishnan", "Sharad Chitlangia", "Maximilian Lam", "Zishen Wan", "Aleksandra Faust", "Vijay Janapa Reddi"] | Recent work has shown that quantization can help reduce the memory, compute, and energy demands of deep neural networks without significantly harming their quality. However, whether these prior techniques, applied traditionally to image-based models, work with the same efficacy to the sequential decision making process... | ["Deep Reinforcement Learning", "Quantization"] | ABSTRACTRecent work has shown that quantization can help reduce the memory, compute,and energy demands of deep neural networks without significantly harming theirquality. However, whether these prior techniques, applied traditionally to image-based models, work with the same efficacy to the sequential decision makingpr... | B1ls5I2aYr | Official Blind Review #3 | 3: Weak Reject | Training and deployment of DRL models is expensive. Quantization has proven useful in supervised learning, however it is yet to be tested thoroughly in DRL. This paper investigates whether quantization can be applied in DRL towards better resource usage (compute, energy) without harming the model quality. Both quantiza... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Quantized Reinforcement Learning (QuaRL)
### Paper Abstract
Recent work has shown that quantization can help reduce the memory, compute, and energy demands of deep neura... | |
HJZiRkZC- | ICLR.cc/2018/Conference | 2018 | Byte-Level Recursive Convolutional Auto-Encoder for Text | ["Xiang Zhang", "Yann LeCun"] | This article proposes to auto-encode text at byte-level using convolutional networks with a recursive architecture. The motivation is to explore whether it is possible to have scalable and homogeneous text generation at byte-level in a non-sequential fashion through the simple task of auto-encoding. We show that non-se... | ["text", "recursive convolutional", "possible", "representation", "results", "model", "article", "convolutional networks", "recursive architecture", "motivation"] | ABSTRACTThis article proposes to auto-encode text at byte-level using convolutional net-works with a recursive architecture. The motivation is to explore whether it ispossible to have scalable and homogeneous text generation at byte-level in a non-sequential fashion through the simple task of auto-encoding. We show tha... | BkLqBW6lG | Related work and other tasks for experiments. | 5: Marginally below acceptance threshold | The authors propose autoencoding text using a byte-level encoding and a convolutional network with shared filters such that the encoder and decoder should exhibit recursive structure. They show that the model can handle various languages and run various experiments testing the ability of the autoencoder to reconstruct ... | 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
Byte-Level Recursive Convolutional Auto-Encoder for Text
### Paper Abstract
This article proposes to auto-encode text at byte-level using convolutional networks with a r... |
BkgiM20cYX | ICLR.cc/2019/Conference | 2019 | A Self-Supervised Method for Mapping Human Instructions to Robot Policies | ["Hsin-Wei Yu", "Po-Yu Wu", "Chih-An Tsao", "You-An Shen", "Shih-Hsuan Lin", "Zhang-Wei Hong", "Yi-Hsiang Chang", "Chun-Yi Lee"] | In this paper, we propose a modular approach which separates the instruction-to-action mapping procedure into two separate stages. The two stages are bridged via an intermediate representation called a goal, which stands for the result after a robot performs a specific task.
The first stage maps an input instruction t... | ["goal", "mapping module", "instruction", "human instructions", "policies", "modular", "mapping procedure", "stages", "robot", "policy"] | ABSTRACTIn this paper, we propose a modular approach which separates the instruction-to-action mapping procedure into two separate stages. The two stages are bridgedvia an intermediate representation called a goal. The first stage maps an inputinstruction to a goal, while the second stage maps the goal to an appropriat... | HJlkE0Bqh7 | Proposed method has several limitations, experimental setup is unclear and the results are not convincing. | 3: Clear rejection | This submission proposes a method for learning to follow instructions by splitting the policy into two stages: human instructions to robot-interpretable goals and goals to actions. The authors claim to achieve better data efficiency, adaptability, and generalization as compared to the baselines.
Here are some comments... | 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
A Self-Supervised Method for Mapping Human Instructions to Robot Policies
### Paper Abstract
In this paper, we propose a modular approach which separates the instruction... |
B1G5ViAqFm | ICLR.cc/2019/Conference | 2019 | Convolutional Neural Networks on Non-uniform Geometrical Signals Using Euclidean Spectral Transformation | ["Chiyu Max Jiang", "Dequan Wang", "Jingwei Huang", "Philip Marcus", "Matthias Niessner"] | Convolutional Neural Networks (CNN) have been successful in processing data signals that are uniformly sampled in the spatial domain (e.g., images). However, most data signals do not natively exist on a grid, and in the process of being sampled onto a uniform physical grid suffer significant aliasing error and informat... | ["Non-uniform Fourier Transform", "3D Learning", "CNN", "surface reconstruction"] | ABSTRACTConvolutional Neural Networks (CNN) have been successful in processing datasignals that are uniformly sampled in the spatial domain (e.g., images). However,most data signals do not natively exist on a grid, and in the process of being sampledonto a uniform physical grid suffer significant aliasing error and inf... | SJlRpk09hQ | Hard to Understand, insufficient results. | 4: Ok but not good enough - rejection | This paper introduces a method for handing input data that is defined on irregular mesh-type domains. If I understand it correctly, the core technique is to perform Fourier analysis to transform the input data to the frequency domain, which is then transformed back to a regular domain before applying a standard neural... | 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
Convolutional Neural Networks on Non-uniform Geometrical Signals Using Euclidean Spectral Transformation
### Paper Abstract
Convolutional Neural Networks (CNN) have been... |
rygp3iRcF7 | ICLR.cc/2019/Conference | 2019 | Area Attention | ["Yang Li", "Lukasz Kaiser", "Samy Bengio", "Si Si"] | Existing attention mechanisms, are mostly item-based in that a model is trained to attend to individual items in a collection (the memory) where each item has a predefined, fixed granularity, e.g., a character or a word. Intuitively, an area in the memory consisting of multiple items can be worth attending to as a whol... | ["Deep Learning", "attentional mechanisms", "neural machine translation", "image captioning"] | ABSTRACTExisting attention mechanisms, are mostly item-based in that a model is trainedto attend to individual items in a collection (the memory) where each item has apredefined, fixed granularity, e.g., a character or a word. Intuitively, an area in thememory consisting of multiple items can be worth attending to as a... | H1gTcC3xnQ | A few concerns | 5: Marginally below acceptance threshold | I prefer the idea of using some statistics (such as variances) of multiple items for attention.
