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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
CF-ZIuSMXRz | ICLR.cc/2021/Conference | 2021 | Spatio-Temporal Graph Scattering Transform | ["Chao Pan", "Siheng Chen", "Antonio Ortega"] | Although spatio-temporal graph neural networks have achieved great empirical success in handling multiple correlated time series, they may be impractical in some real-world scenarios due to a lack of sufficient high-quality training data. Furthermore, spatio-temporal graph neural networks lack theoretical interpretatio... | ["scattering transform", "spatio-temporal graph", "graph neural networks", "skeleton-based action recognition"] | ABSTRACTAlthough spatio-temporal graph neural networks have achieved great empiricalsuccess in handling multiple correlated time series, they may be impractical insome real-world scenarios due to a lack of sufficient high-quality training data.Furthermore, spatio-temporal graph neural networks lack theoretical interpre... | IJZg2Qdgm9 | A very good extension of the graph scattering transform to the spatio-temporal case | 9: Top 15% of accepted papers, strong accept | 1/ Summary of the paper
This paper proposes a novel spatio-temporal graph scattering transform (ST-GST) as an intermediate representation of time-varying signals on graphs, which can then be fed into a simple ML model. Building on previously introduced Graph Scattering Transform (GST), the authors investigate the tech... | 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
Spatio-Temporal Graph Scattering Transform
### Paper Abstract
Although spatio-temporal graph neural networks have achieved great empirical success in handling multiple c... |
S1lwRjR9YX | ICLR.cc/2019/Conference | 2019 | Stability of Stochastic Gradient Method with Momentum for Strongly Convex Loss Functions | ["Ali Ramezani-Kebrya", "Ashish Khisti", "and Ben Liang"] | While momentum-based methods, in conjunction with the stochastic gradient descent, are widely used when training machine learning models, there is little theoretical understanding on the generalization error of such methods. In practice, the momentum parameter is often chosen in a heuristic fashion with little theoreti... | ["Generalization Error", "Stochastic Gradient Descent", "Uniform Stability"] | ABSTRACTWhile momentum-based methods, in conjunction with the stochastic gradient de-scent, are widely used when training machine learning models, there is little theo-retical understanding on the generalization error of such methods. In practice, themomentum parameter is often chosen in a heuristic fashion with little... | Skl1MikYnX | Review | 6: Marginally above acceptance threshold | This paper studies the algorithmic stability of SGD with momentum and provides an upper-bound on true risk through convergence analysis.
This bound clarifies dependencies of convergence speed on the size of dataset and the momentum parameter.
The presentation is easy to follow and technically sounds good.
SGD with mom... | 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
Stability of Stochastic Gradient Method with Momentum for Strongly Convex Loss Functions
### Paper Abstract
While momentum-based methods, in conjunction with the stochas... |
TtYSU29zgR | ICLR.cc/2021/Conference | 2021 | Primal Wasserstein Imitation Learning | ["Robert Dadashi", "Leonard Hussenot", "Matthieu Geist", "Olivier Pietquin"] | Imitation Learning (IL) methods seek to match the behavior of an agent with that of an expert. In the present work, we propose a new IL method based on a conceptually simple algorithm: Primal Wasserstein Imitation Learning (PWIL), which ties to the primal form of the Wasserstein distance between the expert and the agen... | ["Reinforcement Learning", "Inverse Reinforcement Learning", "Imitation Learning", "Optimal Transport", "Wasserstein distance"] | ABSTRACTImitation Learning (IL) methods seek to match the behavior of an agent withthat of an expert. In the present work, we propose a new IL method based on aconceptually simple algorithm: Primal Wasserstein Imitation Learning (PWIL),which ties to the primal form of the Wasserstein distance between the expert andthe ... | xzD-F9c79O | Review for paper "Primal Wasserstein Imitation Learning" | 6: Marginally above acceptance threshold | Summary:
This paper proposes to use Wasserstein distance in the primal form for imitation learning. Compared with its dual form and f-divergence minimization variants, it avoids the unstable minimax optimization. In order to compute the Wasserstein distance in primal form, they also propose a greedy approximation. 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
Primal Wasserstein Imitation Learning
### Paper Abstract
Imitation Learning (IL) methods seek to match the behavior of an agent with that of an expert. In the present wo... |
B1l8SsR9Fm | ICLR.cc/2019/Conference | 2019 | Learning and Data Selection in Big Datasets | ["Hossein S. Ghadikolaei", "Hadi Ghauch", "Carlo Fischione", "Mikael Skoglund"] | Finding a dataset of minimal cardinality to characterize the optimal parameters of a model is of paramount importance in machine learning and distributed optimization over a network. This paper investigates the compressibility of large datasets. More specifically, we propose a framework that jointly learns the input-ou... | ["Data selection", "non-convex optimization", "learning theory", "active learning"] | ABSTRACTFinding a dataset of minimal cardinality to characterize the optimal parameters of amodel is of paramount importance in machine learning and distributed optimizationover a network. This paper investigates the compressibility of large datasets. Morespecifically, we propose a framework that jointly learns the inp... | Skl3TwLU2X | Dataset selection and compression: a work done without much motivation, in isolation, without an interesting theoretical conclusion. | 3: Clear rejection | The paper addresses the problem of selecting a subset of a labeled dataset that retains the performance of training on the full dataset. The techniques presented in the paper are able to select just 5% of the dataset with only a small loss in error.
The paper has several severe issues with it: motivation, related work... | 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 and Data Selection in Big Datasets
### Paper Abstract
Finding a dataset of minimal cardinality to characterize the optimal parameters of a model is of paramount... |
B1lxH20qtX | ICLR.cc/2019/Conference | 2019 | Learning to control self-assembling morphologies: a study of generalization via modularity | ["Deepak Pathak", "Chris Lu", "Trevor Darrell", "Philip Isola", "Alexei A. Efros"] | Much of contemporary sensorimotor learning assumes that one is already given a complex agent (e.g., a robotic arm) and the goal is to learn to control it. In contrast, this paper investigates a modular co-evolution strategy: a collection of primitive agents learns to self-assemble into increasingly complex collectives ... | ["modularity", "compostionality", "graphs", "dynamics", "network"] | ABSTRACTMuch of contemporary sensorimotor learning assumes that one is already given acomplex agent (e.g., a robotic arm) and the goal is to learn to control it. In contrast,this paper investigates a modular co-evolution strategy: a collection of primitiveagents learns to self-assemble into increasingly complex collect... | HJlv_HUI2Q | Interesting ideas and the setup, but virtually no details provided | 4: Ok but not good enough - rejection | Summary:
--------------
The paper considers the problem of constructing compositional robotic morphologies that can solve different continuous control tasks in a (multi-agent) reinforcement learning setting. The authors created an environment where the actor consists of a number of primitive components which interface ... | 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 to control self-assembling morphologies: a study of generalization via modularity
### Paper Abstract
Much of contemporary sensorimotor learning assumes that one... |
KyMw0p9rWL | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | Gaggle: Visual Analytics for Model Space Navigation | ["Subhajit Das", "Dylan Cashman", "Remco Chang", "Alex Endert"] | Recent visual analytics systems make use of multiple machine learning models to better fit the data as opposed to traditional single, pre-defined model systems. However, while multi-model visual analytic systems can be effective, their added complexity poses usability concerns, as users are required to interact with th... | ["visual analytics", "interactive machine learning", "classification", "ranking"] | ABSTRACTRecent visual analytics systems make use of multiple machine learn-ing models to better fit the data as opposed to traditional single,pre-defined model systems. However, while multi-model visual ana-lytic systems can be effective, their added complexity adds usabilityconcerns, as users are required to interact ... | kaavrq84eP | Review | 6: Marginally above acceptance threshold | The authors present a new visual analytic system called Gaggle, which aims to enable non-expert users to interactively navigate a model space by using a demonstration-based approach. An evaluation with 22 non-experts support the claim to simplify the complex model and hyperparameter search by using such an interaction... | 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
Gaggle: Visual Analytics for Model Space Navigation
### Paper Abstract
Recent visual analytics systems make use of multiple machine learning models to better fit the dat... |
S1eOHo09KX | ICLR.cc/2019/Conference | 2019 | Opportunistic Learning: Budgeted Cost-Sensitive Learning from Data Streams | ["Mohammad Kachuee", "Orpaz Goldstein", "Kimmo K\u00e4rkk\u00e4inen", "Sajad Darabi", "Majid Sarrafzadeh"] | In many real-world learning scenarios, features are only acquirable at a cost constrained under a budget. In this paper, we propose a novel approach for cost-sensitive feature acquisition at the prediction-time. The suggested method acquires features incrementally based on a context-aware feature-value function. We for... | ["Cost-Aware Learning", "Feature Acquisition", "Reinforcement Learning", "Stream Learning", "Deep Q-Learning"] | ABSTRACTIn many real-world learning scenarios, features are only acquirable at a cost con-strained under a budget. In this paper, we propose a novel approach for cost-sensitive feature acquisition at the prediction-time. The suggested method acquiresfeatures incrementally based on a context-aware feature-value function... | rygaWIL0h7 | nice results but limited novelty | 6: Marginally above acceptance threshold | The paper presents a RL approach for sequential feature acquisition in a budgeted learning setting, where each feature comes at some cost and the goal is to find a good trade-off between accuracy and cost. Starting with zero feature, the model sequentially acquires new features to update its prediction and stops when 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
Opportunistic Learning: Budgeted Cost-Sensitive Learning from Data Streams
### Paper Abstract
In many real-world learning scenarios, features are only acquirable at a co... |
PiKUvDj5jyN | ICLR.cc/2021/Conference | 2021 | Relational Learning with Variational Bayes | ["Kuang-Hung Liu"] | In psychology, relational learning refers to the ability to recognize and respond to
relationship among objects irrespective of the nature of those objects. Relational
learning has long been recognized as a hallmark of human cognition and a key
question in artificial intelligence research. In this work, we propose an u... | ["Relational learning", "unsupervised learning", "variational inference", "probabilistic graphical model"] | ABSTRACTIn psychology, relational learning refers to the ability to recognize and respond torelationship among objects irrespective of the nature of those objects. Relationallearning has long been recognized as a hallmark of human cognition and a keyquestion in artificial intelligence research. In this work, we propose... | 0SA9ofX4C7p | Review | 6: Marginally above acceptance threshold | The paper proposes variational relational learning by learning relations between two inputs via variational inference on a probabilistic graphical model (PGM). The PGM that they use factors as p(a)p(z)p(b|a,z) where a,b are the two inputs and z is the supposed relationship between them. The example shown in the experim... | 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
Relational Learning with Variational Bayes
### Paper Abstract
In psychology, relational learning refers to the ability to recognize and respond to relationship among obj... |
SkgOzlrKvH | ICLR.cc/2020/Conference | 2020 | The Role of Embedding Complexity in Domain-invariant Representations | ["Ching-Yao Chuang", "Antonio Torralba", "Stefanie Jegelka"] | Unsupervised domain adaptation aims to generalize the hypothesis trained in a source domain to an unlabeled target domain. One popular approach to this problem is to learn domain-invariant embeddings for both domains. In this work, we study, theoretically and empirically, the effect of the embedding complexity on gener... | ["domain adaptation", "domain-invariant representations", "model complexity", "theory", "deep learning"] | ABSTRACTUnsupervised domain adaptation aims to generalize the hypothesis trained in asource domain to an unlabeled target domain. One popular approach to this prob-lem is to learn domain-invariant embeddings for both domains. In this work, westudy, theoretically and empirically, the effect of the embedding complexity o... | SkxxRsGe9H | Official Blind Review #2 | 1: Reject | This paper studies the problem of domain adaptation via learning invariant representations. The main argument here is that when the total depth of layers in a neural network is fixed, tradeoffs exist between feature alignment and prediction power. Furthermore, the authors argue that richer feature extractor can sometim... | <|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 Role of Embedding Complexity in Domain-invariant Representations
### Paper Abstract
Unsupervised domain adaptation aims to generalize the hypothesis trained in a sou... | |
S1lTEh09FQ | ICLR.cc/2019/Conference | 2019 | Combinatorial Attacks on Binarized Neural Networks | ["Elias B Khalil", "Amrita Gupta", "Bistra Dilkina"] | Binarized Neural Networks (BNNs) have recently attracted significant interest due to their computational efficiency. Concurrently, it has been shown that neural networks may be overly sensitive to ``attacks" -- tiny adversarial changes in the input -- which may be detrimental to their use in safety-critical domains. De... | ["binarized neural networks", "combinatorial optimization", "integer programming"] | ABSTRACTBinarized Neural Networks (BNNs) have recently attracted significant interest dueto their computational efficiency. Concurrently, it has been shown that neuralnetworks may be overly sensitive to “attacks” – tiny adversarial changes in theinput – which may be detrimental to their use in safety-critical domains. ... | H1xCvGx3hX | an interesting paper | 6: Marginally above acceptance threshold | This paper proposed a new attack algorithm based on MILP on binary neural networks. In addition to the full MILP formulation, the authors proposed an integer target propagation algorithm (IProp) to find adversarial examples by solving a smaller (instead of the full) MILP.
The topic is important but the clarity shoul... | 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
Combinatorial Attacks on Binarized Neural Networks
### Paper Abstract
Binarized Neural Networks (BNNs) have recently attracted significant interest due to their computat... |
S1lvn0NtwH | ICLR.cc/2020/Conference | 2020 | Mutual Exclusivity as a Challenge for Deep Neural Networks | ["Kanishk Gandhi", "Brenden Lake"] | Strong inductive biases allow children to learn in fast and adaptable ways. Children use the mutual exclusivity (ME) bias to help disambiguate how words map to referents, assuming that if an object has one label then it does not need another. In this paper, we investigate whether or not standard neural architectures ha... | ["Cognitive Science", "Deep Learning", "Word Learning", "Lifelong Learning"] | ABSTRACTStrong inductive biases allow children to learn in fast and adaptable ways. Chil-dren use the mutual exclusivity (ME) bias to help disambiguate how words map toreferents, assuming that if an object has one label then it does not need another. Inthis paper, we investigate whether or not standard neural architect... | SkedBT5ntB | Official Blind Review #3 | 6: Weak Accept | *** Increased to weak accept after discussion of merits of ME bias was improved in the paper ***
This paper investigates whether neural networks exhibit a ‘mutual exclusivity (ME) bias’, whereby novel inputs tend to be associated with previously unseen outputs, an inductive bias that is cited to be present in childre... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Mutual Exclusivity as a Challenge for Deep Neural Networks
### Paper Abstract
Strong inductive biases allow children to learn in fast and adaptable ways. Children use th... | |
n1HD8M6WGn | ICLR.cc/2021/Conference | 2021 | Understanding and Improving Encoder Layer Fusion in Sequence-to-Sequence Learning | ["Xuebo Liu", "Longyue Wang", "Derek F. Wong", "Liang Ding", "Lidia S. Chao", "Zhaopeng Tu"] | Encoder layer fusion (EncoderFusion) is a technique to fuse all the encoder layers (instead of the uppermost layer) for sequence-to-sequence (Seq2Seq) models, which has proven effective on various NLP tasks. However, it is still not entirely clear why and when EncoderFusion should work. In this paper, our main contribu... | ["Encoder layer fusion", "Transformer", "Sequence-to-sequence learning", "Machine translation", "Summarization", "Grammatical error correction"] | ABSTRACTEncoder layer fusion (EncoderFusion) is a technique to fuse all the encoder lay-ers (instead of the uppermost layer) for sequence-to-sequence (Seq2Seq) models,which has proven effective on various NLP tasks. However, it is still not entirelyclear why and when EncoderFusion should work. In this paper, our main c... | Qz0OMIa4aj7 | Thorough analysis of encoder fusion asking for more focus on impact | 5: Marginally below acceptance threshold | The authors perform a thorough analysis of encoder fusion for Transformers: which encoder layer should the N-th decoder layer attend to? It turns out that the final decoder layers often attend to the encoder embeddings, leading the authors to provide them to the last decoder layer which leads to small improvements of p... | 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
Understanding and Improving Encoder Layer Fusion in Sequence-to-Sequence Learning
### Paper Abstract
Encoder layer fusion (EncoderFusion) is a technique to fuse all the ... |
HyydRMZC- | ICLR.cc/2018/Conference | 2018 | Spatially Transformed Adversarial Examples | ["Chaowei Xiao", "Jun-Yan Zhu", "Bo Li", "Warren He", "Mingyan Liu", "Dawn Song"] | Recent studies show that widely used Deep neural networks (DNNs) are vulnerable to the carefully crafted adversarial examples.
Many advanced algorithms have been proposed to generate adversarial examples by leveraging the L_p distance for penalizing perturbations.
