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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Sy4tzwqxe | ICLR.cc/2017/conference | 2017 | Two Methods for Wild Variational Inference | ["Qiang Liu", "Yihao Feng"] | Variational inference provides a powerful tool for approximate probabilistic inference on complex, structured models. Typical variational inference methods, however, require to use inference networks with computationally tractable probability density functions. This largely limits the design and implementation of varia... | ["Theory"] | ABSTRACTVariational inference provides a powerful tool for approximate probabilistic in-ference on complex, structured models. Typical variational inference methods,however, require to use inference networks with computationally tractable proba-bility density functions. This largely limits the design and implementation... | SJztFgf4e | Review: Two Methods for Wild Variational Inference | 3: Clear rejection | The authors propose two variational methods based on the theme of posterior approximations which may not have a tractable density. The first is from another ICLR submission on "amortized SVGD" (Wang and Liu, 2016), where here the innovation is in using SGLD as the inference network. The second is from a NIPS paper (Ran... | 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
Two Methods for Wild Variational Inference
### Paper Abstract
Variational inference provides a powerful tool for approximate probabilistic inference on complex, structur... |
F1vEjWK-lH_ | ICLR.cc/2021/Conference | 2021 | Gradient Vaccine: Investigating and Improving Multi-task Optimization in Massively Multilingual Models | ["Zirui Wang", "Yulia Tsvetkov", "Orhan Firat", "Yuan Cao"] | Massively multilingual models subsuming tens or even hundreds of languages pose great challenges to multi-task optimization. While it is a common practice to apply a language-agnostic procedure optimizing a joint multilingual task objective, how to properly characterize and take advantage of its underlying problem stru... | ["Multi-task Learning", "Multilingual Modeling"] | ABSTRACTMassively multilingual models subsuming tens or even hundreds of languagespose great challenges to multi-task optimization. While it is a common practiceto apply a language-agnostic procedure optimizing a joint multilingual task objec-tive, how to properly characterize and take advantage of its underlying probl... | asPunsSAari | This work takes aim at an interesting problem of optimizing multilingual neural machine translation (MNMT) model. Although MNMT is inherently a multi-task modeling approach, less emphasis have been given on achieving an optimal performance on all of the tasks involved. The proposed approach (GradVac) takes into account... | 8: Top 50% of accepted papers, clear accept | Summary
Taking multilingual NMT (MNMT) into account, this work, investigates better model optimization alternative, that is in part can be attributed as a multi-task optimization problem. MNMT's are quite beneficial from different perspectives (improving low-resource languages, efficiency, etc). However, their inheren... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Gradient Vaccine: Investigating and Improving Multi-task Optimization in Massively Multilingual Models
### Paper Abstract
Massively multilingual models subsuming tens or... |
H1edEyBKDS | ICLR.cc/2020/Conference | 2020 | Plug and Play Language Models: A Simple Approach to Controlled Text Generation | ["Sumanth Dathathri", "Andrea Madotto", "Janice Lan", "Jane Hung", "Eric Frank", "Piero Molino", "Jason Yosinski", "Rosanne Liu"] | Large transformer-based language models (LMs) trained on huge text corpora have shown unparalleled generation capabilities. However, controlling attributes of the generated language (e.g. switching topic or sentiment) is difficult without modifying the model architecture or fine-tuning on attribute-specific data and en... | ["controlled text generation", "generative models", "conditional generative models", "language modeling", "transformer"] | ABSTRACTLarge transformer-based language models (LMs) trained on huge text corporahave shown unparalleled generation capabilities. However, controlling attributesof the generated language (e.g. switching topic or sentiment) is difficult withoutmodifying the model architecture or fine-tuning on attribute-specific data a... | HkegjQxaFH | Official Blind Review #1 | 3: Weak Reject | The paper proposes a Plug and Play LM model for controlled natural language generation. Similar to the idea of the Plug and Play Generative Networks for vision, the model plugs in a discriminator, which is either a bag-of-words model or a single layer classifier. The added simple discriminator is then coupled with a pr... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Plug and Play Language Models: A Simple Approach to Controlled Text Generation
### Paper Abstract
Large transformer-based language models (LMs) trained on huge text corp... | |
SJGCiw5gl | ICLR.cc/2017/conference | 2017 | Pruning Convolutional Neural Networks for Resource Efficient Inference | ["Pavlo Molchanov", "Stephen Tyree", "Tero Karras", "Timo Aila", "Jan Kautz"] | We propose a new formulation for pruning convolutional kernels in neural networks to enable efficient inference. We interleave greedy criteria-based pruning with fine-tuning by backpropagation-a computationally efficient procedure that maintains good generalization in the pruned network. We propose a new criterion base... | ["Deep learning", "Transfer Learning"] | ABSTRACTWe propose a new formulation for pruning convolutional kernels in neural networksto enable efficient inference. We interleave greedy criteria-based pruning with fine-tuning by backpropagation—a computationally efficient procedure that maintainsgood generalization in the pruned network. We propose a new criterio... | SJFoMMKVe | Empirically justified pruning strategy, a few missing comparisons | 7: Good paper, accept | Authors propose a strategy for pruning weights with the eventual goal of reducing GFLOP computations. The pruning strategy is well motivated using the taylor expansion of the neural network function with respect to the feature activations. The obtained strategy removes feature maps that have both a small activation and... | 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 Convolutional Neural Networks for Resource Efficient Inference
### Paper Abstract
We propose a new formulation for pruning convolutional kernels in neural networ... |
rxVXvwpN2fr | icaps-conference.org/ICAPS/2021/Workshop/HSDIP | 2021 | A*+BFHS: A Hybrid Heuristic Search Algorithm | ["Zhaoxing Bu", "Richard Korf"] | We present a new algorithm A*+BFHS for solving problems where A* and IDA* fail due to memory limitations and/or the existence of many short cycles. A*+BFHS is based on A* and breadth-first heuristic search (BFHS). A*+BFHS combines advantages from both algorithms, namely A*’s node ordering, BFHS’s memory savings, and bo... | ["heuristic search", "planning", "artificial intelligence"] | A*+BFHS: A Hybrid Heuristic Search AlgorithmZhaoxing Bu, Richard E. KorfComputer Science DepartmentUniversity of California, Los AngelesLos Angeles, CA 90095fzbu, korf g@cs.ucla.eduAbstractWe present a new algorithm A*+BFHS for solving problemswhere A* and IDA* fail due to memory limitations and/orthe existence of many... | c8FSOREjAN6 | Review | This paper presents an algorithm that performs an A* search until a user-specified memory limit is reached, and then performs (multiple) breadth-first heuristic searches (BFHS). The main advantage of this hybrid search is that it behaves like A* on smaller instances where the exponential memory consumption of A* is not... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A*+BFHS: A Hybrid Heuristic Search Algorithm
### Paper Abstract
We present a new algorithm A*+BFHS for solving problems where A* and IDA* fail due to memory limitations ... | |
B1EjKsRqtQ | ICLR.cc/2019/Conference | 2019 | Hierarchical Attention: What Really Counts in Various NLP Tasks | ["Zehao Dou", "Zhihua Zhang"] | Attention mechanisms in sequence to sequence models have shown great ability and wonderful performance in various natural language processing (NLP) tasks, such as sentence embedding, text generation, machine translation, machine reading comprehension, etc. Unfortunately, existing attention mechanisms only learn eithe... | ["attention", "hierarchical", "machine reading comprehension", "poem generation"] | ABSTRACTAttention mechanisms in sequence to sequence models have shown great abilityand wonderful performance in various natural language processing (NLP) tasks,such as sentence embedding, text generation, machine translation, machine readingcomprehension, etc. Unfortunately, existing attention mechanisms only learneit... | BkeUUGo937 | A simple extension of multi-level attention, but needs more extensive comparison to existing methods | 4: Ok but not good enough - rejection | The paper introduces hierarchical attention, where they propose to weighted combine all the intermediate layers of multi-level attention. The idea is simple and seems to be promising, however originality seems incremental.
