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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
HJlMkTNYvH | ICLR.cc/2020/Conference | 2020 | MODiR: Multi-Objective Dimensionality Reduction for Joint Data Visualisation | ["Tim Repke", "Ralf Krestel"] | Many large text collections exhibit graph structures, either inherent to the content itself or encoded in the metadata of the individual documents.
Example graphs extracted from document collections are co-author networks, citation networks, or named-entity-cooccurrence networks.
Furthermore, social networks can be ext... | ["dimensionality reduction", "visualisation", "text visualisation", "network drawing"] | ABSTRACTMany large text collections exhibit graph structures, either inherent to the contentitself or encoded in the metadata of the individual documents. Example graphsextracted from document collections are co-author networks, citation networks,or named-entity-cooccurrence networks. Furthermore, social networks can b... | SJgoG_OpYS | Official Blind Review #3 | 3: Weak Reject | [Contribution summary]
Authors propose MODiR, a dimensionality reduction approach that utilizes both semantic and structural features. Authors report quantitative (metric: precision of k-means++ clustering & AtEdge-length) and qualitative analysis (visualization).
[Comments]
Experimental analysis could improve e.g. 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
MODiR: Multi-Objective Dimensionality Reduction for Joint Data Visualisation
### Paper Abstract
Many large text collections exhibit graph structures, either inherent to ... | |
rJeidA4KvS | ICLR.cc/2020/Conference | 2020 | Role-Wise Data Augmentation for Knowledge Distillation | ["Jie Fu", "Xue Geng", "Bohan Zhuang", "Xingdi Yuan", "Adam Trischler", "Jie Lin", "Vijay Chandrasekhar", "Chris Pal"] | Knowledge Distillation (KD) is a common method for transferring the ``knowledge'' learned by one machine learning model (the teacher) into another model (the student), where typically, the teacher has a greater capacity (e.g., more parameters or higher bit-widths). To our knowledge, existing methods overlook the fact t... | ["Data Augmentation", "Knowledge Distillation"] | ABSTRACTKnowledge Distillation (KD) is a common method for transferring the “knowl-edge” learned by one machine learning model (the teacher ) into another model(thestudent ), where typically, the teacher has a greater capacity (e.g., more param-eters or higher bit-widths). To our knowledge, existing methods overlook th... | BklZhUMtKH | Official Blind Review #3 | 3: Weak Reject | This paper takes the idea of Population-Based Augmentation (PBA) and extends it to knowledge distillation (KD). The idea is that the ideal augmentation protocol for training-from-scratch (or in this context teacher training) may not be the best for student networks under a KD loss.
I am borderline about this paper and... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Role-Wise Data Augmentation for Knowledge Distillation
### Paper Abstract
Knowledge Distillation (KD) is a common method for transferring the ``knowledge'' learned by on... | |
rklEj2EFvB | ICLR.cc/2020/Conference | 2020 | Estimating Gradients for Discrete Random Variables by Sampling without Replacement | ["Wouter Kool", "Herke van Hoof", "Max Welling"] | We derive an unbiased estimator for expectations over discrete random variables based on sampling without replacement, which reduces variance as it avoids duplicate samples. We show that our estimator can be derived as the Rao-Blackwellization of three different estimators. Combining our estimator with REINFORCE, we ob... | ["gradient", "estimator", "discrete", "categorical", "sampling", "without replacement", "reinforce", "baseline", "variance", "gumbel", "vae", "structured prediction"] | ABSTRACTWe derive an unbiased estimator for expectations over discrete random variablesbased on sampling without replacement , which reduces variance as it avoidsduplicate samples. We show that our estimator can be derived as the Rao-Blackwellization of three different estimators. Combining our estimator with RE-INFORC... | rylJnHURKr | Official Blind Review #1 | 6: Weak Accept | Summary: This paper introduces an gradient estimator for loss functions that are expectations over discrete random variables. The basic idea is that an estimator over a discrete distribution can be Rao-Blackwellized by conditioning on the event that the discrete realization was produced by being the first sample drawn ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Estimating Gradients for Discrete Random Variables by Sampling without Replacement
### Paper Abstract
We derive an unbiased estimator for expectations over discrete rand... | |
Sye57xStvB | ICLR.cc/2020/Conference | 2020 | Never Give Up: Learning Directed Exploration Strategies | ["Adri\u00e0 Puigdom\u00e8nech Badia", "Pablo Sprechmann", "Alex Vitvitskyi", "Daniel Guo", "Bilal Piot", "Steven Kapturowski", "Olivier Tieleman", "Martin Arjovsky", "Alexander Pritzel", "Andrew Bolt", "Charles Blundell"] | We propose a reinforcement learning agent to solve hard exploration games by learning a range of directed exploratory policies. We construct an episodic memory-based intrinsic reward using k-nearest neighbors over the agent's recent experience to train the directed exploratory policies, thereby encouraging the agent to... | ["deep reinforcement learning", "exploration", "intrinsic motivation"] | ABSTRACTWe propose a reinforcement learning agent to solve hard exploration games bylearning a range of directed exploratory policies. We construct an episodic memory-based intrinsic reward using k-nearest neighbors over the agent’s recent experienceto train the directed exploratory policies, thereby encouraging the ag... | SJlUtgf2Yr | Official Blind Review #3 | 8: Accept | The paper proposes a novel intrinsic reward/curiosity metric that combines both episodic and “life-long” novelty. Essentially two competing pressures that push agents to explore as many novel states in a single rollout as possible and to explore as many states as possible as evenly as possible. The primary contribution... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Never Give Up: Learning Directed Exploration Strategies
### Paper Abstract
We propose a reinforcement learning agent to solve hard exploration games by learning a range ... | |
R4aWTjmrEKM | ICLR.cc/2021/Conference | 2021 | Iterative Empirical Game Solving via Single Policy Best Response | ["Max Smith", "Thomas Anthony", "Michael Wellman"] | Policy-Space Response Oracles (PSRO) is a general algorithmic framework for learning policies in multiagent systems by interleaving empirical game analysis with deep reinforcement learning (DRL).
At each iteration, DRL is invoked to train a best response to a mixture of opponent policies.
The repeated application of DR... | ["Empirical Game Theory", "Reinforcement Learning", "Multiagent Learning"] | ABSTRACTPolicy-Space Response Oracles (PSRO) is a general algorithmic framework for learningpolicies in multiagent systems by interleaving empirical game analysis with deepreinforcement learning (Deep RL). At each iteration, Deep RL is invoked to traina best response to a mixture of opponent policies. The repeated appl... | ICQ1OuijJGk | Good ideas, but do not seem significant enough | 7: Good paper, accept | The paper suggests two techniques to improve the calculation of empirically figuring out a Nash equilibrium using an iterative application of best-response dynamics. One method learns the best-response to the previously used strategy. The other uses that technique to model the opponent, and then best-responds to the mo... | 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
Iterative Empirical Game Solving via Single Policy Best Response
### Paper Abstract
Policy-Space Response Oracles (PSRO) is a general algorithmic framework for learning ... |
Hkg8bDqee | ICLR.cc/2017/conference | 2017 | Introspection:Accelerating Neural Network Training By Learning Weight Evolution | ["Abhishek Sinha", "Aahitagni Mukherjee", "Mausoom Sarkar", "Balaji Krishnamurthy"] | Neural Networks are function approximators that have achieved state-of-the-art accuracy in numerous machine learning tasks. In spite of their great success in terms of accuracy, their large training time makes it difficult to use them for various tasks. In this paper, we explore the idea of learning weight evolution pa... | ["Computer vision", "Deep learning", "Optimization"] | ABSTRACTNeural Networks are function approximators that have achieved state-of-the-artaccuracy in numerous machine learning tasks. In spite of their great successin terms of accuracy, their large training time makes it difficult to use them forvarious tasks. In this paper, we explore the idea of learning weight evoluti... | B1F8hRVEg | novel idea but requires more details / experimentation | 8: Top 50% of accepted papers, clear accept | The paper reads well and the idea is new.
Sadly, many details needed for replicating the results (such as layer sizes of the CNNs, learning rates) are missing.
The training of the introspection network could have been described in more detail.
Also, I think that a model, which is closer to the current state-of-the-ar... | 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
Introspection:Accelerating Neural Network Training By Learning Weight Evolution
### Paper Abstract
Neural Networks are function approximators that have achieved state-of... |
Ahtk5SoimmG | NeurIPS.cc/2020/Workshop/SVRHM | 2020 | Learning Translation Invariance in CNNs | ["Valerio Biscione", "Jeffrey Bowers"] | When seeing a new object, humans can immediately recognize it across different retinal locations: we say that the internal object representation is invariant to translation. It is commonly believed that Convolutional Neural Networks (CNNs) are architecturally invariant to translation thanks to the convolution and/or po... | ["Translation Invariance", "Convolution", "Internal Representations", "Equivariance"] | Learning Translation Invariance in CNNsValerio Biscione & Jeffrey BowersDepartment of PsychologyUniversity of BristolBristol, BS81TU, UK{valerio.biscione,j.bowers}@bristol.ac.ukAbstractWhen seeing a new object, humans can immediately recognize it across differentretinal locations: we say that the internal object repres... | 3sLarvWohZZ | Analyses need more development | 5: Marginally below acceptance threshold | CNNs are not fully invariant to translation, as is often assumed. Recent work has found that pretraining on ImageNet allows the network to learn translation invariant representations on a transfer task. In this paper, the authors ask: _What is it about pretraining on ImageNet that endows CNNs with translation invarianc... | 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 Translation Invariance in CNNs
### Paper Abstract
When seeing a new object, humans can immediately recognize it across different retinal locations: we say that ... |
rJD6Xgnoz | MIDL.amsterdam/2018/Conference | 2018 | Recurrent Inference Machines for Accelerated MRI Reconstruction | ["Kai L\u00f8nning", "Patrick Putzky", "Matthan W. A. Caan", "Max Welling"] | Accelerated MRI reconstruction is important for making MRI faster and thus applicable in a broader range of problem domains. Computational tools allow for high-resolution imaging without the need to perform time-consuming measurements. Most recently, deep learning approaches have been applied to this problem. However,... | ["RNNs", "MRI", "Inverse Problem", "Image Reconstruction", "Iterative Methods"] | Recurrent Inference Machinesfor Accelerated MRI ReconstructionKai LønningInformatics InstituteUniversity of Amsterdamkai.lonning@gmail.comPatrick PutzkyAMLAB,University of Amsterdampatrick.putzky@gmail.comMatthan CaanAcademic Medical Center,Spinoza Centre for Neuroimagingm.w.a.caan@amc.nlMax WellingAMLAB, University of... | r1zGUlLaG | Recommended | 4: Top 50% of accepted papers, clear accept | This work clearly demonstrates that their novel method, RIM, is outperforming in MRI reconstruction compared to conventional methods. Overall, the manuscript is well-written and the experiments are well-organized. Highly recommended.
Minor comments:
1) it would be great if it includes some sort of discussion from a c... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Recurrent Inference Machines for Accelerated MRI Reconstruction
### Paper Abstract
Accelerated MRI reconstruction is important for making MRI faster and thus applicable ... |
bjkX6Kzb5H | ICLR.cc/2021/Conference | 2021 | Cut out the annotator, keep the cutout: better segmentation with weak supervision | ["Sarah Hooper", "Michael Wornow", "Ying Hang Seah", "Peter Kellman", "Hui Xue", "Frederic Sala", "Curtis Langlotz", "Christopher Re"] | Constructing large, labeled training datasets for segmentation models is an expensive and labor-intensive process. This is a common challenge in machine learning, addressed by methods that require few or no labeled data points such as few-shot learning (FSL) and weakly-supervised learning (WS). Such techniques, however... | ["Weak supervision", "segmentation", "CNN", "latent variable", "medical imaging"] | ABSTRACTConstructing large, labeled training datasets for segmentation models is an expen-sive and labor-intensive process. This is a common challenge in machine learning,addressed by methods that require few or no labeled data points such as few-shotlearning (FSL) and weakly-supervised learning (WS). Such techniques, ... | yq5lgFz1gJY | This paper provides a weakly supervised method for medical image segmentation. However, it has several weaknesses need to be addressed in the future. | 5: Marginally below acceptance threshold | This paper presents weakly supervised framework for image segmentation tasks with limited annotated data. It first builds several labeling functions with limited annotation and then uses probability graph to fuse the labels. Last, the final output will be generated via CNN network. There are two key challenges in suc... | 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
Cut out the annotator, keep the cutout: better segmentation with weak supervision
### Paper Abstract
Constructing large, labeled training datasets for segmentation model... |
OAdGsaptOXy | ICLR.cc/2021/Conference | 2021 | Pretrain Knowledge-Aware Language Models | ["Corbin L Rosset", "Chenyan Xiong", "Minh Phan", "Xia Song", "Paul N. Bennett", "saurabh tiwary"] | How much knowledge do pretrained language models hold? Recent research observed that pretrained transformers are adept at modeling semantics but it is unclear to what degree they grasp human knowledge, or how to ensure they do so. In this paper we incorporate knowledge-awareness in language model pretraining without ch... | ["Pretraining", "Natural Language Generation", "GPT-2", "QA", "Knowledge Graph"] | ABSTRACTHow much knowledge do pretrained language models hold? Recent research ob-served that pretrained transformers are adept at modeling semantics but it is unclearto what degree they grasp human knowledge, or how to ensure they do so. Inthis paper we incorporate knowledge-awareness in language model pretrainingwith... | a8jGXC5BNc4 | entity information needs further explanation | 6: Marginally above acceptance threshold | The paper introduces to use additional knowledge to pretrain language models, which adds entity information to input and output. And the method improves the performance compared to the corresponding models without the knowledge. This is an interesting piece of work and worth publishing. However, I have some comments an... | 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
Pretrain Knowledge-Aware Language Models
### Paper Abstract
How much knowledge do pretrained language models hold? Recent research observed that pretrained transformers ... |
ByebT3EYvr | ICLR.cc/2020/Conference | 2020 | Single Deep Counterfactual Regret Minimization | ["Eric Steinberger"] | Counterfactual Regret Minimization (CFR) is the most successful algorithm for finding approximate Nash equilibria in imperfect information games. However, CFR's reliance on full game-tree traversals limits its scalability and generality. Therefore, the game's state- and action-space is often abstracted (i.e. simplified... | ["Game Theory", "Deep Reinforcement Learning", "Counterfactual Regret Minimization", "Imperfect Information Games", "Games", "Poker", "Nash Equilibrium"] | ABSTRACTCounterfactual Regret Minimization (CFR) is the most successful algorithm forfinding approximate Nash equilibria in imperfect information games . However,CFR’s reliance on full game-tree traversals limits its scalability and generality.Therefore, the game’s state- and action-space is often abstracted (i.e. simp... | H1xg0YvZoB | Official Blind Review #3 | 3: Weak Reject |
The paper studies " SD-CFR" neural network to approximate the tabular in CFR. It builds on DCFR but computes average strategy from value network buffer and the results show that SD-CFR is better than DCFR in practice.
This paper is similar to DCFR in terms of , and the written should be improved. For example, 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
Single Deep Counterfactual Regret Minimization
### Paper Abstract
Counterfactual Regret Minimization (CFR) is the most successful algorithm for finding approximate Nash ... | |
ry0WOxbRZ | ICLR.cc/2018/Conference | 2018 | IVE-GAN: Invariant Encoding Generative Adversarial Networks | ["Robin Winter", "Djork-Arn\u00e8 Clevert"] | Generative adversarial networks (GANs) are a powerful framework for generative tasks. However, they are difficult to train and tend to miss modes of the true data generation process. Although GANs can learn a rich representation of the covered modes of the data in their latent space, the framework misses an inverse map... | ["Deep learning", "Unsupervised Learning"] | ABSTRACTGenerative adversarial networks (GANs) are a powerful framework for genera-tive tasks. However, they are difficult to train and tend to miss modes of the truedata generation process. Although GANs can learn a rich representation of thecovered modes of the data in their latent space, the framework misses an inve... | SJqfr3qlf | weak evaluation | 4: Ok but not good enough - rejection | This paper presents the IVE-GAN, a model that introduces en encoder to the Generative Adversarial Network (GAN) framework. The model is evaluated qualitatively through samples and reconstructions on a synthetic dataset, MNIST and CelebA.