Different defense methods have also been explored to de... | ["adversarial examples", "spatial transformation"] | ABSTRACTRecent studies show that widely used deep neural networks (DNNs) are vulner-able to carefully crafted adversarial examples. Many advanced algorithms havebeen proposed to generate adversarial examples by leveraging the Lpdistance forpenalizing perturbations. Researchers have explored different defense methods to... | ry_xOQ5ef | Interesting solid paper, could use a few more experiments | 7: Good paper, accept | This paper creates adversarial images by imposing a flow field on an image such that the new spatially transformed image fools the classifier. They minimize a total variation loss in addition to the adversarial loss to create perceptually plausible adversarial images, this is claimed to be better than the normal L2 los... | 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
Spatially Transformed Adversarial Examples
### Paper Abstract
Recent studies show that widely used Deep neural networks (DNNs) are vulnerable to the carefully crafted ad... |
BJgGhiR5KX | ICLR.cc/2019/Conference | 2019 | Learning Cross-Lingual Sentence Representations via a Multi-task Dual-Encoder Model | ["Muthuraman Chidambaram", "Yinfei Yang", "Daniel Cer", "Steve Yuan", "Yun-Hsuan Sung", "Brian Strope", "Ray Kurzweil"] | A significant roadblock in multilingual neural language modeling is the lack of labeled non-English data. One potential method for overcoming this issue is learning cross-lingual text representations that can be used to transfer the performance from training on English tasks to non-English tasks, despite little to no t... | ["sentence", "embeddings", "zero-shot", "multilingual", "multi-task", "cross-lingual"] | ABSTRACTA significant roadblock in multilingual neural language modeling is the lack of la-beled non-English data. One potential method for overcoming this issue is learningcross-lingual text representations that can be used to transfer the performance fromtraining on English tasks to non-English tasks, despite little ... | SJlXDD81pX | A new framework for cross-lingual sentence representation which is an interesting mix of standard building blocks, but more convincing experiments are needed to appreciate the main contributions. | 7: Good paper, accept | This paper proposes a novel cross-lingual multi-tasking framework based on a dual-encoder model that can learn cross-lingual sentence representations which are useful in monolingual tasks and cross-lingual tasks for both languages involved in the training, as observed on the experiments for three language pairs. The ma... | 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 Cross-Lingual Sentence Representations via a Multi-task Dual-Encoder Model
### Paper Abstract
A significant roadblock in multilingual neural language modeling i... |
B1EjKsRqtQ | ICLR.cc/2019/Conference | 2019 | Hierarchical Attention: What Really Counts in Various NLP Tasks | ["Zehao Dou", "Zhihua Zhang"] | Attention mechanisms in sequence to sequence models have shown great ability and wonderful performance in various natural language processing (NLP) tasks, such as sentence embedding, text generation, machine translation, machine reading comprehension, etc. Unfortunately, existing attention mechanisms only learn eithe... | ["attention", "hierarchical", "machine reading comprehension", "poem generation"] | ABSTRACTAttention mechanisms in sequence to sequence models have shown great abilityand wonderful performance in various natural language processing (NLP) tasks,such as sentence embedding, text generation, machine translation, machine readingcomprehension, etc. Unfortunately, existing attention mechanisms only learneit... | B1eHy_bahX | A few major issues | 3: Clear rejection | The paper proposes to enhance existing multi-level attention (self-attention) mechanism by obtaining query and key vectors (= value vectors) from all levels after weighted-averaging them. The paper claims that this is also theoretically beneficial because the loss function will converge to zero as the number of layers ... | 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
Hierarchical Attention: What Really Counts in Various NLP Tasks
### Paper Abstract
Attention mechanisms in sequence to sequence models have shown great ability and wonde... |
rJlUhhVYvS | ICLR.cc/2020/Conference | 2020 | Understanding Isomorphism Bias in Graph Data Sets | ["Ivanov Sergey", "Sviridov Sergey", "Evgeny Burnaev"] | In recent years there has been a rapid increase in classification methods on graph structured data. Both in graph kernels and graph neural networks, one of the implicit assumptions of successful state-of-the-art models was that incorporating graph isomorphism features into the architecture leads to better empirical per... | ["graph classification", "data sets", "graph representation learning"] | ABSTRACTIn recent years there has been a rapid increase in classification methods on graphstructured data. Both in graph kernels and graph neural networks, one of theimplicit assumptions of successful state-of-the-art models was that incorporatinggraph isomorphism features into the architecture leads to better empirica... | SklrAvRctS | Official Blind Review #3 | 6: Weak Accept | The authors discuss here the problem of isomorphism bias in graph dataset, i.e. the overfitting effect in learning networks whenever graph isomorphism features are incorporated within the model. This is a bias which jeopardises the validity and the reproducibility of several studies, and it is theoretically analogous t... | <|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 Isomorphism Bias in Graph Data Sets
### Paper Abstract
In recent years there has been a rapid increase in classification methods on graph structured data.... | |
BylkG20qYm | ICLR.cc/2019/Conference | 2019 | On Meaning-Preserving Adversarial Perturbations for Sequence-to-Sequence Models | ["Paul Michel", "Graham Neubig", "Xian Li", "Juan Miguel Pino"] | Adversarial examples have been shown to be an effective way of assessing the robustness of neural sequence-to-sequence (seq2seq) models, by applying perturbations to the input of a model leading to large degradation in performance. However, these perturbations are only indicative of a weakness in the model if they do n... | ["Sequence-to-sequence", "adversarial attacks", "evaluation", "meaning preservation", "machine translation"] | ABSTRACTAdversarial examples have been shown to be an effective way of assessing therobustness of neural sequence-to-sequence (seq2seq) models, by applying pertur-bations to the input of a model leading to large degradation in performance. How-ever, these perturbations are only indicative of a weakness in the model if ... | Skexz1j3hm | Not enough novelty for acceptance | 4: Ok but not good enough - rejection | The paper is about meaning-preserving adversarial perturbations in the context of Seq2Seq models. The paper proposes two ways of achieving that: (a) kNN - substituting word with nearest neighbors from the word embedding space, and (b) character swapping. It's debatable if character swapping is really meaning preserving... | 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 Meaning-Preserving Adversarial Perturbations for Sequence-to-Sequence Models
### Paper Abstract
Adversarial examples have been shown to be an effective way of assessi... |
S1enmaVFvS | ICLR.cc/2020/Conference | 2020 | Data-Driven Approach to Encoding and Decoding 3-D Crystal Structures | ["Jordan Hoffmann", "Louis Maestrati", "Yoshihide Sawada", "Jian Tang", "Jean Michel Sellier", "Yoshua Bengio"] | Generative models have achieved impressive results in many domains including image and text generation. In the natural sciences, generative models have lead to rapid progress in automated drug discovery. Many of the current methods focus on either 1-D or 2-D representations of typically small, drug-like molecules. How... | ["molecules", "dataset", "atoms", "crystal structures", "impressive results", "many domains", "image", "text generation", "natural sciences"] | ABSTRACTGenerative models have achieved impressive results in many domains includingimage and text generation. In the natural sciences, generative models have lead torapid progress in automated drug discovery. Many of the current methods focus oneither 1-D or 2-D representations of typically small, drug-like molecules.... | Hkl1FAmAFr | Official Blind Review #2 | 8: Accept | The paper deals with accurately encoding and decoding 3D atomic positions and the crystal’s species using 2 sets of neural networks a) a VAE that builds a compressed latent space representation of a crystal and b) a UNET for segmenting the latent space into atoms and assigns each atom to its atomic number. Experiments ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Data-Driven Approach to Encoding and Decoding 3-D Crystal Structures
### Paper Abstract
Generative models have achieved impressive results in many domains including imag... | |
HJg3HyStwB | ICLR.cc/2020/Conference | 2020 | Perturbations are not Enough: Generating Adversarial Examples with Spatial Distortions | ["He Zhao", "Trung Le", "Paul Montague", "Olivier De Vel", "Tamas Abraham", "Dinh Phung"] | Deep neural network image classifiers are reported to be susceptible to adversarial evasion attacks, which use carefully crafted images created to mislead a classifier. Recently, various kinds of adversarial attack methods have been proposed, most of which focus on adding small perturbations to input images. Despite th... | ["adversarial examples", "perturbations", "spatial distortions", "susceptible", "images", "small perturbations", "classifiers", "spatial distortions perturbations", "adversarial evasion attacks"] | ABSTRACTDeep neural network image classifiers are reported to be susceptible to adversarialevasion attacks, which use carefully crafted images created to mislead a classifier.Recently, various kinds of adversarial attack methods have been proposed, mostof which focus on adding small perturbations to input images. Despi... | r1e3Y-QTFH | Official Blind Review #2 | 1: Reject | This paper builds upon the work of AdvGAN and proposes to add spatial transformations on top of it. The resulting attacking framework is demonstrated to outperform AdvGAN on attacking several defense approaches, such as Defense-GAN, AdvCritic and adversarial training. Compared to previous approaches on generating spati... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Perturbations are not Enough: Generating Adversarial Examples with Spatial Distortions
### Paper Abstract
Deep neural network image classifiers are reported to be suscep... | |
EGVxmJKLC2L | ICLR.cc/2021/Conference | 2021 | Learning not to learn: Nature versus nurture in silico | ["Robert Tjarko Lange", "Henning Sprekeler"] | Animals are equipped with a rich innate repertoire of sensory, behavioral and motor skills, which allows them to interact with the world immediately after birth. At the same time, many behaviors are highly adaptive and can be tailored to specific environments by means of learning. In this work, we use mathematical anal... | ["Meta-Learning", "Reinforcement Learning"] | ABSTRACTAnimals are equipped with a rich innate repertoire of sensory, behavioral and motorskills, which allows them to interact with the world immediately after birth. Atthe same time, many behaviors are highly adaptive and can be tailored to specificenvironments by means of learning. In this work, we use mathematical... | F91bjPVYAqN | Review | 5: Marginally below acceptance threshold | Summary: This paper observes that in meta-RL (and evolutionary biology), sometimes it is advantageous to learn behaviors that adapt to the particular task, while other times not adapting to the task, and instead relying on a task-agnostic “hard-coded” behavior is sufficient. While much meta-RL research typically focuse... | 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 not to learn: Nature versus nurture in silico
### Paper Abstract
Animals are equipped with a rich innate repertoire of sensory, behavioral and motor skills, whi... |
B1xu6yStPH | ICLR.cc/2020/Conference | 2020 | Using Explainabilty to Detect Adversarial Attacks | ["Ohad Amosy and Gal Chechik"] | Deep learning models are often sensitive to adversarial attacks, where carefully-designed input samples can cause the system to produce incorrect decisions. Here we focus on the problem of detecting attacks, rather than robust classification, since detecting that an attack occurs may be even more important than avoidin... | ["adversarial", "detection", "explainability"] | Under review as a conference paper at ICLR 2019Using Explainabilty to Detect AdversarialAttacksAnonymous authorsPaper under double-blind reviewAbstractDeep learning models are often sensitive to adversarial attacks, wherecarefully-designed input samples can cause the system to produce incor-rect decisions. Here we focu... | Bke1qCDAtB | Official Blind Review #3 | 3: Weak Reject | A Simple method to detect adversarial examples, but needs more work.
#Summary:
The paper proposed a method that utilizes the model’s explainability to detect adversarial images whose explanations that are not consistent with the predicted class. The explainability is generated by SHAP, which uses Shapley values to id... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Using Explainabilty to Detect Adversarial Attacks
### Paper Abstract
Deep learning models are often sensitive to adversarial attacks, where carefully-designed input samp... | |
SJlj8CNYl | ICLR.cc/2017/workshop | 2017 | Unsupervised Motion Flow estimation by Generative Adversarial Networks | ["Stefano Alletto", "Luca Rigazio"] | In this paper we address the challenging problem of unsupervised motion flow estimation. Under the assumption that image reconstruction is a super-set of the motion flow estimation problem, we train a convolutional neural network to interpolate adjacent video frames and then compute the motion flow via region-based sen... | ["Computer vision", "Deep learning", "Unsupervised Learning"] | ABSTRACTIn this paper we address the challenging problem of unsupervised motion flowestimation. Under the assumption that image reconstruction is a super-set of themotion flow estimation problem, we train a convolutional neural network to in-terpolate adjacent video frames and then compute the motion flow via region-ba... | ryKBGigsg | Official review | 5: Marginally below acceptance threshold | The paper proposes an unsupervised learning approach to image matching. The authors train a deep network for video frame interpolation, and use the trained model to infer the correspondences between frames with backpropagation-based sensitivity analysis. The authors show that adversarial training of the interpolation n... | 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 Motion Flow estimation by Generative Adversarial Networks
### Paper Abstract
In this paper we address the challenging problem of unsupervised motion flow es... |
S1g-qaGhoE | ICML.cc/2019/Workshop/RL4RealLife | 2019 | Off-Policy Policy Gradient with State Distribution Correction | ["Yao Liu", "Adith Swaminathan", "Alekh Agarwal", "Emma Brunskill"] | We study the problem of off-policy policy optimization in Markov decision processes,and develop a novel off-policy policy gradient method. Prior off-policy policy gradient approaches have generally ignored the mismatch between the distribution of states visited under the behavior policy used ... | ["Reinforcement Learning", "Off-Policy Learning", "Policy Gradient", "Actor-Critic"] | Off-Policy Policy Gradient with State Distribution CorrectionYao Liu1Adith Swaminathan2Alekh Agarwal2Emma Brunskill1AbstractWe study the problem of off-policy policy opti-mization in Markov decision processes, and de-velop a novel off-policy policy gradient method.Prior off-policy policy gradient approaches havegeneral... | rygQYUuvpE | Extension on prior work on policy evaluation using state distribution correction to off-policy actor critic | 2: Weak Reject | The work extends prior work (Liu et. al 2018) on Infinite-Horizon Off-Policy Estimation to the case of off-policy policy optimization with actor-critic. The authors compared the proposed method to baseline off-policy actor critic and demonstrated the effectiveness of state-distribution correction in multiple simulation... | 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
Off-Policy Policy Gradient with State Distribution Correction
### Paper Abstract
We study the problem of off-policy policy optimization in Markov decision proc... |
N6JECD-PI5w | ICLR.cc/2021/Conference | 2021 | FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders | ["Pengyu Cheng", "Weituo Hao", "Siyang Yuan", "Shijing Si", "Lawrence Carin"] | Pretrained text encoders, such as BERT, have been applied increasingly in various natural language processing (NLP) tasks, and have recently demonstrated significant performance gains. However, recent studies have demonstrated the existence of social bias in these pretrained NLP models. Although prior works have made p... | ["Fairness", "Contrastive Learning", "Mutual Information", "Pretrained Text Encoders"] | ABSTRACTPretrained text encoders, such as BERT, have been applied increasingly in variousnatural language processing (NLP) tasks, and have recently demonstrated signifi-cant performance gains. However, recent studies have demonstrated the existenceofsocial bias in these pretrained NLP models. Although prior works have ... | XtqbsDdHe2_ | Clearly written, relatively simple, fairly effective | 6: Marginally above acceptance threshold | This paper introduces a novel technique for debiasing pretrained contextual embedding models. Their approach trains a 2 layer fully-connected neural network which takes as input the output from the pretrained model and outputs a new, "debiased" representation. This model is trained by minimizing the InfoNCE between 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
FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders
### Paper Abstract
Pretrained text encoders, such as BERT, have been applied increasingly in va... |
8IbZUle6ieH | ICLR.cc/2021/Conference | 2021 | Graph Neural Network Acceleration via Matrix Dimension Reduction | ["Shunhua Jiang", "Yunze Man", "Zhao Song", "Danyang Zhuo"] | Graph Neural Networks (GNNs) have become the de facto method for machine learning on graph data (e.g., social networks, protein structures, code ASTs), but they require significant time and resource to train. One alternative method is Graph Neural Tangent Kernel (GNTK), a kernel method that corresponds to infinitely wi... | ["Graph Neural Networks", "Deep learning", "Optimization", "Kernel Method"] | Under review as a conference paper at ICLR 2021Graph Neural Network Acceleration viaMatrix Dimension ReductionAnonymous authorsPaper under double-blind reviewAbstractGraph Neural Networks (GNNs) have become the de facto method for ma-chine learning on graph data (e.g., social networks, protein structures, codeASTs), bu... | stUQ4WUMBPD | A good paper on accelerating graph neural tangent kernel | 5: Marginally below acceptance threshold | In this paper, the authors propose two techniques to speed up the training of the graph neural tangent kernel, by matrix decoupling and sketching. Experiments are convincing.
We find the title of the paper inappropriate, because the paper only consider the graph neural tangent kernel.