In order to fully demonstrate the significance of the proposed algorithm, the authors should con... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Hierarchical Attention: What Really Counts in Various NLP Tasks
### Paper Abstract
Attention mechanisms in sequence to sequence models have shown great ability and wonde... |
SklGryBtwr | ICLR.cc/2020/Conference | 2020 | Environmental drivers of systematicity and generalization in a situated agent | ["Felix Hill", "Andrew Lampinen", "Rosalia Schneider", "Stephen Clark", "Matthew Botvinick", "James L. McClelland", "Adam Santoro"] | The question of whether deep neural networks are good at generalising beyond their immediate training experience is of critical importance for learning-based approaches to AI. Here, we consider tests of out-of-sample generalisation that require an agent to respond to never-seen-before instructions by manipulating and p... | ["systematicitiy", "systematic", "generalization", "combinatorial", "agent", "policy", "language", "compositionality"] | ABSTRACTThe question of whether deep neural networks are good at generalising beyondtheir immediate training experience is of critical importance for learning-basedapproaches to AI. Here, we consider tests of out-of-sample generalisation that re-quire an agent to respond to never-seen-before instructions by manipulatin... | rJxuMyU1ir | Official Blind Review #4 | 6: Weak Accept | This work studies factors which promote combinatorial generalization in a "neural network agent" embodied in a 3d simulation environment. The authors present interesting experiments and some insightful empirical findings on how a richer environment and a first-person egocentric perspective can aid a simple neural net 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
Environmental drivers of systematicity and generalization in a situated agent
### Paper Abstract
The question of whether deep neural networks are good at generalising be... | |
S1iiddyDG | ICLR.cc/2018/Workshop | 2018 | Negative eigenvalues of the Hessian in deep neural networks | ["Guillaume Alain", "Nicolas Le Roux", "Pierre-Antoine Manzagol"] | We study the loss function of a deep neural network through the eigendecomposition of its Hessian matrix. We focus on negative eigenvalues, how important they are, and how to best deal with them. The goal is to develop an optimization method specifically tailored for deep neural networks. | ["Optimization", "Hessian matrix", "Neural Networks", "Negative Curvature"] | ABSTRACTWe study the loss function of a deep neural network through the eigendecompo-sition of its Hessian matrix. We focus on negative eigenvalues, how importantthey are, and how to best deal with them. The goal is to develop an optimizationmethod specifically tailored for deep neural networks.1 I NTRODUCTIONThe curre... | Hy5LN8eFG | finds existence proof (for 1 model and 1 dataset) where learning rate should be higher than Dauphin et al's suggested 1 / abs(eigenval) for negative curvature directions | 6: Marginally above acceptance threshold | The main contribution of this work is their simple empirical observation in Figure 4, that the optimal alpha (learning rate) appears to often NOT be 1 / | eigval | when the eigval is negative (contradicting the hypothesis of Dauphin et al. (2014), but instead should be much larger. This seems to this reader to be the o... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Negative eigenvalues of the Hessian in deep neural networks
### Paper Abstract
We study the loss function of a deep neural network through the eigendecomposition of its ... |
H1lxVyStPH | ICLR.cc/2020/Conference | 2020 | Generalized Convolutional Forest Networks for Domain Generalization and Visual Recognition | ["Jongbin Ryu", "Gitaek Kwon", "Ming-Hsuan Yang", "Jongwoo Lim"] | When constructing random forests, it is of prime importance to ensure high accuracy and low correlation of individual tree classifiers for good performance. Nevertheless, it is typically difficult for existing random forest methods to strike a good balance between these conflicting factors. In this work, we propose a g... | ["convolutional forest networks", "domain generalization", "visual recognition", "individual tree classifiers", "feature space", "data", "random forests", "prime importance", "high accuracy", "low correlation"] | ABSTRACTWhen constructing random forests, it is of prime importance to ensure high ac-curacy and low correlation of individual tree classifiers for good performance.Nevertheless, it is typically difficult for existing random forest methods to strikea good balance between these conflicting factors. In this work, we prop... | HklgYgX-5S | Official Blind Review #2 | 6: Weak Accept | The paper aims to improve the random forest performance by iterative constructing & feeding more powerful features that are to be used by the random forest learning processing where a random subset of features are chosen from the current feature pool when making growing/stopping decision at the current split node. The ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Generalized Convolutional Forest Networks for Domain Generalization and Visual Recognition
### Paper Abstract
When constructing random forests, it is of prime importance... | |
B1gi-TVKwB | ICLR.cc/2020/Conference | 2020 | Learning an off-policy predictive state representation for deep reinforcement learning for vision-based steering in autonomous driving | ["Daniel Graves"] | An algorithm is introduced for learning a predictive state representation with off-policy temporal difference (TD) learning that is then used to learn to steer a vehicle with reinforcement learning. There are three components being learned simultaneously: (1) the off-policy predictions as a compact representation of ... | ["Predictive representations", "general value functions", "reinforcement learning", "off-policy learning", "behavior estimation"] | ABSTRACTAn algorithm is introduced for learning a predictive state representation withoff-policy temporal difference (TD) learning that is then used to learn to steer avehicle with reinforcement learning. There are three components being learnedsimultaneously: (1) the off-policy predictions as a compact representation ... | H1eZawypYB | Official Blind Review #1 | 1: Reject | This paper combines predictive state representations (PSRs) with DPG and tests the overall performance on a TORC simulated task. Specifically, the authors propose to train a generalized value function (GVF) where the cumulant is how far the car is from the center of the road as well as the angle of the car. To train th... | <|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 an off-policy predictive state representation for deep reinforcement learning for vision-based steering in autonomous driving
### Paper Abstract
An algorithm is... | |
o-K3HVUeEw | robot-learning.org/CoRL/2023/Conference | 2023 | Composable Part-Based Manipulation | ["Weiyu Liu", "Jiayuan Mao", "Joy Hsu", "Tucker Hermans", "Animesh Garg", "Jiajun Wu"] | In this paper, we propose composable part-based manipulation (CPM), a novel approach that leverages object-part decomposition and part-part correspondences to improve learning and generalization of robotic manipulation skills. By considering the functional correspondences between object parts, we conceptualize function... | ["Manipulation", "Part Decomposition", "Diffusion Model"] | Composable Part-Based ManipulationWeiyu Liu1, Jiayuan Mao2, Joy Hsu1, Tucker Hermans3,4, Animesh Garg3,5, Jiajun Wu11Stanford2MIT3NVIDIA4University of Utah5Georgia TechAbstract: In this paper, we propose composable part-based manipulation (CPM),a novel approach that leverages object-part decomposition and part-part cor... | KL7LOtlRuX | Strengths
- Authors show real-world demonstration showing impressive successful sim-to-real transfer, including generalization to unseen mugs
- The authors proposed method of using part-part correspondences as the basis of modelling manipulation actions, and the integration of this with diffusion models is very interes... | 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
Composable Part-Based Manipulation
### Paper Abstract
In this paper, we propose composable part-based manipulation (CPM), a novel approach that leverages object-part dec... | ||
HyeJmlrFvH | ICLR.cc/2020/Conference | 2020 | Provably Communication-efficient Data-parallel SGD via Nonuniform Quantization | ["Ali Ramezani-Kebrya", "Fartash Faghri", "Ilya Markov", "Vitalii Aksenov", "Dan Alistarh", "Daniel M. Roy"] | As the size and complexity of models and datasets grow, so does the need for communication-efficient variants of stochastic gradient descent that can be deployed on clusters to perform model fitting in parallel. Alistarh et al. (2017) describe two variants of data-parallel SGD that quantize and encode gradients to less... | ["sgd", "nonuniform quantization", "variants", "qsgd", "size", "complexity", "models", "datasets", "need", "stochastic gradient descent"] | ABSTRACTAs the size and complexity of models and datasets grow, so does the need forcommunication-efficient variants of stochastic gradient descent that can be de-ployed on clusters to perform parallel model training. Alistarh et al. (2017) describetwo variants of data-parallel SGD that quantize and encode gradients to... | SJxLkbwatH | Official Blind Review #1 | 3: Weak Reject | In this paper, the authors propose a new gradient compression method, which is called nonuniform quantization. The algorithm is a reasonable variant of SGD with uniform quantization. The paper is well written. The experiments show good performance.