Summary:
The evaluation is superficial, no quantitative evaluation is presented ... | 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
IVE-GAN: Invariant Encoding Generative Adversarial Networks
### Paper Abstract
Generative adversarial networks (GANs) are a powerful framework for generative tasks. Howe... |
bEoxzW_EXsa | ICLR.cc/2021/Conference | 2021 | Wasserstein-2 Generative Networks | ["Alexander Korotin", "Vage Egiazarian", "Arip Asadulaev", "Alexander Safin", "Evgeny Burnaev"] | We propose a novel end-to-end non-minimax algorithm for training optimal transport mappings for the quadratic cost (Wasserstein-2 distance). The algorithm uses input convex neural networks and a cycle-consistency regularization to approximate Wasserstein-2 distance. In contrast to popular entropic and quadratic regular... | ["wasserstein-2 distance", "optimal transport maps", "non-minimax optimization", "cycle-consistency regularization", "input-convex neural networks"] | ABSTRACTWe propose a novel end-to-end non-minimax algorithm for training optimal trans-port mappings for the quadratic cost (Wasserstein-2 distance). The algorithm usesinput convex neural networks and a cycle-consistency regularization to approximateWasserstein-2 distance. In contrast to popular entropic and quadratic ... | S4LferGWJrC | An end-to-end algorithm with a non-minimax objective for training a OT map for quadratic cost | 8: Top 50% of accepted papers, clear accept | From my perspective, this is a much needed and love-to-see work for the line of neural generative modeling. Previous approaches were dominated by GAN based approaches which require solving a minimax optimization which has technical hurdles in practice. This paper reviews the literature of OT theory, and provides a comp... | 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
Wasserstein-2 Generative Networks
### Paper Abstract
We propose a novel end-to-end non-minimax algorithm for training optimal transport mappings for the quadratic cost (... |
vujTf_I8Kmc | ICLR.cc/2021/Conference | 2021 | Attentional Constellation Nets for Few-Shot Learning | ["Weijian Xu", "yifan xu", "Huaijin Wang", "Zhuowen Tu"] | The success of deep convolutional neural networks builds on top of the learning of effective convolution operations, capturing a hierarchy of structured features via filtering, activation, and pooling. However, the explicit structured features, e.g. object parts, are not expressive in the existing CNN frameworks. In th... | ["few-shot learning", "constellation models"] | ABSTRACTThe success of deep convolutional neural networks builds on top of the learningof effective convolution operations, capturing a hierarchy of structured featuresvia filtering, activation, and pooling. However, the explicit structured features, e.g.object parts, are not expressive in the existing CNN frameworks. ... | Cn2xWFqe3hJ | Review | 5: Marginally below acceptance threshold | The proposed contellation module is an improved version of non-local block [a], which contains a cell feature clustering module and a self-attention module for modeling pixel-wise (cell-wise) relationships. Inserting this block to the backbones could improve the performance for few-shot learning setting.
Concerns:
1. ... | 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
Attentional Constellation Nets for Few-Shot Learning
### Paper Abstract
The success of deep convolutional neural networks builds on top of the learning of effective conv... |
HJrJpzZRZ | ICLR.cc/2018/Conference | 2018 | Self-Supervised Learning of Object Motion Through Adversarial Video Prediction | ["Alex X. Lee", "Frederik Ebert", "Richard Zhang", "Chelsea Finn", "Pieter Abbeel", "Sergey Levine"] | Can we build models that automatically learn about object motion from raw, unlabeled videos? In this paper, we study the problem of multi-step video prediction, where the goal is to predict a sequence of future frames conditioned on a short context. We focus specifically on two aspects of video prediction: accurately m... | ["adversarial", "video prediction", "flow"] | ABSTRACTCan we build models that automatically learn about object motion from raw, un-labeled videos? In this paper, we study the problem of multi-step video pre-diction, where the goal is to predict a sequence of future frames conditionedon a short context. We focus specifically on two aspects of video prediction:accu... | HJ99sdVbG | Marginal contributions and missing comparison with state of the art | 3: Clear rejection | In this paper a neural-network based method for multi-frame video prediction is proposed. It builds on the previous work of [Finn et al. 2016] that uses a neural network to predict transformation parameters of an affine image transformation for future frame prediction, an idea akin to the Spatial Transformer Network pa... | 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 of Object Motion Through Adversarial Video Prediction
### Paper Abstract
Can we build models that automatically learn about object motion from r... |
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... | H1g0t0mMnQ | budgeted feature acquisition to train networks - seems similar to RADIN | 7: Good paper, accept | This paper presents a novel method for budgeted cost sensitive learning from Data Streams.
This paper seems very similar to the work of Contrado’s RADIN algorithm which similarly evaluates sequential datapoints with a recurrent neural network by adaptively “purchasing” the most valuable features for the current datapoi... | 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... |
S1Dh8Tg0- | ICLR.cc/2018/Conference | 2018 | Fix your classifier: the marginal value of training the last weight layer | ["Elad Hoffer", "Itay Hubara", "Daniel Soudry"] | Neural networks are commonly used as models for classification for a wide variety of tasks. Typically, a learned affine transformation is placed at the end of such models, yielding a per-class value used for classification. This classifier can have a vast number of parameters, which grows linearly with the number of po... | ["classifier", "marginal value", "models", "classification", "tasks", "last weight layer", "last weight", "neural networks", "wide variety", "learned affine transformation"] | ABSTRACTNeural networks are commonly used as models for classification for a wide varietyof tasks. Typically, a learned affine transformation is placed at the end of suchmodels, yielding a per-class value used for classification. This classifier can havea vast number of parameters, which grows linearly with the number ... | S1kGhTKez | . | 6: Marginally above acceptance threshold | This paper proposes replacing the weights of the final classifier layer in a CNN with a fixed projection matrix. In particular a Hadamard matrix can be used, which can be represented implicitly.
I'd have liked to see some discussion of how to efficiently implement the Hadamard transform when the number of penultimate... | 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
Fix your classifier: the marginal value of training the last weight layer
### Paper Abstract
Neural networks are commonly used as models for classification for a wide va... |
vU1QL3jGmV_ | NeurIPS.cc/2020/Workshop/DL-IG | 2020 | The Volume of Non-Restricted Boltzmann Machines and Their Double Descent Model Complexity | ["Prasad Cheema", "Mahito Sugiyama"] | The double descent risk phenomenon has received much interest in the machine learning and statistics community. Motivated through Rissanen's minimum description length (MDL) principle, and Amari's information geometry, we investigate how a double descent-like behavior may manifest by considering the $\log V$ modeling t... | ["Minimum description length", "geometric volume", "Boltzmann machine", "double descent", "generalization"] | The Volume of Non-Restricted Boltzmann Machinesand Their Double Descent Model ComplexityAnonymous Author(s)AffiliationAddressemailAbstractThe double descent risk phenomenon has received much interest in the machine 1learning and statistics community. Motivated through Rissanen’s minimum descrip- 2tion length (MDL) prin... | h-KjOvr9O4 | Review of "The Volume of Non-Restricted Boltzmann Machines and Their Double Descent Model Complexity" | 8: Top 50% of accepted papers, clear accept | This paper explores a phenomenon of intense recent interest, double descent, where for increasing model complexity we see test error falls, rises, then falls again for extremely over-parametrized models. The paper studies a particular class of lattice models for which some theoretical observations can be made. In parti... | 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 Volume of Non-Restricted Boltzmann Machines and Their Double Descent Model Complexity
### Paper Abstract
The double descent risk phenomenon has received much interes... |
rygDeZqap7 | AKBC.ws/2019/Conference | 2019 | Semi-supervised Ensemble Learning with Weak Supervision for Biomedical Relationship Extraction | ["Antonios Minas Krasakis", "Evangelos Kanoulas", "George Tsatsaronis"] | Natural language understanding research has recently shifted towards complex Machine Learning and Deep Learning algorithms. Such models often outperform their simpler counterparts significantly. However, their performance relies on the availability of large amounts of labeled data, which are rarely available. To tackle... | ["weak supervision", "meta-learning", "biomedical relationship extraction", "semi-supervised learning", "ensemble learning"] | Automated Knowledge Base Construction (2019) Conference paperSemi-supervised Ensemble Learning with Weak Supervisionfor Biomedical Relation ExtractionAntonios Minas Krasakis amkrasakis@gmail.comUniversity of Amsterdam, Science Park 904,1098XH Amsterdam, The NetherlandsEvangelos Kanoulas e.kanoulas@uva.nlUniversity of A... | B1lGTzl7fV | The paper proposes a potentially interesting direction, though needs to improve in many aspects. | 5: Marginally below acceptance threshold | The authors propose to semi-supervised method for relation classification, which trains multiple base learners using a small labeled dataset, and applies an ensemble of them to annotate unlabeled examples for semi-supervised learning. They conducted experiments on a BioCreative shared task for detecting chemical-protei... | 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
Semi-supervised Ensemble Learning with Weak Supervision for Biomedical Relationship Extraction
### Paper Abstract
Natural language understanding research has recently sh... |
6deUA11mOJ5 | ICLR.cc/2021/Conference | 2021 | A Large-scale Study on Training Sample Memorization in Generative Modeling | ["Ching-Yuan Bai", "Hsuan-Tien Lin", "Colin Raffel", "Wendy Kan"] | Many recent developments on generative models for natural images have relied on heuristically-motivated metrics that can be easily gamed by memorizing a small sample from the true distribution or training a model directly to improve the metric.
In this work, we critically evaluate the gameability of the benchmarking pr... | ["GAN", "generative adversarial networks", "generative modeling", "memorization"] | ABSTRACTMany recent developments on generative models for natural images have relied onheuristically-motivated metrics that can be easily gamed by memorizing a smallsample from the true distribution or training a model directly to improve the metric.In this work, we critically evaluate the gameability of such metrics b... | j7tdwf426T7 | Review of "A Large-scale Study on Training Sample Memorization in Generative Modeling" | 4: Ok but not good enough - rejection | The paper investigates memorization/overfitting in GANs and proposes a new metric (MiFID) to identify memorization in trained GAN models automatically. The topic of memorization in GANs is an important one and certainly one that does require more work and research.
Contributions/Novelty:
- The authors introduce a new ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Large-scale Study on Training Sample Memorization in Generative Modeling
### Paper Abstract
Many recent developments on generative models for natural images have relie... |
zWvMjL6o60V | ICLR.cc/2021/Conference | 2021 | Improving Mutual Information based Feature Selection by Boosting Unique Relevance | ["Shiyu Liu", "Mehul Motani"] | Mutual Information (MI) based feature selection makes use of MI to evaluate each feature and eventually shortlist a relevant feature subset, in order to address issues associated with high-dimensional datasets.
Despite the effectiveness of MI in feature selection, we have noticed that many state-of-the-art algorithms d... | ["Feature Selection", "Mutual Information", "Unique Relevance"] | ABSTRACTMutual Information (MI) based feature selection makes use of MI to evaluate eachfeature and eventually shortlist a relevant feature subset, in order to address issuesassociated with high-dimensional datasets. Despite the effectiveness of MI infeature selection, we have noticed that many state-of-the-art algorit... | IG_7fQst1PA | This paper is a slightly expanded version of work published elsewhere. | 2: Strong rejection | Update: The author response has not changed my opinion that there is insufficient new material in this paper vs the ISIT paper, and the presentation of the material from the ISIT paper does not note that this material was previously presented there. Without clarity in what is the novel material claimed in this paper it... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Improving Mutual Information based Feature Selection by Boosting Unique Relevance
### Paper Abstract
Mutual Information (MI) based feature selection makes use of MI to e... |
HJguLo0cKQ | ICLR.cc/2019/Conference | 2019 | Strength in Numbers: Trading-off Robustness and Computation via Adversarially-Trained Ensembles | ["Edward Grefenstette", "Robert Stanforth", "Brendan O'Donoghue", "Jonathan Uesato", "Grzegorz Swirszcz", "Pushmeet Kohli"] | While deep learning has led to remarkable results on a number of challenging problems, researchers have discovered a vulnerability of neural networks in adversarial settings, where small but carefully chosen perturbations to the input can make the models produce extremely inaccurate outputs. This makes these models par... | ["adversarial examples", "adversarial robustness", "visualisation", "ensembles"] | ABSTRACTWhile deep learning has led to remarkable results on a number of challengingproblems, researchers have discovered a vulnerability of neural networks in ad-versarial settings, where small but carefully chosen perturbations to the input canmake the models produce extremely inaccurate outputs. This makes these mod... | HklJi6Ox37 | Empirical Study of Robustness of (small) Ensembles of Neural Nets | 4: Ok but not good enough - rejection | Summary. The paper considers the robustness of neural nets against adversarial attacks. More precisely, the authors experimentally investigate the robustness of ensembles of neural nets. They empirically show that adversarially trained ensembles of 2 neural nets are more robust than ensembles of 2 adversarially trained... | 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
Strength in Numbers: Trading-off Robustness and Computation via Adversarially-Trained Ensembles
### Paper Abstract
While deep learning has led to remarkable results on a... |
r1eIiCNYwS | ICLR.cc/2020/Conference | 2020 | Transformer-XH: Multi-Evidence Reasoning with eXtra Hop Attention | ["Chen Zhao", "Chenyan Xiong", "Corby Rosset", "Xia Song", "Paul Bennett", "Saurabh Tiwary"] | Transformers have achieved new heights modeling natural language as a sequence of text tokens. However, in many real world scenarios, textual data inherently exhibits structures beyond a linear sequence such as trees and graphs; many tasks require reasoning with evidence scattered across multiple pieces of texts. This ... | ["Transformer-XH", "multi-hop QA", "fact verification", "extra hop attention", "structured modeling"] | ABSTRACTTransformers have achieved new heights modeling natural language as a sequenceof text tokens. However, in many real world scenarios, textual data inherently ex-hibits structures beyond a linear sequence such as trees and graphs; many tasksrequire reasoning with evidence scattered across multiple pieces of texts... | BJlHcZMptB | Official Blind Review #2 | 8: Accept | Summary: This paper introduces a way to train transformer models over document graphs, where each node is a document and edges connect related documents. It is inspired by the transformer-XL model, as well as Graph Neural networks. They apply this model to answer multi-hop questions on the HotPotQA dataset and outperfo... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Transformer-XH: Multi-Evidence Reasoning with eXtra Hop Attention
### Paper Abstract
Transformers have achieved new heights modeling natural language as a sequence of te... | |
rJxXDsCqYX | ICLR.cc/2019/Conference | 2019 | Sentence Encoding with Tree-Constrained Relation Networks | ["Lei Yu", "Cyprien de Masson d'Autume", "Chris Dyer", "Phil Blunsom", "Lingpeng Kong", "Wang Ling"] | The meaning of a sentence is a function of the relations that hold between its words. We instantiate this relational view of semantics in a series of neural models based on variants of relation networks (RNs) which represent a set of objects (for us, words forming a sentence) in terms of representations of pairs of obj... | ["sentence encoder", "relation networks", "tree", "machine translation"] | ABSTRACTThe meaning of a sentence is a function of the relations that hold between itswords. We instantiate this relational view of semantics in a series of neural modelsbased on variants of relation networks (RNs) which represent a set of objects (forus, words forming a sentence) in terms of representations of pairs o... | BJerp7cc2m | The actual effectiveness of the proposed method is unclear | 5: Marginally below acceptance threshold |
[Summary]
The main purpose of this paper is to propose an extension of relation networks.
The proposal consists of two parts: 1) to integrate constraints of dependency syntax to control which relations influence the representation, and 2) utilize a recurrent computation to capture higher-order relations.
[clarity]
Th... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Sentence Encoding with Tree-Constrained Relation Networks
### Paper Abstract
The meaning of a sentence is a function of the relations that hold between its words. We ins... |
HkEI22jeg | ICLR.cc/2017/conference | 2017 | Multilayer Recurrent Network Models of Primate Retinal Ganglion Cell Responses | ["Eleanor Batty", "Josh Merel", "Nora Brackbill", "Alexander Heitman", "Alexander Sher", "Alan Litke", "E.J. Chichilnisky", "Liam Paninski"] | Developing accurate predictive models of sensory neurons is vital to understanding sensory processing and brain computations. The current standard approach to modeling neurons is to start with simple models and to incrementally add interpretable features. An alternative approach is to start with a more complex model th... | ["Deep learning", "Applications"] | ABSTRACTDeveloping accurate predictive models of sensory neurons is vital to understandingsensory processing and brain computations. The current standard approach tomodeling neurons is to start with simple models and to incrementally add inter-pretable features. An alternative approach is to start with a more complex m... | S1dHogMEg | a clearly written paper with nice, if straightforward, results | 7: Good paper, accept | This is a clearly written paper with a nice, if straightforward, result: RNNs can be good predictive models of neuron firing rates in the retina.