In general, the paper is diffic... | 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
Graph Neural Network Acceleration via Matrix Dimension Reduction
### Paper Abstract
Graph Neural Networks (GNNs) have become the de facto method for machine learning on ... |
S1tLzLkvf | ICLR.cc/2018/Workshop | 2018 | USING WORD EMBEDDING TO SELECTIVELY DISCLOSE DATABASE INFORMATION | ["Rajesh Bordawekar", "Oded Shmueli"] | Database information may be disclosed in a variety of ways depending on the sensitivity of the stored information and the recipient’s need to know. Traditionally, researchers have been concerned with preventing a recipient of the information from associating sensitive information (e.g., a disease) with specific individ... | ["Word Embedding", "Data Privacy", "Relational Databases"] | ABSTRACTDatabase information may be disclosed in a variety of ways depending on thesensitivity of the stored information and the recipient’s need to know. Tradition-ally, researchers have been concerned with preventing a recipient of the infor-mation from associating sensitive information (e.g., a disease) with specifi... | ByhnpUytG | I sort of liked the idea of word embeddings as semantic hashes, but not sure there's a good fit here | 3: Clear rejection | Summary:
This paper suggest that encoding via word embeddings is a way to hide information in a database while still disclosing some information such as identity and similarity. (I could not speak how this relates to prior work in similar areas - outside my domain.)
Novelty:
This doesn't feel to me that novel, but t... | 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
USING WORD EMBEDDING TO SELECTIVELY DISCLOSE DATABASE INFORMATION
### Paper Abstract
Database information may be disclosed in a variety of ways depending on the sensitiv... |
Skl1HCNKDr | ICLR.cc/2020/Conference | 2020 | Learning Generative Models using Denoising Density Estimators | ["Siavash Bigdeli", "Geng Lin", "Tiziano Portenier", "Andrea Dunbar", "Matthias Zwicker"] | Learning generative probabilistic models that can estimate the continuous density given a set of samples, and that can sample from that density is one of the fundamental challenges in unsupervised machine learning. In this paper we introduce a new approach to obtain such models based on what we call denoising density e... | ["generative probabilistic models", "denoising autoencoders", "neural density estimation"] | ABSTRACTLearning generative probabilistic models that can estimate the continuous den-sity given a set of samples, and that can sample from that density, is one of thefundamental challenges in unsupervised machine learning. In this paper we in-troduce a new approach to obtain such models based on what we call denoising... | ryg_dkuQqB | Official Blind Review #2 | 6: Weak Accept | This paper provides a method for density estimation in high-dimensional domains. The proposed algorithm combines ideas from denoising autoencoders The noise estimator procedure developed in this paper estimates the logarithm of the unnormalized density. The inference procedure is stochastic gradient descent w.r.t. the ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Generative Models using Denoising Density Estimators
### Paper Abstract
Learning generative probabilistic models that can estimate the continuous density given ... | |
1uOTdL2H9i | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | Becoming Cat People: Animal-like Human Experiences with a Sensory Augmenting Whisker Wearable | ["Anonymous"] | Humans have a natural curiosity to imagine what it feels like to exist as someone or something else. This curiosity becomes even stronger for the pets we care for. Humans cannot truly know what it is like to be our pets, but we can deepen our understanding of what it is like to perceive and explore the world like them.... | ["Wearables", "biosensing", "empathy", "exploration", "exploitation"] | ABSTRACTHumans have a natural curiosity to imagine what it feels liketo exist as someone or something else. This curiosity becomeseven stronger for the pets we care for. Humans cannot trulyknow what it is like to be our pets, but we can deepen ourunderstanding of what it is like to perceive and explore theworld like th... | OttbUyzl5d | An intriguing study with good motivation and most of the procedures done right, but falls short on findings and does not deliver as promised. | 4: Ok but not good enough - rejection | In this paper, the authors present a whisker-like wearable device (Whisker Beard) and conducted a study to investigate how we can create technologies and environments that remediate non-human experiences (cat in this case), and human’s impressions towards them. Main findings include participants did exhibit navigation ... | 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
Becoming Cat People: Animal-like Human Experiences with a Sensory Augmenting Whisker Wearable
### Paper Abstract
Humans have a natural curiosity to imagine what it feels... |
H1gL3RVtwr | ICLR.cc/2020/Conference | 2020 | CURSOR-BASED ADAPTIVE QUANTIZATION FOR DEEP NEURAL NETWORK | ["Bapu Li(*)", "Yanwen Fan(*)", "Zhiyu Cheng", "Yingze Bao (* means equal contribution)"] | Deep neural network (DNN) has rapidly found many applications in different scenarios.
However, its large computational cost and memory consumption are barriers
to computing restrained applications. DNN model quantization is a widely
used method to reduce the DNN storage and computation burden by decreasing
the bit widt... | ["adaptive quantization", "dnn", "das", "multiple bits", "das process", "cursor", "deep neural network", "many applications", "different scenarios"] | ABSTRACTDeep neural network (DNN) has rapidly found many applications in different sce-narios. However, its large computational cost and memory consumption are bar-riers to computing restrained applications. DNN model quantization is a widelyused method to reduce the DNN storage and computation burden by decreasingthe ... | Syx6-A1ZqH | Official Blind Review #3 | 6: Weak Accept | The paper proposes a novel way to determine automatically the quantization level at each layer of
deep neural networks while training. The quantization can result in a more compact network requiring
less memory for storing the parameters (weights), with minimal drop in accuracy.
If I understand correctly, the main id... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
CURSOR-BASED ADAPTIVE QUANTIZATION FOR DEEP NEURAL NETWORK
### Paper Abstract
Deep neural network (DNN) has rapidly found many applications in different scenarios. Howev... | |
SyxfEn09Y7 | ICLR.cc/2019/Conference | 2019 | G-SGD: Optimizing ReLU Neural Networks in its Positively Scale-Invariant Space | ["Qi Meng", "Shuxin Zheng", "Huishuai Zhang", "Wei Chen", "Qiwei Ye", "Zhi-Ming Ma", "Nenghai Yu", "Tie-Yan Liu"] | It is well known that neural networks with rectified linear units (ReLU) activation functions are positively scale-invariant. Conventional algorithms like stochastic gradient descent optimize the neural networks in the vector space of weights, which is, however, not positively scale-invariant. This mismatch may lead to... | ["optimization", "neural network", "irreducible positively scale-invariant space", "deep learning"] | ABSTRACTIt is well known that neural networks with rectified linear units (ReLU) activationfunctions are positively scale-invariant. Conventional algorithms like stochasticgradient descent optimize the neural networks in the vector space of weights, whichis, however, not positively scale-invariant. This mismatch may le... | B1gKT5yq37 | interesting idea, but not convincing exps | 7: Good paper, accept | This paper proposed a new training algorithm, G-SGD, by exploring the positively scale-invariant space for relu neural networks. The basic idea is to identify the basis paths in the path graph, and convert the weight update in SGD to the weight rescaling in G-SGD. My major concerns are as follows:
1. Empirical signifi... | 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
G-SGD: Optimizing ReLU Neural Networks in its Positively Scale-Invariant Space
### Paper Abstract
It is well known that neural networks with rectified linear units (ReLU... |
BJleciCcKQ | ICLR.cc/2019/Conference | 2019 | EXPLORATION OF EFFICIENT ON-DEVICE ACOUSTIC MODELING WITH NEURAL NETWORKS | ["Wonyong Sung", "Lukas Lee", "Jinwhan Park"] | Real-time speech recognition on mobile and embedded devices is an important application of neural networks. Acoustic modeling is the fundamental part of speech recognition and is usually implemented with long short-term memory (LSTM)-based recurrent neural networks (RNNs). However, the single thread execution of an LST... | ["Parallelization", "Speech Recognition", "Sequence Modeling", "Recurrent Neural Network", "Embedded Systems"] | ABSTRACTReal-time speech recognition on mobile and embedded devices is an important ap-plication of neural networks. Acoustic modeling is the fundamental part of speechrecognition and is usually implemented with long short-term memory (LSTM)-based recurrent neural networks (RNNs). However, the single thread execution o... | HklSyeodh7 | review of the paper | 4: Ok but not good enough - rejection | This paper investigates a number of techniques and neural network architectures for embedded acoustic modeling. The goal is to reduce the memory access and make efficient computation, in the meantime, to sustain good ASR performance. Overall, the paper is well motivated and well written. However, I have following co... | 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
EXPLORATION OF EFFICIENT ON-DEVICE ACOUSTIC MODELING WITH NEURAL NETWORKS
### Paper Abstract
Real-time speech recognition on mobile and embedded devices is an important ... |
bsRjn0RH620 | ICLR.cc/2021/Conference | 2021 | Towards Understanding the Cause of Error in Few-Shot Learning | ["Liang Song", "Jinlu Liu", "Yongqiang Qin"] | Few-Shot Learning (FSL) is a challenging task of recognizing novel classes from scarce labeled samples. Many existing researches focus on learning good representations that generalize well to new categories. However, given low-data regime, the restricting factors of performance on novel classes has not been well studie... | ["upper bound of error", "feature separability", "classifier discrepancy", "few-shot learning"] | ABSTRACTFew-Shot Learning (FSL) is a challenging task of recognizing novel classes fromscarce labeled samples. Many existing researches focus on learning good repre-sentations that generalize well to new categories. However, given low-data regime,the restricting factors of performance on novel classes has not been well... | rzOaGvLwQFW | This paper shows mathematically and experimentally that reducing classifier discrepancy leads to better results. | 6: Marginally above acceptance threshold | I have not had the time to write a detailed review but I have gone through the paper carefully.
Motivation
The paper is motivated well and presents an interesting mathematical derivation that drives the subsequent approach based on reducing classifier discrepancy. The manuscript has a number of typos and grammatical er... | 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
Towards Understanding the Cause of Error in Few-Shot Learning
### Paper Abstract
Few-Shot Learning (FSL) is a challenging task of recognizing novel classes from scarce ... |
H1lJJnR5Ym | ICLR.cc/2019/Conference | 2019 | Exploration by random network distillation | ["Yuri Burda", "Harrison Edwards", "Amos Storkey", "Oleg Klimov"] | We introduce an exploration bonus for deep reinforcement learning methods that is easy to implement and adds minimal overhead to the computation performed. The bonus is the error of a neural network predicting features of the observations given by a fixed randomly initialized neural network. We also introduce a method ... | ["reinforcement learning", "exploration", "curiosity"] | ABSTRACTWe introduce an exploration bonus for deep reinforcement learning methods thatis easy to implement and adds minimal overhead to the computation performed.The bonus is the error of a neural network predicting features of the observationsgiven by a fixed randomly initialized neural network. We also introduce a me... | rylpqi3tnX | A simple yet surprisingly effective take on intrinsic motivation for exploration in sparse reward RL tasks | 7: Good paper, accept | The algorithm proposed in this paper consists in driving exploration in RL through an intrinsic reward, computed as the prediction error of a neural network whose target is the output of a randomly initialized network (with the state reached by the agent as input). The intuition is that rarely seen states will have a l... | 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
Exploration by random network distillation
### Paper Abstract
We introduce an exploration bonus for deep reinforcement learning methods that is easy to implement and add... |
ryl-RTEYvB | ICLR.cc/2020/Conference | 2020 | Robust Learning with Jacobian Regularization | ["Judy Hoffman", "Daniel A. Roberts", "Sho Yaida"] | Design of reliable systems must guarantee stability against input perturbations. In machine learning, such guarantee entails preventing overfitting and ensuring robustness of models against corruption of input data. In order to maximize stability, we analyze and develop a computationally efficient implementation of Jac... | ["Supervised Representation Learning", "Few-Shot Learning", "Regularization", "Adversarial Defense", "Deep Learning"] | ABSTRACTDesign of reliable systems must guarantee stability against input perturbations.In machine learning, such guarantee entails preventing overfitting and ensuringrobustness of models against corruption of input data. In order to maximizestability, we analyze and develop a computationally efficient implementation o... | SygfkBvaYH | Official Blind Review #2 | 3: Weak Reject | This paper proposes an efficient method to (differentiably) estimate input-output Jacobian. The method is useful for Jacobian regularization. The regularization improves robustness and generalization of networks.
I tend to vote for rejection. There are two concerns. 1) This paper needs to demonstrate the effectiveness... | <|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 Learning with Jacobian Regularization
### Paper Abstract
Design of reliable systems must guarantee stability against input perturbations. In machine learning, suc... | |
r7qgus1bZ2 | ICLR.cc/2021/Conference | 2021 | FOC OSOD: Focus on Classification One-Shot Object Detection | ["Hanqing Yang", "Huaijin Pi", "SABA GHORBANI BARZEGAR", "Yu Zhang"] | One-shot object detection (OSOD) aims at detecting all instances that are consistent with the category of the single reference image. OSOD achieves object detection by comparing the query image and the reference image. We observe that the essential problem behind the limited performance of OSOD is that OSOD generates a... | ["one-shot object detection", "false positives", "focus on classification"] | ABSTRACTOne-shot object detection (OSOD) aims at detecting all instances that are con-sistent with the category of the single reference image. OSOD achieves objectdetection by comparing the query image and the reference image. We observethat the essential problem behind the limited performance of OSOD is that OSODgener... | LqIoNyS0Lv5 | Review | 4: Ok but not good enough - rejection | 1. Summary
The paper identifies false positives as the major problem in one-shot object detection and proposes a model (FOC OSOD) to overcome this problem. The method is tested on two one-hot object detection benchmarks on Pascal VOC and COCO improving upon previous work.
2. Strengths
+ Having worked on one-shot obje... | 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
FOC OSOD: Focus on Classification One-Shot Object Detection
### Paper Abstract
One-shot object detection (OSOD) aims at detecting all instances that are consistent with ... |
HJMXTsCqYQ | ICLR.cc/2019/Conference | 2019 | Constrained Bayesian Optimization for Automatic Chemical Design | ["Ryan-Rhys Griffiths", "Jos\u00e9 Miguel Hern\u00e1ndez-Lobato"] | Automatic Chemical Design provides a framework for generating novel molecules with optimized molecular properties. The current model suffers from the pathology that it tends to produce invalid molecular structures. By reformulating the search procedure as a constrained Bayesian optimization problem, we showcase improve... | ["Bayesian Optimization", "Generative Models"] | ABSTRACTAutomatic Chemical Design provides a framework for generating novel moleculeswith optimized molecular properties. The current model suffers from the pathol-ogy that it tends to produce invalid molecular structures. By reformulating thesearch procedure as a constrained Bayesian optimization problem, we showcasei... | rJlRry6x2Q | Novelty is limited. No comparison with SOTA models | 4: Ok but not good enough - rejection | This paper proposes to improve the chemical compound generation by the Bayes optimization strategy, not by the new models.
The main proposal is to use the acquisition that switches the function based on the violation of a constraint, estimated via a BNN.
I understand that the objective function, J_{comp}^{QED} is ne... | 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
Constrained Bayesian Optimization for Automatic Chemical Design
### Paper Abstract
Automatic Chemical Design provides a framework for generating novel molecules with opt... |
UZ0SqYpH4ot | iscaconf.org/ISCA/2022/Workshop/MLArchSys | 2022 | Towards GPU-driven Code Execution for Distributed Deep Learning | ["Changho Hwang", "KyoungSoo Park", "Ran Shu", "Xinyuan Qu", "Peng Cheng", "Yongqiang Xiong"] | Modern state-of-the-art deep learning (DL) applications tend to scale out to a large number of parallel GPUs. Unfortunately, we observe that the collective communication overhead across GPUs is often the key limiting factor of performance for distributed DL. It under-utilizes the networking bandwidth by frequent transf... | ["machine learning", "GPU", "communication"] | Towards GPU-driven Code Execution forDistributed Deep LearningChangho Hwang∗KAISTKyoungSoo ParkKAISTRan ShuMicrosoft ResearchXinyuan Qu†UCASPeng ChengMicrosoft ResearchYongqiang XiongMicrosoft ResearchAbstract —Modern state-of-the-art deep learning (DL) appli-cations tend to scale out to a large number of parallel GPUs... | iKSgn5KJUOv | The paper is well-written. The current sets of evaluations are solid. However, more evaluations are needed to support the paper. | 7: Good paper, accept | Summary:
The authors identified the latency challenge when executing parallelism in SOTA large-scaled DNN models. The authors propose a GPU-driven DMA method to communicate data between GPUs, saving latency and increasing throughput.