However, there are several weakness in this paper:
1. In this 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
Provably Communication-efficient Data-parallel SGD via Nonuniform Quantization
### Paper Abstract
As the size and complexity of models and datasets grow, so does the nee... | |
S1emOTNKvS | ICLR.cc/2020/Conference | 2020 | Robust Graph Representation Learning via Neural Sparsification | ["Cheng Zheng", "Bo Zong", "Wei Cheng", "Dongjin Song", "Jingchao Ni", "Wenchao Yu", "Haifeng Chen", "Wei Wang"] | Graph representation learning serves as the core of many important prediction tasks, ranging from product recommendation in online marketing to fraud detection in financial domain. Real-life graphs are usually large with complex local neighborhood, where each node is described by a rich set of features and easily conne... | ["graphs", "complexity", "risk", "neuralsparse", "graph neural networks", "robust graph representation", "serves", "core"] | ABSTRACTGraph representation learning serves as the core of many important predictiontasks, ranging from product recommendation in online marketing to fraud detec-tion in financial domain. Real-life graphs are usually large with complex localneighborhood, where each node is described by a rich set of features and easil... | r1xkiEt2Fr | Official Blind Review #2 | 1: Reject | The authors argue that existing GCN-based approaches may pose non-trival overfitting risk during the training phase, especially when high-dimensional features and high-degree entities are observed in the graphs. To address the issue, the authors integrate graph sparsification with conventional graph neural nets. Experi... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust Graph Representation Learning via Neural Sparsification
### Paper Abstract
Graph representation learning serves as the core of many important prediction tasks, ra... | |
6Tm1mposlrM | ICLR.cc/2021/Conference | 2021 | Sharpness-aware Minimization for Efficiently Improving Generalization | ["Pierre Foret", "Ariel Kleiner", "Hossein Mobahi", "Behnam Neyshabur"] | In today's heavily overparameterized models, the value of the training loss provides few guarantees on model generalization ability. Indeed, optimizing only the training loss value, as is commonly done, can easily lead to suboptimal model quality. Motivated by the connection between geometry of the loss landscape and g... | ["Sharpness Minimization", "Generalization", "Regularization", "Training Method", "Deep Learning"] | ABSTRACTIn today’s heavily overparameterized models, the value of the training loss pro-vides few guarantees on model generalization ability. Indeed, optimizing onlythe training loss value, as is commonly done, can easily lead to suboptimalmodel quality. Motivated by prior work connecting the geometry of the losslandsc... | SMQLOQzD-T | Interesting work with good results, concern is on selecting the right $\rho$ | 7: Good paper, accept | Motivated by the connection between the flatness of minima and its generalization ability, the authors propose Sharpness-aware Minimization (SAM), which explicitly minimizes both loss value and loss sharpness during training deep neural networks. They find SAM improves generalization for a range of image classification... | 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
Sharpness-aware Minimization for Efficiently Improving Generalization
### Paper Abstract
In today's heavily overparameterized models, the value of the training loss prov... |
rCztBS_Uqxc | ICLR.cc/2022/Workshop/OSC | 2022 | Learning Two-Step Hybrid Policy for Graph-Based Interpretable Reinforcement Learning | ["Tongzhou Mu", "Kaixiang Lin", "Feiyang Niu", "Govind Thattai"] | We present a two-step hybrid reinforcement learning (RL) policy that is designed to generate interpretable and robust hierarchical policies
on the RL problem with graph-based input. Unlike prior deep reinforcement learning policies parameterized by an end-to-end black-box graph neural network, our approach disentangles... | ["Graph-based Reinforcement Learning", "Interpretable Reinforcement Learning", "Generalization", "Robustness"] | ABSTRACTWe present a two-step hybrid reinforcement learning (RL) policy that is designedto generate interpretable and robust hierarchical policies on the RL problem withgraph-based input. Unlike prior deep reinforcement learning policies parameterizedby an end-to-end black-box graph neural network, our approach disenta... | H5eT1_2Cbc | Review | 2: Good workshop paper, accept | The paper is well-written and the main approach could be understood.
Pros:
* The approach of separating out a policy intro a graph-based classification problem and then rule-mining based approach seems novel.