On the one hand, the primary scientific contribution seems to just be to confirm that this approach works. On this particular stimulus locked task the gains from using the R... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Multilayer Recurrent Network Models of Primate Retinal Ganglion Cell Responses
### Paper Abstract
Developing accurate predictive models of sensory neurons is vital to un... |
9GsFOUyUPi | ICLR.cc/2021/Conference | 2021 | Progressive Skeletonization: Trimming more fat from a network at initialization | ["Pau de Jorge", "Amartya Sanyal", "Harkirat Behl", "Philip Torr", "Gr\u00e9gory Rogez", "Puneet K. Dokania"] | Recent studies have shown that skeletonization (pruning parameters) of networks at initialization provides all the practical benefits of sparsity both at inference and training time, while only marginally degrading their performance. However, we observe that beyond a certain level of sparsity (approx 95%), these approa... | ["Pruning", "Pruning at initialization", "Sparsity"] | ABSTRACTRecent studies have shown that skeletonization (pruning parameters) of networksat initialization provides all the practical benefits of sparsity both at inference andtraining time, while only marginally degrading their performance. However, weobserve that beyond a certain level of sparsity (approx 95%), these a... | nRMdudFlSo | A good paper, nice problem fromulation | 7: Good paper, accept | Summary:
This paper presented a new sensitivity-based pruning methods to cut network connections at initialisation. The proposed method FORCE shows great performance when compared to rencelty proposed SNIP and GRASP. In general, I liked the formulation of the problem, its in-depth analysis and detailed comparison to e... | 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
Progressive Skeletonization: Trimming more fat from a network at initialization
### Paper Abstract
Recent studies have shown that skeletonization (pruning parameters) of... |
HyljY04YDB | ICLR.cc/2020/Conference | 2020 | Towards Interpretable Molecular Graph Representation Learning | ["Emmanuel Noutahi", "Dominique Beani", "Julien Horwood", "Prudencio Tossou"] | Recent work in graph neural networks (GNNs) has led to improvements in molecular activity and property prediction tasks. Unfortunately, GNNs often fail to capture the relative importance of interactions between molecular substructures, in part due to the absence of efficient intermediate pooling steps. To address these... | ["molecular graphs", "graph pooling", "hierarchical", "GNN", "Laplacian", "drug discovery"] | ABSTRACTRecent work in graph neural networks (GNNs) has led to improvements in molecu-lar activity and property prediction tasks. Unfortunately, GNNs often fail to capturethe relative importance of interactions between molecular substructures, in partdue to the absence of efficient intermediate pooling steps. To addres... | ByxxBKYotB | Official Blind Review #1 | 1: Reject | This paper proposes a graph down-sampling component named LaPool. It uses the graph signal to dynamically select some "important" centroids and learn a sparse assignment matrix for clustering the remaining nodes.
This paper should be rejected due to the following reasons.
1. Strange new results compared with its prev... | <|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 Interpretable Molecular Graph Representation Learning
### Paper Abstract
Recent work in graph neural networks (GNNs) has led to improvements in molecular activit... | |
fmSC3olLjkb | icaps-conference.org/ICAPS/2021/Workshop/XAIP | 2021 | Planning for Proactive Assistance in Environments with Partial Observability | ["Anagha Kulkarni", "Siddharth Srivastava", "Subbarao Kambhampati"] | This paper addresses the problem of synthesizing the behavior of an AI agent that provides proactive task assistance to a human in settings like factory floors where they may coexist in a common environment. Unlike in the case of requested assistance, the human may not be expecting proactive assistance and hence it is ... | ["proactive assistance", "controlling observability", "legibility", "obfuscation"] | Planning for Proactive Assistance inEnvironments with Partial ObservabilityAnagha Kulkarni, Siddharth Srivastava, Subbarao KambhampatiDepartment of Computer Science, Arizona State Universityfanaghak, siddharths, rao g@asu.eduAbstractThis paper addresses the problem of synthesizing the behav-ior of an AI agent that prov... | Ghphx1ZiLvN | Interesting ideas on proactive assistance | 7: Good paper, accept |
The paper presents an algorithm for an autonomous planning-based agent to proactively help another agent (i.e. Human) to perform their task while trying to achieve its own goals. This proactive assistant is such that does help the human agent achieve their goals in a more efficient and legible manner. The problem is p... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Planning for Proactive Assistance in Environments with Partial Observability
### Paper Abstract
This paper addresses the problem of synthesizing the behavior of an AI ag... |
MjvduJCsE4 | ICLR.cc/2021/Conference | 2021 | Exploring the Uncertainty Properties of Neural Networks’ Implicit Priors in the Infinite-Width Limit | ["Ben Adlam", "Jaehoon Lee", "Lechao Xiao", "Jeffrey Pennington", "Jasper Snoek"] | Modern deep learning models have achieved great success in predictive accuracy for many data modalities. However, their application to many real-world tasks is restricted by poor uncertainty estimates, such as overconfidence on out-of-distribution (OOD) data and ungraceful failing under distributional shift. Previous b... | ["Deep Learning", "Uncertainty", "Infinite-Width Limit", "Neural Network Gaussian Process", "Bayesian Neural Networks", "Gaussian Process"] | ABSTRACTModern deep learning models have achieved great success in predictive accuracyfor many data modalities. However, their application to many real-world tasksis restricted by poor uncertainty estimates, such as overconfidence on out-of-distribution (OOD) data and ungraceful failing under distributional shift. Prev... | 2sw3ggn3TWt | Review | 5: Marginally below acceptance threshold | This is an interesting paper which evaluates the calibration of NNGPs, including around OOD detection.
Major points:
1. It is worth noting that in the appendix for arXiv:1808.05587, they explicitly use the full categorical log-likelihood with softmax probabilities (albeit with inducing points), and they find that 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
Exploring the Uncertainty Properties of Neural Networks’ Implicit Priors in the Infinite-Width Limit
### Paper Abstract
Modern deep learning models have achieved great s... |
S1Euwz-Rb | ICLR.cc/2018/Conference | 2018 | Compositional Attention Networks for Machine Reasoning | ["Drew A. Hudson", "Christopher D. Manning"] | We present Compositional Attention Networks, a novel fully differentiable neural network architecture, designed to facilitate explicit and expressive reasoning. While many types of neural networks are effective at learning and generalizing from massive quantities of data, this model moves away from monolithic black-box... | ["Deep Learning", "Reasoning", "Memory", "Attention", "VQA", "CLEVR", "Recurrent Neural Networks", "Module Networks", "Compositionality"] | ABSTRACTWe present Compositional Attention Networks, a novel fully differentiable neu-ral network architecture, designed to facilitate explicit and expressive reasoning.While many types of neural networks are effective at learning and generalizingfrom massive quantities of data, this model moves away from monolithic bl... | Sk0oVNYlM | Review | 7: Good paper, accept | This paper describes a new model architecture for machine reasoning. In contrast
to previous approaches that explicitly predict a question-specific module
network layout, the current paper introduces a monolithic feedforward network
with iterated rounds of attention and memory. On a few variants of the CLEVR
dataset, i... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Compositional Attention Networks for Machine Reasoning
### Paper Abstract
We present Compositional Attention Networks, a novel fully differentiable neural network archit... |
SJeFNoRcFQ | ICLR.cc/2019/Conference | 2019 | Traditional and Heavy Tailed Self Regularization in Neural Network Models | ["Charles H. Martin", "Michael W. Mahoney"] | Random Matrix Theory (RMT) is applied to analyze the weight matrices of Deep Neural Networks (DNNs), including both production quality, pre-trained models such as AlexNet and Inception, and smaller models trained from scratch, such as LeNet5 and a miniature-AlexNet. Empirical and theoretical results clearly indicate t... | ["statistical mechanics", "self-regularization", "random matrix", "glassy behavior", "heavy-tailed"] | ABSTRACTRandom Matrix Theory (RMT) is applied to analyze the weight matrices of DeepNeural Networks (DNNs), including both production quality, pre-trained modelssuch as AlexNet and Inception, and smaller models trained from scratch, suchas LeNet5 and a miniature-AlexNet. Empirical and theoretical results clearly in-dic... | H1gn4x1MaX | Although the experiments seem interesting, theoretical results are not clearly presented. | 4: Ok but not good enough - rejection | This manuscript studies the implicit regularization of neural networks from the perspective of random matrix theory. The authors provide both empirical and theoretical results that aim to show that the empirical spectral density of weights of DNN captures the implicit regularization phenomenon. However, the results are... | 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
Traditional and Heavy Tailed Self Regularization in Neural Network Models
### Paper Abstract
Random Matrix Theory (RMT) is applied to analyze the weight matrices of Deep... |
T6AxtOaWydQ | ICLR.cc/2021/Conference | 2021 | $i$-Mix: A Domain-Agnostic Strategy for Contrastive Representation Learning | ["Kibok Lee", "Yian Zhu", "Kihyuk Sohn", "Chun-Liang Li", "Jinwoo Shin", "Honglak Lee"] | Contrastive representation learning has shown to be effective to learn representations from unlabeled data. However, much progress has been made in vision domains relying on data augmentations carefully designed using domain knowledge. In this work, we propose i-Mix, a simple yet effective domain-agnostic regularizatio... | ["self-supervised learning", "unsupervised representation learning", "contrastive representation learning", "data augmentation", "MixUp"] | ABSTRACTContrastive representation learning has shown to be effective to learn represen-tations from unlabeled data. However, much progress has been made in visiondomains relying on data augmentations carefully designed using domain knowledge.In this work, we propose i-Mix, a simple yet effective domain-agnostic regula... | PEivHdGFTbF | Strong empirical results, but a few clarifications are needed. | 7: Good paper, accept | In essence, the premise and contribution of the paper is rather simple -- contrastive learning and mixup are both very effective methods in their respective domains, and this paper proposes the use of mixup to improve contrastive learning, and the results seem to indicate this.
Positives:
- Strong empirical results, v... | 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
$i$-Mix: A Domain-Agnostic Strategy for Contrastive Representation Learning
### Paper Abstract
Contrastive representation learning has shown to be effective to learn rep... |
0owsv3F-fM | ICLR.cc/2021/Conference | 2021 | Cross-Modal Domain Adaptation for Reinforcement Learning | ["Xiong-Hui Chen", "Shengyi Jiang", "Feng Xu", "Yang Yu"] | Domain adaptation is a promising direction for deploying RL agents in real-world applications, where vision-based robotics tasks constitute an important part. Cur-rent methods that train polices on simulated images not only require a delicately crafted simulator, but also add extra burdens to the training process. In t... | ["Domain Adaptation", "Reinforcement Learning"] | ABSTRACTTo overcome the unbearable reinforcement training of agents in the real-world, thesim-to-real approach, i.e., training in simulators and adapting to target environ-ments, is a promising direction. However, crafting a delicately simulator can alsobe difficult and costly. For example, to simulate vision-based rob... | WKCvShjBMYk | New topic in RL, but experimental results are missing | 5: Marginally below acceptance threshold | ## Cross-modal Domain Adaptation for Reinforcement Learning
### Summary
The authors propose CODAS, a domain adaptation algorithm for transferring policies learned in state-space to the image domain. They learn a variational RNN to map image inputs to state space with an adversarial loss to ensure that the distributions... | 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
Cross-Modal Domain Adaptation for Reinforcement Learning
### Paper Abstract
Domain adaptation is a promising direction for deploying RL agents in real-world applications... |
Syx7A3NFvH | ICLR.cc/2020/Conference | 2020 | Multi-agent Reinforcement Learning for Networked System Control | ["Tianshu Chu", "Sandeep Chinchali", "Sachin Katti"] | This paper considers multi-agent reinforcement learning (MARL) in networked system control. Specifically, each agent learns a decentralized control policy based on local observations and messages from connected neighbors. We formulate such a networked MARL (NMARL) problem as a spatiotemporal Markov decision process and... | ["deep reinforcement learning", "multi-agent reinforcement learning", "decision and control"] | ABSTRACTThis paper considers multi-agent reinforcement learning (MARL) in networkedsystem control. Specifically, each agent learns a decentralized control policy basedon local observations and messages from connected neighbors. We formulatesuch a networked MARL (NMARL) problem as a spatiotemporal Markov decisionprocess... | Skg3YJBCFH | Official Blind Review #1 | 6: Weak Accept | 1. Summary
The authors use decentralized MARL for networked system control. Each agent might control a traffic light (exp 1) or a car in traffic (exp 2). Some features of their approach are a spatial Markov assumption (only neighborhood matters), a spatial discount factor, and NeurComm: a general message passing schem... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Multi-agent Reinforcement Learning for Networked System Control
### Paper Abstract
This paper considers multi-agent reinforcement learning (MARL) in networked system con... | |
B18WgG-CZ | ICLR.cc/2018/Conference | 2018 | Learning General Purpose Distributed Sentence Representations via Large Scale Multi-task Learning | ["Sandeep Subramanian", "Adam Trischler", "Yoshua Bengio", "Christopher J Pal"] | A lot of the recent success in natural language processing (NLP) has been driven by distributed vector representations of words trained on large amounts of text in an unsupervised manner. These representations are typically used as general purpose features for words across a range of NLP problems. However, extending th... | ["distributed sentence representations", "multi-task learning"] | ABSTRACTA lot of the recent success in natural language processing (NLP) has been drivenby distributed vector representations of words trained on large amounts of textin an unsupervised manner. These representations are typically used as generalpurpose features for words across a range of NLP problems. However, extend-... | BJOtf_JlG | Generally positive review | 8: Top 50% of accepted papers, clear accept | Follow-Up Comments
----
I continue to argue that this paper makes a contribution to a major open question, and clearly warrants acceptance.
I agree with R1 that the results do not tell a completely clear story, and that the benefits of pretraining are occasionally minimal or absent. However, R1 uses this as the basi... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning General Purpose Distributed Sentence Representations via Large Scale Multi-task Learning
### Paper Abstract
A lot of the recent success in natural language proc... |
IM8zOC94HF | robot-learning.org/CoRL/2023/Conference | 2023 | Fighting Uncertainty with Gradients: Offline Reinforcement Learning via Diffusion Score Matching | ["H.J. Terry Suh", "Glen Chou", "Hongkai Dai", "Lujie Yang", "Abhishek Gupta", "Russ Tedrake"] | Gradient-based methods enable efficient search capabilities in high dimensions. However, in order to apply them effectively in offline optimization paradigms such as offline Reinforcement Learning (RL) or Imitation Learning (IL), we require a more careful consideration of how uncertainty estimation interplays with firs... | ["Model-Based Reinforcement Learning", "Offline Reinforcement Learning", "Planning under Uncertainty", "Diffusion", "Score Matching"] | Fighting Uncertainty with Gradients: OfflineReinforcement Learning via Diffusion Score MatchingH.J. Terry SuhCSAIL, MITCambridge, MA 02139hjsuh@mit.eduGlen Chou∗CSAIL, MITCambridge, MA 02139gchou@mit.eduHongkai Dai∗Toyota Research InstituteLos Altos, CA 94022hongkai.dai@tri.globalLujie Yang∗CSAIL, MITCambridge, MA 0213... | DOFDRZsg7A_ | The reinterpretation of score-matching for a perturbed empirical distribution as learning gradients of smoothed distance to data is a clever idea, and the paper is well-written with convincing experiments supporting the authors' main claims. I am pleased to recommend its acceptance, however I believe that there are two... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Fighting Uncertainty with Gradients: Offline Reinforcement Learning via Diffusion Score Matching
### Paper Abstract
Gradient-based methods enable efficient search capabi... | ||
Hyffti0ctQ | ICLR.cc/2019/Conference | 2019 | PRUNING WITH HINTS: AN EFFICIENT FRAMEWORK FOR MODEL ACCELERATION | ["Wei Gao", "Yi Wei", "Quanquan Li", "Hongwei Qin", "Wanli Ouyang", "Junjie Yan"] | In this paper, we propose an efficient framework to accelerate convolutional neural networks. We utilize two types of acceleration methods: pruning and hints. Pruning can reduce model size by removing channels of layers. Hints can improve the performance of student model by transferring knowledge from teacher model. We... | ["model acceleration", "mimic", "knowledge distillation", "channel pruning"] | ABSTRACTIn this paper, we propose an efficient framework to accelerate convolutional neuralnetworks. We utilize two types of acceleration methods: pruning and hints. Prun-ing can reduce model size by removing channels of layers. Hints can improvethe performance of student model by transferring knowledge from teacher mo... | SJl2i4dA3X | a pruning-with-hints framework for model acceleration | 5: Marginally below acceptance threshold | In this paper, the authors propose a pruning-with-hints framework for model acceleration. Via performing pruning and hints iteratively, one can leverage the complementary characteristics of these two approaches to boost performance.
Here are the comments:
1 The framework seems to be technically sound. However, the 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
PRUNING WITH HINTS: AN EFFICIENT FRAMEWORK FOR MODEL ACCELERATION
### Paper Abstract
In this paper, we propose an efficient framework to accelerate convolutional neural ... |
ryh_8f9lg | ICLR.cc/2017/conference | 2017 | Classless Association using Neural Networks | ["Federico Raue", "Sebastian Palacio", "Andreas Dengel", "Marcus Liwicki"] | The goal of this paper is to train a model based on the relation between two instances that represent the same unknown class. This scenario is inspired by the Symbol Grounding Problem and the association learning in infants. We propose a novel model called Classless Association. It has two parallel Multilayer Percep... | ["model", "classless association", "network", "addition", "neural networks", "goal", "relation", "instances", "unknown class"] | ABSTRACTThe goal of this paper is to train a model based on the relation between two in-stances that represent the same unknown class. This scenario is inspired by theSymbol Grounding Problem and the association learning in infants. We propose anovel model called Classless Association . It has two parallel Multilayer P... | r1uSbyMVx | Classes association | 6: Marginally above acceptance threshold | The paper looks correct but still i am not convinced about the experimentation performed. Perhaps another experiment with more challenging data would be welcome. Honestly i don't find a clear motivation for this work however it could have some potential and it would be interested to be presented in conference.
| 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
Classless Association using Neural Networks
### Paper Abstract
The goal of this paper is to train a model based on the relation between two instances that represent the ... |
H1gFuiA9KX | ICLR.cc/2019/Conference | 2019 | Skip-gram word embeddings in hyperbolic space | ["Matthias Leimeister", "Benjamin J. Wilson"] | Embeddings of tree-like graphs in hyperbolic space were recently shown to surpass their Euclidean counterparts in performance by a large margin.