Pro:
* The authors base their arguments on real-system evaluation results. The data-d... | 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
Towards GPU-driven Code Execution for Distributed Deep Learning
### Paper Abstract
Modern state-of-the-art deep learning (DL) applications tend to scale out to a large n... |
Y-Wl1l0Va- | ICLR.cc/2021/Conference | 2021 | Shortest-Path Constrained Reinforcement Learning for Sparse Reward Tasks | ["Sungryull Sohn", "Sungtae Lee", "Jongwook Choi", "Harm van Seijen", "Honglak Lee", "Mehdi Fatemi"] | We propose the k-Shortest-Path (k-SP) constraint: a novel constraint on the agent’s trajectory that improves the sample-efficiency in sparse-reward MDPs. We show that any optimal policy necessarily satisfies the k-SP constraint. Notably, the k-SP constraint prevents the policy from exploring state-action pairs along th... | ["reinforcement learning", "exploration", "sample efficient reinforcement learning", "sparse rewards"] | ABSTRACTWe propose the k-Shortest-Path ( k-SP) constraint: a novel constraint on the agent’strajectory that improves the sample-efficiency in sparse-reward MDPs. We showthat any optimal policy necessarily satisfies the k-SP constraint. Notably, thek-SP constraint prevents the policy from exploring state-action pairs al... | gnMTIIHcnV | A well written paper introducing shortest path constraint for sparse reward MDPs | 6: Marginally above acceptance threshold | **Summary**
This paper proposes a new constraint for constrained MDP, based on k-shortest path, which helps improve sample efficiency for (model-free) RL algorithms in sparse-reward MDP, while theoretically proving that the constraint retains the same optimal policy in the original MDP. Intuitively, for sparse (positiv... | 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
Shortest-Path Constrained Reinforcement Learning for Sparse Reward Tasks
### Paper Abstract
We propose the k-Shortest-Path (k-SP) constraint: a novel constraint on the a... |
bNdohBx9sPa | ICLR.cc/2021/Conference | 2021 | Unsupervised Domain Adaptation via Minimized Joint Error | ["Dexuan Zhang", "Tatsuya Harada"] | Unsupervised domain adaptation transfers knowledge from learned source domain to a different (but related) target distribution, for which only few or no labeled data is available. Some researchers proposed upper bounds for the target error when transferring the knowledge, i.e.,Ben-David et al. (2010) established a theo... | ["joint error", "marginal distributions", "unsupervised domain adaptation", "domain adaptation", "domain adaptation transfers", "source domain", "different", "target distribution", "data", "available"] | June 21, 2021 | f4LGm4G8t0F | new theory for unsupervised domain adaptation; good empirical results | 7: Good paper, accept | This paper studies the problem of unsupervised domain adaption, giving a theoretical analysis that yields a new upper bound on the target error. In particular, in contrast to much existing work, this paper focuses on the role of the ideal joint source/target error in the upper bound. The paper proposes two methods fo... | 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
Unsupervised Domain Adaptation via Minimized Joint Error
### Paper Abstract
Unsupervised domain adaptation transfers knowledge from learned source domain to a different ... |
H1acq85gx | ICLR.cc/2017/conference | 2017 | Maximum Entropy Flow Networks | ["Gabriel Loaiza-Ganem *", "Yuanjun Gao *", "John P. Cunningham"] | Maximum entropy modeling is a flexible and popular framework for formulating statistical models given partial knowledge. In this paper, rather than the traditional method of optimizing over the continuous density directly, we learn a smooth and invertible transformation that maps a simple distribution to the desired ma... | ["flexible", "popular framework", "statistical models", "partial knowledge", "traditional", "continuous density", "smooth", "invertible transformation", "simple distribution"] | ABSTRACTMaximum entropy modeling is a flexible and popular framework for formulat-ing statistical models given partial knowledge. In this paper, rather than the tra-ditional method of optimizing over the continuous density directly, we learn asmooth and invertible transformation that maps a simple distribution to the d... | ryrU9ztre | Application of normalizing flows to max-ent | 6: Marginally above acceptance threshold | This paper applies the idea of normalizing flows (NFs), which allows us to build complex densities with tractable likelihoods, to maximum entropy constrained optimization.
The paper is clearly written and is easy to follow.
Novelty is a weak factor in this paper. The main contributions come from (1) applying previous... | 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
Maximum Entropy Flow Networks
### Paper Abstract
Maximum entropy modeling is a flexible and popular framework for formulating statistical models given partial knowledge.... |
ryl3blSFPr | ICLR.cc/2020/Conference | 2020 | Denoising Improves Latent Space Geometry in Text Autoencoders | ["Tianxiao Shen", "Jonas Mueller", "Regina Barzilay", "Tommi Jaakkola"] | Neural language models have recently shown impressive gains in unconditional text generation, but controllable generation and manipulation of text remain challenging. In particular, controlling text via latent space operations in autoencoders has been difficult, in part due to chaotic latent space geometry. We propose ... | ["controllable text generation", "autoencoders", "denoising", "latent space geometry"] | ABSTRACTNeural language models have recently shown impressive gains in unconditional textgeneration, but controllable generation and manipulation of text remain challenging.In particular, controlling text via latent space operations in autoencoders has beendifficult, in part due to chaotic latent space geometry. We pro... | BylI49ChFr | Official Blind Review #3 | 6: Weak Accept | The paper argues that adding noise to the inputs of an adversarial autoencoder for text improves the geometry of the learned latent space (in terms of mapping similar input sentences to nearby points in the latent space). The authors present a mathematical argument for why adding noise to the inputs would enforce laten... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Denoising Improves Latent Space Geometry in Text Autoencoders
### Paper Abstract
Neural language models have recently shown impressive gains in unconditional text genera... | |
dwIudqXzBCL | AAAI.org/2023/Bridge/CCBridge | 2023 | Continually Updating Neural Causal Models | ["Florian Peter Busch", "Jonas Seng", "Moritz Willig", "Matej Ze\u010devi\u0107"] | A common assumption in causal modelling is that the relations between variables are fixed mechanisms. But in reality, these mechanisms often change over time and new data might not fit the original model as well. But is it reasonable to regularly train new models or can we update a single model continually instead? We ... | ["neural causal models", "mechanisms", "time", "common assumption", "causal modelling", "relations", "variables", "reality", "new data", "original model"] | Continually Updating Neural Causal ModelsFlorian Peter Busch†1,2Jonas Seng1Moritz Willig1Matej Ze ˇcevi ́c11Computer Science Department, AIML, TU Darmstadt, Germany2Hessian Center for AI (hessian.AI)†correspondence: florian peter.busch@tu-darmstadt.deAbstractA common assumption in causal modelling is that the rela-tion... | zuNhLj03Sg2 | Suggestions | 6: Marginally above acceptance threshold | Given the length requirements, I think this paper is a bit challenging to read coming from a continual learning background. I think briefly introducing causal models, and specifically NCMs would be useful. Then motivate why they would benefit from continual learning. While there are a plethora of continual learning par... | 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
Continually Updating Neural Causal Models
### Paper Abstract
A common assumption in causal modelling is that the relations between variables are fixed mechanisms. But in... |
iQxS0S9ir1a | ICLR.cc/2021/Conference | 2021 | Distributional Generalization: A New Kind of Generalization | ["Preetum Nakkiran", "Yamini Bansal"] | We introduce a new notion of generalization--- Distributional Generalization--- which roughly states that outputs of a classifier at train and test time are close as distributions, as opposed to close in just their average error. For example, if we mislabel 30% of dogs as cats in the train set of CIFAR-10, then a ResNe... | ["understanding deep learning", "generalization", "interpolating methods", "empirical investigation"] | ABSTRACTWe introduce a new notion of generalization— Distributional Generalization—which roughly states that outputs of a classifier at train and test time are close asdistributions , as opposed to close in just their average error. For example, if wemislabel 30% of dogs as cats in the train set of CIFAR-10, then a Res... | _623CXN_77 | Review | 7: Good paper, accept | This paper proposes an extended notion of generalization. The new proposed notion asks that for a family of tests $T: X \times Y → [0,1]$, $T(x, f(x))$ will be similar for train/test examples. The paper proposes three interesting conjectures that are related to distributional generalization. The paper proves Conjectur... | 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
Distributional Generalization: A New Kind of Generalization
### Paper Abstract
We introduce a new notion of generalization--- Distributional Generalization--- which roug... |
7pgFL2Dkyyy | ICLR.cc/2021/Conference | 2021 | Class Normalization for (Continual)? Generalized Zero-Shot Learning | ["Ivan Skorokhodov", "Mohamed Elhoseiny"] | Normalization techniques have proved to be a crucial ingredient of successful training in a traditional supervised learning regime. However, in the zero-shot learning (ZSL) world, these ideas have received only marginal attention. This work studies normalization in ZSL scenario from both theoretical and practical persp... | ["zero-shot learning", "normalization", "continual learning", "initialization"] | ABSTRACTNormalization techniques have proved to be a crucial ingredient of successfultraining in a traditional supervised learning regime. However, in the zero-shotlearning (ZSL) world, these ideas have received only marginal attention. Thiswork studies normalization in ZSL scenario from both theoretical and practicalp... | rJAddEroK-W | An excellent paper investigating the normalization effect in the zero-shot learning context | 8: Top 50% of accepted papers, clear accept | ================
Summary:
This paper provides a thorough analysis in the perspective of data variances on the widely used normalization tricks in the zero-shot learning research: normalize+scale and attribute normalization. It also demonstrates these tricks are not enough w.r.t. normalizing the variance in a non-line... | 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
Class Normalization for (Continual)? Generalized Zero-Shot Learning
### Paper Abstract
Normalization techniques have proved to be a crucial ingredient of successful trai... |
r1lKJlSiK4 | MIDL.io/2019/Conference/Abstract | 2019 | Template Transformer Networks for Image Segmentation | ["Matthew Chung Hai Lee", "Kersten Petersen", "Nick Pawlowski", "Ben Glocker", "Michiel Schaap"] | In this paper we introduce and compare different approaches for incorporating shape prior information into neural network based image segmentation. Specifically, we introduce the concept of template transformer networks where a shape template is deformed to match the underlying structure of interest through an end-to-e... | ["Image Segmentation", "Shape Priors", "Neural Networks", "Template Deformation", "Image Registration"] | Proceedings of Machine Learning Research { Under Review:1{4, 2019 Extended Abstract { MIDL 2019 submissionTemplate Transformer Networks for Image SegmentationMatthew Chung Hai Lee1;2mlee@heartflow.comKersten Petersen1kpetersen@heartflow.comNick Pawlowski2n.pawlowski16@imperial.ac.ukBen Glocker1;2b.glocker@imperial.ac.u... | Bkg7rTV8oN | Interesting work on incorporating shape priors into neural networks | 4: Strong accept | This paper presents a method for incorporating shape prior into neural networks to improve segmentation.
This is achieved by introducing template transformer network where the shape is transformed to match the target using spatial transformer networks. In addition, a method to incorporate priors into standard CNNs is... | 3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Template Transformer Networks for Image Segmentation
### Paper Abstract
In this paper we introduce and compare different approaches for incorporating shape prior informa... |
BGMfS7tgIWq | aclweb.org/ACL/2022/Workshop/CMCL | 2022 | Seeing the advantage: visually grounding word embeddings to better capture human semantic knowledge | ["Anonymous"] | Distributional semantic models capture word-level meaning that is useful in many natural language processing tasks and have even been shown to capture cognitive aspects of word meaning. The majority of these models are purely text based, even though the human sensory experience is much richer. In this paper we create v... | ["Distributional semantics", "Computational modelling", "Multi-modal learning"] | Seeing the advantage: visually grounding word embeddings to bettercapture human semantic knowledgeDanny MerkxRadboud UniversityNijmegen, The Netherlandsdanny.merkx@ru.nlStefan L. FrankRadboud UniversityNijmegen, The Netherlandsstefan.frank@ru.nlMirjam ErnestusRadboud UniversityNijmegen, The Netherlandsmirjam.ernestus@r... | r4heOuEdof9 | Careful experiments that disentangle visual information in text and actual image data | 7: Good paper, accept | This is a well-written paper that carefully experiments with VGEs and compares the additional variance that can be explained using VGEs compared to text-based models.
Strong points:
The paper makes a contribution to previous work by disentangling the visual information that may be contained in the text part of a da... | 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
Seeing the advantage: visually grounding word embeddings to better capture human semantic knowledge
### Paper Abstract
Distributional semantic models capture word-level ... |
_QdvdkxOii6 | ICLR.cc/2021/Conference | 2021 | Measuring Progress in Deep Reinforcement Learning Sample Efficiency | ["Florian E. Dorner"] | Sampled environment transitions are a critical input to deep reinforcement learning (DRL) algorithms. Current DRL benchmarks often allow for the cheap and easy generation of large amounts of samples such that perceived progress in DRL does not necessarily correspond to improved sample efficiency. As simulating real wor... | ["Deep Reinforcement Learning", "Sample Efficiency"] | ABSTRACTSampled environment transitions are a critical input to deep reinforcement learn-ing (DRL) algorithms. Current DRL benchmarks often allow for the cheap andeasy generation of large amounts of samples such that perceived progress in DRLdoes not necessarily correspond to improved sample efficiency. As simulating r... | bQkvpLiFj0u | Review | 5: Marginally below acceptance threshold | The paper is trying to make extensive and systematic investigation in deep RL papers to measure the recent progress in the broad literatures. The authors look into published state-of-the-art results in Atari, state-based continuous control and pixel-based continuous control. Inferring from papers' published training cu... | 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
Measuring Progress in Deep Reinforcement Learning Sample Efficiency
### Paper Abstract
Sampled environment transitions are a critical input to deep reinforcement learni... |
VKmBj45zWs | IEEE.org/2022/Workshop/altVIS | 2022 | Changing Lanes: Towards Symbolizing Visualizations | ["Anonymous"] | The data visualization community often focuses on identifying and putting best practices into action when thinking about chart and dashboard authoring. These practices often incorporate recommendations informed by human perception and are often prescriptive around precision at the expense of semantics. Drawing from the... | ["Symbols", "icons", "register", "visual design", "semantics", "intent"] | Changing Lanes: Towards Symbolizing VisualizationsVidya Setlur and Bridget CogleyFig. 1. Icons and symbols in modern pop culture. Row-wise from left to right: dining, slippery road, resist, wash hands, wear a mask, Cfor Croissant in an alphabet card, all-gender bathroom, shh - be quiet!, “I love you” in ASL, finger hea... | 08JnR9fgJWC | In this work, the authors present a position suggesting that visualization may benefit from greater use of symbolization. This position is supported by a series of examples (flatten the curve, maps, and road signs). While the ultimate call to action derived from this stance is somewhat pedestrian---and well supported 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
Changing Lanes: Towards Symbolizing Visualizations
### Paper Abstract
The data visualization community often focuses on identifying and putting best practices into actio... | |||
HyY4Owjll | ICLR.cc/2017/conference | 2017 | Boosted Generative Models | ["Aditya Grover", "Stefano Ermon"] | We propose a new approach for using boosting to create an ensemble of generative models, where models are trained in sequence to correct earlier mistakes. Our algorithm can leverage many existing base learners, including recent latent variable models. Further, our approach allows the ensemble to leverage discriminativ... | ["Theory", "Deep learning", "Unsupervised Learning"] | ABSTRACTWe propose a new approach for using boosting to create an ensemble of gener-ative models, where models are trained in sequence to correct earlier mistakes.Our algorithm can leverage many existing base learners, including recent latentvariable models. Further, our approach allows the ensemble to leverage discrim... | SkmDS6lEg | Interesting work, many weak points | 5: Marginally below acceptance threshold | The paper proposes two approaches to boosting generative models, both based on likelihood ratio estimates. The approaches are evaluated on synthetic data, as well as on MNIST dataset for the tasks of generating samples and semi-supervised learning.
While the idea of boosting generative models and the proposed methods a... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Boosted Generative Models
### Paper Abstract
We propose a new approach for using boosting to create an ensemble of generative models, where models are trained in sequenc... |
HxZZV3MQ20Y | ML_Reproducibility_Challenge/2021/Fall | 2021 | Transparent Object Tracking Benchmark | ["\u017diga Trojer"] | Scope of Reproducibility
In the article, the authors of the Transparent Object Tracking Benchmark compare the performance of 25 state-of-the-art tracking algorithms, evaluated on the TOTB dataset, with a new proposed algorithm for tracking transparent objects called TransATOM. Authors claim that it outperforms all othe... | ["totb dataset", "algorithms", "trackers", "transparent object", "performance", "algorithm", "results", "authors", "transparent objects", "transatom"] | Transparent Object Tracking BenchmarkAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2In the article, the authors of the Transparent Object Tracking Benchmark compare the performance of 25 state-of- 3the-art tracking algorithms, evaluated on the TOTB dataset, with a new propo... | r6feWTJt5x9 | Official Review | 8: Top 50% of accepted papers, clear accept | **Scope of reproducibility**
The report defines well the scope of the reproducibility.
**Code**
The authors have re-used public implementations of the different trackers as well as the metrics.