* The experiment results are promising
Cons:
* Discussion regarding how such an approach could translate to ... | 2: The reviewer is somewhat 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 Two-Step Hybrid Policy for Graph-Based Interpretable Reinforcement Learning
### Paper Abstract
We present a two-step hybrid reinforcement learning (RL) policy t... |
ryxWIgBFPS | ICLR.cc/2020/Conference | 2020 | A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms | ["Yoshua Bengio", "Tristan Deleu", "Nasim Rahaman", "Nan Rosemary Ke", "Sebastien Lachapelle", "Olexa Bilaniuk", "Anirudh Goyal", "Christopher Pal"] | We propose to use a meta-learning objective that maximizes the speed of transfer on a modified distribution to learn how to modularize acquired knowledge. In particular, we focus on how to factor a joint distribution into appropriate conditionals, consistent with the causal directions. We explain when this can work, us... | ["meta-learning", "transfer learning", "structure learning", "modularity", "causality"] | ABSTRACTWe propose to use a meta-learning objective that maximizes the speed of transferon a modified distribution to learn how to modularize acquired knowledge anddiscover causal dependencies. In particular, we focus on how to factor a jointdistribution into appropriate conditionals, consistent with the causal directi... | rkl6smfDtB | Official Blind Review #2 | 8: Accept |
Summary:
The paper first shows that, in a very simple two-variable task, the model with the correct underlying structure will adapt faster to a causal intervention than the model with the incorrect structure. This idea is used to develop a “meta-transfer” objective function for which gradient ascent on a continuous r... | <|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 Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms
### Paper Abstract
We propose to use a meta-learning objective that maximizes the speed of transf... | |
v_1Soh8QUNc | ICLR.cc/2021/Conference | 2021 | Learning Energy-Based Models by Diffusion Recovery Likelihood | ["Ruiqi Gao", "Yang Song", "Ben Poole", "Ying Nian Wu", "Diederik P Kingma"] | While energy-based models (EBMs) exhibit a number of desirable properties, training and sampling on high-dimensional datasets remains challenging. Inspired by recent progress on diffusion probabilistic models, we present a diffusion recovery likelihood method to tractably learn and sample from a sequence of EBMs traine... | ["energy-based model", "EBM", "recovery likelihood", "generative model", "diffusion process", "MCMC", "Langevin dynamics", "HMC"] | ABSTRACTWhile energy-based models (EBMs) exhibit a number of desirable properties,training and sampling on high-dimensional datasets remains challenging. Inspiredby recent progress on diffusion probabilistic models, we present a diffusion re-covery likelihood method to tractably learn and sample from a sequence of EBMs... | 0N7YHITYd0e | Interesting approach for training EBMs as a sequence of conditional EBMs | 7: Good paper, accept | This paper describes training a sequence of conditional EBMs (inspired by Ho et al. (2020)) instead of training unconditional EBMs. Each conditional energy describes the probability of recovering x, given its noisy version \hat{x}. The noisy version of x can be described as a normal distribution centered at x, so 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
Learning Energy-Based Models by Diffusion Recovery Likelihood
### Paper Abstract
While energy-based models (EBMs) exhibit a number of desirable properties, training and ... |
Byk-VI9eg | ICLR.cc/2017/conference | 2017 | Generative Multi-Adversarial Networks | ["Ishan Durugkar", "Ian Gemp", "Sridhar Mahadevan"] | Generative adversarial networks (GANs) are a framework for producing a generative model by way of a two-player minimax game. In this paper, we propose the \emph{Generative Multi-Adversarial Network} (GMAN), a framework that extends GANs to multiple discriminators. In previous work, the successful training of GANs requ... | ["Deep learning", "Unsupervised Learning", "Games"] | ABSTRACTGenerative adversarial networks (GANs) are a framework for producing a gen-erative model by way of a two-player minimax game. In this paper, we proposetheGenerative Multi-Adversarial Network (GMAN), a framework that extendsGANs to multiple discriminators. In previous work, the successful training ofGANs require... | HkYCUhr4g | 6: Marginally above acceptance threshold | This work brings multiple discriminators into GAN. From the result, multiple discriminators is useful for stabilizing.
The main problem of stabilizing seems is from gradient signal from discriminator, the authors motivation is using multiple discriminators to reduce this effect.
I think this work indicates the direc... | 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
Generative Multi-Adversarial Networks
### Paper Abstract
Generative adversarial networks (GANs) are a framework for producing a generative model by way of a two-player m... | |
B1ecYsqSuN | ICLR.cc/2019/Workshop/LLD | 2019 | INCORPORATING BILINGUAL DICTIONARIES FOR LOW RESOURCE SEMI-SUPERVISED NEURAL MACHINE TRANSLATION | ["Mihir Kale", "Sreyashi Nag", "Varun Lakshinarasimhan", "Swapnil Singhavi"] | We explore ways of incorporating bilingual dictionaries to enable semi-supervised
neural machine translation. Conventional back-translation methods have shown
success in leveraging target side monolingual data. However, since the quality of
back-translation models is tied to the size of the available parallel corpora, ... | ["bilingual dictionaries", "neural machine translation", "low resource", "ways", "conventional", "methods", "success", "quality", "models"] | ABSTRACTWe explore ways of incorporating bilingual dictionaries to enable semi-supervisedneural machine translation. Conventional back-translation methods have shownsuccess in leveraging target side monolingual data. However, since the quality ofback-translation models is tied to the size of the available parallel corp... | BkghQVbBYE | Simple technique for improving low-resource translation when bilingual dictionaries are given. | 3: Marginally above acceptance threshold | This paper investigates the idea of using bilingual dictionaries to create synthetic sources for target-side monolingual data in order to improve over NMT models trained with small amounts of parallel data.
This strategy is compared with back-translation and copying the target to the source side and evaluated on TED d... | 3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
INCORPORATING BILINGUAL DICTIONARIES FOR LOW RESOURCE SEMI-SUPERVISED NEURAL MACHINE TRANSLATION
### Paper Abstract
We explore ways of incorporating bilingual dictionari... |
YPm0fzy_z6R | ICLR.cc/2021/Conference | 2021 | Signed Graph Diffusion Network | ["Jinhong Jung", "Jaemin Yoo", "U Kang"] | Given a signed social graph, how can we learn appropriate node representations to infer the signs of missing edges?
Signed social graphs have received considerable attention to model trust relationships.
Learning node representations is crucial to effectively analyze graph data, and various techniques such as network e... | ["graph neural network", "signed graph analysis", "representation learning", "graph diffusion", "random walk", "link sign prediction"] | ABSTRACTGiven a signed social graph, how can we learn appropriate node representations toinfer the signs of missing edges? Signed social graphs have received considerableattention to model trust relationships. Learning node representations is crucial toeffectively analyze graph data, and various techniques such as netw... | -DrNlWUCLi8 | The paper propose SGDNet, Signed Graph Diffusion Network, to perform end-to-end node presentation and signed link prediction. It also considers the over-smoothing issue and use local feature injection to prevent over-smoothing. | 4: Ok but not good enough - rejection | The strength:
- Empirical results show improvements of the proposed method over the baseline methods, including methods for unsigned graphs, signed embedding emthods, and signed GCN methods not addressing over-smoothing.
The weakness:
- The proposed method is a straightforward integration of existing methods on sign... | 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
Signed Graph Diffusion Network
### Paper Abstract
Given a signed social graph, how can we learn appropriate node representations to infer the signs of missing edges? Sig... |
r1esys0Nd4 | ICLR.cc/2019/Workshop/LLD | 2019 | Train Neural Network by Embedding Space Probabilistic Constraint | ["Kaiyuan Chen", "Zhanyuan Yin"] | Using higher order knowledge to reduce training data has become a popular research topic. However, the ability for available methods to draw effective decision boundaries is still limited: when training set is small, neural networks will be biased to certain labels. Based on this observation, we consider constraining o... | ["probability", "constraint", "constraint learning", "weak supervision", "embedding", "deep neural network"] | ABSTRACTUsing higher order knowledge to reduce training data has become a popular re-search topic. However, the ability for available methods to draw effective deci-sion boundaries is still limited: when training set is small, neural networks willbe biased to certain labels. Based on this observation, we consider const... | Skgk0KR9FV | Leveraging additional supervision to handle label distribution shift | 3: Marginally above acceptance threshold |
This paper tries to incorporate label distribution into model learning, when a limited number of training instances is available.
Intuitively, the output label distribution could be wrongly biased, and the prior information like label distribution could be helpful.