Inspired by these results, we present an algorithm for learning word embeddings in hyperbolic space from free text. An objective function based on the hyperbolic distance is d... | ["word embeddings", "hyperbolic", "skip-gram"] | ABSTRACTEmbeddings of tree-like graphs in hyperbolic space were recently shown to sur-pass their Euclidean counterparts in performance by a large margin. Inspired bythese results, we present an algorithm for learning word embeddings in hyper-bolic space from free text. An objective function based on the hyperbolic dis-... | rJlysr_ka7 | skip-gram + Minkowski distance -> unclear why it should work | 5: Marginally below acceptance threshold | The present paper aims to apply recent developments in hyperbolic embeddings of graphs to the area of word embeddings.
The paper is relatively clearly written and looks technically correct. The main contribution of the paper is in suggesting the usage of Minkowski dot-product instead of Eucledian dot-product in skip-g... | 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
Skip-gram word embeddings in hyperbolic space
### Paper Abstract
Embeddings of tree-like graphs in hyperbolic space were recently shown to surpass their Euclidean counte... |
gwnoVHIES05 | ICLR.cc/2021/Conference | 2021 | Creative Sketch Generation | ["Songwei Ge", "Vedanuj Goswami", "Larry Zitnick", "Devi Parikh"] | Sketching or doodling is a popular creative activity that people engage in. However, most existing work in automatic sketch understanding or generation has focused on sketches that are quite mundane. In this work, we introduce two datasets of creative sketches -- Creative Birds and Creative Creatures -- containing 10k ... | ["creativity", "sketches", "part-based", "GAN", "dataset", "generative art"] | ABSTRACTSketching or doodling is a popular creative activity that people engage in. How-ever, most existing work in automatic sketch understanding or generation hasfocused on sketches that are quite mundane. In this work, we introduce twodatasets of creative sketches – Creative Birds and Creative Creatures – contain-in... | NtraifPTXTo | Introduces two novel doodle sketch datasets as well as a generative model for producing new sketches but does not seem to provide a substantial advantage over existing work | 6: Marginally above acceptance threshold | ### Summary
This paper introduces two creative sketch datasets of birds and creatures, segmented into parts, each with ~10k doodles collected from Amazon MTurk workers. In a user study, people tend to favor sketches from their dataset over the similar Gogole QuickDraw sketches. Additionally, the authors propose a GAN 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
Creative Sketch Generation
### Paper Abstract
Sketching or doodling is a popular creative activity that people engage in. However, most existing work in automatic sketch... |
S1xitgHtvS | ICLR.cc/2020/Conference | 2020 | Making Sense of Reinforcement Learning and Probabilistic Inference | ["Brendan O'Donoghue", "Ian Osband", "Catalin Ionescu"] | Reinforcement learning (RL) combines a control problem with statistical estimation: The system dynamics are not known to the agent, but can be learned through experience. A recent line of research casts ‘RL as inference’ and suggests a particular framework to generalize the RL problem as probabilistic inference. Our pa... | ["Reinforcement learning", "Bayesian inference", "Exploration"] | ABSTRACTReinforcement learning (RL) combines a control problem with statistical estima-tion: The system dynamics are not known to the agent, but can be learned throughexperience. A recent line of research casts ‘RL as inference’ and suggests a partic-ular framework to generalize the RL problem as probabilistic inferenc... | ryez-G1Lqr | Official Blind Review #3 | 6: Weak Accept | The authors develop a criticism of the "RL as inference" standard approximations and propose a simple modification that solves its main issues while keeping hold of its advantage. Even though this modification ends up relating to a previously published algorithm, I judge the submission to be worthwhile publishing for 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
Making Sense of Reinforcement Learning and Probabilistic Inference
### Paper Abstract
Reinforcement learning (RL) combines a control problem with statistical estimation:... | |
0F_OC_oROWb | ICLR.cc/2021/Conference | 2021 | RSO: A Gradient Free Sampling Based Approach For Training Deep Neural Networks | ["Rohun Tripathi", "Bharat Singh"] | We propose RSO (random search optimization), a gradient free, sampling based approach for training deep neural networks. To this end, RSO adds a perturbation to a weight in a deep neural network and tests if it reduces the loss on a mini-batch. If this reduces the loss, the weight is updated, otherwise the existing wei... | ["rso", "weight", "deep neural networks", "loss", "gradient free sampling", "deep neural network", "times", "mnist", "random search optimization"] | ABSTRACTWe propose RSO (random search optimization), a gradient free, sampling basedapproach for training deep neural networks. To this end, RSO adds a perturba-tion to a weight in a deep neural network and tests if it reduces the loss on amini-batch. If this reduces the loss, the weight is updated, otherwise the exist... | JYuw83ZSc1L | Very simple but potentially very practical idea. | 7: Good paper, accept | In this paper, rather than training a DNN using SGD, the proposed idea is to perturb the weights of the network and accept the perturbation if it improves the performance. This naive idea seems to perform almost as well as SGD, and "an order of magnitude faster" (see below).
While I commend the authors for bringing up... | 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
RSO: A Gradient Free Sampling Based Approach For Training Deep Neural Networks
### Paper Abstract
We propose RSO (random search optimization), a gradient free, sampling ... |
rkaqxm-0b | ICLR.cc/2018/Conference | 2018 | Neural Compositional Denotational Semantics for Question Answering | ["Nitish Gupta", "Mike Lewis"] | Answering compositional questions requiring multi-step reasoning is challenging for current models. We introduce an end-to-end differentiable model for interpreting questions, which is inspired by formal approaches to semantics. Each span of text is represented by a denotation in a knowledge graph, together with a vect... | ["question answering", "knowledge graph", "compositional model", "semantics"] | ABSTRACTAnswering compositional questions requiring multi-step reasoning is challengingfor current models. We introduce an end-to-end differentiable model for inter-preting questions, which is inspired by formal approaches to semantics. Eachspan of text is represented by a denotation in a knowledge graph, together with... | B1uoZsYlM | Not convinced by only synthetic data evaluation | 4: Ok but not good enough - rejection | The paper describes an end to end differentiable model to answer questions based on a knowledge base. They learn the composition modules which combine representations for parts of the question to generate a representation of the whole question.
My major complaint is the evaluation on a synthetically generated data ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Compositional Denotational Semantics for Question Answering
### Paper Abstract
Answering compositional questions requiring multi-step reasoning is challenging for... |
ryxsCiAqKm | ICLR.cc/2019/Conference | 2019 | Spectral Convolutional Networks on Hierarchical Multigraphs | ["Boris Knyazev", "Xiao Lin", "Mohamed R. Amer", "Graham W. Taylor"] | Spectral Graph Convolutional Networks (GCNs) are a generalization of convolutional networks to learning on graph-structured data. Applications of spectral GCNs have been successful, but limited to a few problems where the graph is fixed, such as shape correspondence and node classification. In this work, we address thi... | ["graph convolution", "hierarchical models", "neural networks", "multigraph", "deep learning"] | ABSTRACTSpectral Graph Convolutional Networks (GCNs) are a generalization of convo-lutional networks to learning on graph-structured data. Applications of spectralGCNs have been successful, but limited to a few problems where the graph isfixed, such as shape correspondence and node classification. In this work, we ad-d... | H1ew2c9uhQ | Highly-problematic experimental setup | 3: Clear rejection | # Summary of the paper
This paper presents a method for graph classification based on spectral convolutional neural networks. At its heart, the paper applies a standard decomposition in terms of the graph Laplacian and uses Chebyshev polynomials to obtain an approximation of node feature convolution. This is now exten... | 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
Spectral Convolutional Networks on Hierarchical Multigraphs
### Paper Abstract
Spectral Graph Convolutional Networks (GCNs) are a generalization of convolutional network... |
H1M7soActX | ICLR.cc/2019/Conference | 2019 | The Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects | ["Zhanxing Zhu", "Jingfeng Wu", "Bing Yu", "Lei Wu", "Jinwen Ma"] | Understanding the behavior of stochastic gradient descent (SGD) in the context of deep neural networks has raised lots of concerns recently. Along this line, we theoretically study a general form of gradient based optimization dynamics with unbiased noise, which unifies SGD and standard Langevin dynamics. Through inve... | ["Stochastic gradient descent", "anisotropic noise", "regularization"] | ABSTRACTUnderstanding the behavior of stochastic gradient descent (SGD) in the contextof deep neural networks has raised lots of concerns recently. Along this line, wetheoretically study a general form of gradient based optimization dynamics withunbiased noise, which unifies SGD and standard Langevin dynamics. Through ... | H1lJasFPnX | Interesting but lacks clarity | 5: Marginally below acceptance threshold | The paper studies the benefit of an anisotropic gradient covariance matrix in SGD optimization for training deep network in terms of escaping sharp minima (which has been discussed to correlate with poor generalization in recent literature).
In order to do so, SGD is studied as a discrete approximation of stochastic ... | 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 Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects
### Paper Abstract
Understanding the behavior of s... |
BJlRKgkDwN | ICLR.cc/2019/Workshop/LLD | 2019 | WEAKLY SEMI-SUPERVISED NEURAL TOPIC MODELS | ["Ian Gemp", "Ramesh Nallapati", "Ran Ding", "Feng Nan", "Bing Xiang"] | We consider the problem of topic modeling in a weakly semi-supervised setting. In this scenario, we assume that the user knows a priori a subset of the topics she wants the model to learn and is able to provide a few exemplar documents for those topics. In addition, while each document may typically consist of multiple... | ["topic models", "variational inference", "variational autoencoder", "semi-supervised", "deep learning"] | ABSTRACTWe consider the problem of topic modeling in a weakly semi-supervised setting.In this scenario, we assume that the user knows a priori a subset of the topics shewants the model to learn and is able to provide a few exemplar documents forthose topics. In addition, while each document may typically consist of mul... | HkxY9HPIKN | unclear | 2: Marginally below acceptance threshold | This short paper investigates weak supervision for neural-based topic modelling (NTM) by modifying both the priors and the posteriors, in a VAE setting.
The priors are modified to using the logits of the probability for each topic to appear in a document. The use of the logits is meant to strengthen "ground truth" inf... | 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
WEAKLY SEMI-SUPERVISED NEURAL TOPIC MODELS
### Paper Abstract
We consider the problem of topic modeling in a weakly semi-supervised setting. In this scenario, we assume ... |
B1xxAJHFwS | ICLR.cc/2020/Conference | 2020 | A Finite-Time Analysis of Q-Learning with Neural Network Function Approximation | ["Pan Xu", "Quanquan Gu"] | Q-learning with neural network function approximation (neural Q-learning for short) is among the most prevalent deep reinforcement learning algorithms. Despite its empirical success, the non-asymptotic convergence rate of neural Q-learning remains virtually unknown. In this paper, we present a finite-time analysis of a... | ["Reinforcement Learning", "Neural Networks"] | ABSTRACTQ-learning with neural network function approximation (neural Q-learning forshort) is among the most prevalent deep reinforcement learning algorithms. De-spite its empirical success, the non-asymptotic convergence rate of neural Q-learning remains virtually unknown. In this paper, we present a finite-time analy... | B1l0WBG3tB | Official Blind Review #2 | 3: Weak Reject | [Summary]
This paper studies the convergence of Q-Learning when a wide multi-layer network in the Neural Tangent Kernel (NTK) regime is used as the function approximator. Concretely, it shows that Q-learning converges with rate O(1/T) with data sampled from a single trajectory (non-i.i.d.) of an infinite-horizon MDP + ... | <|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 Finite-Time Analysis of Q-Learning with Neural Network Function Approximation
### Paper Abstract
Q-learning with neural network function approximation (neural Q-learn... | |
ECuvULjFQia | ICLR.cc/2021/Conference | 2021 | A teacher-student framework to distill future trajectories | ["Alexander Neitz", "Giambattista Parascandolo", "Bernhard Sch\u00f6lkopf"] | By learning to predict trajectories of dynamical systems, model-based methods can make extensive use of all observations from past experience. However, due to partial observability, stochasticity, compounding errors, and irrelevant dynamics, training to predict observations explicitly often results in poor models. Mode... | ["meta-learning", "privileged information"] | ABSTRACTBy learning to predict trajectories of dynamical systems, model-based methodscan make extensive use of all observations from past experience. However, due topartial observability, stochasticity, compounding errors, and irrelevant dynamics,training to predict observations explicitly often results in poor models.... | Gr9x8mxy_r | Recommendation to Accept | 6: Marginally above acceptance threshold | This paper proposes a teacher-student training scheme to incorporate the useful information of trajectory to improve the predictive performance of model-free methods. The teacher network tries to "guide" the student network at the training stage by presenting an interpretation of the trajectory. The guidance is impleme... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A teacher-student framework to distill future trajectories
### Paper Abstract
By learning to predict trajectories of dynamical systems, model-based methods can make exte... |
BJgZGeHFPH | ICLR.cc/2020/Conference | 2020 | Dynamics-Aware Embeddings | ["William Whitney", "Rajat Agarwal", "Kyunghyun Cho", "Abhinav Gupta"] | In this paper we consider self-supervised representation learning to improve sample efficiency in reinforcement learning (RL). We propose a forward prediction objective for simultaneously learning embeddings of states and actions. These embeddings capture the structure of the environment's dynamics, enabling efficient ... | ["representation learning", "reinforcement learning", "rl"] | ABSTRACTIn this paper we consider self-supervised representation learning to improve sam-ple efficiency in reinforcement learning (RL). We propose a forward predictionobjective for simultaneously learning embeddings of states and action sequences.These embeddings capture the structure of the environment’s dynamics, ena... | ByxvmHURtr | Official Blind Review #3 | 6: Weak Accept | The author propose a representation learning method based on predictive information. They compress a start state and an action sequence to predict the following state. Since the latent space is factorized between state and action sequence, it can be used as an abstract action space to accelerate model-free algorithms.
... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Dynamics-Aware Embeddings
### Paper Abstract
In this paper we consider self-supervised representation learning to improve sample efficiency in reinforcement learning (RL... | |
6zaTwpNSsQ2 | ICLR.cc/2021/Conference | 2021 | A Block Minifloat Representation for Training Deep Neural Networks | ["Sean Fox", "Seyedramin Rasoulinezhad", "Julian Faraone", "david boland", "Philip Leong"] | Training Deep Neural Networks (DNN) with high efficiency can be difficult to achieve with native floating-point representations and commercially available hardware. Specialized arithmetic with custom acceleration offers perhaps the most promising alternative. Ongoing research is trending towards narrow floating-point r... | ["deep neural networks", "representations", "block minifloat representation", "operations", "accuracy", "dnn", "high efficiency", "difficult", "native"] | ABSTRACTTraining Deep Neural Networks (DNN) with high efficiency can be difficult toachieve with native floating-point representations and commercially availablehardware. Specialized arithmetic with custom acceleration offers perhaps the mostpromising alternative. Ongoing research is trending towards narrow floating-po... | pYHwUaesTTd | A good submission but need to provide more information | 6: Marginally above acceptance threshold | This paper introduced a new representation (Block Minifloat) for training DNNs with low precisions of 8-bit or less. This new representation combines FP8 formats and the shared exponent bias concept to cover the dynamic range of tensors needed for DNN training. Compared to other published FP8 format, this representatio... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Block Minifloat Representation for Training Deep Neural Networks
### Paper Abstract
Training Deep Neural Networks (DNN) with high efficiency can be difficult to achiev... |
HyeSin4FPB | ICLR.cc/2020/Conference | 2020 | Learning to Control PDEs with Differentiable Physics | ["Philipp Holl", "Nils Thuerey", "Vladlen Koltun"] | Predicting outcomes and planning interactions with the physical world are long-standing goals for machine learning. A variety of such tasks involves continuous physical systems, which can be described by partial differential equations (PDEs) with many degrees of freedom. Existing methods that aim to control the dynamic... | ["Differentiable physics", "Optimal control", "Deep learning"] | ABSTRACTPredicting outcomes and planning interactions with the physical world are long-standing goals for machine learning. A variety of such tasks involves contin-uous physical systems, which can be described by partial differential equations(PDEs) with many degrees of freedom. Existing methods that aim to control the... | Sye8KBAe9B | Official Blind Review #3 | 8: Accept | ## Summary
The authors propose a method for training physical systems whose behavior is governed by partial differential equations. They consider situations where only partial observations available, and where control is indirect.