**Communication with original authors**
The authors of the report reached out the authors of the original article, which ... | 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
Transparent Object Tracking Benchmark
### Paper Abstract
Scope of Reproducibility In the article, the authors of the Transparent Object Tracking Benchmark compare the pe... |
2FQgZhV-LDL | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | scikit-learn Pipelines meet Knowledge Graphs - The Python kgextension Package | ["Tabea-Clara Bucher", "Xuehui Jiang", "Ole Meyer", "Stephan Waitz", "Sven Hertling", "Heiko Paulheim"] | Python is currently the most used platform for data science and machine learning. At the same time, public knowledge graphs have been identified as a valuable source of background knowledge in many data science tasks. In this paper, we introduce the kgextension package for Python, which allows for using knowledge graph... | ["Python", "scikit-learn", "Knowledge Graph", "Background Knowledge", "Data Mining"] | scikit-learn Pipelines meet Knowledge GraphsThe Python kgextension PackageTabea-Clara Bucher, Xuehui Jiang, Ole Meyer, Stephan Waitz,Sven Hertling[0000000303335888], and Heiko Paulheim[0000000343868195]Data and Web Science Group, University of Mannheim, Germanyftbucher,xujiang,olmeyer,swaitz g@mail.uni-mannheim.de,fhei... | yIZJYOfK8he | A "facilitator" for KG research | 8: Top 50% of accepted papers, clear accept | The paper introduces kgextension, which make available a set of utilities for link exploration and feature generation.
The paper is clear, the code available and well documented. There is no evaluation in the paper, but a jupyter notebook in the repository (clearly linked in the paper) serves to the scope.
The libra... | 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
scikit-learn Pipelines meet Knowledge Graphs - The Python kgextension Package
### Paper Abstract
Python is currently the most used platform for data science and machine ... |
-I8lSfnXFPI | icaps-conference.org/ICAPS/2021/Workshop/HSDIP | 2021 | Classical Planning with Avoid Conditions | ["Marcel Steinmetz", "Joerg Hoffmann", "Stefan Borgwardt", "Alisa Kovtunova"] | It is often natural in planning to specify conditions that should be avoided, characterizing dangerous or highly undesirable behavior. PDDL3 supports this with temporal-logic state constraints. Here we focus on the simpler case where the constraint is a non-temporal formula $phi$ -- the avoid condition -- that must be ... | ["classical planning", "heuristic search", "state constraints"] | Classical Planning with Avoid ConditionsMarcel Steinmetz1, J ̈org Hoffmann1, Alisa Kovtunova2, Stefan Borgwardt21Saarland University, Saarland Informatics Campus, Saarbr ̈ucken, Germany2Institute of Theoretical Computer Science, Technische Universit ̈at Dresden, Germanylastname@cs.uni-saarland.de, firstname.lastname@tu... | gX4wDimRbjb | Review | The paper investigates planning tasks with avoid conditions, i.e. tasks where
we are given a formula $\varphi$ which must be avoided in all states. While
this is a special type of PDDL3 constraint (and thus existing PDDL3
compilations can handle it), an approach tailored for avoid conditions seems
more promising in ter... | 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
Classical Planning with Avoid Conditions
### Paper Abstract
It is often natural in planning to specify conditions that should be avoided, characterizing dangerous or hig... | |
TmkN9JmDJx1 | ICLR.cc/2021/Conference | 2021 | Thinking Like Transformers | ["Gail Weiss", "Yoav Goldberg", "Eran Yahav"] | What is the computational model behind a transformer? Where recurrent neural networks have direct parallels in finite state machines, allowing clear discussion and thought around architecture variants or trained models, transformers have no such familiar parallel. In this paper we aim to change that, proposing a comput... | ["transformers", "interpretability"] | ABSTRACTWhat is the computational model behind a transformer? Where recurrent neuralnetworks have direct parallels in finite state machines, allowing clear discussion andthought around architecture variants or trained models, transformers have no suchfamiliar parallel. In this paper we aim to change that, proposing a c... | nXDj_IJqITL | Very interesting but underdeveloped | 4: Ok but not good enough - rejection | The authors introduce a DSL, the Restricted Access Sequence Processing (RASP) language, that they claim can serve as a computational model for the transformer-encoder. They develop the reader's intuition for RASP by providing RASP implementations of many basic operations such as computing histograms, sorting, and reve... | 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
Thinking Like Transformers
### Paper Abstract
What is the computational model behind a transformer? Where recurrent neural networks have direct parallels in finite state... |
Syt0r4bRZ | ICLR.cc/2018/Conference | 2018 | Tree2Tree Learning with Memory Unit | ["Ning Miao", "Hengliang Wang", "Ran Le", "Chongyang Tao", "Mingyue Shang", "Rui Yan", "Dongyan Zhao"] | Traditional recurrent neural network (RNN) or convolutional neural net- work (CNN) based sequence-to-sequence model can not handle tree structural data well. To alleviate this problem, in this paper, we propose a tree-to-tree model with specially designed encoder unit and decoder unit, which recursively encodes tree in... | ["model", "embeddings", "tree", "rnn", "convolutional neural", "work", "cnn"] | ABSTRACTTraditional recurrent neural network (RNN) or convolutional neural net-work (CNN) based sequence-to-sequence model can not handle tree structuraldata well. To alleviate this problem, in this paper, we propose a tree-to-tree modelwith specially designed encoder unit and decoder unit, which recursively encodestre... | HJA-R_Fxf | review | 2: Strong rejection | Summary: the paper proposes a tree2tree architecture for NLP tasks. Both the encoder and decoder of this architecture make use of memory cells: the encoder looks like a tree-lstm to encode a tree bottom-up, the decoder generates a tree top-down by predicting the number of children first. The objective function is a lin... | 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
Tree2Tree Learning with Memory Unit
### Paper Abstract
Traditional recurrent neural network (RNN) or convolutional neural net- work (CNN) based sequence-to-sequence mode... |
DRnNTPsjTCQ | MICCAI.org/2021/Workshop/COMPAY | 2021 | Self supervised learning improves dMMR/MSI detection from histology slides across multiple cancers | ["Charlie Saillard", "Olivier Dehaene", "Tanguy Marchand", "Olivier Moindrot", "Aur\u00e9lie Kamoun", "Benoit Schmauch", "Simon Jegou"] | Microsatellite instability (MSI) is a tumor phenotype whose diagnosis largely impacts patient care in colorectal cancers (CRC), and is associated with response to immunotherapy in all solid tumors. Deep learning models detecting MSI tumors directly from H&E stained slides have shown promise in improving diagnosis of MS... | ["Self-supervised learning", "Microsatellite Instability (MSI)"] | Proceedings of Machine Learning Research 156, 2021 MICCAI Computational Pathlogy (COMPAY) WorkshopSelf-supervised learning improves dMMR/MSI detectionfrom histology slides across multiple cancersCharlie Saillard charlie.saillard@owkin.comOwkin, Inc.Olivier Dehaene olivier.dehaene@gmail.comOwkin, Inc.Tanguy Marchand tan... | RP2m0di8GVD | Self Supervised Learning for MSI Detection | 5: Marginally below acceptance threshold | The paper's main claim is that self-supervised learning helps improve the performance of MSI detection.
The paper is well written and the problem is introduced well.
Kather et al, 2019 and Bilal et al, 2020 both learn only from tumour patches, whereas you learn from all tissue patches. As far as I'm aware, MSI relat... | 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
Self supervised learning improves dMMR/MSI detection from histology slides across multiple cancers
### Paper Abstract
Microsatellite instability (MSI) is a tumor phenoty... |
r1MVuoCctX | ICLR.cc/2019/Conference | 2019 | MAJOR-MINOR LSTMS FOR WORD-LEVEL LANGUAGE MODEL | ["Kai Shuang", "Rui Li", "Mengyu Gu", "Qianqian Yang", "Jonathan", "Sen Su"] | As a widely-accepted evaluation criterion, complexity has attracted more and more attention in the design of language models. The parameter count is a proxy for complexity, which is often reported and compared in research papers. In general, more parameters means better model performance, but higher complexity. Therefo... | ["Language model", "LSTM", "Deep Learning", "NLP"] | ABSTRACTAs a widely-accepted evaluation criterion, complexity has attracted more andmore attention in the design of language models. The parameter count is a proxyfor complexity, which is often reported and compared in research papers. In gen-eral, more parameters means better model performance, but higher complexity.T... | HJgpV8XK3X | Problems of clarity , novelty and significance | 3: Clear rejection | The paper proposes an architecture for language modeling composed of a bigger and smaller LSTM. The authors propose to add skip connections directly from the input layer to the layers of the "major" lstm. The skip-connections are computed using the small LSTM. The authors claim to reach better performance with smaller ... | 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
MAJOR-MINOR LSTMS FOR WORD-LEVEL LANGUAGE MODEL
### Paper Abstract
As a widely-accepted evaluation criterion, complexity has attracted more and more attention in the des... |
BJg9DoR9t7 | ICLR.cc/2019/Conference | 2019 | Max-MIG: an Information Theoretic Approach for Joint Learning from Crowds | ["Peng Cao*", "Yilun Xu*", "Yuqing Kong", "Yizhou Wang"] | Eliciting labels from crowds is a potential way to obtain large labeled data. Despite a variety of methods developed for learning from crowds, a key challenge remains unsolved: \emph{learning from crowds without knowing the information structure among the crowds a priori, when some people of the crowds make highly corr... | ["crowdsourcing", "information theory"] | ABSTRACTEliciting labels from crowds is a potential way to obtain large labeled data. De-spite a variety of methods developed for learning from crowds, a key challengeremains unsolved: learning from crowds without knowing the information structureamong the crowds a priori, when some people of the crowds make highly cor... | rygf8BMF37 | More details on actual learning are required | 6: Marginally above acceptance threshold | Update after feedback: I would like to thank the authors for their detailed answers, it would be great to see some revisions in the paper also though (except new experimental results).
Especially thank you for providing details of a training procedure which I was missing in the initial draft. I hope to see them in 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
Max-MIG: an Information Theoretic Approach for Joint Learning from Crowds
### Paper Abstract
Eliciting labels from crowds is a potential way to obtain large labeled data... |
H92-E4kFwbR | ICLR.cc/2021/Conference | 2021 | Composite Adversarial Training for Multiple Adversarial Perturbations and Beyond | ["Xinyang Zhang", "Zheng Zhang", "Ting Wang"] | One intriguing property of deep neural networks (DNNs) is their vulnerability to adversarial perturbations. Despite the plethora of work on defending against individual perturbation models, improving DNN robustness against the combinations of multiple perturbations is still fairly under-studied. In this paper, we propo... | ["adversarial examples", "deep learning", "robustness"] | ABSTRACTOne intriguing property of deep neural networks (DNNs) is their vulnerability toadversarial perturbations. Despite the plethora of work on defending against indi-vidual perturbation models, improving DNN robustness against the combinationsof multiple perturbations is still fairly under-studied. In this paper, w... | AVCMFSvCEIr | Nice contribution to adversarial training literature but with mixed results | 5: Marginally below acceptance threshold | The authors propose a method for dealing with *composite* adversarial attacks, which are defined as a sequence of perturbation operators each applying some constrained perturbation to the output of the previous operator. Their method models the composed adversarial examples $x^*$ as the sum of the unperturbed example w... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Composite Adversarial Training for Multiple Adversarial Perturbations and Beyond
### Paper Abstract
One intriguing property of deep neural networks (DNNs) is their vulne... |
3ZeGLibhFo0 | ICLR.cc/2021/Conference | 2021 | Enabling counterfactual survival analysis with balanced representations | ["Paidamoyo Chapfuwa", "Serge Assaad", "Shuxi Zeng", "Michael Pencina", "Lawrence Carin", "Ricardo Henao"] | Balanced representation learning methods have been applied successfully to counterfactual
inference from observational data. However, approaches that account for
survival outcomes are relatively limited. Survival data are frequently encountered
across diverse medical applications, i.e., drug development, risk profiling... | ["survival analysis", "time-to-event", "counterfactual inference", "causal survival analysis"] | ABSTRACTBalanced representation learning methods have been applied successfully to coun-terfactual inference from observational data. However, approaches that account forsurvival outcomes are relatively limited. Survival data are frequently encounteredacross diverse medical applications, i.e., drug development, risk pr... | ncuqnOmWwsy | Important problem but limited methodological contribution. | 5: Marginally below acceptance threshold | **Summary and key claims**
This paper repurposes the balanced representation learning framework for estimating treatment effects, originally proposed in (Shalit et al., 2017), for the survival prediction setup. The proposed model deals with selection bias induced by confounded treatment variables, and in addition, dea... | 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
Enabling counterfactual survival analysis with balanced representations
### Paper Abstract
Balanced representation learning methods have been applied successfully to cou... |
rJgdHs05FQ | ICLR.cc/2019/Conference | 2019 | Nonlinear Channels Aggregation Networks for Deep Action Recognition | ["Zhigang Zhu", "Hongbing Ji", "Wenbo Zhang", "Cheng Ouyang"] | We introduce the concept of channel aggregation in ConvNet architecture, a novel compact representation of CNN features useful for explicitly modeling the nonlinear channels encoding especially when the new unit is embedded inside of deep architectures for action recognition. The channel aggregation is based on multipl... | ["action recognition", "convolutional neural network", "network training"] | Under review as a conference paper at ICLR 2019Nonlinear Channels Aggregation Networksfor Deep Action RecognitionAnonymous authorsPaper under double-blind reviewAbstractW e introduce the concept of channel aggregation in ConvNet architecture, anovel compact representation of CNN features useful for explicitly modelingt... | BJxMaQHFhQ | Non linear 'channel aggregation' to improve on 3D convs and various 3D pooling methods. Badly written paper with low novelty and limited and unconvincing empirical evaluation. | 3: Clear rejection | Summary
The paper proposes a method for channel aggregation in CNNs for the task of human action
recognition. The main component proposed is the nonlinear channel aggregation layer (NCAL) which can
be potentially inserted in any CNN architecture. Empirical results are reported on UCF101 and
HMDB51.
General 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
Nonlinear Channels Aggregation Networks for Deep Action Recognition
### Paper Abstract
We introduce the concept of channel aggregation in ConvNet architecture, a novel c... |
VX0swzJEzpg | AKBC.ws/2021/Conference | 2021 | Abstractified Multi-instance Learning (AMIL) for Biomedical Relation Extraction | ["William P Hogan", "Molly Huang", "Yannis Katsis", "Tyler Baldwin", "Ho-Cheol Kim", "Yoshiki Baeza", "Andrew Bartko", "Chun-Nan Hsu"] | Relation extraction in the biomedical domain is a challenging task due to a lack of labeled data and a long-tail distribution of fact triples. Many works leverage distant supervision which automatically generates labeled data by pairing a knowledge graph with raw textual data. Distant supervision produces noisy labels ... | ["Information Extraction", "Relation Extraction", "Biomedical NLP", "Machine Learning"] | Automated Knowledge Base Construction (2021) Conference paperAbstractied Multi-instance Learning (AMIL)for Biomedical Relation ExtractionWilliam Hogan1,2whogan@ucsd.eduMolly Huang2m7huang@ucsd.eduYannis Katsis3yannis.katsis@ibm.comTyler Baldwin3tbaldwin@us.ibm.comHo-Cheol Kim3hckim@us.ibm.comYoshiki Vazquez Baeza2yoshi... | sOoOslWtPVF | Valid contribution and encouraging results | 6: Marginally above acceptance threshold | The paper presents explores the benefits of abstractifying relationships, i.e., taking into account the type signatures of its instances, to improve relation extraction from biomedical text corpora. The paper applies this technique on a state-of-the-art method for relation extraction based on BioBERT. The experimental ... | 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
Abstractified Multi-instance Learning (AMIL) for Biomedical Relation Extraction
### Paper Abstract
Relation extraction in the biomedical domain is a challenging task due... |
Ef1nNHQHZ20 | ICLR.cc/2021/Conference | 2021 | Layer-wise Adversarial Defense: An ODE Perspective | ["Zonghan Yang", "Yang Liu", "Chenglong Bao", "Zuoqiang Shi"] | Deep neural networks are observed to be fragile against adversarial attacks, which have dramatically limited their practical applicability. On improving model robustness, the adversarial training techniques have proven effective and gained increasing attention from research communities. Existing adversarial training ap... | ["adversarial training", "robustness", "ODE"] | ABSTRACTDeep neural networks are observed to be fragile against adversarial attacks, whichhave dramatically limited their practical applicability. On improving model ro-bustness, the adversarial training techniques have proven effective and gained in-creasing attention from research communities. Existing adversarial tr... | jHGVFM8A1KZ | Interesting ODE interpretations of layer-wise adversarial training | 5: Marginally below acceptance threshold | The authors proposed a new adversarial training scheme based on ODE techniques. The standard adversarial training seeks for small perturbations in the input space. But the perturbations can be found in the feature space. From the observation and the similarity between the layer-wise adversarial attacks and ODE formulat... | 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
Layer-wise Adversarial Defense: An ODE Perspective
### Paper Abstract
Deep neural networks are observed to be fragile against adversarial attacks, which have dramaticall... |
Bye6weHFvB | ICLR.cc/2020/Conference | 2020 | Plan2Vec: Unsupervised Representation Learning by Latent Plans | ["Ge Yang", "Amy Zhang", "Ari Morcos", "Joelle Pineau", "Pieter Abbeel", "Roberto Calandra"] | Creating a useful representation of the world takes more than just rote memorization of individual data samples. This is because fundamentally, we use our internal representation to plan, to solve problems, and to navigate the world. For a representation to be amenable to planning, it is critical for it to embody some ... | ["Unsupervised Learning", "Reinforcement Learning", "Manifold Learning"] | Under review as a conference paper at ICLR 2020Plan2Vec: Unsupervised RepresentationLearning by Latent PlansAnonymous authorsPaper under double-blind reviewAbstractCreating a useful representation of the world takes more than just rotememorization of individual data samples. This is because fundamentally,we use our int... | H1g_F7Z0KB | Official Blind Review #3 | 1: Reject | The paper proposes to learn a representation space that is useful for planning. This is done by a 2-step process: (1) learn a metric to reflect locality, (2) utilize that metric to learn a global metric by utilizing UVFAs from Reinforcement Learning literature. While the idea may be an interesting direction to improve ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Plan2Vec: Unsupervised Representation Learning by Latent Plans
### Paper Abstract
Creating a useful representation of the world takes more than just rote memorization of... | |
BJ8c3f-0b | ICLR.cc/2018/Conference | 2018 | Auto-Encoding Sequential Monte Carlo | ["Tuan Anh Le", "Maximilian Igl", "Tom Rainforth", "Tom Jin", "Frank Wood"] | We build on auto-encoding sequential Monte Carlo (AESMC): a method for model and proposal learning based on maximizing the lower bound to the log marginal likelihood in a broad family of structured probabilistic models. Our approach relies on the efficiency of sequential Monte Carlo (SMC) for performing inference in st... | ["Variational Autoencoders", "Inference amortization", "Model learning", "Sequential Monte Carlo", "ELBOs"] | ABSTRACTWe build on auto-encoding sequential Monte Carlo ( AESMC ):1a method for modeland proposal learning based on maximizing the lower bound to the log marginallikelihood in a broad family of structured probabilistic models. Our approachrelies on the efficiency of sequential Monte Carlo ( SMC) for performing inferen... | Syi1Z34ZG | Some very interesting ideas and empirical results. But there are a couple of issues. | 7: Good paper, accept | Update:
On further consideration (and reading the other reviews), I'm bumping my rating up to a 7. I think there are still some issues, but this work is both valuable and interesting, and it deserves to be published (alongside the Naesseth et al. and Maddison et al. work).