To handle this problem, the authors propose two dif... | 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
Train Neural Network by Embedding Space Probabilistic Constraint
### Paper Abstract
Using higher order knowledge to reduce training data has become a popular research t... |
rkdF0ZNKl | ICLR.cc/2017/workshop | 2017 | Fast Generation for Convolutional Autoregressive Models | ["Prajit Ramachandran", "Tom Le Paine", "Pooya Khorrami", "Mohammad Babaeizadeh", "Shiyu Chang", "Yang Zhang", "Mark A. Hasegawa-Johnson", "Roy H. Campbell", "Thomas S. Huang"] | Convolutional autoregressive models have recently demonstrated state-of-the-art performance on a number of generation tasks. While fast, parallel training methods have been crucial for their success, generation is typically implemented in a naive fashion where redundant computations are unnecessarily repeated. This res... | ["Deep learning", "Applications"] | ABSTRACTConvolutional autoregressive models have recently demonstrated state-of-the-artperformance on a number of generation tasks. While fast, parallel training methodshave been crucial for their success, generation is typically implemented in a na ̈ıvefashion where redundant computations are unnecessarily repeated. T... | rkpqv3xoe | simple and good | 7: Good paper, accept | This is a nice workshop paper. its a simple idea but people will be interested in it. If nothing else, the released code is valuable, and having the poster to advertise it is a good use of workshop poster space. | 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
Fast Generation for Convolutional Autoregressive Models
### Paper Abstract
Convolutional autoregressive models have recently demonstrated state-of-the-art performance on... |
rkEfPeZRb | ICLR.cc/2018/Conference | 2018 | Variance-based Gradient Compression for Efficient Distributed Deep Learning | ["Yusuke Tsuzuku", "Hiroto Imachi", "Takuya Akiba"] | Due to the substantial computational cost, training state-of-the-art deep neural networks for large-scale datasets often requires distributed training using multiple computation workers. However, by nature, workers need to frequently communicate gradients, causing severe bottlenecks, especially on lower bandwidth conne... | ["distributed deep learning", "gradient compression", "collective communication", "data parallel distributed sgd", "image classification"] | ABSTRACTDue to the substantial computational cost, training state-of-the-art deep neuralnetworks for large-scale datasets often requires distributed training using multiplecomputation workers. However, by nature, workers need to frequently communi-cate gradients, causing severe bottlenecks, especially on lower bandwidt... | B1O_32YeM | The idea to adopt approximated variances of gradients to reduce communication cost seems to be interesting. However, there also exist several major issues in the paper. | 6: Marginally above acceptance threshold | This paper proposes a variance-based gradient compression method to reduce the communication overhead of distributed deep learning. Experiments on real datasets are used for evaluation.
The idea to adopt approximated variances of gradients to reduce communication cost seems to be interesting. However, there also exis... | 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
Variance-based Gradient Compression for Efficient Distributed Deep Learning
### Paper Abstract
Due to the substantial computational cost, training state-of-the-art deep ... |
S1D8MPxA- | ICLR.cc/2018/Conference | 2018 | Viterbi-based Pruning for Sparse Matrix with Fixed and High Index Compression Ratio | ["Dongsoo Lee", "Daehyun Ahn", "Taesu Kim", "Pierce I. Chuang", "Jae-Joon Kim"] | Weight pruning has proven to be an effective method in reducing the model size and computation cost while not sacrificing the model accuracy. Conventional sparse matrix formats, however, involve irregular index structures with large storage requirement and sequential reconstruction process, resulting in inefficient use... | ["pruning", "sparse matrix", "memory footprint", "model size", "model compression"] | ABSTRACTWeight pruning has proven to be an effective method of reducing the model sizeand computation cost without sacrificing its model accuracy. Conventional sparsematrix formats, however, involve irregular index structures with large storage re-quirement and a sequential reconstruction process, resulting in ineffici... | SktHYC_xM | The authors use Viterbi encoding to dramatically compress the sparse matrix index of a pruned network, reducing one of the main memory overheads of a pruned neural network and speeding up inference in the parallel setting. | 7: Good paper, accept | quality: this paper is of good quality
clarity: this paper is very clear but contains a few minor typos/grammatical mistakes (missing -s for plurals, etc.)
originality: this paper is original
significance: this paper is significant
PROS
- Using ECC theory for reducing the memory footprint of a neural network seems bot... | 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
Viterbi-based Pruning for Sparse Matrix with Fixed and High Index Compression Ratio
### Paper Abstract
Weight pruning has proven to be an effective method in reducing th... |
f75kMo1dnKD | thecvf.com/ECCV/2020/Workshop/VIPriors | 2020 | Injecting Prior Knowledge into Image Caption Generation | ["Arushi Goel", "Basura Fernando", "Thanh-Son Nguyen", "Hakan Bilen"] | Automatically generating natural language descriptions from an image is a challenging problem in artificial intelligence that requires a good understanding of the visual and textual signals and the correlations between them. The state-of-the-art methods in image captioning struggle to approach human level performance, ... | ["Image Captioning", "Prior", "Attention", "Regularization"] | 000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044ECCV#8ECCV#8Injecting Prior Knowledge into Image C... | HTjATmcpkQJ | Injecting Prior Knowledge into Image Caption Generation | 9: Top 15% of accepted papers, strong accept | 1. [Summary] In 2-3 sentences, describe the key ideas, experiments, and their significance.
The paper tries to mitigate overfitting and generating easy captions by introducing prior knowledge from the dataset during training. To this end, authors propose to add visual-semantic relation prior knowledge by defining a s... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Injecting Prior Knowledge into Image Caption Generation
### Paper Abstract
Automatically generating natural language descriptions from an image is a challenging problem ... |
32B5lOqZUiO | ICLR.cc/2021/Conference | 2021 | Pareto-Frontier-aware Neural Architecture Search | ["Yong Guo", "Yaofo Chen", "Yin Zheng", "Peilin Zhao", "Jian Chen", "Junzhou Huang", "Mingkui Tan"] | Designing feasible and effective architectures is essential for deploying deep models to real-world scenarios. In practice, one has to consider multiple objectives (e.g., model performance and computational cost) and diverse constraints incurred by different computation resources. To address this, most methods seek to ... | ["Neural Architecture Search", "Pareto Frontier Learning", "Resource Constraint"] | ABSTRACTDesigning feasible and effective architectures is essential for deploying deep modelsto real-world scenarios. In practice, one has to consider multiple objectives ( e.g.,model performance and computational cost) and diverse constraints incurred bydifferent computation resources. To address this, most methods se... | Sdp9_TojDx5 | Official Review | 4: Ok but not good enough - rejection | Paper Summary
The paper considers the problem of Neural Architecture Search (NAS) when multiple objectives needs to be optimized jointly. An approach called Pareto-Frontier-aware Neural Architecture Search (PF-NAS) is proposed for optimizing over two objectives (specifically latency and accuracy). The approach consist... | 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
Pareto-Frontier-aware Neural Architecture Search
### Paper Abstract
Designing feasible and effective architectures is essential for deploying deep models to real-world s... |
u4WfreuXxnk | ICLR.cc/2021/Conference | 2021 | Single-Node Attack for Fooling Graph Neural Networks | ["Ben Finkelshtein", "Chaim Baskin", "Evgenii Zheltonozhskii", "Uri Alon"] | Graph neural networks (GNNs) have shown broad applicability in a variety of domains.