This direction may lead to better attention units for future works.
I do not fully understand the argument, "Attention mechanisms are designed to focus on a single item in the entire memory".
In my understanding, the att... | 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
Area Attention
### Paper Abstract
Existing attention mechanisms, are mostly item-based in that a model is trained to attend to individual items in a collection (the memo... |
SkNEsmJwf | ICLR.cc/2018/Workshop | 2018 | 3D-Scene-GAN: Three-dimensional Scene Reconstruction with Generative Adversarial Networks | ["Chong Yu", "Young Wang"] | Three-dimensional (3D) Reconstruction is a vital and challenging research topic in advanced computer graphics and computer vision due to the intrinsic complexity and computation cost. Existing methods often produce holes, distortions and obscure parts in the reconstructed 3D models which are not adequate for real usage... | ["3D Reconstruction", "Generative Adversarial Network", "semi-supervised model"] | ABSTRACTThree-dimensional (3D) Reconstruction is a vital and challenging research topic inadvanced computer graphics and computer vision due to the intrinsic complexityand computation cost. Existing methods often produce holes, distortions and ob-scure parts in the reconstructed 3D models which are not adequate for rea... | ByMOwBlKM | Interesting idea but the description of the method lacks of clarity and details. | 5: Marginally below acceptance threshold | The idea of using GAN method to 3D scene reconstruction is novel and interesting. The proposed method performs well especially for the BRE Metric presented in Table 1.
Major concerns:
However, the method is presented as a 3D reconstruction technique, but it is not clear how the method can deal with a complete scene ... | 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
3D-Scene-GAN: Three-dimensional Scene Reconstruction with Generative Adversarial Networks
### Paper Abstract
Three-dimensional (3D) Reconstruction is a vital and challen... |
ETBc_MIMgoX | ICLR.cc/2021/Conference | 2021 | Learning with AMIGo: Adversarially Motivated Intrinsic Goals | ["Andres Campero", "Roberta Raileanu", "Heinrich Kuttler", "Joshua B. Tenenbaum", "Tim Rockt\u00e4schel", "Edward Grefenstette"] | A key challenge for reinforcement learning (RL) consists of learning in environments with sparse extrinsic rewards. In contrast to current RL methods, humans are able to learn new skills with little or no reward by using various forms of intrinsic motivation. We propose AMIGo, a novel agent incorporating -- as form of ... | ["reinforcement learning", "exploration", "meta-learning"] | ABSTRACTA key challenge for reinforcement learning (RL) consists of learning in environ-ments with sparse extrinsic rewards. In contrast to current RL methods, humansare able to learn new skills with little or no reward by using various forms ofintrinsic motivation. We propose AMIG O, a novel agent incorporating—as for... | uke4JYM7ftx | SoTA results on MiniGrid, similar ideas in previous methods, additional discussion about related works is needed. | 6: Marginally above acceptance threshold | This paper proposes a setter-solver or teacher-student scheme for training goal-conditioned agents (student/solver) in a discrete environment MiniGrid. In their method called AMIGo, the teacher takes the initial state of the student at the beginning of each episode as its input and proposes an intrinsic goal for the st... | 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 with AMIGo: Adversarially Motivated Intrinsic Goals
### Paper Abstract
A key challenge for reinforcement learning (RL) consists of learning in environments with... |
rkxwVZPpPV | icaps-conference.org/ICAPS/2019/Workshop/HSDIP | 2019 | Beyond Cost-to-go Estimates in Situated Temporal Planning | ["Anonymous"] | Heuristic search research often deals with finding algorithms for offline planning which aim to minimize the number of expanded nodes or planning time. In online planning, algorithms for real-time search or deadline-aware search have been considered before. However, in this paper, we are interested in the problem of {\... | ["heuristic search", "automated planning", "temporal planning", "real time search"] | Beyond Cost-to-go Estimates in Situated Temporal PlanningAndrew Coles,2Shahaf S. Shperberg,1Erez Karpas,4Solomon E. Shimony1Wheeler Ruml,31Ben-Gurion University, Israel;2King’s College London, UK;3University of New Hampshire, USA;4Technion, Israel;andrew.coles@kcl.ac.uk, shperbsh@post.bgu.ac.il, karpase@technion.ac.ils... | HyxbzM57FN | Interesting problem, encouraging results | 7: Good paper, accept | The paper addresses the problem of situated temporal planning, in which the planning agent needs not only to compute a plan, but to do so while taking into account the time it takes to compute that plan, as plans might have e.g. deadlines, real clock times where the validity of the plan expires.