Indirectly controlling physical systems is a very important problem with applications 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
Learning to Control PDEs with Differentiable Physics
### Paper Abstract
Predicting outcomes and planning interactions with the physical world are long-standing goals for... | |
TBIzh9b5eaz | ICLR.cc/2021/Conference | 2021 | Risk-Averse Offline Reinforcement Learning | ["N\u00faria Armengol Urp\u00ed", "Sebastian Curi", "Andreas Krause"] | Training Reinforcement Learning (RL) agents in high-stakes applications might be too prohibitive due to the risk associated to exploration. Thus, the agent can only use data previously collected by safe policies. While previous work considers optimizing the average performance using offline data, we focus on optimizing... | ["offline", "reinforcement learning", "risk-averse", "risk sensitive", "robust", "safety", "safe"] | ABSTRACTTraining Reinforcement Learning (RL) agents online in high-stakes applicationsis often prohibitive due to the risk associated with exploration. Thus, the agentcan only use data previously collected by safe policies. While previous workconsiders optimizing the average performance using offline data, we focus ono... | a4hTtmA_K4 | Study the problem of safety in offline RL, and propose a novel algorithm for learning risk-averse policy | 6: Marginally above acceptance threshold | Summary
The paper studies the problem of safe reinforcement learning, where we want to learn risk-averse policies in the offline setting. It proposes “Offline Risk Averse Actor Critic” (ORAAC) which performs competitively as risk-neural agent, and outperforms D4PG based baseline as a risk-averse agent on D4RL benchmark... | 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
Risk-Averse Offline Reinforcement Learning
### Paper Abstract
Training Reinforcement Learning (RL) agents in high-stakes applications might be too prohibitive due to the... |
rsf1z-JSj87 | ICLR.cc/2021/Conference | 2021 | End-to-end Adversarial Text-to-Speech | ["Jeff Donahue", "Sander Dieleman", "Mikolaj Binkowski", "Erich Elsen", "Karen Simonyan"] | Modern text-to-speech synthesis pipelines typically involve multiple processing stages, each of which is designed or learnt independently from the rest. In this work, we take on the challenging task of learning to synthesise speech from normalised text or phonemes in an end-to-end manner, resulting in models which oper... | ["text-to-speech", "speech synthesis", "adversarial", "GAN", "end-to-end", "feed-forward", "generative model"] | ABSTRACTModern text-to-speech synthesis pipelines typically involve multiple processingstages, each of which is designed or learnt independently from the rest. In thiswork, we take on the challenging task of learning to synthesise speech fromnormalised text or phonemes in an end-to-end manner, resulting in models which... | 8i0Rsbolbje | Review | 7: Good paper, accept | This paper presents an approach for end-to-end speech synthesis, where every step is learned jointly with the others. Specifically, the proposed model takes a character sequence as input and outputs an audio signal directly. The model is trained using an combination of losses, including an adversarial loss. In the 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
End-to-end Adversarial Text-to-Speech
### Paper Abstract
Modern text-to-speech synthesis pipelines typically involve multiple processing stages, each of which is designe... |
rygPm64tDH | ICLR.cc/2020/Conference | 2020 | Learning Explainable Models Using Attribution Priors | ["Gabriel Erion", "Joseph D. Janizek", "Pascal Sturmfels", "Scott M. Lundberg", "Su-In Lee"] | Two important topics in deep learning both involve incorporating humans into the modeling process: Model priors transfer information from humans to a model by regularizing the model's parameters; Model attributions transfer information from a model to humans by explaining the model's behavior. Previous work has taken i... | ["Deep Learning", "Interpretability", "Attributions", "Explanations", "Biology", "Health", "Computational Biology"] | ABSTRACTTwo important topics in deep learning both involve incorporating humans into themodeling process: Model priors transfer information from humans to a model byregularizing the model’s parameters; Model attributions transfer information froma model to humans by explaining the model’s behavior. Previous work has ta... | Syer6jjDYH | Official Blind Review #2 | 8: Accept | The authors proposed the attribution priors framework to incorporate human domain knowledge as constraints when training deep neural networks. This is a general framework that the users can define different attribution priors for different tasks. For example, in this work, the authors proposed three reasonable priors f... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Explainable Models Using Attribution Priors
### Paper Abstract
Two important topics in deep learning both involve incorporating humans into the modeling process... | |
JI2TGOehNT0 | ICLR.cc/2021/Conference | 2021 | Combining Imitation and Reinforcement Learning with Free Energy Principle | ["Ryoya Ogishima", "Izumi Karino", "Yasuo Kuniyoshi"] | Imitation Learning (IL) and Reinforcement Learning (RL) from high dimensional sensory inputs are often introduced as separate problems, but a more realistic problem setting is how to merge the techniques so that the agent can reduce exploration costs by partially imitating experts at the same time it maximizes its retu... | ["Imitation", "Reinforcement Learning", "Free Energy Principle"] | ABSTRACTImitation Learning (IL) and Reinforcement Learning (RL) from high dimensionalsensory inputs are often introduced as separate problems, but a more realistic prob-lem setting is how to merge the techniques so that the agent can reduce explorationcosts by partially imitating experts at the same time it maximizes i... | JTk4S_Fv1EY | An interesting paper shows theoretical framework to explain imitation learning and RL with free energy principle | 6: Marginally above acceptance threshold | This manuscript develops RL algorithms that can learn reward from expert data and even exceed the expert performance. It introduces free energy framework that combines ideas of imitation learning and reinforcement learning in a Bayesian probabilistic way. The methodology is well written. It is interesting that they fin... | 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
Combining Imitation and Reinforcement Learning with Free Energy Principle
### Paper Abstract
Imitation Learning (IL) and Reinforcement Learning (RL) from high dimensiona... |
PeT5p3ocagr | ICLR.cc/2021/Conference | 2021 | PGPS : Coupling Policy Gradient with Population-based Search | ["Namyong Kim", "Hyunsuk Baek", "Hayong Shin"] | Gradient-based policy search algorithms (such as PPO, SAC or TD3) in deep reinforcement learning (DRL) have shown successful results on a range of challenging control tasks. However, they often suffer from flat or deceptive gradient problems. As an alternative to policy gradient methods, population-based evolutionary a... | ["Reinforcement Learning", "Population-based Search", "Policy Gradient", "Combining PG with PS"] | ABSTRACTGradient-based policy search algorithms (such as PPO, SAC, or TD3) in deep rein-forcement learning (DRL) have shown successful results on a range of challengingcontrol tasks. However, they often suffer from deceptive gradient problems in flator gentle regions of the objective function. As an alternative to poli... | nVwyhJoRgD5 | Combination of existing algorithmic components, evaluation is not strong enough | 5: Marginally below acceptance threshold | Summary:
Recently, several researchers have been trying to combine the goodnesses of direct policy search approaches (mostly based on evolutionary computation approaches) and those of policy gradient approaches in control tasks. This paper proposes a novel combination of an evolutionary direct policy search and an 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
PGPS : Coupling Policy Gradient with Population-based Search
### Paper Abstract
Gradient-based policy search algorithms (such as PPO, SAC or TD3) in deep reinforcement l... |
Whq-nTgCbNR | ICLR.cc/2021/Conference | 2021 | Anomaly detection in dynamical systems from measured time series | ["Andrei Ivanov", "Anna Golovkina"] | The paper addresses a problem of abnormalities detection in nonlinear processes represented by measured time series. Anomaly detection problem is usually formulated as finding outlier data points relative to some usual signals such as unexpected spikes, drops, or trend changes. In nonlinear dynamical systems, there are... | ["dynamical systems", "polynomial neural networks", "anomaly detection"] | ABSTRACTThe paper addresses a problem of abnormalities detection in nonlinear processesrepresented by measured time series. Anomaly detection problem is usually for-mulated as finding outlier data points relative to some usual signals such as un-expected spikes, drops, or trend changes. In nonlinear dynamical systems, ... | Gl0TTKlnMJE | Official Blind Review #4 | 5: Marginally below acceptance threshold | This paper is exploring anomaly detection problems where a time series does not contain statistical outliers. It shows the proposed polynomial neural network-based approach, which uses the fitted weights as extracted features, outperforms both classical statistical features and an LSTM autoencoder-based neural network.... | 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
Anomaly detection in dynamical systems from measured time series
### Paper Abstract
The paper addresses a problem of abnormalities detection in nonlinear processes repre... |
ey1XXNzcIZS | ICLR.cc/2021/Conference | 2021 | Exploring Routing Strategies for Multilingual Mixture-of-Experts Models | ["Sneha Kudugunta", "Yanping Huang", "Ankur Bapna", "Maxim Krikun", "Dmitry Lepikhin", "Thang Luong", "Orhan Firat"] | Sparsely-Gated Mixture-of-Experts (MoE) has been a successful approach for scaling multilingual translation models to billions of parameters without a proportional increase in training computation. These models, however, are prohibitively large for serving deployment and there is no easy way to extract a sub-network to... | ["Mixture-of-Experts", "Neural Machine Translation", "Multilingual", "Multi-Task Learning", "Conditional Computation", "Natural Language Processing"] | ABSTRACTSparsely-Gated Mixture-of-Experts (MoE) has been a successful approach forscaling multilingual translation models to billions of parameters without a propor-tional increase in training computation. These models, however, are prohibitivelylarge for serving deployment and there is no easy way to extract a sub-net... | FVahvqP-LrJ | Nice observations on task-level routing | 6: Marginally above acceptance threshold | In this paper, the authors explore alternatives to the standard token-based routing in sparsely-gated MoE models for multilingual NMT. This exploration is motivated by the need for efficient inference in MoE models, for which token-based routing is a limitation. The alternative is task-based routing, where examples for... | 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
Exploring Routing Strategies for Multilingual Mixture-of-Experts Models
### Paper Abstract
Sparsely-Gated Mixture-of-Experts (MoE) has been a successful approach for sca... |
bkincnjT8zx | ICLR.cc/2021/Conference | 2021 | Neural Dynamical Systems: Balancing Structure and Flexibility in Physical Prediction | ["Viraj Mehta", "Ian Char", "Willie Neiswanger", "Youngseog Chung", "Andrew Oakleigh Nelson", "Mark D Boyer", "Egemen Kolemen", "Jeff Schneider"] | We introduce Neural Dynamical Systems (NDS), a method of learning dynamical models in various gray-box settings which incorporates prior knowledge in the form of systems of ordinary differential equations. NDS uses neural networks to estimate free parameters of the system, predicts residual terms, and numerically int... | ["nuclear fusion", "physics", "differential equations", "dynamical systems", "control", "dynamics"] | ABSTRACTWe introduce Neural Dynamical Systems (NDS), a method of learning dynamicalmodels in various gray-box settings which incorporates prior knowledge in theform of systems of ordinary differential equations. NDS uses neural networks toestimate free parameters of the system, predicts residual terms, and numericallyi... | 4r10zGcQvxd | interesting application, novelty of the contribution under question | 4: Ok but not good enough - rejection | This work proposes an hybrid framework where the dynamical system inferred in Neural ODE (NODE) is parameterised by two separated components: a component implementing known dynamics provided by a given ODE, and a free component parameterised by a neural network. The rationale behind this approach is to exploit known pr... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Dynamical Systems: Balancing Structure and Flexibility in Physical Prediction
### Paper Abstract
We introduce Neural Dynamical Systems (NDS), a method of learning... |
r1nmx5l0W | ICLR.cc/2018/Conference | 2018 | SIC-GAN: A Self-Improving Collaborative GAN for Decoding Sketch RNNs | ["Chi-Chun Chuang", "Zheng-Xin Weng", "Shan-Hung Wu"] | Variational RNNs are proposed to output “creative” sequences. Ideally, a collection of sequences produced by a variational RNN should be of both high quality and high variety. However, existing decoders for variational RNNs suffer from a trade-off between quality and variety. In this paper, we seek to learn a variation... | ["RNNs", "GANs", "Variational RNNs", "Sketch RNNs"] | ABSTRACTVariational RNNs are proposed to output “creative” sequences. Ideally, a collec-tion of sequences produced by a variational RNN should be of both high qualityand high variety. However, existing decoders for variational RNNs suffer froma trade-off between quality and variety. In this paper, we seek to learn a va... | Skie1qFxM | A novel architecture for generating greater variety on QuickDraw dataset, but seems confused about what it's actually doing. | 4: Ok but not good enough - rejection | This paper baffles me. It appears to be a stochastic RNN with skip connections (so it's conditioned on the last two states rather than last one) trained by an adversarial objective (which is no small feat to make work for sequential tasks) with results shown on the firetruck category of the QuickDraw dataset. Yet the a... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
SIC-GAN: A Self-Improving Collaborative GAN for Decoding Sketch RNNs
### Paper Abstract
Variational RNNs are proposed to output “creative” sequences. Ideally, a collecti... |
04cII6MumYV | ICLR.cc/2021/Conference | 2021 | A Universal Representation Transformer Layer for Few-Shot Image Classification | ["Lu Liu", "William L. Hamilton", "Guodong Long", "Jing Jiang", "Hugo Larochelle"] | Few-shot classification aims to recognize unseen classes when presented with only a small number of samples. We consider the problem of multi-domain few-shot image classification, where unseen classes and examples come from diverse data sources. This problem has seen growing interest and has inspired the development of... | ["image classification", "unseen classes", "problem", "urt", "classification aims", "small number", "samples", "examples", "diverse data sources"] | ABSTRACTFew-shot classification aims to recognize unseen classes when presented with onlya small number of samples. We consider the problem of multi-domain few-shotimage classification, where unseen classes and examples come from diverse datasources. This problem has seen growing interest and has inspired the developme... | X5fspbD-lNp | This paper proposes a transformer based exploitation of multiple domain-specific backbones to achieve better performance across all the domains at hand. | 7: Good paper, accept | The review is brief because of time pressure. However, I have gone through the paper carefully.
Motivation
The paper is well motivated. It is keenly aware of previous work in the field and establishes its advancement of the state of the art clearly. It reviews past work in meta-learning as well as universal representat... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Universal Representation Transformer Layer for Few-Shot Image Classification
### Paper Abstract
Few-shot classification aims to recognize unseen classes when presented... |
r1lFYoRcFm | ICLR.cc/2019/Conference | 2019 | Quantile Regression Reinforcement Learning with State Aligned Vector Rewards | ["Oliver Richter", "Roger Wattenhofer"] | Learning from a scalar reward in continuous action space environments is difficult and often requires millions if not billions of interactions. We introduce state aligned vector rewards, which are easily defined in metric state spaces and allow our deep reinforcement learning agent to tackle the curse of dimensionalit... | ["deep reinforcement learning", "quantile regression", "vector reward"] | ABSTRACTLearning from a scalar reward in continuous action space environments is diffi-cult and often requires millions if not billions of interactions. We introduce statealigned vector rewards, which are easily defined in metric state spaces and allowour deep reinforcement learning agent to tackle the curse of dimensi... | rkxnj29uhQ | Paper is smooth to read, but important issues remian. | 3: Clear rejection | The paper addresses RL in the continuous action space. The scope is right for ICLR. The main two claimed contributions of the paper are the use of a re-parametrised policy and the use of a novel vector-based training objective (similar to solving each component of a factored MDP).
I have the following concerns about 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
Quantile Regression Reinforcement Learning with State Aligned Vector Rewards
### Paper Abstract
Learning from a scalar reward in continuous action space environments is ... |
NxvF-PleYy | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | A Baseline Study of Emphasis Effects in Information Visualization | ["Anonymous"] | Emphasis effects – visual changes that make certain elements more
prominent – are commonly used in information visualization to draw
the user’s attention or to indicate importance. Although theoretical
frameworks of emphasis exist (that link visually diverse emphasis
effects through the idea of visual prominence compar... | ["Human-centered computing", "Visualization", "Visualization techniques", "Perception", "Visualization design and evaluation methods"] | ABSTRACTEmphasis effects – visual changes that make certain elements moreprominent – are commonly used in information visualization to drawthe user’s attention or to indicate importance. Although theoreticalframeworks of emphasis exist (that link visually diverse emphasiseffects through the idea of visual prominence co... | 4oaxWP8lk1 | An interesting but essentially flawed paper | 5: Marginally below acceptance threshold | The authors of this paper carried out two studies on the efficacy of emphasis effects, one basic in assessing the levels of useful differences and one more applied using actual different visualizations for a more ecologically valid investigation. They considered three so-called emphasis effects: blurring the backgroun... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Baseline Study of Emphasis Effects in Information Visualization
### Paper Abstract
Emphasis effects – visual changes that make certain elements more prominent – are co... |
BkbOsNeSM | ICLR.cc/2018/Conference | 2018 | FastNorm: Improving Numerical Stability of Deep Network Training with Efficient Normalization | ["Sadhika Malladi", "Ilya Sharapov"] | We propose a modification to weight normalization techniques that provides the same convergence benefits but requires fewer computational operations. The proposed method, FastNorm, exploits the low-rank properties of weight updates and infers the norms without explicitly calculating them, replacing an $O(n^2)$ computat... | ["Neural networks", "Training", "Convergence"] | ABSTRACTWe propose a modification to weight normalization techniques that provides thesame convergence benefits but requires fewer computational operations. The pro-posed method, FastNorm, exploits the low-rank properties of weight updates andinfers the norms without explicitly calculating them, replacing an O(n2)compu... | HyeZ8SLHG | This authors proposed to use an implicit weight normalization approach to replace the explicit weight normalization used in training of neural networks. The contribution is limited. | 4: Ok but not good enough - rejection | This authors proposed to use an implicit weight normalization approach to replace the explicit weight normalization used in training of neural networks. The authors claimed to obtain efficiency improvement and better numerical stability.