Some of these domains, such as social networks and product recommendations, are fertile ground for malicious users and behavior.
In this paper, we show that GNNs are vulnerable to the extremely limited scenario of a single-node adversa... | ["graphs", "GNN", "adversarial", "attack"] | Under review as a conference paper at ICLR 2021SI N G L E - NO D E AT TA C K F O R FO O L I N GGR A P H NE U R A L NE T W O R K SAnonymous authorsPaper under double-blind reviewAB S T R A C TGraph neural networks (GNNs) have shown broad applicability in a variety ofdomains. Some of these domains, such as social network... | JVikAF-wTG | the paper lacks novelty | 5: Marginally below acceptance threshold | In this paper, the authors mainly show that the adversary can force the GNN to classify any target node to a chosen label by perturbing another single arbitrary node’s feature in the graph. The paper is well written and easy to understand. However, there are several concerns about the paper:
1. The novelty of the pape... | 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
Single-Node Attack for Fooling Graph Neural Networks
### Paper Abstract
Graph neural networks (GNNs) have shown broad applicability in a variety of domains. Some of thes... |
Brxlk6EuPW9 | ICLR.cc/2022/Workshop/DGM4HSD | 2022 | Deep Generative Neural Embeddings for High Dimensional Data Visualization | ["Anonymous"] | We propose an embedding-based visualization method along with a data generation model. In particular, corresponding locations of data points in the visualization are optimized as embeddings along with a generative network such that the network could reconstruct the original data. The generalization aspect of the neural... | ["Deep Generative Models", "Embedding"] | ABSTRACTWe propose an embedding-based visualization method along with a data gener-ation model. In particular, corresponding locations of data points in the visual-ization are optimized as embeddings along with a generative network such thatthe network could reconstruct the original data. The generalization aspect of t... | rgblTZfOLz5 | Existing method, limited evaluation | 3: Clear rejection | The authors consider an auto-encoder where we parametrize the latent codes of training points _directly_, instead of parametrizing an encoder. The model is evaluated on data visualization, where we aim to visualize the structure in the data by plotting the two-dimensional latent codes.
Unfortunately, the proposed mode... | 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
Deep Generative Neural Embeddings for High Dimensional Data Visualization
### Paper Abstract
We propose an embedding-based visualization method along with a data generat... |
SygKyeHKDH | ICLR.cc/2020/Conference | 2020 | Making Efficient Use of Demonstrations to Solve Hard Exploration Problems | ["Caglar Gulcehre", "Tom Le Paine", "Bobak Shahriari", "Misha Denil", "Matt Hoffman", "Hubert Soyer", "Richard Tanburn", "Steven Kapturowski", "Neil Rabinowitz", "Duncan Williams", "Gabriel Barth-Maron", "Ziyu Wang", "Nando de Freitas", "Worlds Team"] | This paper introduces R2D3, an agent that makes efficient use of demonstrations to solve hard exploration problems in partially observable environments with highly variable initial conditions. We also introduce a suite of eight tasks that combine these three properties, and show that R2D3 can solve several of the tasks... | ["imitation learning", "deep learning", "reinforcement learning"] | ABSTRACTThis paper introduces R2D3, an agent that makes efficient use of demonstrations tosolve hard exploration problems in partially observable environments with highlyvariable initial conditions. We also introduce a suite of eight tasks that combinethese three properties, and show that R2D3 can solve several of the ... | rJl2wiVt9S | Official Blind Review #4 | 6: Weak Accept | In this work, R2D3 (Recurrent Replay Distributed DQN from Demonstration), which combines R2D2 [1] with imitation learning (IL), is proposed. Similar to the existing works on “reinforcement learning (RL) with demonstration” such as DQfD, DDPGfD, policy optimization with demonstration (POfD) [2], hard exploration conditi... | <|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 Efficient Use of Demonstrations to Solve Hard Exploration Problems
### Paper Abstract
This paper introduces R2D3, an agent that makes efficient use of demonstrati... | |
mo3Uqtnvz_ | ICLR.cc/2021/Conference | 2021 | Multi-scale Network Architecture Search for Object Detection | ["Yuxin Yue", "Quanquan Li", "Yujie Wang"] | Many commonly-used detection frameworks aim to handle the multi-scale object detection problem. The input image is always encoded to multi-scale features and objects grouped by scale range are assigned to the corresponding features. However, the design of multi-scale feature production is quite hand-crafted or partiall... | ["Object Detection", "Neural Architecture Search"] | ABSTRACTMany commonly-used detection frameworks aim to handle the multi-scale objectdetection problem. The input image is always encoded to multi-scale features andobjects grouped by scale range are assigned to the corresponding features. How-ever, the design of multi-scale feature production is quite hand-crafted or p... | jVcwNQp7MCw | Good results with limited novelty | 4: Ok but not good enough - rejection | This paper argues searching for both encoder and anchor assignment (feature utilization) in a unified NAS framework would lead to better detector performance. Experiments have been conducted on COCO dataset to show proposed method outperform baseline FPN by a significant margin, and both searching for better encoder 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
Multi-scale Network Architecture Search for Object Detection
### Paper Abstract
Many commonly-used detection frameworks aim to handle the multi-scale object detection pr... |
SJGvns0qK7 | ICLR.cc/2019/Conference | 2019 | Bayesian Policy Optimization for Model Uncertainty | ["Gilwoo Lee", "Brian Hou", "Aditya Mandalika", "Jeongseok Lee", "Sanjiban Choudhury", "Siddhartha S. Srinivasa"] | Addressing uncertainty is critical for autonomous systems to robustly adapt to the real world. We formulate the problem of model uncertainty as a continuous Bayes-Adaptive Markov Decision Process (BAMDP), where an agent maintains a posterior distribution over latent model parameters given a history of observations and ... | ["Bayes-Adaptive Markov Decision Process", "Model Uncertainty", "Bayes Policy Optimization"] | Published as a conference paper at ICLR 2019Bayesian Policy Optimization forModel UncertaintyGilwoo Lee, Brian Hou, Aditya Mandalika Vamsikrishna, Jeongseok Lee,Sanjiban Choudhury, Siddhartha S. SrinivasaPaul G. Allen School of Computer Science & EngineeringUniversity of Washington{gilwoo,bhou,adityavk,jslee02,sanjibac... | Hklm9vec3m | Experiment results are not much convincing. | 7: Good paper, accept | Summary: This paper proposes a policy optimization framework for Bayesian RL (BPO). BPO is based on a Bayesian model-based RL formulation. Using a Bayesian approach, it is expected to have better trade-off between exploration and exploitation in RL, and be able to deal with model uncertainty as well. Experiments are do... | 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
Bayesian Policy Optimization for Model Uncertainty
### Paper Abstract
Addressing uncertainty is critical for autonomous systems to robustly adapt to the real world. We f... |
D_I6trPKwlt | ICLR.cc/2021/Conference | 2021 | Spectrally Similar Graph Pooling | ["Kyoung-Woon On", "Eun-Sol Kim", "Il-Jae Kwon", "Sangwoong Yoon", "Byoung-Tak Zhang"] | We consider the problem of learning compositional hierarchies of graphs. Even though structural characteristics of graphs can be learned by Graph Neural Networks (GNNs), it is difficult to find an overall compositional hierarchy using such flat operators.