The paper builds on pr... | 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
Beyond Cost-to-go Estimates in Situated Temporal Planning
### Paper Abstract
Heuristic search research often deals with finding algorithms for offline planning which aim... |
Yz-XtK5RBxB | ICLR.cc/2021/Conference | 2021 | Deep Repulsive Clustering of Ordered Data Based on Order-Identity Decomposition | ["Seon-Ho Lee", "Chang-Su Kim"] | We propose the deep repulsive clustering (DRC) algorithm of ordered data for effective order learning. First, we develop the order-identity decomposition (ORID) network to divide the information of an object instance into an order-related feature and an identity feature. Then, we group object instances into clusters ac... | ["Clustering", "order learning", "age estimation", "aesthetic assessment", "historical color image classification"] | ABSTRACTWe propose the deep repulsive clustering (DRC) algorithm of ordered data for ef-fective order learning. First, we develop the order-identity decomposition (ORID)network to divide the information of an object instance into an order-related fea-ture and an identity feature. Then, we group object instances into cl... | w8AJ0yjLzIx | The proposed method has good indicators and visualization effects. Compared with the previous methods, a better framework is proposed. The expression of the article is very clear, but some basic theories need not be explained in detail. | 7: Good paper, accept | The novelty of the network structure is marginal. The decomposition way of feature is very common in computer vision. Just utilizing the latent vector of the encoder with only the comparator loss to decompose the feature into two feature types is limited. The authors should show the visual differences between these two... | 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
Deep Repulsive Clustering of Ordered Data Based on Order-Identity Decomposition
### Paper Abstract
We propose the deep repulsive clustering (DRC) algorithm of ordered da... |
SkkC41HYl | ICLR.cc/2017/workshop | 2017 | De novo drug design with deep generative models : an empirical study | ["Mehdi cherti", "Balazs Kegl", "Akin kazakci"] | We present an empirical study about the usage of RNN generative models for stochastic optimization in the context of de novo drug design. We study different kinds of architectures and we find models that can generate molecules with higher values than ones seen in the training set. Our results suggest that we can improv... | ["Unsupervised Learning", "Optimization"] | ABSTRACTWe present an empirical study about the usage of RNN generative models forstochastic optimization in the context of de novo drug design. We study differ-ent kinds of architectures and we find models that can generate molecules withhigher values than ones seen in the training set. Our results suggest that we can... | rJ7kIm1og | Good work, but somewhat incremental | 6: Marginally above acceptance threshold | This paper presents an approach to generating novel small molecule structures (SMILES representations) with drug-like properties. Rather than generating random molecules using a set of hand-crafted rules (the space of possible small molecules is astronomical), the authors aim to learn common characteristics of drug-lik... | 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
De novo drug design with deep generative models : an empirical study
### Paper Abstract
We present an empirical study about the usage of RNN generative models for stocha... |
5SST78xEh4A | ICLR.cc/2021/Conference | 2021 | ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks | ["Xinshao Wang", "Yang Hua", "Elyor Kodirov", "David A. Clifton", "Neil M. Robertson"] | To train robust deep neural networks (DNNs), we systematically study several target modification approaches, which include output regularisation, self and non-self label correction (LC). Two key issues are discovered: (1) Self LC is the most appealing as it exploits its own knowledge and requires no extra models. Howev... | ["label correction", "entropy minimisation", "maximum entropy", "confidence penalty", "knowledge distillation", "regularization", "label noise"] | ABSTRACTTo train robust deep neural networks (DNNs), we systematically study severaltarget modification approaches, which include output regularisation, self and non-self label correction (LC). Two key issues are discovered: (1) Self LC is the mostappealing as it exploits its own knowledge and requires no extra models.... | txpix_Okuoz | An adaptive bootstrapping method that progressively modifies the target predictions, intuitive somewhat heuristic | 5: Marginally below acceptance threshold | The paper proposes Progressive Self Label Correction, which modifies the target predictions adaptively during the training. The method is based on label correction/bootstrapping (Reed et al., 2015), where the training targets are the combination of noisy labels and the predictions themselves. Instead of fixing the coef... | 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
ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks
### Paper Abstract
To train robust deep neural networks (DNNs), we systematically s... |
nXSDybDWV3 | ICLR.cc/2021/Conference | 2021 | Einstein VI: General and Integrated Stein Variational Inference in NumPyro | ["Ahmad Salim Al-Sibahi", "Ola R\u00f8nning", "Christophe Ley", "Thomas Wim Hamelryck"] | Stein Variational Inference is a technique for approximate Bayesian inferencethat is recently gaining popularity since it combines the scalability of traditionalVariational Inference (VI) with the flexibility of non-parametric particle basedinference methods. While there has been considerable progress in developmentof... | ["Stein variational inference", "variational inference", "probabilistic programming", "Pyro", "deep probabilistic programming", "deep learning"] | ABSTRACTStein Variational Inference is a technique for approximate Bayesian inference thatis gaining popularity since it combines the scalability of traditional Variational In-ference (VI) with the flexibility of non-parametric particle based inference methods.While there has been considerable progress in development o... | YM69dKGw_Xv | A novel library for Stein VI, but the paper lacks details of the examples and evaluation | 4: Ok but not good enough - rejection | ### Summary