This is a short paper that contains five pages. The idea of the proposed implicit... | 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
FastNorm: Improving Numerical Stability of Deep Network Training with Efficient Normalization
### Paper Abstract
We propose a modification to weight normalization techni... |
SkxcZCNKDS | ICLR.cc/2020/Conference | 2020 | If MaxEnt RL is the Answer, What is the Question? | ["Benjamin Eysenbach", "Sergey Levine"] | Experimentally, it has been observed that humans and animals often make decisions that do not maximize their expected utility, but rather choose outcomes randomly, with probability proportional to expected utility. Probability matching, as this strategy is called, is equivalent to maximum entropy reinforcement learning... | ["reinforcement learning", "maximum entropy", "POMDP"] | ABSTRACTExperimentally, it has been observed that humans and animals often make de-cisions that do not maximize their expected utility, but rather choose outcomesrandomly, with probability proportional to expected utility. Probability matching,as this strategy is called, is equivalent to maximum entropy reinforcement l... | rylwfsfRFH | Official Blind Review #1 | 3: Weak Reject | Summary :
The paper discusses the use of maximum entropy in Reinforcement Learning. Specifically, it relates the solution of the maximum entropy RL problem to the solutions of two different settings, 1) a ‘meta-POMDP’ regret minimization problem and 2) a ‘robust reward control’ problem. Both cases follow with simple ex... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
If MaxEnt RL is the Answer, What is the Question?
### Paper Abstract
Experimentally, it has been observed that humans and animals often make decisions that do not maximi... | |
r1gbP4BqWr | MICCAI.org/2019/Workshop/COMPAY | 2019 | An Interpretable Multi-scale Deep Network for Structure Localization in Medical Images | ["Yongfei Liu", "Bo Wan", "Xuming He"] | Anatomical structure localization plays an important role in image-based medical diagnosis. Despite recent progress achieved by deep learning methods, most existing neural networks for localization lack an interpretable inference process and thus are difficult to use for critical diagnosis in practice.
In this paper, ... | ["Interpretable structure localization", "Coarse-to-fine", "Nuchal Translucency"] | An Interpretable Multi-scale Deep Network forStructure Localization in Medical ImagesYongfei Liu, Bo Wan, and Xuming HeShanghaiTech University, Shanghai Chinafliuyf3, wanbo, hexm g@shanghaitech.edu.cnAbstract. Anatomical structure localization plays an important rolein image-based medical diagnosis. Despite recent prog... | SJe920Af4H | Review | 6: Marginally above acceptance threshold | The paper proposes a deep two-branch architecture with better interpretability for structure localization and evaluates it on a private medical dataset.
The contribution is of interest, since interpretability is especially important in the medical domain.
The paper can be improved by
- explaining the local and contex... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
An Interpretable Multi-scale Deep Network for Structure Localization in Medical Images
### Paper Abstract
Anatomical structure localization plays an important role in im... |
RCGBA1i5MF | ICLR.cc/2021/Conference | 2021 | The Unreasonable Effectiveness of the Class-reversed Sampling in Tail Sample Memorization | ["Benyi Hu", "Chi Zhang", "Yuehu Liu", "Le Wang", "Li Liu"] | Long-tailed visual recognition poses significant challenges to traditional machine learning and emerging deep networks due to its inherent class imbalance. A common belief is that tail classes with few samples cannot exhibit enough regularity for pattern extraction. What makes things worse, the limited cardinality may ... | ["long-tailed learning", "re-sampling", "sample memorization"] | ABSTRACTLong-tailed visual recognition poses significant challenges to traditional machinelearning and emerging deep networks due to its inherent class imbalance. A com-mon belief is that tail classes with few samples cannot exhibit enough regularityfor pattern extraction. What makes things worse, the limited cardinali... | zQmq228Rp-p | Official Blind Review #3 | 5: Marginally below acceptance threshold | Summary:
This paper aims to address the problem of long-tail classification by revising the data resampling strategy. The main contribution is that it considers the representation trade-off between the head and tail classes. To this end, it combines instance-balanced sampling and class-reversed sampling, by first cond... | 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 Unreasonable Effectiveness of the Class-reversed Sampling in Tail Sample Memorization
### Paper Abstract
Long-tailed visual recognition poses significant challenges ... |
-6vS_4Kfz0 | ICLR.cc/2021/Conference | 2021 | Optimizing Memory Placement using Evolutionary Graph Reinforcement Learning | ["Shauharda Khadka", "Estelle Aflalo", "Mattias Marder", "Avrech Ben-David", "Santiago Miret", "Shie Mannor", "Tamir Hazan", "Hanlin Tang", "Somdeb Majumdar"] | For deep neural network accelerators, memory movement is both energetically expensive and can bound computation. Therefore, optimal mapping of tensors to memory hierarchies is critical to performance. The growing complexity of neural networks calls for automated memory mapping instead of manual heuristic approaches; ye... | ["Reinforcement Learning", "Memory Mapping", "Device Placement", "Evolutionary Algorithms"] | ABSTRACTFor deep neural network accelerators, memory movement is both energeticallyexpensive and can bound computation. Therefore, optimal mapping of tensorsto memory hierarchies is critical to performance. The growing complexity ofneural networks calls for automated memory mapping instead of manual heuristicapproaches... | dCCxpALWJ4 | Interesting extension of the Graph Optimization using DRL line of work | 7: Good paper, accept | Optimizing the execution of deep neural networks has tremendous impact on the cost and performance in many industries due to the proliferation of "Deep Learning". There has recently been an interesting line of work of using learning to optimize policies related to placement and scheduling of the neural network computat... | 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
Optimizing Memory Placement using Evolutionary Graph Reinforcement Learning
### Paper Abstract
For deep neural network accelerators, memory movement is both energeticall... |
BJeAHkrYDS | ICLR.cc/2020/Conference | 2020 | Fast Task Inference with Variational Intrinsic Successor Features | ["Steven Hansen", "Will Dabney", "Andre Barreto", "David Warde-Farley", "Tom Van de Wiele", "Volodymyr Mnih"] | It has been established that diverse behaviors spanning the controllable subspace of a Markov decision process can be trained by rewarding a policy for being distinguishable from other policies. However, one limitation of this formulation is the difficulty to generalize beyond the finite set of behaviors being explicit... | ["Reinforcement Learning", "Variational Intrinsic Control", "Successor Features"] | Published as a conference paper at ICLR 2020Fast Task Inference withVariational Intrinsic Successor FeaturesSteven HansenDeepMindstevenhansen@google.comWill DabneyDeepMindwdabney@google.comAndré BarretoDeepMindandrebarreto@google.comDavid Warde-FarleyDeepMinddwf@google.comTom Van de WieleDeepMind (Former)tvdwiele@gmai... | rygRKCvOtr | Official Blind Review #2 | 8: Accept | The paper builds upon the idea of Successor Features; this is a principle used to facilitate generalization beyond the finite set of behaviors being explicitly learned by an MDP. The proposed paper ameliorates the need of defining the reward function as linear in some grounded feature space by resorting to variational ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Fast Task Inference with Variational Intrinsic Successor Features
### Paper Abstract
It has been established that diverse behaviors spanning the controllable subspace of... | |
SlrqM9_lyju | ICLR.cc/2021/Conference | 2021 | AutoLRS: Automatic Learning-Rate Schedule by Bayesian Optimization on the Fly | ["Yuchen Jin", "Tianyi Zhou", "Liangyu Zhao", "Yibo Zhu", "Chuanxiong Guo", "Marco Canini", "Arvind Krishnamurthy"] | The learning rate (LR) schedule is one of the most important hyper-parameters needing careful tuning in training DNNs. However, it is also one of the least automated parts of machine learning systems and usually costs significant manual effort and computing. Though there are pre-defined LR schedules and optimizers with... | ["steps", "autolrs", "lr schedules", "schedule", "bayesian optimization", "automatic", "dnns", "validation loss"] | ABSTRACTThe learning rate (LR) schedule is one of the most important hyper-parametersneeding careful tuning in training DNNs. However, it is also one of the least auto-mated parts of machine learning systems and usually costs significant manual effortand computing. Though there are pre-defined LR schedules and optimize... | 0q5Le39isi6 | Review | 5: Marginally below acceptance threshold | The paper presents an algorithm (AutoLRS) to tune learning rate schedules based on Bayesian Optimization with Gaussian processes and loss curve forecasting with exponential functions. The authors conduct experiments on three large scales settings in computer vision and natural language processing.
Strengths:
- This is... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
AutoLRS: Automatic Learning-Rate Schedule by Bayesian Optimization on the Fly
### Paper Abstract
The learning rate (LR) schedule is one of the most important hyper-param... |
fn5K7VfI3MV | AKBC.ws/2021/Workshop/CSKB | 2021 | No Need to Know Everything! Efficiently Augmenting Language Models With External Knowledge | ["Jivat Neet Kaur", "Sumit Bhatia", "Milan Aggarwal", "Rachit Bansal", "Balaji Krishnamurthy"] | Large transformer-based pre-trained language models have achieved impressive performance on a variety of knowledge-intensive tasks and can capture semantic, syntactic, and factual knowledge in their parameters. However, storing large amounts of factual knowledge in the parameters of the model is sub-optimal given the r... | ["Language models", "knowledge bases", "structured knowledge", "knowledge probes"] | Workshop on Commonsense Reasoning and Knowledge Bases (CSKB) at AKBC (2021) Workshop paperNo Need to Know Everything!Efficiently Augmenting Language Models With External KnowledgeJivat Neet Kaur1JIVATNEET @GMAIL .COMSumit Bhatia2SUMIT .BHATIA @ADOBE .COMMilan Aggarwal2MILAGGAR @ADOBE .COMRachit Bansal3RACBANSA @GMAIL .... | mQYFIzlKtRM | Very well presented , clear and nice work | 7: Good paper, accept | Even though large transformer-based pre-trained language models memorize factual knowledge storing large amounts of it in the parameters of the model is sub-optimal given the resource requirements and ever-growing amounts of knowledge. Instead of packing all the knowledge in the model parameters, the authors present an... | 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
No Need to Know Everything! Efficiently Augmenting Language Models With External Knowledge
### Paper Abstract
Large transformer-based pre-trained language models have ac... |
IgXOXUVObLB | NeurIPS.cc/2022/Workshop/TEA | 2022 | Characterising the Robustness of Reinforcement Learning for Continuous Control using Disturbance Injection | ["Catherine Glossop", "Jacopo Panerati", "Amrit Krishnan", "Zhaocong Yuan", "Angela P. Schoellig"] | In this study, we leverage the deliberate and systematic fault-injection capabilities of an open-source benchmark suite to perform a series of experiments on state-of-the-art deep and robust reinforcement learning algorithms. We aim to benchmark robustness in the context of continuous action spaces---crucial for deploy... | ["reinforcement learning", "robust control", "continuous control", "robotics", "safety", "benchmarks"] | Characterising the Robustness of Reinforcement Learning forContinuous Control using Disturbance InjectionCatherine R. Glossop∗Department of Engineering ScienceUniversity of Torontocatherine.glossop@robotics.utias.utoronto.caJacopo PaneratiInstitute for Aerospace StudiesUniversity of Torontojacopo.panerati@utoronto.caAm... | 6QgQH5NeRD | Useful comparison of state-of-the-art RL methods with injected disturbances; however, results are in simple settings | 6: Marginally above acceptance threshold | **Summary:** The authors test the robustness and safety of a set of RL methods on the standard cart-pole problem by injecting various kinds of noise to the observation, action, and dynamics. The authors show that, on cart-pole, the robust RL agents do not perform significantly better than the standard RL agents and 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
Characterising the Robustness of Reinforcement Learning for Continuous Control using Disturbance Injection
### Paper Abstract
In this study, we leverage the deliberate a... |
Al01q787m71 | NeurIPS.cc/2022/Workshop/GMML | 2022 | Learning to count visual objects by combining "what" and "where" in recurrent memory | ["Jessica A F Thompson", "Hannah Sheahan", "Christopher Summerfield"] | Counting the number of objects in a visual scene is easy for humans but challenging for modern deep neural networks. Here we explore what makes this problem hard and study the neural computations that allow transfer of counting ability to new objects and contexts. Previous work has implicated posterior parietal cortex ... | ["numerosity", "counting", "gaze", "attention", "RNN", "dorsal stream"] | Proceedings of Machine Learning Research 1:1–20, 2023 NeurIPS 2022 Gaze Meets ML WorkshopLearning to count visual objects by combining “what” and“where” in recurrent memoryJessica A.F. Thompson jessica.thompson@psy.ox.ac.ukHannah Sheahan sheahan.hannah@gmail.comChristopher Summerfield christopher.summerfield@psy.ox.ac.... | ZWFQsyvoxpc | Too simplistic | 3: Clear rejection | The authors propose to use a two input recurrent neural network to count the number of objects in an image. For this, they use a sequence of randomly selected bounding boxes as one input and the coordinates of the bounding boxes as another input.They conclude that the use of these two sources of information as "what" (... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to count visual objects by combining "what" and "where" in recurrent memory
### Paper Abstract
Counting the number of objects in a visual scene is easy for huma... |
gnAGreyEzP2 | ICLR.cc/2021/Conference | 2021 | Recurrently Controlling a Recurrent Network with Recurrent Networks Controlled by More Recurrent Networks | ["Yi Tay", "Yikang Shen", "Alvin Chan", "Aston Zhang", "Shuai Zhang"] | This paper explores an intriguing idea of recursively parameterizing recurrent nets. Simply speaking, this refers to recurrently controlling a recurrent network with recurrent networks controlled by recurrent networks. The proposed architecture recursively parameterizes its gating functions whereby gating mechanisms of... | ["Deep Learning", "Recurrent Neural Networks"] | ABSTRACTThis paper explores an intriguing idea of recursively parameterizing recurrent nets.Simply speaking, this refers to recurrently controlling a recurrent network withrecurrent networks controlled by recurrent networks. The proposed architecturerecursively parameterizes its gating functions whereby gating mechanis... | nrs2IZsE1Ip | missing theoretical or empirical justification for the improvements | 5: Marginally below acceptance threshold |
The main topic of this paper is the enhancement of recurrent networks for improving the task performance.
The proposed method combines the idea of external meta-controllers with recursive architectures. As a result, recurrent networks can be controlled by themselves by selecting the gate functions.