-----------
This paper proposes a version 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
Auto-Encoding Sequential Monte Carlo
### Paper Abstract
We build on auto-encoding sequential Monte Carlo (AESMC): a method for model and proposal learning based on maxim... |
H43MpnN_vZ9 | ICLR.cc/2022/Workshop/DGM4HSD | 2022 | Annealed Importance Sampling meets Score Matching | ["Arnaud Doucet", "Will Sussman Grathwohl", "Alexander G. de G. Matthews", "Heiko Strathmann"] | Annealed Importance Sampling (AIS) is one of the most effective methods for marginal likelihood estimation. It relies on a sequence of distributions interpolating between a tractable initial distribution and the posterior of interest which we simulate from approximately using a non-homogeneous Markov chain. To obtain a... | ["importance sampling", "monte carlo", "annealing", "score matching"] | ABSTRACTAnnealed Importance Sampling (AIS) is one of the most effective methods formarginal likelihood estimation. It relies on a sequence of distributions interpolat-ing between a tractable initial distribution and the posterior of interest which wesimulate from approximately using a non-homogeneous Markov chain. To o... | BzxdmE4KG9 | Improving AIS with a better extended target distribution | 7: Good paper, accept | This paper proposes a novel AIS procedure for Bayesian evidence estimation. The central idea is that, while the extended target distribution used by standard AIS (eq 2) is convenient, it is not optimal. The authors suggest a better extended target distribution motivated by score matching. On the three benchmark example... | 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
Annealed Importance Sampling meets Score Matching
### Paper Abstract
Annealed Importance Sampling (AIS) is one of the most effective methods for marginal likelihood esti... |
PQ2Cel-1rJh | ICLR.cc/2021/Conference | 2021 | Pea-KD: Parameter-efficient and accurate Knowledge Distillation | ["IKHYUN CHO", "U Kang"] | How can we efficiently compress a model while maintaining its performance? Knowledge Distillation (KD) is one of the widely known methods for model compression. In essence, KD trains a smaller student model based on a larger teacher model and tries to retain the teacher model's level of performance as much as possible... | ["BERT", "Deep Learning", "Natural Language Processing", "Transformer", "Knowledge Distillation", "Parameter Sharing"] | ABSTRACTHow can we efficiently compress a model while maintaining its performance?Knowledge Distillation (KD) is one of the widely known methods for model com-pression. In essence, KD trains a smaller student model based on a larger teachermodel and tries to retain the teacher model’s level of performance as much aspos... | JhWFWPoSzy | The paper is poorly written. The motivation and the proposed method are problematic. | 5: Marginally below acceptance threshold | This paper proposes a distillation method for BERT. The work is based on two-fold main ideas. First, as the student model is usually smaller in the number of parameters, the model capacity is limited. The authors propose to stack the layers that share parameters to counter this limitation. Second, the authors argue 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
Pea-KD: Parameter-efficient and accurate Knowledge Distillation
### Paper Abstract
How can we efficiently compress a model while maintaining its performance? Knowledge ... |
SyYe6k-CW | ICLR.cc/2018/Conference | 2018 | Deep Bayesian Bandits Showdown: An Empirical Comparison of Bayesian Deep Networks for Thompson Sampling | ["Carlos Riquelme", "George Tucker", "Jasper Snoek"] | Recent advances in deep reinforcement learning have made significant strides in performance on applications such as Go and Atari games. However, developing practical methods to balance exploration and exploitation in complex domains remains largely unsolved. Thompson Sampling and its extension to reinforcement learning... | ["exploration", "Thompson Sampling", "Bayesian neural networks", "bandits", "reinforcement learning", "variational inference", "Monte Carlo"] | ABSTRACTRecent advances in deep reinforcement learning have made significant strides inperformance on applications such as Go and Atari games. However, developingpractical methods to balance exploration and exploitation in complex domainsremains largely unsolved. Thompson Sampling and its extension to reinforcementlear... | H11if2uxf | Benchmark most useful if accompanying code is well engineered | 7: Good paper, accept | If two major questions below are answered affirmatively, I believe this article could be very good contribution to the field and deserve publication in ICLR.
In this article the authors provide a service to the community by comparing the current most used algorithms for Thompson Sampling-based contextual (parametric) ... | 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 Bayesian Bandits Showdown: An Empirical Comparison of Bayesian Deep Networks for Thompson Sampling
### Paper Abstract
Recent advances in deep reinforcement learnin... |
SylR6n4tPS | ICLR.cc/2020/Conference | 2020 | Learning to Generate Grounded Visual Captions without Localization Supervision | ["Chih-Yao Ma", "Yannis Kalantidis", "Ghassan AlRegib", "Peter Vajda", "Marcus Rohrbach", "Zsolt Kira"] | When automatically generating a sentence description for an image or video, it often remains unclear how well the generated caption is grounded, or if the model hallucinates based on priors in the dataset and/or the language model. The most common way of relating image regions with words in caption models is through an... | ["image captioning", "video captioning", "self-supervised learning", "visual grounding"] | ABSTRACTWhen automatically generating a sentence description for an image or video, itoften remains unclear how well the generated caption is grounded, or if the modelhallucinates based on priors in the dataset and/or the language model. The mostcommon way of relating image regions with words in caption models is throu... | Hkg1_5l1cB | Official Blind Review #3 | 6: Weak Accept | This paper addresses captioning generation for images and videos, by proposing a novel cyclical training regimen consisting of three steps: decoding, localization, and reconstruction. The experimental results show that the performance on image captioning and video captioning are improved without grounding supervision.
... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Generate Grounded Visual Captions without Localization Supervision
### Paper Abstract
When automatically generating a sentence description for an image or vi... | |
HylfPgHYvr | ICLR.cc/2020/Conference | 2020 | Occlusion resistant learning of intuitive physics from videos | ["Ronan Riochet", "Josef Sivic", "Ivan Laptev", "Emmanuel Dupoux"] | To reach human performance on complex tasks, a key ability for artificial
systems is to understand physical interactions between objects, and predict
future outcomes of a situation. This ability, often referred to as
intuitive
physics
, has recently received attention and several methods were proposed to
learn these ph... | ["intuitive physics", "objects", "occlusions", "object dynamics", "segmentation masks", "results", "occlusion resistant learning", "videos", "human performance"] | Under review as a conference paper at ICLR 2020Occlusion resistant learning of intuitivephysics from videosAnonymous authorsPaper under double-blind reviewAbstractTo reach human performance on complex tasks, a key ability for artificialsystems is to understand physical interactions between objects, and predictfuture ou... | Bylu2aeMoB | Official Blind Review #3 | 3: Weak Reject | The key contribution of this paper is a model that can predict the dynamics of pre-segmented image patches under multiple frames of occlusion. The input image is processed by a CNN, the dynamics are predicted by a recurrent interaction net, and the output image is generated by a (deconv) CNN.
The key weaknesses I see... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Occlusion resistant learning of intuitive physics from videos
### Paper Abstract
To reach human performance on complex tasks, a key ability for artificial systems is... | |
BJxWx0NYPr | ICLR.cc/2020/Conference | 2020 | Adaptive Structural Fingerprints for Graph Attention Networks | ["Kai Zhang", "Yaokang Zhu", "Jun Wang", "Jie Zhang"] | Graph attention network (GAT) is a promising framework to perform convolution and massage passing on graphs. Yet, how to fully exploit rich structural information in the attention mechanism remains a challenge. In the current version, GAT calculates attention scores mainly using node features and among one-hop neighbor... | ["Graph attention networks", "graph neural networks", "node classification"] | ABSTRACTGraph attention network (GAT) is a promising framework to perform convolutionand massage passing on graphs. Yet, how to fully exploit rich structural informa-tion in the attention mechanism remains a challenge. In the current version, GATcalculates attention scores mainly using node features and among one-hop n... | H1eSNZHrKr | Official Blind Review #4 | 6: Weak Accept | This work suggests a graph structure based methodology to augment the attention mechanism of graph neural networks.
The main idea is to explore the interaction between different types of nodes of the local neighborhood of a root node. One component of this "fingerprinting" is a node weighting (closeness) that is comp... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adaptive Structural Fingerprints for Graph Attention Networks
### Paper Abstract
Graph attention network (GAT) is a promising framework to perform convolution and massag... | |
Sy3nGCYXz | ICLR.cc/2018/Conference | 2018 | THE LOCAL DIMENSION OF DEEP MANIFOLD | ["Mengxiao Zhang", "Wangquan Wu", "Yanren Zhang", "Kun He", "Tao Yu", "Huan Long", "John E. Hopcroft"] | Based on our observation that there exists a dramatic drop for the singular values of the fully connected layers or a single feature map of the convolutional layer, and that the dimension of the concatenated feature vector almost equals the summation of the dimension on each feature map, we propose a singular value dec... | ["activation manifold", "dimension", "deep neural network", "singular value decomposition"] | ABSTRACTBased on our observation that there exists a dramatic drop for the singular valuesof the fully connected layers or a single feature map of the convolutional layer, andthat the dimension of the concatenated feature vector almost equals the summa-tion of the dimension on each feature map, we propose a singular va... | HJyDHIZVf | Unclear motivation, insufficient theory and small evaluation | 3: Clear rejection | This paper analyzed the dimensionality of feature maps and fully connected layers of pre-trained CNN on images within a same category. The local dimensionality of this paper is the SVD dimensionality on the augmented images of the same class images which classified by the neural network as high probability. However, th... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
THE LOCAL DIMENSION OF DEEP MANIFOLD
### Paper Abstract
Based on our observation that there exists a dramatic drop for the singular values of the fully connected layers ... |
KRKGJrbPcKE | ICLR.cc/2021/Conference | 2021 | Distribution Based MIL Pooling Filters are Superior to Point Estimate Based Counterparts | ["Mustafa Umit Oner", "Jared Marc Song", "Hwee Kuan Lee", "Wing-Kin Sung"] | Multiple instance learning (MIL) is a machine learning paradigm which learns the mapping between bags of instances and bag labels. There are different MIL tasks which can be solved by different MIL methods. One common component of all MIL methods is the MIL pooling filter, which obtains bag level representations from e... | ["multiple instance learning", "mil", "mil pooling filters", "distribution pooling", "point estimate based pooling"] | ABSTRACTMultiple instance learning (MIL) is a machine learning paradigm which learns themapping between bags of instances and bag labels. There are different MIL taskswhich can be solved by different MIL methods. One common component of allMIL methods is the MIL pooling filter, which obtains bag level representationsfr... | IPTM2Mz_MX0 | An OK paper that extends two previous works for MIL. | 5: Marginally below acceptance threshold | Methodology:
The paper presents an approach for MIL. Two previous pooling operators (attention and distribution) are combined to capture more information with the attention scores and raw instance-level score. Some theoretical results were provided to support the claim of information gain in the proposed operator. Expe... | 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
Distribution Based MIL Pooling Filters are Superior to Point Estimate Based Counterparts
### Paper Abstract
Multiple instance learning (MIL) is a machine learning paradi... |
S1-6egSFl | ICLR.cc/2017/workshop | 2017 | Unsupervised and Scalable Algorithm for Learning Node Representations | ["Tiago Pimentel", "Adriano Veloso", "Nivio Ziviani"] | Representation learning is one of the foundations of Deep Learning and allowed big improvements on several Machine Learning fields, such as Neural Machine Translation, Question Answering and Speech Recognition. Recent works have proposed new methods for learning representations for nodes and edges in graphs. In this wo... | ["Unsupervised Learning"] | ABSTRACTRepresentation learning is one of the foundations of Deep Learning and allowedbig improvements on several Machine Learning fields, such as Neural MachineTranslation, Question Answering and Speech Recognition. Recent works haveproposed new methods for learning representations for nodes and edges in graphs.In thi... | rywV8Flie | 7: Good paper, accept | Essentially the goal of the contribution is to adapt ideas from Word2Vec to learn node embeddings. I.e., like Node2Vec but borrowing ideas from SkipGrams rather than random walks. This is claimed to lead to faster training times and more general-purpose embeddings.