This paper introduces EinStein VI: a lightweight composable library for Stein Variational Inference (Stein VI). The library is built on top of NumPyro and can take advantage of many of NumPyro's capabilities. It supports recent techniques associated with Stein VI, as well as novel features. The paper provid... | 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
Einstein VI: General and Integrated Stein Variational Inference in NumPyro
### Paper Abstract
Stein Variational Inference is a technique for approximate Bayesian infer... |
SJSVuReCZ | ICLR.cc/2018/Conference | 2018 | SHADE: SHAnnon DEcay Information-Based Regularization for Deep Learning | ["Michael Blot", "Thomas Robert", "Nicolas Thome", "Matthieu Cord"] | Regularization is a big issue for training deep neural networks. In this paper, we propose a new information-theory-based regularization scheme named SHADE for SHAnnon DEcay. The originality of the approach is to define a prior based on conditional entropy, which explicitly decouples the learning of invariant represent... | ["shannon decay", "shade", "regularization", "learning", "deep learning shade", "deep learning regularization", "big issue", "deep neural networks", "new", "regularization scheme"] | ABSTRACTRegularization is a big issue for training deep neural networks. In this paper, wepropose a new information-theory-based regularization scheme named SHADEfor SHAnnon DEcay. The originality of the approach is to define a prior basedon conditional entropy, which explicitly decouples the learning of invariant repr... | Hkd4Dn2eG | Review | 4: Ok but not good enough - rejection | Summary:
The paper presents an information theoretic regularizer for deep learning
algorithms. The regularizer aims to enforce compression of the learned
representation while conditioning upon the class label so preventing the
learned code from being constant across classes. The presentation of the Z
latent variable u... | 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
SHADE: SHAnnon DEcay Information-Based Regularization for Deep Learning
### Paper Abstract
Regularization is a big issue for training deep neural networks. In this paper... |
I-VfjSBzi36 | ICLR.cc/2021/Conference | 2021 | EarlyBERT: Efficient BERT Training via Early-bird Lottery Tickets | ["Xiaohan Chen", "Yu Cheng", "Shuohang Wang", "Zhe Gan", "Zhangyang Wang", "Jingjing Liu"] | Deep, heavily overparameterized language models such as BERT, XLNet and T5 have achieved impressive success in many NLP tasks. However, their high model complexity requires enormous computation resources and extremely long training time for both pre-training and fine-tuning. Many works have studied model compression on... | ["Natural Language Processing", "Lottery Tickets Hypothesis", "Efficient Training"] | ABSTRACTDeep, heavily overparameterized language models such as BERT, XLNet and T5have achieved impressive success in many natural language processing (NLP)tasks. However, their high model complexity requires enormous computation re-sources and extremely long training time for both pre-training and fine-tuning.Many wor... | WJCsET4sSBq | Nice work | 7: Good paper, accept | Summary:
This paper uses the Lottery Ticket Hypothesis to compress BERT. More specifically, they adapt EarlyBird lottery tickets to the BERT setting in order to find winning configurations in early stages of training combine it with structured pruning methods to ensure the resulting network is more efficient to train. ... | 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
EarlyBERT: Efficient BERT Training via Early-bird Lottery Tickets
### Paper Abstract
Deep, heavily overparameterized language models such as BERT, XLNet and T5 have achi... |
VymtVcEKLVh | CUHK.edu.hk/2021/Course/IERG5350 | 2021 | A Coarse-to-Fine Auto-Sampler For Long-tailed Image Recognition | ["Tong Wu", "Hao Li"] | The long-tail distributed data in the real world has always been a great challenge for deep learning. Current approaches to mitigate the long-tail issue include re-sampling and data augmentation. However, the hand-crafted re-sampling and augmentation strategies are sub-optimal. In this paper, we propose a coarse-to-fin... | ["Reinforcement Learning", "AutoML"] | A Coarse-to-Fine Auto-Sampler For Long-tailedImage RecognitionTong WuDepartment of Information EngineeringThe Chinese University of Hong KongShatin, Hong Kongtongwu@link.cuhk.edu.hkHao LiDepartment of Electronic EngineeringThe Chinese University of Hong KongShatin, Hong Konghaoli@link.cuhk.edu.hkAbstractThe long-tail d... | 3ZM_Bxd6VS3 | RL-based sampler for Long-tailed Image Recognition | 7: Good paper, accept | Summary: This paper proposes a coarse-to-fine RL-based auto-sampler that includes 1) a patch sampler for augmenting data and 2) a class-wise weighted sampler for classifier learning. Both phases resort to the RL approach to eliminate hand-crafted/random policies. Experiments on CIFAR-10/100-LT and ImageNet-LTT shows 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
A Coarse-to-Fine Auto-Sampler For Long-tailed Image Recognition
### Paper Abstract
The long-tail distributed data in the real world has always been a great challenge for... |
fpJX0O5bWKJ | ICLR.cc/2021/Conference | 2021 | Estimating Example Difficulty using Variance of Gradients | ["Chirag Agarwal", "Sara Hooker"] | In machine learning, a question of great interest is understanding what examples are challenging for a model to classify. Identifying atypical examples helps inform safe deployment of models, isolates examples that require further human inspection, and provides interpretability into model behavior. In this work, we pro... | ["interpretability", "human in the loop learning", "atypical examples"] | ABSTRACTIn machine learning, a question of great interest is understanding what examples arechallenging for a model to classify. Identifying atypical examples helps inform safedeployment of models, isolates examples that require further human inspection, andprovides interpretability into model behavior. In this work, w... | rORb0u5dCC | Computing variance of gradient for estimating example difficulty during neural network training | 6: Marginally above acceptance threshold | ##########################################################################