The authors show 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
Recurrently Controlling a Recurrent Network with Recurrent Networks Controlled by More Recurrent Networks
### Paper Abstract
This paper explores an intriguing idea of re... |
0Zxk3ynq7jE | ICLR.cc/2021/Conference | 2021 | An Empirical Exploration of Open-Set Recognition via Lightweight Statistical Pipelines | ["Shu Kong", "Deva Ramanan"] | Machine-learned safety-critical systems need to be self-aware and reliably know their unknowns in the open-world. This is often explored through the lens of anomaly/outlier detection or out-of-distribution modeling. One popular formulation is that of open-set classification, where an image classifier trained for 1-of-$... | ["open-set recognition", "anomaly detection", "statistical models", "Gaussian Mixture Models", "open-world image classification", "open-world semantic segmentation"] | ABSTRACTMachine-learned safety-critical systems need to be self-aware and reliably knowtheir unknowns in the open-world. This is often explored through the lens ofanomaly/outlier detection or out-of-distribution modeling. One popular formula-tion is that of open-set classification, where an image classifier trained for... | FtJVUjKewGi | Good Empirical Study Falls Short | 4: Ok but not good enough - rejection | ## Summary and Contributions
The paper presents an empirical study of the open set image recognition (OSR) problem. It considers two classification setups -- single-domain and cross-domain, and, one semantic segmentation setup. In particular, the paper evaluates simple statistical models like the Nearest Class Means (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
An Empirical Exploration of Open-Set Recognition via Lightweight Statistical Pipelines
### Paper Abstract
Machine-learned safety-critical systems need to be self-aware a... |
Im8G9R1boQi | logconference.io/LOG/2022/Conference | 2022 | A Survey on Deep Graph Generation: Methods and Applications | ["Yanqiao Zhu", "Yuanqi Du", "Yinkai Wang", "Yichen Xu", "Jieyu Zhang", "Qiang Liu", "Shu Wu"] | Graphs are ubiquitous in encoding relational information of real-world objects in many domains. Graph generation, whose purpose is to generate new graphs from a distribution similar to the observed graphs, has received increasing attention thanks to the recent advances of deep learning models. In this paper, we conduct... | ["deep graph generation", "methods", "survey", "applications", "applications graphs", "ubiquitous", "relational information", "objects", "many domains", "graph generation"] | A Survey on Deep Graph Generation:Methods and ApplicationsYanqiao Zhu∗UCLAyzhu@cs.ucla.eduYuanqi Du∗Cornellyd392@cornell.eduYinkai Wang∗Tuftsyinkai.wang@tufts.eduYichen XuEPFLyichen.xu@epfl.chJieyu ZhangUWjieyuz2@cs.washington.eduQiang LiuCASIAqiang.liu@nlpr.ia.ac.cnShu WuCASIAshu.wu@nlpr.ia.ac.cnAbstractGraphs are ubi... | 0MdswnVXic | > (1) Summary
This paper provides a survey on graph generation, including a summary of methods and applications. This survey presents two mainstream methods called latent variable approaches and reinforcement learning approaches and focuses more on the first one. The authors summarize latent variable approaches into a... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Survey on Deep Graph Generation: Methods and Applications
### Paper Abstract
Graphs are ubiquitous in encoding relational information of real-world objects in many dom... | |||
h0de3QWtGG | ICLR.cc/2021/Conference | 2021 | Learning "What-if" Explanations for Sequential Decision-Making | ["Ioana Bica", "Daniel Jarrett", "Alihan H\u00fcy\u00fck", "Mihaela van der Schaar"] | Building interpretable parameterizations of real-world decision-making on the basis of demonstrated behavior--i.e. trajectories of observations and actions made by an expert maximizing some unknown reward function--is essential for introspecting and auditing policies in different institutions. In this paper, we propose... | ["counterfactuals", "explaining decision-making", "preference learning"] | ABSTRACTBuilding interpretable parameterizations of real-world decision-making on the basisof demonstrated behavior—i.e. trajectories of observations and actions made by anexpert maximizing some unknown reward function—is essential for introspectingand auditing policies in different institutions. In this paper, we prop... | VvmiBePAuM7 | Review for Learning What If Explanations for Sequential Decision Making | 6: Marginally above acceptance threshold | ## Summary
The authors address the problem of comparing the effects of different treatments in medical decision making. They formulate this as learning a reward function via inverse reinforcement learning in a batch setting, where the only admissible data is in the form of medical records. Their goal is to develop ... | 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
Learning "What-if" Explanations for Sequential Decision-Making
### Paper Abstract
Building interpretable parameterizations of real-world decision-making on the basis of ... |
MpStQoD73Mj | ICLR.cc/2021/Conference | 2021 | Differentiable Weighted Finite-State Transducers | ["Awni Hannun", "Vineel Pratap", "Jacob Kahn", "Wei-Ning Hsu"] | We introduce a framework for automatic differentiation with weighted finite-state transducers (WFSTs) allowing them to be used dynamically at training time. Through the separation of graphs from operations on graphs, this framework enables the exploration of new structured loss functions which in turn eases the encodin... | ["weighted automata", "automatic differentiation", "sequence models"] | ABSTRACTWe introduce a framework for automatic differentiation with weighted finite-statetransducers (WFSTs) allowing them to be used dynamically at training time.Through the separation of graphs from operations on graphs, this framework en-ables the exploration of new structured loss functions which in turn eases the ... | YoJJ5jbi48W | Interesting software with limited scientific novelty | 4: Ok but not good enough - rejection | This is a paper describing a software package for common calculations on WFSTs (like forward score or best path score), in a way that they are differentiable w.r.t. the arc weights.
It also contains several examples of WFSTs and WFSAs specifically for speech recognition or handwriting recognition, like for the CTC or ... | 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
Differentiable Weighted Finite-State Transducers
### Paper Abstract
We introduce a framework for automatic differentiation with weighted finite-state transducers (WFSTs)... |
SkgE8sRcK7 | ICLR.cc/2019/Conference | 2019 | Sample Efficient Deep Neuroevolution in Low Dimensional Latent Space | ["Bin Zhou", "Jiashi Feng"] | Current deep neuroevolution models are usually trained in a large parameter search space for complex learning tasks, e.g. playing video games, which needs billions of samples and thousands of search steps to obtain significant performance. This raises a question of whether we can make use of sequential data generated d... | ["Neuroevolution", "Reinforcement Learning"] | ABSTRACTCurrent deep neuroevolution models are usually trained in a large parametersearch space for complex learning tasks, e.g. playing video games, which needsbillions of samples and thousands of search steps to obtain significant perfor-mance. This raises a question of whether we can make use of sequential datagener... | B1ehYPDtn7 | Review for "Sample Efficient Deep Neuroevolution in Low Dimensional Latent Space" | 5: Marginally below acceptance threshold | Summary: They propose "Sample Efficient Deep Neuroevolution" (SEDN) model and experiment on Atari games. In this model, they use a Variational Encoder (VAE) to encode state frames into a latent vector, and use an LSTM to encode the current latent vector and action to predict the next latent space. A policy network (tra... | 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
Sample Efficient Deep Neuroevolution in Low Dimensional Latent Space
### Paper Abstract
Current deep neuroevolution models are usually trained in a large parameter searc... |
enoVQWLsfyL | ICLR.cc/2021/Conference | 2021 | Viewmaker Networks: Learning Views for Unsupervised Representation Learning | ["Alex Tamkin", "Mike Wu", "Noah Goodman"] | Many recent methods for unsupervised representation learning train models to be invariant to different "views," or distorted versions of an input. However, designing these views requires considerable trial and error by human experts, hindering widespread adoption of unsupervised representation learning methods across d... | ["unsupervised learning", "self-supervised", "representation learning", "contrastive learning", "views", "data augmentation"] | ABSTRACTMany recent methods for unsupervised representation learning train models to beinvariant to different “views,” or distorted versions of an input. However, design-ing these views requires considerable trial and error by human experts, hinderingwidespread adoption of unsupervised representation learning methods a... | gI17l27mIZJ | A well-written paper on a new generative model to generate data for contrastive learning, with some deficiency in experiments | 7: Good paper, accept | The paper presents a generative model to automatically generate data that is needed for contrastive learning, with a focus on the SimCLR framework, while the method itself is general. Experiments were conducted across multiple modalities, including image, speech and wearable sensor data. The results demonstrate the eff... | 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
Viewmaker Networks: Learning Views for Unsupervised Representation Learning
### Paper Abstract
Many recent methods for unsupervised representation learning train models ... |
V8jrrnwGbuc | ICLR.cc/2021/Conference | 2021 | On the geometry of generalization and memorization in deep neural networks | ["Cory Stephenson", "suchismita padhy", "Abhinav Ganesh", "Yue Hui", "Hanlin Tang", "SueYeon Chung"] | Understanding how large neural networks avoid memorizing training data is key to explaining their high generalization performance. To examine the structure of when and where memorization occurs in a deep network, we use a recently developed replica-based mean field theoretic geometric analysis method. We find that all ... | ["deep learning theory", "representation learning", "statistical physics methods", "double descent"] | ABSTRACTUnderstanding how large neural networks avoid memorizing training data is keyto explaining their high generalization performance. To examine the structureof when and where memorization occurs in a deep network, we use a recentlydeveloped replica-based mean field theoretic geometric analysis method. We findthat ... | hXWN4UqX-yh | Review | 7: Good paper, accept | This paper analyses memorization in DNNs, from the lens of memorization = fitting random labels, and finds that it seems to happen in later layers. These results are obtained using the MFTMA framework, a manifold analysis tool, testing geometric properties of individual layers. The analysis also attempts to explain why... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the geometry of generalization and memorization in deep neural networks
### Paper Abstract
Understanding how large neural networks avoid memorizing training data is k... |
JsRdc90lpws | NeurIPS.cc/2020/Workshop/DL-IG | 2020 | Comparing Fisher Information Regularization with Distillation for DNN Quantization | ["Prad Kadambi", "Karthikeyan Natesan Ramamurthy", "Visar Berisha"] | A large body of work addresses deep neural network (DNN) quantization and pruning to mitigate the high computational burden of deploying DNNs. We analyze two prominent classes of methods; the first class uses regularization based on the Fisher Information Matrix (FIM) of parameters, whereas the other uses a student-tea... | ["Fisher Information", "Knowledge distillation", "network quantization", "loss landscape"] | Comparing Fisher Information Regularization withDistillation for DNN QuantizationPrad KadambiArizona State Universitypadambi@asu.eduKarthikeyan Natesan RamamurthyIBM Researchknatesa@us.ibm.comVisar BerishaArizona State Universityvisar@asu.eduAbstractA large body of work addresses deep neural network (DNN) quantization ... | sYfme_XMU4M | Review of "Comparing Fisher Information Regularization with Distillation for DNN Quantization" | 7: Good paper, accept | This paper compares the loss landscape for two different schemes to quantizing neural networks. Although I wasn't very familiar with work in this area, I found the introduction very clear and appreciated the thoroughness of the related work.
Eq. 6 seems intuitive, in the high temperature limit we get a uniform predic... | 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
Comparing Fisher Information Regularization with Distillation for DNN Quantization
### Paper Abstract
A large body of work addresses deep neural network (DNN) quantizati... |
asLT0W1w7Li | ICLR.cc/2021/Conference | 2021 | Efficient Exploration for Model-based Reinforcement Learning with Continuous States and Actions | ["Ying Fan", "Yifei Ming"] | Balancing exploration and exploitation is crucial in reinforcement learning (RL). In this paper, we study the model-based posterior sampling algorithm in continuous state-action spaces theoretically and empirically. First, we improve the regret bound: with the assumption that reward and transition functions can be mode... | ["Model-based reinforcement learning", "posterior sampling", "Bayesian reinforcement learning"] | ABSTRACTBalancing exploration and exploitation is crucial in reinforcement learning (RL). Inthis paper, we study the model-based posterior sampling algorithm in continuousstate-action spaces theoretically and empirically. First, we improve the regretbound: with the assumption that reward and transition functions can be... | UaLb_g15XwI | Restricted contribution with no fundamental limit | 5: Marginally below acceptance threshold | The paper proposes a model-based posterior sampling algorithms with regret guarantees when the model is assumed to be drawn from a distribution randomly. The authors also provide numerical evaluations of the proposed method.
- The contribution of this work as theoretical work is limited. There is no study on fundament... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Efficient Exploration for Model-based Reinforcement Learning with Continuous States and Actions
### Paper Abstract
Balancing exploration and exploitation is crucial in r... |
rylDfnCqF7 | ICLR.cc/2019/Conference | 2019 | Lagging Inference Networks and Posterior Collapse in Variational Autoencoders | ["Junxian He", "Daniel Spokoyny", "Graham Neubig", "Taylor Berg-Kirkpatrick"] | The variational autoencoder (VAE) is a popular combination of deep latent variable model and accompanying variational learning technique. By using a neural inference network to approximate the model's posterior on latent variables, VAEs efficiently parameterize a lower bound on marginal data likelihood that can be opti... | ["variational autoencoders", "posterior collapse", "generative models"] | ABSTRACTThe variational autoencoder (V AE) is a popular combination of deep latent vari-able model and accompanying variational learning technique. By using a neuralinference network to approximate the model’s posterior on latent variables, V AEsefficiently parameterize a lower bound on marginal data likelihood that ca... | Byl9JZ0u27 | Neither the objective nor the model but the optimization procedure may be the key to training VAE | 8: Top 50% of accepted papers, clear accept | General:
The paper tackles one of the most important problems of learning VAEs, namely, the posterior collapse. Typically, this problem is attacked by either proposing a new model or modifying the objective. Interestingly, the authors considered a third option, i.e., changing the training procedure only, leaving the mo... | 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
Lagging Inference Networks and Posterior Collapse in Variational Autoencoders
### Paper Abstract
The variational autoencoder (VAE) is a popular combination of deep laten... |
H1xX5TkseQ | ICML.cc/2018/Workshop/NAMPI | 2018 | Towards Neural Theorem Proving at Scale | ["Anonymous"] | Neural models combining representation learning and reasoning in an end-to-end trainable manner are receiving increasing interest. However, their use is severely limited by their computational complexity, which renders them unusable on real world datasets. We focus on the Neural Theorem Prover model proposed by Rocktäs... | ["Neural Theorem Provers", "Approximate Nearest Neighbours", "Neural-Symbolic Integration", "Program Induction"] | 000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044045046047048049050051052053054Towards Neural Theorem Proving at ScaleAnonymous Authors1AbstractNeural models combining representation learningand reasoning in an end-to-end trainable ma... | BylT5S1g-X | Nice trick to avoid complexity blowup | 7: Good paper, accept | The paper describes an extension to the Neural Theorem Proving work that allows scaling to substantially larger knowledge bases. The core insight is to use an approximate nearest neighbour search (in embedding space) to filter the set of facts that are used in the full logic programming-like routine. By aggressively fi... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Neural Theorem Proving at Scale
### Paper Abstract
Neural models combining representation learning and reasoning in an end-to-end trainable manner are receiving ... |
BJxZ3ZH1-S | MICCAI.org/2019/Workshop/COMPAY | 2019 | Texture-based classification of confocal laser endomicroscopy images for Barrett’s esophagus surveillance | ["Giacomo Nardi", "Marzieh Kohandani Tafreshi", "Jessie Mah\u00e9", "Nicholas Ayache", "Fran\u00e7ois Lacombe"] | Barrett’s esophagus is a complication of gastroesophageal reflux diseases that generates a transformation of esophagus epithelium turning into adenocarcinoma with a high risk. The surveillance of the changes in the esophageal mucosa is primordial to estimate the cancer progression. Confocal laser endomicroscopy is a no... | ["Barrett\u2019s esophagus", "confocal laser endomicroscopy", "texture analysis", "fractal local density", "image classification"] | Texture-based classication of confocal laserendomicroscopy images for Barrett's esophagussurveillanceGiacomo Nardi1, Marzieh Kohandani Tafreshi1, Jessie Mah e1, NicholasAyache2, and Fran cois Lacombe11Mauna Kea Technologies, 9 Rue d'Enghien, 75010 Paris, Francefgiacomo, marzieh, jessie, francois g@maunakeatech.com2INRI... | rkxMGscc7H | Interesting in vivo real-time histological data collected by confocal endomicroscopy, classified by fractal features. Several gaps in the methodology description as well as poor evaluation in relation to other methods makes the study less interesting. | 5: Marginally below acceptance threshold | Concerns:
The authors confuse classification and segmentation, where the paper is about image classification while references focus on segmentation (which is a different problem and would need a different type of ground truth and evaluation).
The authors make an incomplete comparison to previous methods; presenting th... | 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
Texture-based classification of confocal laser endomicroscopy images for Barrett’s esophagus surveillance
### Paper Abstract
Barrett’s esophagus is a complication of gas... |
dHOb7BE-uxY | eswc-conferences.org/ESWC/2021/Workshop/SeDiT | 2021 | Semantic Digital Twins for Organizational Development | ["Alfonso Diez Rubio", "Juan de Lara"] | Semantic-based Digital Twins can be applied to the enactment of the digital counterparts of enterprises and organizations, and not only for physical devices or machines. Digital Twins for Organizations have a profound impact in the architecture of corporate management platforms, in the design of business processes and ... | ["Digital Twins for Organizations", "Digital Twins", "Enactive Systems"] | Semantic Digital Twins for Organizational Development Alfonso Díez1 and Juan de Lara2 1 UGROUND GLOBAL, Madrid, Spain 2 Universidad Autónoma de Madrid, Madrid, Spain Abstract. Semantic -based Digital Twins can be applied to the enact ment of the digital counterparts of enterprises and organizations, and not only for ... | 0Mx3xFZo8IR | Nice theoretical argumentation for DTOs | 7: Good paper, accept | The paper summarized authors experience in the creation of DT for organization based on semantic modeling and model-driven engineering. The paper introduces the notion of DTOs including a elaborated theoretical framework for their definition.