The basic idea is to form "sentences" based on random... | 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 and Scalable Algorithm for Learning Node Representations
### Paper Abstract
Representation learning is one of the foundations of Deep Learning and allowed b... | |
rkl3-hA5Y7 | ICLR.cc/2019/Conference | 2019 | Towards Decomposed Linguistic Representation with Holographic Reduced Representation | ["Jiaming Luo", "Yuan Cao", "Yonghui Wu"] | The vast majority of neural models in Natural Language Processing adopt a form of structureless distributed representations. While these models are powerful at making predictions, the representational form is rather crude and does not provide insights into linguistic structures. In this paper we introduce novel languag... | ["holographic", "representations", "linguistic representation", "representation", "models", "hrr", "towards", "representation towards", "vast majority", "neural models"] | ABSTRACTThe vast majority of neural models in Natural Language Processing adopt a formof structureless distributed representations. While these models are powerful atmaking predictions, the representational form is rather crude and does not pro-vide insights into linguistic structures. In this paper we introduce novel ... | S1xJZx1i2X | Back to the past | 5: Marginally below acceptance threshold | This paper is very interesting as it seems to bring the clock back to Holographic Reduced Representations (HRRs) and their role in Deep Learning. It is an important paper as it is always important to learn from the past. HRRs have been introduced as a form of representation that is invertible. There are two important a... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Decomposed Linguistic Representation with Holographic Reduced Representation
### Paper Abstract
The vast majority of neural models in Natural Language Processing... |
TDDZxmr6851 | ICLR.cc/2021/Conference | 2021 | The large learning rate phase of deep learning | ["Aitor Lewkowycz", "Yasaman Bahri", "Ethan Dyer", "Jascha Sohl-Dickstein", "Guy Gur-Ari"] | The choice of initial learning rate can have a profound effect on the performance of deep networks. We present empirical evidence that networks exhibit sharply distinct behaviors at small and large learning rates. In the small learning rate phase, training can be understood using the existing theory of infinitely wide ... | ["deep learning", "wide networks", "training dynamics"] | ABSTRACTThe choice of initial learning rate can have a profound effect on the performanceof deep networks. We present empirical evidence that networks exhibit sharplydistinct behaviors at small and large learning rates. In the small learning ratephase, training can be understood using the existing theory of infinitely ... | H5lsQgDLMX | The Large Learning Rate Phase of Deep Learning | 3: Clear rejection | This paper analyzes the effect of choosing a large step-size on the generalization of deep networks. They suggest that starting with a large learning rate, the loss initially increases before converging to a flatter minimum with improved generalization (catapult effect). When the learning rate is above a certain thresh... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The large learning rate phase of deep learning
### Paper Abstract
The choice of initial learning rate can have a profound effect on the performance of deep networks. We ... |
SyedHyBFwS | ICLR.cc/2020/Conference | 2020 | Relative Pixel Prediction For Autoregressive Image Generation | ["Wang Ling", "Chris Dyer", "Lei Yu", "Lingpeng Kong", "Dani Yogatama", "Susannah Young"] | In natural images, transitions between adjacent pixels tend to be smooth and gradual, a fact that has long been exploited in image compression models based on predictive coding. In contrast, existing neural autoregressive image generation models predict the absolute pixel intensities at each position, which is a more c... | ["Image Generation", "Autoregressive"] | ABSTRACTIn natural images, transitions between adjacent pixels tend to be smooth and gradual,a fact that has long been exploited in image compression models based on predictivecoding. In contrast, existing neural autoregressive image generation models predictthe absolute pixel intensities at each position, which is a m... | rJgWhRKyjH | Official Blind Review #4 | 3: Weak Reject | The paper bases its methodology on well known developments in image analysis/synthesis about similarity of pixel values in adjacent locations. Many techniques have been used for modelling this similarity, including predictive models, cliques and graphs. The paper uses a simple autoregressive model for generating pixel ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Relative Pixel Prediction For Autoregressive Image Generation
### Paper Abstract
In natural images, transitions between adjacent pixels tend to be smooth and gradual, a ... | |
LFjnKhTNNQD | ICLR.cc/2021/Conference | 2021 | Prepare for the Worst: Generalizing across Domain Shifts with Adversarial Batch Normalization | ["Manli Shu", "Zuxuan Wu", "Micah Goldblum", "Tom Goldstein"] | Adversarial training is the industry standard for producing models that are robust to small adversarial perturbations. However, machine learning practitioners need models that are robust to other kinds of changes that occur naturally, such as changes in the style or illumination of input images. Such changes in input ... | ["adversarial training", "distributional shifts"] | ABSTRACTAdversarial training is the industry standard for producing models that are robust tosmall adversarial perturbations. However, machine learning practitioners need mod-els that are robust to other kinds of changes that occur naturally, such as changes inthe style or illumination of input images. Such changes in ... | V1VXCJEVdLR | Official Blind Review #3 | 5: Marginally below acceptance threshold | This paper proposes adversarial batch normalization (ABN) to perturb feature statistics. It is to makes CNNs more robust to image style or appearance changes. An ABN layer can plug into the middle of a CNN to finetune a pre-trained model. The CNN layers after the ABN layer can learn more robust representations from 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
Prepare for the Worst: Generalizing across Domain Shifts with Adversarial Batch Normalization
### Paper Abstract
Adversarial training is the industry standard for produc... |
Skluy2RcK7 | ICLR.cc/2019/Conference | 2019 | Selectivity metrics can overestimate the selectivity of units: a case study on AlexNet | ["Ella M. Gale", "Anh Nguyen", "Ryan Blything", "Nicholas Martin and Jeffrey S. Bowers"] | Various methods of measuring unit selectivity have been developed in order to understand the representations learned by neural networks (NNs). Here we undertake a comparison of four such measures on AlexNet, namely, localist selectivity, \precision (Zhou et al, ICLR 2015), class-conditional mean activity selectivity C... | ["AlexNet", "neural networks", "selectivity", "localist", "distributed", "represenataion", "precision", "measures of selectivity", "object detectors", "single directions", "network analysis"] | ABSTRACTVarious methods of measuring unit selectivity have been developed in order tounderstand the representations learned by neural networks (NNs). Here we under-take a comparison of four such measures on AlexNet, namely, localist selectivityBowers et al. (2014), precision (Zhou et al., 2015), class-conditional mean ... | rkgdcTkchX | Poorly written with unclear contributions | 3: Clear rejection | This is a paper with scattered potentially interesting ideas. But the execution is limited and the writing poor with critical details lacking. A major limitation of the paper is that it is not clear what contribution it makes. Some of the analyses are indeed interesting but 1) these analyses are mostly descriptive and... | 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
Selectivity metrics can overestimate the selectivity of units: a case study on AlexNet
### Paper Abstract
Various methods of measuring unit selectivity have been develop... |
a8OpoWkLA1 | ICLR.cc/2021/Conference | 2021 | ACDC: Weight Sharing in Atom-Coefficient Decomposed Convolution | ["Ze Wang", "Xiuyuan Cheng", "Guillermo Sapiro", "Qiang Qiu"] | Convolutional Neural Networks (CNNs) are known to be significantly over-parametrized, and difficult to interpret, train and adapt. In this paper, we introduce a structural regularization across convolutional kernels in a CNN. In our approach, each convolution kernel is first decomposed as 2D dictionary atoms linearly c... | ["parameters", "weight", "cnns", "dictionary atoms", "decomposed coefficients", "models", "network structures", "acdc", "decomposed convolution acdc"] | ABSTRACTConvolutional Neural Networks (CNNs) are known to be significantly over-parametrized, and difficult to interpret, train, and adapt. In this work, we introducea structural regularization across convolutional kernels in a CNN. In our approach,each convolution kernel is first decomposed into 2D dictionary atoms li... | Y9xfV5_sapf | ConvNet compression | 4: Ok but not good enough - rejection | The paper explores a form of model compression that reduces the number of parameters and FLOPs for VGG/ResNet type models.
In isolation, the results look promising but there is a lack of comparison to related work.
Some relevant points of comparison would be:
1) Models using 'hand-designed' group or depthwise convoluti... | 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
ACDC: Weight Sharing in Atom-Coefficient Decomposed Convolution
### Paper Abstract
Convolutional Neural Networks (CNNs) are known to be significantly over-parametrized, ... |
rkx1b64Fvr | ICLR.cc/2020/Conference | 2020 | A New Multi-input Model with the Attention Mechanism for Text Classification | ["Junhao Qiu", "Ronghua Shi", "Fangfang Li (the corresponding author)", "Jinjing Shi", "Wangmin Liao"] | Recently, deep learning has made extraordinary achievements in text classification. However, most of present models, especially convolutional neural network (CNN), do not extract long-range associations, global representations, and hierarchical features well due to their relatively shallow and simple structures. This c... | ["Natural Language Processing", "Text Classification", "Densent", "Multi-input Model", "Attention Mechanism"] | ABSTRACTRecently, deep learning has made extraordinary achievements in text classifica-tion. However, most of present models, especially convolutional neural network(CNN), do not extract long-range associations, global representations, and hier-archical features well due to their relatively shallow and simple structure... | BJg7Gw_0KB | Official Blind Review #1 | 1: Reject | This paper presents a multi-input model for text classification. The presentation of this paper is very poor. The novelty is very limited. The justification of the proposed architecture is not persuasive. The experiment design has many flaws. All the compared baselines are extremely weak. The authors need to foll... | <|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 New Multi-input Model with the Attention Mechanism for Text Classification
### Paper Abstract
Recently, deep learning has made extraordinary achievements in text class... | |
rk4Qso0cKm | ICLR.cc/2019/Conference | 2019 | Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network | ["Xuanqing Liu", "Yao Li", "Chongruo Wu", "Cho-Jui Hsieh"] | We present a new algorithm to train a robust neural network against adversarial attacks.
Our algorithm is motivated by the following two ideas. First, although recent work has demonstrated that fusing randomness can improve the robustness of neural networks (Liu 2017), we noticed that adding noise blindly to all the l... | ["randomness", "adversarial defense", "adversarial attacks", "algorithm", "liu", "bnn", "new algorithm", "robust neural network", "following"] | ABSTRACTWe present a new algorithm to train a robust neural network against adversarialattacks. Our algorithm is motivated by the following two ideas. First, althoughrecent work has demonstrated that fusing randomness can improve the robustnessof neural networks (Liu et al., 2017), we noticed that adding noise blindly ... | Byxndvbkh7 | A nice paper that bridges adversarial training and Bayesian neural nets | 7: Good paper, accept | The paper extends the PGD adversarial training method (Madry et al., 2017) to Bayesian Neural Nets (BNNs).
The proposed method defines a generative process that ties the prediction output and the adversarial input
pattern via a set of shared neural net weights. These weights are then assinged a prior and
the resulta... | 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
Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network
### Paper Abstract
We present a new algorithm to train a robust neural network against adver... |
B1lsXREYvr | ICLR.cc/2020/Conference | 2020 | One-Shot Neural Architecture Search via Compressive Sensing | ["Minsu Cho", "Mohammadreza Soltani", "Chinmay Hegde"] | Neural architecture search (NAS), or automated design of neural network models, remains a very challenging meta-learning problem. Several recent works (called "one-shot" approaches) have focused on dramatically reducing NAS running time by leveraging proxy models that still provide architectures with competitive perfor... | ["deep learning", "autoML", "neural architecture search", "image classification", "language modeling"] | ABSTRACTNeural architecture search (NAS), or automated design of neural network models,remains a very challenging meta-learning problem. Several recent works (called“one-shot” approaches) have focused on dramatically reducing NAS running timeby leveraging proxy models that still provide architectures with competitive p... | HJxy8jjAcH | Official Blind Review #4 | 3: Weak Reject | This paper proposes a new algorithm for one-shot neural architecture search (NAS) via compressive sensing. The authors propose a new search strategy, as well as a slightly different search space compared to DARTS [1], ProxylessNAS [2], etc. They use architecture samples from the one-shot model evaluated with the search... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
One-Shot Neural Architecture Search via Compressive Sensing
### Paper Abstract
Neural architecture search (NAS), or automated design of neural network models, remains a ... | |
MDsQkFP1Aw | ICLR.cc/2021/Conference | 2021 | Into the Wild with AudioScope: Unsupervised Audio-Visual Separation of On-Screen Sounds | ["Efthymios Tzinis", "Scott Wisdom", "Aren Jansen", "Shawn Hershey", "Tal Remez", "Dan Ellis", "John R. Hershey"] | Recent progress in deep learning has enabled many advances in sound separation and visual scene understanding. However, extracting sound sources which are apparent in natural videos remains an open problem. In this work, we present AudioScope, a novel audio-visual sound separation framework that can be trained without ... | ["Audio-visual sound separation", "in-the-wild data", "unsupervised learning", "self-supervised learning", "universal sound separation"] | ABSTRACTRecent progress in deep learning has enabled many advances in sound separationand visual scene understanding. However, extracting sound sources which areapparent in natural videos remains an open problem. In this work, we presentAudioScope , a novel audio-visual sound separation framework that can be trainedwit... | SGqi9o6sY4P | Official Blind Review #1 | 6: Marginally above acceptance threshold | This paper proposed an unsupervised method for open-domain, audio-visual separation system. The proposed model was optimized using the newly suggested mixture invariant training (MixIT) together with a cross entropy loss function. The authors suggest to separately process the audio and video, and next align them with a... | 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
Into the Wild with AudioScope: Unsupervised Audio-Visual Separation of On-Screen Sounds
### Paper Abstract
Recent progress in deep learning has enabled many advances in ... |
rkRwGg-0Z | ICLR.cc/2018/Conference | 2018 | Beyond Word Importance: Contextual Decomposition to Extract Interactions from LSTMs | ["W. James Murdoch", "Peter J. Liu", "Bin Yu"] | The driving force behind the recent success of LSTMs has been their ability to learn complex and non-linear relationships. Consequently, our inability to describe these relationships has led to LSTMs being characterized as black boxes. To this end, we introduce contextual decomposition (CD), an interpretation algorithm... | ["interpretability", "LSTM", "natural language processing", "sentiment analysis", "interactions"] | ABSTRACTThe driving force behind the recent success of LSTMs has been their ability tolearn complex and non-linear relationships. Consequently, our inability to de-scribe these relationships has led to LSTMs being characterized as black boxes.To this end, we introduce contextual decomposition (CD), an interpretation al... | r1k_ETYlM | An interesting fine-grained model which lacks of a proper comparison with the state of the art | 7: Good paper, accept | This article aims at understanding the role played by the different words in a sentence, taking into account their order in the sentence. In sentiment analysis for instance, this capacity is critical to model properly negation.
As state-of-the-art approaches rely on LSTM, the authors want to understand which informatio... | 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 Word Importance: Contextual Decomposition to Extract Interactions from LSTMs
### Paper Abstract
The driving force behind the recent success of LSTMs has been the... |
BylalAEtvB | ICLR.cc/2020/Conference | 2020 | Lipschitz Lifelong Reinforcement Learning | ["Erwan Lecarpentier", "David Abel", "Kavosh Asadi", "Yuu Jinnai", "Emmanuel Rachelson", "Michael L. Littman"] | We consider the problem of reusing prior experience when an agent is facing a series of Reinforcement Learning (RL) tasks. We introduce a novel metric between Markov Decision Processes and focus on the study and exploitation of the optimal value function's Lipschitz continuity in the task space with respect to that met... | ["Reinforcement Learning", "Lifelong Learning"] | ABSTRACTWe consider the problem of knowledge transfer when an agent is facing a seriesof Reinforcement Learning (RL) tasks. We introduce a novel metric betweenMarkov Decision Processes and establish that close MDPs have close optimal valuefunctions, that is that optimal value functions are Lipschitz continuous with res... | BJx61rA3KH | Official Blind Review #3 | 6: Weak Accept | Updated review: Thanks to the authors for response. I think it clarifies a few points for me except one, which is that I still dont understand how the ATARI games share the same state space? Under what model is this true? And if the paper really wants to use this as an example, why are there no results on even a single... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Lipschitz Lifelong Reinforcement Learning
### Paper Abstract
We consider the problem of reusing prior experience when an agent is facing a series of Reinforcement Learni... | |
HJIoJWZCZ | ICLR.cc/2018/Conference | 2018 | Adversarial Dropout Regularization | ["Kuniaki Saito", "Yoshitaka Ushiku", "Tatsuya Harada", "Kate Saenko"] | We present a domain adaptation method for transferring neural representations from label-rich source domains to unlabeled target domains. Recent adversarial methods proposed for this task learn to align features across domains by ``fooling'' a special domain classifier network. However, a drawback of this approach is t... | ["domain adaptation", "computer vision", "generative models"] | ABSTRACTWe present a domain adaptation method for transferring neural representationsfrom label-rich source domains to unlabeled target domains. Recent adversarialmethods proposed for this task learn to align features across domains by “fool-ing” a special domain classifier network. However, a drawback of this approach... | rJO3y_qgz | An interesting method for domain adaptation | 7: Good paper, accept |
Unsupervised Domain adaptation is the problem of training a classifier without labels in some target domain if we have labeled data from a (hopefully) similar dataset with labels. For example, training a classifier using simulated rendered images with labels, to work on real images.
Learning discriminative features f... | 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
Adversarial Dropout Regularization
### Paper Abstract
We present a domain adaptation method for transferring neural representations from label-rich source domains to unl... |
HygTE309t7 | ICLR.cc/2019/Conference | 2019 | Outlier Detection from Image Data | ["Lei Cao", "Yizhou Yan", "Samuel Madden", "Elke Rundensteiner"] | Modern applications from Autonomous Vehicles to Video Surveillance generate massive amounts of image data. In this work we propose a novel image outlier detection approach (IOD for short) that leverages the cutting-edge image classifier to discover outliers without using any labeled outlier. We observe that although in... | ["Image outlier", "CNN", "Deep Neural Forest"] | Under review as a conference paper at ICLR 2019Outlier Detection from Image DataAnonymous authorsPaper under double-blind reviewAbstractModern applications from Autonomous Vehicles to Video Surveillance gener-ate massive amounts of image data. In this work we propose a novel imageoutlier detection approach (IOD for sho... | HylM6a_W6Q | Review of submission 1492 | 4: Ok but not good enough - rejection | Summary: This paper modifies an existing technique designed for image classification to make it applicable to outlier detection.
Strengths: The outlined problem is of significant practical importance.