Summary:
The authors propose to use the variance of the gradient (VOG) values measured during neural network training as an estimate of the difficulty of examples. For each example, the gradient of the loss function is computed with respect 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
Estimating Example Difficulty using Variance of Gradients
### Paper Abstract
In machine learning, a question of great interest is understanding what examples are challen... |
rJVruWZRW | ICLR.cc/2018/Conference | 2018 | Dense Recurrent Neural Network with Attention Gate | ["Yong-Ho Yoo", "Kook Han", "Sanghyun Cho", "Kyoung-Chul Koh", "Jong-Hwan Kim"] | We propose the dense RNN, which has the fully connections from each hidden state to multiple preceding hidden states of all layers directly. As the density of the connection increases, the number of paths through which the gradient flows can be increased. It increases the magnitude of gradients, which help to prevent t... | ["recurrent neural network", "language modeling", "dense connection"] | ABSTRACTWe propose the dense RNN, which has fully connections from each hidden statedirectly to multiple preceding hidden states of all layers. As the density of theconnection increases, the number of paths through which the gradient flows isincreased. The magnitude of gradients is also increased, which helps to preven... | SkLFMZ9gG | Review | 2: Strong rejection | This paper proposes a new type of RNN architectures called Dense RNNs. The authors combine several different RNN architectures and claim that their RNN can model long-term dependencies better, can learn multiscale representation of the sequential data, and can sidestep the exploding or vanishing gradients problem by u... | 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
Dense Recurrent Neural Network with Attention Gate
### Paper Abstract
We propose the dense RNN, which has the fully connections from each hidden state to multiple preced... |
HjD70ArLTQt | ICLR.cc/2021/Conference | 2021 | Generating unseen complex scenes: are we there yet? | ["Arantxa Casanova", "Michal Drozdzal", "Adriana Romero"] | Although recent complex scene conditional generation models generate increasingly appealing scenes, it is very hard to assess which models perform better and why. This is often due to models being trained to fit different data splits, and defining their own experimental setups. In this paper, we propose a methodology t... | ["generative adversarial networks", "conditional scene generation", "zero-shot generalization", "out of distribution"] | ABSTRACTAlthough recent complex scene conditional generation models generate increas-ingly appealing scenes, it is very hard to assess which models perform better andwhy. This is often due to models being trained to fit different data splits, anddefining their own experimental setups. In this paper, we propose a method... | LyPVBdpBD2B | Official Blind Review #4 | 5: Marginally below acceptance threshold | Summary:
This paper discusses the achievements and limitations of existing conditional generative models. As expected, while existing methods can generalize reasonably well to unseen layouts of seen object combinations, none of the methods can work well to unseen object combinations.
Strengths:
The paper provides exte... | 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
Generating unseen complex scenes: are we there yet?
### Paper Abstract
Although recent complex scene conditional generation models generate increasingly appealing scenes... |
H1gNHs05FX | ICLR.cc/2019/Conference | 2019 | Clinical Risk: wavelet reconstruction networks for marked point processes | ["Jeremy C. Weiss"] | Timestamped sequences of events, pervasive in domains with data logs, e.g., health records, are often modeled as point processes with rate functions over time. Leading classical methods for risk scores such as Cox and Hawkes processes use such data but make strong assumptions about the shape and form of multivariate in... | ["point processes", "wavelets", "temporal neural networks", "Hawkes processes"] | ABSTRACTTimestamped sequences of events, pervasive in domains with data logs, e.g, healthrecords, are often modeled as point processes with rate functions over time. Lead-ing classical methods for risk scores such as Cox and Hawkes processes use suchdata but make strong assumptions about the shape and form of multivari... | Hyg8Wkua3m | It introduces wavelet reconstruction to the temporal point process literature | 7: Good paper, accept | The authors of this paper propose a point process model that uses wavelet reconstruction to capture complicated dependencies between events. They motivate this approach using experiments in the medical domain, which show how certain dependencies between events is better captured by their proposed model. The primary con... | 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
Clinical Risk: wavelet reconstruction networks for marked point processes
### Paper Abstract
Timestamped sequences of events, pervasive in domains with data logs, e.g., ... |
B1gi0TEFDB | ICLR.cc/2020/Conference | 2020 | Understanding Top-k Sparsification in Distributed Deep Learning | ["Shaohuai Shi", "Xiaowen Chu", "Ka Chun Cheung", "Simon See"] | Distributed stochastic gradient descent (SGD) algorithms are widely deployed in training large-scale deep learning models, while the communication overhead among workers becomes the new system bottleneck. Recently proposed gradient sparsification techniques, especially Top-$k$ sparsification with error compensation (To... | ["Distributed Deep Learning", "SGD", "Gradient Sparsification", "Communication-efficient SGD", "Top-k"] | ABSTRACTDistributed stochastic gradient descent (SGD) algorithms are widely deployedin training large-scale deep learning models, while the communication overheadamong workers becomes the new system bottleneck. Recently proposed gradientsparsification techniques, especially Top- ksparsification with error compensation(... | SkeyfEe1cH | Official Blind Review #3 | 3: Weak Reject | This paper makes two contributions to gradient sparsification to reduce the communication bottleneck in distributed SGD.