Even though the paper stays at a general level, it represents a very good ... | 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
Semantic Digital Twins for Organizational Development
### Paper Abstract
Semantic-based Digital Twins can be applied to the enactment of the digital counterparts of ente... |
B1xgWpEKvH | ICLR.cc/2020/Conference | 2020 | ADASAMPLE: ADAPTIVE SAMPLING OF HARD POSITIVES FOR DESCRIPTOR LEARNING | ["Xin-Yu Zhang", "Jia-Wang Bian", "Le Zhang", "Zao-Yi Zheng", "Yun Liu", "Ming-Ming Cheng", "Ian Reid"] | Triplet loss is commonly used in descriptor learning, where the performance heavily relies on mining triplets. Typical solution to that is first picking pairs of intra-class patches (positives) from the dataset to form batches, and then selecting in-batch negatives to construct triplets. For high-informativeness triple... | ["Descriptor", "Correspondence"] | ABSTRACTTriplet loss is commonly used in descriptor learning, where the performance heav-ily relies on mining triplets. Typical solution to that is first picking pairs of intra-class patches (positives) from the dataset to form batches, and then selecting in-batch negatives to construct triplets. For high-informativene... | HyeqUgyCur | Official Blind Review #3 | 3: Weak Reject | This paper presents a method for adaptively selecting the positive samples in triplet loss. Based on the assumption that the norm of the gradient of the loss correlates with the informativeness of each sample, the authors theoretically motivate and propose a method to adaptively select the positive examples. On an eval... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ADASAMPLE: ADAPTIVE SAMPLING OF HARD POSITIVES FOR DESCRIPTOR LEARNING
### Paper Abstract
Triplet loss is commonly used in descriptor learning, where the performance hea... | |
zd32pf6RLng | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | Ontology for Informatics Research Artifacts | ["Viet Bach Nguyen", "Vojt\u011bch Sv\u00e1tek"] | The IRAO ontology, as a new contribution to the network of ontologies for the scholarly domain, aims to model the most tangible aspect of research in computing disciplines – the research artifacts. It consists of parts focusing on the concepts of researcher, research artifact classification, research artifact meta info... | ["ontology", "artifact", "resource", "thesis", "NeOn", "competency questions"] | Ontology for Informatics Research Artifacts?Viet Bach Nguyen and Vojtěch SvátekDepartment of Information and Knowledge EngineeringPrague University of Economics and Business, Czech Republic{nguv03,svatek}@vse.czAbstract. The IRAO ontology, as a new contribution to the networkof ontologies for the scholarly domain, ai... | gzgnwk_5A5p | The validation is not well explained, but the ontology meets the requirements | 6: Marginally above acceptance threshold | The authors' previous studio revealed that current ontologies for research domains are not enough for modeling artifacts (e.g., benchmark datasets, ontologies, software, etc.). With this motivation, this paper proposes the ontology IRAO for modeling artifacts and their researcher environments. The authors created a set... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Ontology for Informatics Research Artifacts
### Paper Abstract
The IRAO ontology, as a new contribution to the network of ontologies for the scholarly domain, aims to mo... |
rJgE13soM | MIDL.amsterdam/2018/Conference | 2018 | MILD-Net: Minimal Information Loss Dilated Network for Gland Instance Segmentation in Colon Histology Images | ["Simon Graham", "Hao Chen", "Qi Dou", "Pheng-Ann Heng", "Nasir Rajpoot"] | The analysis of glandular morphology within colon histopathology images is a crucial step in determining the stage of colon cancer. Despite the importance of this task, manual segmentation is laborious, time-consuming and can suffer from subjectivity among pathologists. The rise of computational pathology has led to th... | ["Computational Pathology", "Gland Segmentation", "Deep Learning"] | MILD-Net: Minimal Information Loss DilatedNetwork for Gland Instance Segmentation in ColonHistology ImagesSimon Graham1;2Hao Chen3Qi Dou3Pheng Ann-Heng3Nasir M. Rajpoot2;4;51Mathematics for Real-World Systems Centre for Doctoral Training, University of Warwick, UK2Department of Computer Science, University of Warwick, ... | B14f1VWRM | Well designed study with a novel architecture | 4: Top 50% of accepted papers, clear accept | This paper proposes a novel deep learning architecture and applies that method to the task of colon gland segmentation. The presented method is compared to numerous state of the art approaches and is shown to outperform them. The paper is well written with descriptive and clear figures. I recommend acceptance of this s... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
MILD-Net: Minimal Information Loss Dilated Network for Gland Instance Segmentation in Colon Histology Images
### Paper Abstract
The analysis of glandular morphology with... |
RgDq8-AwvtN | ICLR.cc/2021/Conference | 2021 | Model-Based Robust Deep Learning: Generalizing to Natural, Out-of-Distribution Data | ["Alexander Robey", "Hamed Hassani", "George J. Pappas"] | While deep learning (DL) has resulted in major breakthroughs in many applications, the frameworks commonly used in DL remain fragile to seemingly innocuous changes in the data. In response, adversarial training has emerged as a principled approach for improving the robustness of DL against norm-bounded perturbations. ... | ["robustness", "out-of-distribution generalization", "natural variation", "deep learning"] | ABSTRACTWhile deep learning (DL) has resulted in major breakthroughs in many appli-cations, the frameworks commonly used in DL remain fragile to seemingly in-nocuous changes in the data. In response, adversarial training has emerged as aprincipled approach for improving the robustness of DL against norm-boundedperturba... | uN7_9dG4MVD | Unclear novelty and very similar to a broken robustness method ( defense ) | 5: Marginally below acceptance threshold | The paper looks at the idea of building models of natural variation of an input and then using these models to develop robust training algorithms that are less susceptible to outliers ( testable for worst-cases via adversarial attacks ).
Strength:
+ The paper addresses a very important topic of adversarial robustnes... | 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
Model-Based Robust Deep Learning: Generalizing to Natural, Out-of-Distribution Data
### Paper Abstract
While deep learning (DL) has resulted in major breakthroughs in ma... |
r1D4bs1Wz | ICLR.cc/2018/Conference | 2018 | Dense Transformer Networks | ["Jun Li", "Yongjun Chen", "Lei Cai", "Ian Davidson", "Shuiwang Ji"] | The key idea of current deep learning methods for dense prediction
is to apply a model on a regular patch centered on each pixel to
make pixel-wise predictions. These methods are limited in the sense
that the patches are determined by network architecture instead of
learned from data. In this work, we propose the dense... | ["dense transformer networks", "patches", "data", "dense prediction", "methods", "work", "shapes", "sizes", "dense transformer modules", "transformer networks"] | ABSTRACTThe key idea of current deep learning methods for dense prediction is to apply amodel on a regular patch centered on each pixel to make pixel-wise predictions.These methods are limited in the sense that the patches are determined by networkarchitecture instead of learned from data. In this work, we propose the ... | r1bFVm1fz | Review for Dense Transformer Networks | 4: Ok but not good enough - rejection | In this paper, the authors propose a version of Transformer Networks to deal with convolutional neural network architectures that contains upsampling (eg U-Net). The original Transformer Network does not take in consideration pixel level alignment, as it was designed for classification task. The authors propose a laye... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Dense Transformer Networks
### Paper Abstract
The key idea of current deep learning methods for dense prediction is to apply a model on a regular patch centered on each ... |
e1RgaHCQrT | Interspeech.org/2023/Workshop/SSW | 2023 | Voice Cloning: Training Speaker Selection with Limited Multi-Speaker Corpus | ["David Guennec", "Lily Wadoux", "Aghilas Sini", "Nelly Barbot", "Damien Lolive"] | Text-to-speech synthesis with few data is a challenging task, in particular when choosing the target speaker is not an option. Voice cloning is a popular method to alleviate these issues using only a few minutes of target speech. To do this, the model must first be trained on a large corpus of thousands of hours and hu... | ["Voice Cloning", "Speaker Adaptation", "Speaker Encoding", "Speaker Similarity", "Data Selection", "Neural Speech Synthesis"] | Voice Cloning: Training Speaker Selection in Limited Multi-Speaker CorpusDavid Guennec, Lily Wadoux, Aghilas Sini, Nelly Barbot, Damien LoliveUniv Rennes, CNRS, IRISA, France{firstname.secondname }@irisa.frAbstractText-To-Speech synthesis with few data is a challengingtask, in particular when choosing the target speake... | FOVFRP7Soc | Interesting experiment but not really conclusive | 4: Ok but not good enough - rejection | The authors consider the scenario of voice cloning when there is not a large dataset of speakers to train the initial multispeaker model, and investigate whether it is useful or not to select the training speakers based on their similarity to the target. It's an interesting question although it might be argued that sel... | 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
Voice Cloning: Training Speaker Selection with Limited Multi-Speaker Corpus
### Paper Abstract
Text-to-speech synthesis with few data is a challenging task, in particula... |
3MrDT4bTycn | graphicsinterface.org/Graphics_Interface/2022/Conference | 2022 | VR Stylus and Controller Combination Was Preferred Over the Mouse or Hand Tracking for Object Manipulation and Marking Tasks in Virtual Reality | ["Anonymous"] | For medical surgery planning, virtual reality (VR) provides a new kind of user experience, where 3D images of the operation area can be utilized. Using VR, it is possible to view the 3D models in a more realistic 3D environment, which would reduce the perception problems and increase spatial understanding. In this expe... | ["3D visualization", "virtual reality", "hand tracking", "controller interaction", "mouse interaction", "3D imaging"] | ABSTRACTFor medical surgery planning, virtual reality (VR) provides a newkind of user experience, where 3D images of the operation area canbe utilized. Using VR, it is possible to view the 3D models in amore realistic 3D environment, which would reduce the perceptionproblems and increase spatial understanding. In this ... | AlzWKNVfsOR | comparison of selection techniques for VR | 5: Marginally below acceptance threshold | The submission presents an experimental comparison of 3 user interfaces for selecting points on a 3D model, motivated by potential applications in jaw osteotomy surgery planning (section 2.3). The submission refers to 'manipulation' and 'marking' as tasks, where 'manipulation' seems to mean rotating and translating 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
VR Stylus and Controller Combination Was Preferred Over the Mouse or Hand Tracking for Object Manipulation and Marking Tasks in Virtual Reality
### Paper Abstract
For me... |
Hye_V0NKwr | ICLR.cc/2020/Conference | 2020 | Locality and Compositionality in Zero-Shot Learning | ["Tristan Sylvain", "Linda Petrini", "Devon Hjelm"] | In this work we study locality and compositionality in the context of learning representations for Zero Shot Learning (ZSL).
In order to well-isolate the importance of these properties in learned representations, we impose the additional constraint that, differently from most recent work in ZSL, no pre-training on dif... | ["Zero-shot learning", "Compositionality", "Locality", "Deep Learning"] | ABSTRACTIn this work we study locality andcompositionality in the context of learningrepresentations for Zero Shot Learning (ZSL). In order to well-isolate the importanceof these properties in learned representations, we impose the additional constraintthat, differently from most recent work in ZSL, no pre-training on ... | HJgLUdTnYB | Official Blind Review #1 | 6: Weak Accept | The paper proposes an evaluation framework for Zero-Shot Learning (ZSL) methods called zero-shot learning from scratch (ZFS) where the model is not allowed to be pretrained on other datasets such as ImageNet. The main motivation of this approach is that it is difficult to understand what is useful for generalization in... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Locality and Compositionality in Zero-Shot Learning
### Paper Abstract
In this work we study locality and compositionality in the context of learning representations for... | |
kqDCPX7eWS | ICLR.cc/2021/Conference | 2021 | Local SGD Meets Asynchrony | ["Bapi Chatterjee", "Vyacheslav Kungurtsev", "Dan Alistarh"] | Distributed variants of stochastic gradient descent (SGD) are central to training deep neural networks on massive datasets.
Several scalable versions of data-parallel SGD have been developed, leveraging asynchrony, communication-compression, and local gradient steps. Current research seeks a balance between distribut... | ["SGD", "Data-parallel", "Asynchronous", "Optimization", "Non-convex", "Deep Neural Network"] | ABSTRACTDistributed variants of stochastic gradient descent (SGD) are central to train-ing deep neural networks on massive datasets. Several scalable versions ofdata-parallel SGD have been developed, leveraging asynchrony, communication-compression, and local gradient steps. Current research seeks a balance be-tween di... | 1z4SokANabG | Official Blind Review #2 | 4: Ok but not good enough - rejection | In this paper, the authors argue that the mini-batch method and local SGD method suffers generalization performance degradation for large local mini-batch size. An asynchronous method is proposed to improve the generalization performance. A sublinear convergence rate is provided for the non-convex objective. As there 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
Local SGD Meets Asynchrony
### Paper Abstract
Distributed variants of stochastic gradient descent (SGD) are central to training deep neural networks on massive datasets.... |
H1aKXVNKx | ICLR.cc/2017/workshop | 2017 | Predicting Surgery Duration with Neural Heteroscedastic Regression | ["Nathan Ng", "Rodney A Gabriel", "Julian McAuley", "Charles Elkan", "Zachary C Lipton"] | Scheduling surgeries is a challenging task due to the fundamental uncertainty of the clinical environment, as well as the risks and costs associated with under- and over-booking. We investigate neural regression algorithms to estimate the parameters of surgery case durations, focusing on the issue of heteroscedasticity... | ["Deep learning", "Supervised Learning"] | ABSTRACTScheduling surgeries is a challenging task due to the fundamental uncertainty ofthe clinical environment, as well as the risks and costs associated with under- andover-booking. We investigate neural regression algorithms to estimate the param-eters of surgery case durations, focusing on the issue of heterosceda... | B1PJU9gix | Review | 5: Marginally below acceptance threshold | Summary:
This work models the distribution of surgery durations using unimodal
parameteric distributions (viz., Gaussian and Laplace) by regressing their
parameters using multi-layer perceptrons based on patient and clinical
environment attributes.
Using the uncertainty (or standard-deviation) estimates, they report
... | 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
Predicting Surgery Duration with Neural Heteroscedastic Regression
### Paper Abstract
Scheduling surgeries is a challenging task due to the fundamental uncertainty of t... |
Hkjg7N1Pz | ICLR.cc/2018/Workshop | 2018 | Reconstructing evolutionary trajectories of mutations in cancer | ["Yulia Rubanova", "Ruian Shi", "Roujia Li", "Jeff Wintersinger", "Amit Deshwar", "Nil Sahin", "Quaid Morris"] | We present a new method, TrackSig, to estimate evolutionary trajectories in cancer. Our method represents cancer evolution in terms of mutational signatures -- multinomial distributions over mutation types. TrackSig infers an approximate order in which mutations accumulated in cancer genome, and then fits the signature... | ["cancer", "bioinformatics", "topic modelling", "time series"] | ABSTRACTWe present a new method, TrackSig, to estimate evolutionary trajectories in can-cer. Our method represents cancer evolution in terms of mutational signatures –multinomial distributions over mutation types. TrackSig infers an approximateorder in which mutations accumulated in cancer genome, and then fits the sig... | rJYwQ-7KG | Interesting model | 5: Marginally below acceptance threshold | This submission applies topic models to infer mutational signature exposures (i.e. topic proportions) from cancer mutation data across time. They apply their method to simulations and data from PCAWG. Overall, the model is interesting and potentially useful. I do have some comments on the experiments. Also, not knowin... | 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
Reconstructing evolutionary trajectories of mutations in cancer
### Paper Abstract
We present a new method, TrackSig, to estimate evolutionary trajectories in cancer. Ou... |
rx19UMFbC9u | ICLR.cc/2021/Conference | 2021 | Waste not, Want not: All-Alive Pruning for Extremely Sparse Networks | ["Daejin Kim", "Hyunjung Shim", "Jongwuk Lee"] | Network pruning has been widely adopted for reducing computational cost and memory consumption in low-resource devices. Recent studies show that saliency-based pruning can achieve high compression ratios (e.g., 80-90% of the parameters in original networks are removed) without sacrificing much accuracy loss. Neverthele... | ["Model compression", "Network pruning", "Iterative pruning", "Dead connections"] | ABSTRACTNetwork pruning has been widely adopted for reducing computational cost andmemory consumption in low-resource devices. Recent studies show that saliency-based pruning can achieve high compression ratios ( e.g., 80–90% of the parametersin original networks are removed) without sacrificing much accuracy loss. Nev... | hywksm4uyD0 | Simple yet effective pruning approach focusing on removing dead connections in a network | 7: Good paper, accept | Summary:
This paper talks about a novel network pruning method, called all-alive pruning (AAP), which seeks to effectively remove dead connections in a network. The proposed approach aims at enhancing the saliency-based pruning, and technically the approach searches for the dead neurons by inspecting their gradient flo... | 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
Waste not, Want not: All-Alive Pruning for Extremely Sparse Networks
### Paper Abstract
Network pruning has been widely adopted for reducing computational cost and memor... |
NNd0J677PN | ICLR.cc/2021/Conference | 2021 | Voting-based Approaches For Differentially Private Federated Learning | ["Yuqing Zhu", "Xiang Yu", "Yi-Hsuan Tsai", "Francesco Pittaluga", "Masoud Faraki", "Manmohan Chandraker", "Yu-Xiang Wang"] | While federated learning (FL) enables distributed agents to collaboratively train a centralized model without sharing data with each other, it fails to protect users against inference attacks that mine private information from the centralized model. Thus, facilitating federated learning methods with differential privac... | ["approaches", "private", "learning", "centralized model", "dpfl", "gradients", "agents", "distributed agents", "data"] | ABSTRACTWhile federated learning (FL) enables distributed agents to collaboratively traina centralized model without sharing data with each other, it fails to protect usersagainst inference attacks that mine private information from the centralized model.Thus, facilitating federated learning methods with differential p... | -xEbLI3yp1U | Simple and effective ideas, some details are missing | 5: Marginally below acceptance threshold | Summary:
The paper proposes two approaches (i.e., PATE-FL and Private-kNN-FL) to train a differentially private global model in a federated setting based on [1] and [2]. In PATE-FL, each client first trains a teacher model using their local dataset. The teacher models are used to make noisy predictions on a public data... | 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
Voting-based Approaches For Differentially Private Federated Learning
### Paper Abstract
While federated learning (FL) enables distributed agents to collaboratively trai... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.