In this paper, we propose a new graph pooling algorithm, Spectra... | ["Graph Neural Networks", "Graph Pooling", "Spectral Similarity on Graph"] | ABSTRACTWe consider the problem of learning compositional hierarchies of graphs. Eventhough structural characteristics of graphs can be learned by Graph Neural Net-works (GNNs), it is difficult to find an overall compositional hierarchy using suchflat operators. In this paper, we propose a new graph pooling algorithm, ... | N6stbIy6rGE | This paper tackles an interesting problem on GNNs. However, the current presentation and evaluation of the paper is not good enough | 4: Ok but not good enough - rejection | Graph Neural Networks (GNNs) is an increasing popular topic of research in machine learning on irregular graph-structured data. This paper tackles the problem of graph pooling and presents Spectrally Similar Graph Pooling (SSGPool) algorithm for learning hierarchical representations of the graphs. The main idea of 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
Spectrally Similar Graph Pooling
### Paper Abstract
We consider the problem of learning compositional hierarchies of graphs. Even though structural characteristics of gr... |
H1ecDoR5Y7 | ICLR.cc/2019/Conference | 2019 | Local Stability and Performance of Simple Gradient Penalty $\mu$-Wasserstein GAN | ["Cheolhyeong Kim", "Seungtae Park", "Hyung Ju Hwang"] | Wasserstein GAN(WGAN) is a model that minimizes the Wasserstein distance between a data distribution and sample distribution. Recent studies have proposed stabilizing the training process for the WGAN and implementing the Lipschitz constraint. In this study, we prove the local stability of optimizing the simple gradien... | ["WGAN", "gradient penalty", "stability", "measure valued differentiation"] | ABSTRACTWasserstein GAN(WGAN) is a model that minimizes the Wasserstein distancebetween a data distribution and sample distribution. Recent studies have proposedstabilizing the training process for the WGAN and implementing the Lipschitzconstraint. In this study, we prove the local stability of optimizing the simple gr... | BylhohL6hQ | assumptions need better justification | 4: Ok but not good enough - rejection | This paper shows that an ideal equilibrium point of a SGP-WGAN is stable. It makes several assumptions that, while clear why they are needed in the proof, is unjustified in practice. The authors should elaborate on these assumptions and comment on why they are reasonable.
Assumptions 1 and 3 essentially say that ther... | 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 Stability and Performance of Simple Gradient Penalty $\mu$-Wasserstein GAN
### Paper Abstract
Wasserstein GAN(WGAN) is a model that minimizes the Wasserstein dista... |
SklcyJBtvB | ICLR.cc/2020/Conference | 2020 | Off-policy Bandits with Deficient Support | ["Noveen Sachdeva", "Yi Su", "Thorsten Joachims"] | Off-policy training of contextual-bandit policies is attractive in online systems (e.g. search, recommendation, ad placement), since it enables the reuse of large amounts of log data from the production system. State-of-the-art methods for off-policy learning, however, are based on inverse propensity score (IPS) weight... | ["Recommender System", "Search Engine", "Counterfactual Learning"] | ABSTRACTOff-policy training of contextual-bandit policies is attractive in online systems(e.g. search, recommendation, ad placement), since it enables the reuse of largeamounts of log data. State-of-the-art methods for off-policy learning, however, arebased on inverse propensity score (IPS) weighting, which requires th... | H1glUK7pFr | Official Blind Review #2 | 3: Weak Reject | This paper considers a new off-policy contextual-bandit method that can learn even when the logging policy has deficient support. Three approaches are explored, namely restricting the action space, reward extrapolation, and restricting the policy space.
This paper is well written and it considers an important problem... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Off-policy Bandits with Deficient Support
### Paper Abstract
Off-policy training of contextual-bandit policies is attractive in online systems (e.g. search, recommendati... | |
e-ZdxsIwweR | ICLR.cc/2021/Conference | 2021 | Robust Constrained Reinforcement Learning for Continuous Control with Model Misspecification | ["Daniel J Mankowitz", "Dan Andrei Calian", "Rae Jeong", "Cosmin Paduraru", "Nicolas Heess", "Sumanth Dathathri", "Martin Riedmiller", "Timothy Mann"] | Many real-world physical control systems are required to satisfy constraints upon deployment. Furthermore, real-world systems are often subject to effects such as non-stationarity, wear-and-tear, uncalibrated sensors and so on. Such effects effectively perturb the system dynamics and can cause a policy trained successf... | ["reinforcement learning", "constraints", "robustness"] | ABSTRACTMany real-world physical control systems are required to satisfy constraints upondeployment. Furthermore, real-world systems are often subject to effects suchas non-stationarity, wear-and-tear, uncalibrated sensors and so on. Such effectseffectively perturb the system dynamics and can cause a policy trained suc... | 4P9LkndnRQ_ | Encouraging results, but the theory section needs more work | 4: Ok but not good enough - rejection | The standard Reinforcement Learning framework is limited in many ways, and numerous variants have been introduced to deal with aspects such as partial observability, temporal abstraction, safety, domain transfer, etc. Yet, these issues are often studied separately and it is often unclear how to combine them together. T... | 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
Robust Constrained Reinforcement Learning for Continuous Control with Model Misspecification
### Paper Abstract
Many real-world physical control systems are required to ... |
bFb3V8ALx4W | MIDL.io/2023/Short_Paper_Track | 2023 | Visualizing chest X-ray dataset biases using GANs | ["Hao Liang", "Kangqi Ni", "Guha Balakrishnan"] | Recent work demonstrates that images from various chest X-ray datasets contain visual features that are strongly correlated with protected demographic attributes like race and gender. This finding raises issues of fairness, since some of these factors may be used by downstream algorithms for clinical predictions. In th... | ["Chest X-rays", "fairness", "bias", "explainability", "generative adversarial networks (GANs)"] | Medical Imaging with Deep LearningVisualizing chest X-ray dataset biases using GANsHao Liang hl106@rice.eduKangqi Ni kn22@rice.eduGuha Balakrishnan guha@rice.eduDepartment of Electrical and Computer Engineering, Rice University, USAAbstractRecent work demonstrates that images from various chest X-ray datasets contain v... | 5VcC3frEqaZ | Interesting findings | 6: Marginally above acceptance threshold | This work presents a novel approach that highlights the most different features between different scanners that belong to different demographic groups.
Strengths:
- This method can be useful in practice, as it can potentially mitigate biases on downstream tasks that make decisions using data that contains multiple de... | 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
Visualizing chest X-ray dataset biases using GANs
### Paper Abstract
Recent work demonstrates that images from various chest X-ray datasets contain visual features that ... |
rJg851rYwH | ICLR.cc/2020/Conference | 2020 | Making the Shoe Fit: Architectures, Initializations, and Tuning for Learning with Privacy | ["Nicolas Papernot", "Steve Chien", "Shuang Song", "Abhradeep Thakurta", "Ulfar Erlingsson"] | Because learning sometimes involves sensitive data, standard machine-learning algorithms have been extended to offer strong privacy guarantees for training data. However, in practice, this has been mostly an afterthought, with privacy-preserving models obtained by re-running training with a different optimizer, but usi... | ["differential privacy", "deep learning"] | ABSTRACTBecause learning sometimes involves sensitive data, standard machine-learningalgorithms have been extended to offer strong privacy guarantees for trainingdata. However, in practice, this has been mostly an afterthought, with privacy-preserving models obtained by re-running training with a different optimizer, b... | B1gcmtH15B | Official Blind Review #2 | 3: Weak Reject | Overall, this work empirically evaluates different techniques used in privacy learning and suggest useful methods to stabilize or improve performance.