1) Based on an assumption on the distribution of gradient coordinate values that are backed by an empirical study, the paper derives a tighter bound on the approximation quality of top-k gradient spa... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Understanding Top-k Sparsification in Distributed Deep Learning
### Paper Abstract
Distributed stochastic gradient descent (SGD) algorithms are widely deployed in traini... | |
r1e_FpNFDr | ICLR.cc/2020/Conference | 2020 | Generalization bounds for deep convolutional neural networks | ["Philip M. Long", "Hanie Sedghi"] | We prove bounds on the generalization error of convolutional networks.
The bounds are in terms of the training loss, the number of
parameters, the Lipschitz constant of the loss and the distance from
the weights to the initial weights. They are independent of the
number of pixels in the input, and the height and width... | ["generalization", "convolutional networks", "statistical learning theory"] | ABSTRACTWe prove bounds on the generalization error of convolutional networks. Thebounds are in terms of the training loss, the number of parameters, the Lipschitzconstant of the loss and the distance from the weights to the initial weights. Theyare independent of the number of pixels in the input, and the height and w... | Byx_mr2g9H | Official Blind Review #3 | 6: Weak Accept |
The paper describes new norm-based generalization bounds that were specifically adapted to convolutional neural networks. Since convolutional neural networks do not explicitly depend on the input dimension, these bounds share the same property. Further additional improvement over Bartlett et al. ‘17 bound, is that thi... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Generalization bounds for deep convolutional neural networks
### Paper Abstract
We prove bounds on the generalization error of convolutional networks. The bounds are in ... | |
Z532uNJyG5y | ICLR.cc/2021/Conference | 2021 | Iterative Graph Self-Distillation | ["Hanlin Zhang", "Shuai Lin", "Weiyang Liu", "Pan Zhou", "Jian Tang", "Xiaodan Liang", "Eric Xing"] | How to discriminatively vectorize graphs is a fundamental challenge that attracts increasing attentions in recent years. Motivated by the recent success of unsupervised contrastive learning, we aim to learn graph-level representation in an unsupervised manner. Specifically, we propose a novel unsupervised graph learnin... | ["graph-level representation learning", "knowledge distillation"] | ABSTRACTHow to discriminatively vectorize graphs is a fundamental challenge that attractsincreasing attentions in recent years. Inspired by the recent success of unsupervisedcontrastive learning, we aim to learn graph-level representation in an unsupervisedmanner. Specifically, we propose a novel unsupervised graph lea... | g0FAaD_-Pp | Experiments need to be improved | 5: Marginally below acceptance threshold | ------------------------------------
update after reading the authors' response.
The authors didn't address my question "Did the authors perform a significance study?" A significance test such as double-sided t-test is needed to verifying whether the proposed method is significantly better than baselines.
----------... | 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
Iterative Graph Self-Distillation
### Paper Abstract
How to discriminatively vectorize graphs is a fundamental challenge that attracts increasing attentions in recent ye... |
pdsec2YIOCx | ICLR.cc/2021/Conference | 2021 | Untangle: Critiquing Disentangled Recommendations | ["Preksha Nema", "Alexandros Karatzoglou", "Filip Radlinski"] | The core principle behind most collaborative filtering methods is to embed users and items in latent spaces, where individual dimensions are learned independently of any particular item attributes. It is thus difficult for users to control their recommendations based on particular aspects (critiquing). In this work, we... | ["Disentangling", "Recommender Systems", "VAE", "Critiquing", "Explainability"] | Under review as a conference paper at ICLR 2021Untangle :Critiquing Disentangled RecommendationsAnonymous authorsPaper under double-blind reviewAbstractThe core principle behind most collaborative filtering methods is to embedusers and items in latent spaces, where individual dimensions are learnedindependently of any ... | yotvgl7SNXB | Good motivation, weak technical contribution, and weak evaluation | 5: Marginally below acceptance threshold | The submission aims to develop a disentangled recommender that explicitly grants the users the ability to control a chosen aspect of the recommendation list. This is achieved by (1) learning disentangled user/item representations via beta-VAE and (2) aligning text tags with certain dimensions via an auxiliary loss (the... | 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
Untangle: Critiquing Disentangled Recommendations
### Paper Abstract
The core principle behind most collaborative filtering methods is to embed users and items in latent... |
S1ghzlHFPS | ICLR.cc/2020/Conference | 2020 | Informed Temporal Modeling via Logical Specification of Factorial LSTMs | ["Hongyuan Mei", "Guanghui Qin", "Minjie Xu", "Jason Eisner"] | Consider a world in which events occur that involve various entities. Learning how to predict future events from patterns of past events becomes more difficult as we consider more types of events. Many of the patterns detected in the dataset by an ordinary LSTM will be spurious since the number of potential pairwise co... | ["factorized LSTM", "temporal point process", "event streams", "structural bias", "Datalog"] | ABSTRACTConsider a world in which events occur that involve various entities. Learninghow to predict future events from patterns of past events becomes more difficultas we consider more types of events. Many of the patterns detected in the datasetby an ordinary LSTM will be spurious since the number of potential pairwi... | Skx3yBR8qH | Official Blind Review #1 | 1: Reject | Summary:
The paper proposed to use Datalog rules to specify the design of the LSTM architecture for event data in continuous time. The LSTM module will be used to model the rate of the events. By incorporating Datalog rules, the paper aims to encode informed inductive biases into the model.