Weaknesses:
- The improvement over the existing method is incremental;
- The regularization on routing decision m... | 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
Outlier Detection from Image Data
### Paper Abstract
Modern applications from Autonomous Vehicles to Video Surveillance generate massive amounts of image data. In this w... |
Skg9aAEKwH | ICLR.cc/2020/Conference | 2020 | Visual Hide and Seek | ["Boyuan Chen", "Shuran Song", "Hod Lipson", "Carl Vondrick"] | We train embodied agents to play Visual Hide and Seek where a prey must navigate in a simulated environment in order to avoid capture from a predator. We place a variety of obstacles in the environment for the prey to hide behind, and we only give the agents partial observations of their environment using an egocentric... | ["Embodied Learning", "Self-supervised Learning", "Reinforcement Learning"] | ABSTRACTWe train embodied agents to play Visual Hide and Seek where a prey must nav-igate in a simulated environment in order to avoid capture from a predator. Weplace a variety of obstacles in the environment for the prey to hide behind, andwe only give the agents partial observations of their environment using an ego... | HJezg2rAuB | Official Blind Review #1 | 3: Weak Reject |
Paper Summary: The paper studies what an embodied agent that is trained using RL learns in the context of the hide and seek game. There are two agents in a simple environment (one hider and one seeker). The strategy for the seeker is fixed, but the hider learns a policy based on ego-centric visual input. The paper att... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Visual Hide and Seek
### Paper Abstract
We train embodied agents to play Visual Hide and Seek where a prey must navigate in a simulated environment in order to avoid cap... | |
VZs3gFWv7Y | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | Lean-Interaction: passive image manipulation in concurrent multitasking | ["Danny Schott", "Benjamin Hatscher", "Fabian Joeres", "Mareike Gabele", "Steffi Hu\u00dflein", "Christian Hansen"] | Complex bi-manual tasks often benefit from supporting visual information and guidance. Controlling the system that provides this information is a secondary task that forces the user to perform concurrent multitasking, which in turn may affect the main task performance. Interactions based on natural behavior are a promi... | ["Multimodal Interaction", "Multitasking", "Hands-free Interaction", "Radiology", "Medical Domain"] | ABSTRACTComplex bi-manual tasks often benefit from supporting visual in-formation and guidance. Controlling the system that provides thisinformation is a secondary task that forces the user to performconcurrent multitasking, which in turn may affect the main taskperformance. Interactions based on natural behavior are a... | nG4MRx2Cqo | Well structured and easy to read. However, the necessity of task abstraction and usefulness of the evaluation outcomes are unclear. | 5: Marginally below acceptance threshold | The paper proposes to design natural interaction for secondary tasks where primary task requires bimanual interaction. The goal is to make sure that the task performance of a complex primary task is not impacted when simultaneously executing the associated secondary task. The paper selected interventional radiology to ... | 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
Lean-Interaction: passive image manipulation in concurrent multitasking
### Paper Abstract
Complex bi-manual tasks often benefit from supporting visual information and g... |
SkGuG2R5tm | ICLR.cc/2019/Conference | 2019 | Spreading vectors for similarity search | ["Alexandre Sablayrolles", "Matthijs Douze", "Cordelia Schmid", "Herv\u00e9 J\u00e9gou"] | Discretizing floating-point vectors is a fundamental step of modern indexing methods. State-of-the-art techniques learn parameters of the quantizers on training data for optimal performance, thus adapting quantizers to the data. In this work, we propose to reverse this paradigm and adapt the data to the quantizer: we t... | ["dimensionality reduction", "similarity search", "indexing", "differential entropy"] | ABSTRACTDiscretizing multi-dimensional data distributions is a fundamental step of modernindexing methods. State-of-the-art techniques learn parameters of quantizers ontraining data for optimal performance, thus adapting quantizers to the data. In thiswork, we propose to reverse this paradigm and adapt the data to the ... | B1x0Ka5M2X | Well motivated novel idea; excellent results | 7: Good paper, accept | Pros
----
[Originality]
The authors propose a novel idea of learning representations that improves the performance of the subsequent fixed discretization method.
[Clarity]
The authors clearly motivate their solution and explain the different ideas and enhancements introduced. The manuscript is fairly easy to follow. ... | 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
Spreading vectors for similarity search
### Paper Abstract
Discretizing floating-point vectors is a fundamental step of modern indexing methods. State-of-the-art techniq... |
rSwTMomgCz | ICLR.cc/2021/Conference | 2021 | Decoupling Exploration and Exploitation for Meta-Reinforcement Learning without Sacrifices | ["Evan Zheran Liu", "Aditi Raghunathan", "Percy Liang", "Chelsea Finn"] | The goal of meta-reinforcement learning (meta-RL) is to build agents that can quickly learn new tasks by leveraging prior experience on related tasks. Learning a new task often requires both exploring to gather task-relevant information and exploiting this information to solve the task. In principle, optimal exploratio... | ["meta-reinforcement learning", "reinforcement learning", "exploration"] | ABSTRACTThe goal of meta-reinforcement learning (meta-RL) is to build agents that canquickly learn new tasks by leveraging prior experience on related tasks. Learninga new task often requires both exploring to gather task-relevant information andexploiting this information to solve the task. In principle, optimal explo... | 2ViZxEWPsw9 | Review | 6: Marginally above acceptance threshold | The paper investigates the exploration-exploitation problem in meta-learning. The authors explain the problem of coupled exploration and validate it through a toy example. To overcome this issue, the paper introduces DREAM, a meta-algorithm decoupling exploration and exploitation. In the first step, DREAM learns an exp... | 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
Decoupling Exploration and Exploitation for Meta-Reinforcement Learning without Sacrifices
### Paper Abstract
The goal of meta-reinforcement learning (meta-RL) is to bui... |
ryza73R9tQ | ICLR.cc/2019/Conference | 2019 | Machine Translation With Weakly Paired Bilingual Documents | ["Lijun Wu", "Jinhua Zhu", "Di He", "Fei Gao", "Xu Tan", "Tao Qin", "Tie-Yan Liu"] | Neural machine translation, which achieves near human-level performance in some languages, strongly relies on the availability of large amounts of parallel sentences, which hinders its applicability to low-resource language pairs. Recent works explore the possibility of unsupervised machine translation with monolingual... | ["Natural Language Processing", "Machine Translation", "Unsupervised Learning"] | ABSTRACTNeural machine translation, which achieves near human-level performance insome languages, strongly relies on the availability of large amounts of parallelsentences, which hinders its applicability to low-resource language pairs. Recentworks explore the possibility of unsupervised machine translation with monoli... | r1lZazqO3m | nice contribution | 7: Good paper, accept | Summary
The authors propose a relatively simple approach to mine noisy parallel sentences which are useful to greatly improve performance of purely unsupervised MT algorithms.
The method consists of a) mining documents that refer to the same topic, b) extracting from these documents parallel sentences, c) training 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
Machine Translation With Weakly Paired Bilingual Documents
### Paper Abstract
Neural machine translation, which achieves near human-level performance in some languages, ... |
r1gzaCEtPS | ICLR.cc/2020/Conference | 2020 | Posterior Control of Blackbox Generation | ["Xiang Lisa Li", "Alexander M. Rush"] | Many tasks in natural language processing and related domains require high precision output that obeys dataset-specific constraints. This level of fine-grained control can be difficult to obtain in large-scale neural network models. In this work, we propose a structured latent-variable approach that adds discrete contr... | ["nlp", "text generation", "deep latent variable models", "variational autoencoders"] | ABSTRACTMany tasks in natural language processing and related domains require high pre-cision output that obeys dataset-specific constraints. This level of fine-grainedcontrol can be difficult to obtain in large-scale neural network models. In this work,we propose a structured latent-variable approach that adds discret... | Bkx9vS125H | Official Blind Review #1 | 6: Weak Accept | The paper presents a method that modifies an output of an arbitrary neural decoder to follow weak supervision. This is done through posterior regularization within an amortized structured variational framework. The method is evaluated on two problems: text generation and unsupervised tagging.
The paper is well written... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Posterior Control of Blackbox Generation
### Paper Abstract
Many tasks in natural language processing and related domains require high precision output that obeys datas... | |
rkMW1hRqKX | ICLR.cc/2019/Conference | 2019 | Optimal Completion Distillation for Sequence Learning | ["Sara Sabour", "William Chan", "Mohammad Norouzi"] | We present Optimal Completion Distillation (OCD), a training procedure for optimizing sequence to sequence models based on edit distance. OCD is efficient, has no hyper-parameters of its own, and does not require pre-training or joint optimization with conditional log-likelihood. Given a partial sequence generated by t... | ["Sequence Learning", "Edit Distance", "Speech Recognition", "Deep Reinforcement Learning"] | ABSTRACTWe present Optimal Completion Distillation (OCD), a training procedure foroptimizing sequence to sequence models based on edit distance. OCD is efficient,has no hyper-parameters of its own, and does not require pretraining or jointoptimization with conditional log-likelihood. Given a partial sequence generatedb... | rJe4U-Bcnm | Exciting approach | 7: Good paper, accept | The authors propose an alternative approach to training seq2seq models, which addresses concerns about exposure bias and about the typical MLE objective being different from the final evaluation metric. In particular, the authors propose to use a dynamic program to compute the optimal continuations of predicted prefixe... | 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
Optimal Completion Distillation for Sequence Learning
### Paper Abstract
We present Optimal Completion Distillation (OCD), a training procedure for optimizing sequence t... |
SJefGpEtDB | ICLR.cc/2020/Conference | 2020 | A Dynamic Approach to Accelerate Deep Learning Training | ["John Osorio", "Adri\u00e0 Armejach", "Eric Petit", "Marc Casas"] | Mixed-precision arithmetic combining both single- and half-precision operands in the same operation have been successfully applied to train deep neural networks. Despite the advantages of mixed-precision arithmetic in terms of reducing the need for key resources like memory bandwidth or register file size, it has a lim... | ["reduced precision", "bfloat16", "CNN", "DNN", "dynamic precision", "mixed precision"] | ABSTRACTMixed-precision arithmetic combining both single- and half-precision operandson the same operation has been successfully applied to train deep neural net-works. Despite the advantages of mixed-precision arithmetic in terms of reducingthe need for key resources like memory bandwidth or register file size, it has... | r1g83xPntS | Official Blind Review #2 | 3: Weak Reject | The author(s) propose to accelerate the training of deep neural networks while also maintain the performance of the trained model by switching between fully half-precision computation and mixed-precision computation. Compared to the commonly-used mixed-precision training strategy, the proposed method can accelerate the... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Dynamic Approach to Accelerate Deep Learning Training
### Paper Abstract
Mixed-precision arithmetic combining both single- and half-precision operands in the same oper... | |
JyDnXkeJpjU | ICLR.cc/2021/Conference | 2021 | Task-similarity Aware Meta-learning through Nonparametric Kernel Regression | ["Arun Venkitaraman", "Anders Hansson", "Bo Wahlberg"] | This paper investigates the use of nonparametric kernel-regression to obtain a task- similarity aware meta-learning algorithm. Our hypothesis is that the use of task- similarity helps meta-learning when the available tasks are limited and may contain outlier/ dissimilar tasks. While existing meta-learning approaches im... | ["Task-similarity", "Meta-learning", "Kernel regression", "Nonparametric regression", "Task-descriptors"] | ABSTRACTThis paper investigates the use of nonparametric kernel-regression to obtain a task-similarity aware meta-learning algorithm. Our hypothesis is that the use of task-similarity helps meta-learning when the available tasks are limited and may con-tain outlier/ dissimilar tasks. While existing meta-learning approa... | D_I0_i7WbcD | Overall a sound algorithm, but framed/motivated in a strange way and hardly supported by relevant/realistic experiments | 4: Ok but not good enough - rejection | === Summary ===
The paper proposes a meta-learning method based on a notion of task similarity/dissimilarity. In particular, the paper motivates its proposed method TANML through a generalization of Meta-SGD wherein the learnable parameter-wise learning rate in the inner update of Meta-SGD is replaced by a quadratic p... | 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
Task-similarity Aware Meta-learning through Nonparametric Kernel Regression
### Paper Abstract
This paper investigates the use of nonparametric kernel-regression to obta... |
Hynn8SHOx | ICLR.cc/2017/workshop | 2017 | DL-gleaning: An approach for Improving inference speed and accuracy | ["HyunYong Lee and Byung-Tak Lee"] | Improving inference speed and accuracy is one of the main objectives of current deep learning-related research. In this paper, we introduce our approach using middle output layer for this purpose. From the feasibility study using Inception-v4, we found that our approach has potential to reduce the average inference tim... | ["Deep learning", "Supervised Learning"] | ABSTRACTImproving inference speed and accuracy is one of the main objectives of currentdeep learning-related research. In this paper, we introduce our approach usingmiddle output layer for this purpose. From the feasibility study using Inception-v4, we found that our approach has potential to reduce the average inferen... | B1Ymc5lol | No promising results | 3: Clear rejection | This work tries to speed up inference by taking a shortcut from the network. The paper shows that an accuracy increase is possible with such a shortcut. However, the way the accuracy is calculated for the proposed model is very unrealistic. The accuracy is calculated by only accepting the correct answers from the mid-l... | 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
DL-gleaning: An approach for Improving inference speed and accuracy
### Paper Abstract
Improving inference speed and accuracy is one of the main objectives of current de... |
efaE7iJ2GJv | robot-learning.org/CoRL/2023/Conference | 2023 | PolarNet: 3D Point Clouds for Language-Guided Robotic Manipulation | ["Shizhe Chen", "Ricardo Garcia Pinel", "Cordelia Schmid", "Ivan Laptev"] | The ability for robots to comprehend and execute manipulation tasks based on natural language instructions is a long-term goal in robotics. The dominant approaches for language-guided manipulation use 2D image representations, which face difficulties in combining multi-view cameras and inferring precise 3D positions an... | ["Robotic manipulation", "3D point clouds", "language-guided policy"] | PolarNet: 3D Point Clouds for Language-GuidedRobotic ManipulationShizhe Chen∗, Ricardo Garcia∗, Cordelia Schmid, Ivan LaptevInria, ́Ecole normale sup ́erieure, CNRS, PSL Research Universityhttps://www.di.ens.fr/willow/research/polarnet/Abstract: The ability for robots to comprehend and execute manipulation tasksbased o... | iYhfcs7O-P | Strengths:
- The authors propose a new method that addresses the weaknesses of both 2D-image-based approaches (lack of precision) and 3D-based approaches (quantization, computation inefficiency)
- Ablations on the system design regarding the design choices are interesting
Weaknesses:
- No evaluation on real hardware. ... | 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
PolarNet: 3D Point Clouds for Language-Guided Robotic Manipulation
### Paper Abstract
The ability for robots to comprehend and execute manipulation tasks based on natura... | ||
BJgEd6NYPH | ICLR.cc/2020/Conference | 2020 | Ellipsoidal Trust Region Methods for Neural Network Training | ["Leonard Adolphs", "Jonas Kohler", "Aurelien Lucchi"] | We investigate the use of ellipsoidal trust region constraints for second-order optimization of neural networks. This approach can be seen as a higher-order counterpart of adaptive gradient methods, which we here show to be interpretable as first-order trust region methods with ellipsoidal constraints. In particular, w... | ["non-convex", "optimization", "neural networks", "trust-region"] | ABSTRACTWe investigate the use of ellipsoidal trust region constraints for second-orderoptimization of neural networks. This approach can be seen as a higher-ordercounterpart of adaptive gradient methods, which we here show to be interpretable asfirst-order trust region methods with ellipsoidal constraints. In particul... | H1e970N2Fr | Official Blind Review #3 | 3: Weak Reject | This paper proposes ellipsoidal trust region methods for optimization on neural networks. This approach is motivated by the adaptive gradient methods and classical trust region methods. The idea of the design is reasonable, but the theoretical and empirical results are not strong. I can not support acceptance for curre... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Ellipsoidal Trust Region Methods for Neural Network Training
### Paper Abstract
We investigate the use of ellipsoidal trust region constraints for second-order optimizat... | |
BkisuzWRW | ICLR.cc/2018/Conference | 2018 | Zero-Shot Visual Imitation | ["Deepak Pathak", "Parsa Mahmoudieh", "Guanghao Luo", "Pulkit Agrawal", "Dian Chen", "Yide Shentu", "Evan Shelhamer", "Jitendra Malik", "Alexei A. Efros", "Trevor Darrell"] | The current dominant paradigm for imitation learning relies on strong supervision of expert actions to learn both 'what' and 'how' to imitate. We pursue an alternative paradigm wherein an agent first explores the world without any expert supervision and then distills its experience into a goal-conditioned skill policy ... | ["imitation", "zero-shot", "self-supervised", "robotics", "skills", "navigation", "manipulation", "vizdoom", "reinforcement"] | ABSTRACTThe current dominant paradigm for imitation learning relies on strong supervisionof expert actions to learn both what andhow to imitate. We pursue an alternativeparadigm wherein an agent first explores the world without any expert supervisionand then distills its experience into a goal-conditioned skill policy ... | BJdG429gz | Why is this not simply doing RL in the real world and then using the learnt policy (with no exploration)? | 7: Good paper, accept | One of the main problems with imitation learning in general is the expense of expert demonstration. The authors here propose a method for sidestepping this issue by using the random exploration of an agent to learn generalizable skills which can then be applied without any specific pretraining on any new task.
The pr... | 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
Zero-Shot Visual Imitation
### Paper Abstract
The current dominant paradigm for imitation learning relies on strong supervision of expert actions to learn both 'what' an... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.