Comments:
After ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Informed Temporal Modeling via Logical Specification of Factorial LSTMs
### Paper Abstract
Consider a world in which events occur that involve various entities. Learning... | |
LIJZpJix3MN | graphicsinterface.org/Graphics_Interface/2023/Conference | 2023 | Exploring the Needs of Informal Learners of Computational Skills: Probe-Based Elicitation for the Design of Self-Monitoring Interventions | ["Rimika Chaudhury", "Taha Liaqat", "Parmit Chilana"] | Informal learners of computational skills often find it difficult to self-direct their learning pursuits, which may be spread across different mediums and study sessions. Inspired by self-monitoring interventions from domains such as health and productivity, we investigate key requirements for helping informal learners... | ["Self-monitoring interventions", "informal learning", "visualization"] | ABSTRACTInformal learners of computational skills often find it difficult to self-direct their learning pursuits, which may be spread across differentmediums and study sessions. Inspired by self-monitoring interven-tions from domains such as health and productivity, we investigatekey requirements for helping informal l... | 5RTxO0Zrrrn | This paper presents two qualitative studies, using paper-mock-ups and interactive probes respectively, to investigate the design of a self-monitoring visualization tool for informal learners to track and reflect on their learning experiences of technical contents from a variety of resources and media. This is a very we... | 9: Top 15% of accepted papers, strong accept | The related work is well framed to motivate the current research. The two studies were well-designed with clear goals. The data collection and data analysis methods are both clearly described. The findings are also clearly presented with good insights. There are however a few minor issues that should be fixed/improved:... | 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
Exploring the Needs of Informal Learners of Computational Skills: Probe-Based Elicitation for the Design of Self-Monitoring Interventions
### Paper Abstract
Informal lea... |
o81ZyBCojoA | ICLR.cc/2021/Conference | 2021 | On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning | ["Ren Wang", "Kaidi Xu", "Sijia Liu", "Pin-Yu Chen", "Tsui-Wei Weng", "Chuang Gan", "Meng Wang"] | Model-agnostic meta-learning (MAML) has emerged as one of the most successful meta-learning techniques in few-shot learning. It enables us to learn a $\textit{meta-initialization}$ of model parameters (that we call $\textit{meta-model}$) to rapidly adapt to new tasks using a small amount of labeled training data. Despi... | ["maml", "learning", "adversarial robustness", "robustness", "stage", "successful", "techniques", "model parameters", "new tasks"] | ABSTRACTModel-agnostic meta-learning (MAML) has emerged as one of the most suc-cessful meta-learning techniques in few-shot learning. It enables us to learnameta-initialization of model parameters (that we call meta-model ) to rapidlyadapt to new tasks using a small amount of labeled training data. Despite thegeneraliz... | fpH2KlYuoVd | clear motivation and reasonable method, but marginal novelty | 6: Marginally above acceptance threshold | This paper explores a way to promote the adversarial robustness in Model-agnostic meta-learning (MAML). It conducts extensive experiments to show regularizing adversarial robustness at meta-update level is sufficient to offer fast and effective robustness adaptation on few-shot test tasks. However, it lacks the theoret... | 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 Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning
### Paper Abstract
Model-agnostic meta-learning (MAML) has emerged as one of the most successfu... |
BygWBRNtvH | ICLR.cc/2020/Conference | 2020 | Benchmarking Adversarial Robustness | ["Yinpeng Dong", "Qi-An Fu", "Xiao Yang", "Tianyu Pang", "Hang Su", "Jun Zhu"] | Deep neural networks are vulnerable to adversarial examples, which becomes one of the most important problems in the development of deep learning. While a lot of efforts have been made in recent years, it is of great significance to perform correct and complete evaluations of the adversarial attack and defense algorith... | ["Adversarial Example", "Robustness", "Benchmark"] | ABSTRACTDeep neural networks are vulnerable to adversarial examples, which becomes oneof the most important problems in the development of deep learning. While a lot ofefforts have been made in recent years, it is of great significance to perform correctand complete evaluations of the adversarial attack and defense alg... | rylDRTg3tB | Official Blind Review #3 | 1: Reject | The paper presents an evaluation of a number of kinds of classification models under various adversarial attack methods.
For classification models, the authors consider a collection of defenses ranging from those trained with robust optimization to those utilizing input transformations. The set of models includes tho... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Benchmarking Adversarial Robustness
### Paper Abstract
Deep neural networks are vulnerable to adversarial examples, which becomes one of the most important problems in t... | |
HJe_yR4Fwr | ICLR.cc/2020/Conference | 2020 | Improved Sample Complexities for Deep Neural Networks and Robust Classification via an All-Layer Margin | ["Colin Wei", "Tengyu Ma"] | For linear classifiers, the relationship between (normalized) output margin and generalization is captured in a clear and simple bound – a large output margin implies good generalization. Unfortunately, for deep models, this relationship is less clear: existing analyses of the output margin give complicated bounds whic... | ["deep learning theory", "generalization bounds", "adversarially robust generalization", "data-dependent generalization bounds"] | ABSTRACTFor linear classifiers, the relationship between (normalized) output margin andgeneralization is captured in a clear and simple bound – a large output marginimplies good generalization. Unfortunately, for deep models, this relationship isless clear: existing analyses of the output margin give complicated bounds... | rye0FNG0tB | Official Blind Review #2 | 8: Accept | The paper proposes a new generalization bound for deep neural networks and develops a regularizer which optimize quantities related to the bound and improve generalization error on competitive baselines. The paper treats deep neural network as a composition of functions (i.e. the layers) and introduces a new complexity... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Improved Sample Complexities for Deep Neural Networks and Robust Classification via an All-Layer Margin
### Paper Abstract
For linear classifiers, the relationship betwe... | |
OCRKCul3eKN | ICLR.cc/2021/Conference | 2021 | Addressing Extrapolation Error in Deep Offline Reinforcement Learning | ["Caglar Gulcehre", "Sergio G\u00f3mez Colmenarejo", "ziyu wang", "Jakub Sygnowski", "Thomas Paine", "Konrad Zolna", "Yutian Chen", "Matthew Hoffman", "Razvan Pascanu", "Nando de Freitas"] | Reinforcement learning (RL) encompasses both online and offline regimes. Unlike its online counterpart, offline RL agents are trained using logged-data only, without interaction with the environment. Therefore, offline RL is a promising direction for real-world applications, such as healthcare, where repeated interac... | ["Addressing Extrapolation Error in Deep Offline Reinforcement Learning"] | ABSTRACTReinforcement learning (RL) encompasses both online and offline regimes. Unlikeits online counterpart, offline RL agents are trained using logged-data only, withoutinteraction with the environment. Therefore, offline RL is a promising directionfor real-world applications, such as healthcare, where repeated inte... | Dws57_LhFLK | no sound knowledge to rely on | 3: Clear rejection | Summary:
The paper deals with offline aka batch RL for discrete actions. Three techniques ((i) behavior value estimation, (ii) ranking regularization, and (iii) reparametrization of the value function), which can be combined with each other, are presented. These techniques are compared with other methods in different e... | 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
Addressing Extrapolation Error in Deep Offline Reinforcement Learning
### Paper Abstract
Reinforcement learning (RL) encompasses both online and offline regimes. Unlike... |
HJlTpCEKvS | ICLR.cc/2020/Conference | 2020 | Which Tasks Should Be Learned Together in Multi-task Learning? | ["Trevor Standley", "Amir R. Zamir", "Dawn Chen", "Leonidas Guibas", "Jitendra Malik", "Silvio Savarese"] | Many computer vision applications require solving multiple tasks in real-time. A neural network can be trained to solve multiple tasks simultaneously using 'multi-task learning'. This saves computation at inference time as only a single network needs to be evaluated. Unfortunately, this often leads to inferior overall ... | ["multi-task learning", "Computer Vision"] | ABSTRACTMany computer vision applications require solving multiple tasks in real-time. Aneural network can be trained to solve multiple tasks simultaneously using multi-task learning . This saves computation at inference time as only a single networkneeds to be evaluated. Unfortunately, this often leads to inferior ove... | HJgJ83u0KS | Official Blind Review #1 | 6: Weak Accept | This paper focuses on how to partition a bunch of tasks in several groups and then it use multi-task learning to improve the performance. The paper makes an observation that multi-task relationships are not entirely correlated to transfer relationships and proposes a computational framework to optimize the assignment ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Which Tasks Should Be Learned Together in Multi-task Learning?
### Paper Abstract
Many computer vision applications require solving multiple tasks in real-time. A neural... | |
ryxsS3A5Km | ICLR.cc/2019/Conference | 2019 | Continual Learning via Explicit Structure Learning | ["Xilai Li", "Yingbo Zhou", "Tianfu Wu", "Richard Socher", "Caiming Xiong"] | Despite recent advances in deep learning, neural networks suffer catastrophic forgetting when tasks are learned sequentially. We propose a conceptually simple and general framework for continual learning, where structure optimization is considered explicitly during learning. We implement this idea by separating the str... | ["continuous learning", "catastrophic forgetting", "architecture learning"] | ABSTRACTDespite recent advances in deep learning, neural networks suffer catastrophic for-getting when tasks are learned sequentially. We propose a conceptually simple andgeneral framework for continual learning, where structure optimization is consid-ered explicitly during learning. We implement this idea by separatin... | B1e--EHYnQ | review | 4: Ok but not good enough - rejection |
This paper proposes a new approach to mitigate the catastrophic forgetting for continual learning. The model is composed to the neural architecture search and parameter learning based on the intuition that largely different tasks should allow to use different network structure to train them. In structure learning, the... | 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
Continual Learning via Explicit Structure Learning
### Paper Abstract
Despite recent advances in deep learning, neural networks suffer catastrophic forgetting when tasks... |
SygLHbcapm | AKBC.ws/2019/Conference | 2019 | Integrating User Feedback under Identity Uncertainty in Knowledge Base Construction | ["Ari Kobren", "Nicholas Monath", "Andrew McCallum"] | Users have tremendous potential to aid in the construction and maintenance of knowledges bases (KBs) through the contribution of feedback that identifies incorrect and missing entity attributes and relations. However, as new data is added to the KB, the KB entities, which are constructed by running entity resolution (E... | ["user feedback", "entity resolution", "identity uncertainty"] | Automated Knowledge Base Construction (2019) Conference paperIntegrating User Feedback under Identity Uncertaintyin Knowledge Base ConstructionAri Kobren akobren@cs.umass.eduNicholas Monath nmonath@cs.umass.eduAndrew McCallum mccallum@cs.umass.eduUniversity of Massachusetts AmherstCollege of Information and Computer Sc... | Bye0ls1mfV | A nice algorithmic contribution for integrating user feedback for KBC | 7: Good paper, accept | This paper introduces a method to integrate user feedback into KBs in the presence of identity uncertainty, a problem that arises in the integration of new data in knowledge base construction. The proposed method represents user feedback as feedback mentions and uses an online algorithm for integrating these mentions i... | 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
Integrating User Feedback under Identity Uncertainty in Knowledge Base Construction
### Paper Abstract
Users have tremendous potential to aid in the construction and mai... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.