Detail comments:
Strength:
Despite the progress of privacy-preserving learning in theory, there are few works providing learning details for better training. Especiall... | <|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 the Shoe Fit: Architectures, Initializations, and Tuning for Learning with Privacy
### Paper Abstract
Because learning sometimes involves sensitive data, standard... | |
B1gF56VYPH | ICLR.cc/2020/Conference | 2020 | Deep 3D Pan via local adaptive "t-shaped" convolutions with global and local adaptive dilations | ["Juan Luis Gonzalez Bello", "Munchurl Kim"] | Recent advances in deep learning have shown promising results in many low-level vision tasks. However, solving the single-image-based view synthesis is still an open problem. In particular, the generation of new images at parallel camera views given a single input image is of great interest, as it enables 3D visualizat... | ["Deep learning", "Stereoscopic view synthesis", "Monocular depth", "Deep 3D Pan"] | ABSTRACTRecent advances in deep learning have shown promising results in many low-levelvision tasks. However, solving the single-image-based view synthesis is still anopen problem. In particular, the generation of new images at parallel camera viewsgiven a single input image is of great interest, as it enables 3D visua... | B1euNLVnFH | Official Blind Review #1 | 6: Weak Accept | The paper considers the problem of performing stereoscopic view synthesis (i.e., generating a new view seen from a different camera position) at an arbitrary position along the X-axis from a single input image only. This is an important problem as it enables 3D visualization of a 2D input scene. The paper focuses on th... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deep 3D Pan via local adaptive "t-shaped" convolutions with global and local adaptive dilations
### Paper Abstract
Recent advances in deep learning have shown promising ... | |
B1e8CsRctX | ICLR.cc/2019/Conference | 2019 | Generative Ensembles for Robust Anomaly Detection | ["Hyunsun Choi", "Eric Jang"] | Deep generative models are capable of learning probability distributions over large, high-dimensional datasets such as images, video and natural language. Generative models trained on samples from p(x) ought to assign low likelihoods to out-of-distribution (OoD) samples from q(x), making them suitable for anomaly detec... | ["Anomaly Detection", "Uncertainty", "Out-of-Distribution", "Generative Models"] | ABSTRACTDeep generative models are capable of learning probability distributions overlarge, high-dimensional datasets such as images, video and natural language. Gen-erative models trained on samples from p(x)ought to assign low likelihoods toout-of-distribution (OoD) samples from q(x), making them suitable for anomaly... | SJxjubiqhm | Needs a lot of work on improving technical rigor and clarity | 5: Marginally below acceptance threshold | Note to Area Chair: Another paper submitted to ICLR under the title “Do Deep Generative Models Know What They Don’t Know?” shares several similarities with the current submission.
This paper highlights a deficiency of current generative models in detecting out-of-distribution based samples based on likelihoods assigne... | 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
Generative Ensembles for Robust Anomaly Detection
### Paper Abstract
Deep generative models are capable of learning probability distributions over large, high-dimensiona... |
AJTAcS7SZzf | ICLR.cc/2021/Conference | 2021 | AUTOSAMPLING: SEARCH FOR EFFECTIVE DATA SAMPLING SCHEDULES | ["Ming Sun", "Haoxuan Dou", "Baopu Li", "Junjie Yan", "Wanli Ouyang"] | Data sampling acts as a pivotal role in training deep learning models. However, an effective sampling schedule is difficult to learn due to its inherent high-dimension as a hyper-parameter. In this paper, we propose the AutoSampling method to automatically learn sampling schedules for model training, which consists of ... | ["Hyper-parameter Learning", "AutoML", "Computer Vision"] | ABSTRACTData sampling acts as a pivotal role in training deep learning models. However, aneffective sampling schedule is difficult to learn due to its inherent high-dimensionas a hyper-parameter. In this paper, we propose the AutoSampling method toautomatically learn sampling schedules for model training, which consist... | HcUj9JCXDX | A paper marginally above average | 6: Marginally above acceptance threshold | The authors mainly concentrate on data sampling. To address the issue of optimizing high-dimensional sampling hyper-parameter in data sampling and release the requirement of prior knowledge from current methods, the authors introduce a searching-based method named AutoSampling. This method is comprised of exploration s... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
AUTOSAMPLING: SEARCH FOR EFFECTIVE DATA SAMPLING SCHEDULES
### Paper Abstract
Data sampling acts as a pivotal role in training deep learning models. However, an effectiv... |
r1f78iAcFm | ICLR.cc/2019/Conference | 2019 | GRAPH TRANSFORMATION POLICY NETWORK FOR CHEMICAL REACTION PREDICTION | ["Kien Do", "Truyen Tran", "Svetha Venkatesh"] | We address a fundamental problem in chemistry known as chemical reaction product prediction. Our main insight is that the input reactant and reagent molecules can be jointly represented as a graph, and the process of generating product molecules from reactant molecules can be formulated as a sequence of graph transform... | ["Chemical Reaction", "Graph Transformation", "Reinforcement Learning"] | ABSTRACTWe address a fundamental problem in chemistry known as chemical reaction prod-uct prediction. Our main insight is that the input reactant and reagent molecules canbe jointly represented as a graph, and the process of generating product moleculesfrom reactant molecules can be formulated as a sequence of graph tr... | BylmJnLcnX | Nice application. Some edits are needed. | 6: Marginally above acceptance threshold | Update:
Score increased.
___________________________________
Original review:
The paper presents an approach to predict the products of chemical reactions, given the reactants and reagents. It works by stepwise predicting the atom pairs that change their bonds in course of reaction, and then adjusting the bonds bet... | 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
GRAPH TRANSFORMATION POLICY NETWORK FOR CHEMICAL REACTION PREDICTION
### Paper Abstract
We address a fundamental problem in chemistry known as chemical reaction product ... |
3SqrRe8FWQ- | ICLR.cc/2021/Conference | 2021 | WrapNet: Neural Net Inference with Ultra-Low-Precision Arithmetic | ["Renkun Ni", "Hong-min Chu", "Oscar Castaneda", "Ping-yeh Chiang", "Christoph Studer", "Tom Goldstein"] | Low-precision neural networks represent both weights and activations with few bits, drastically reducing the cost of multiplications. Meanwhile, these products are accumulated using high-precision (typically 32-bit) additions. Additions dominate the arithmetic complexity of inference in quantized (e.g., binary) nets, ... | ["quantization", "efficient inference"] | ABSTRACTLow-precision neural networks represent both weights and activations with fewbits, drastically reducing the cost of multiplications. Meanwhile, these productsare accumulated using high-precision (typically 32-bit) additions. Additions dom-inate the arithmetic complexity of inference in quantized (e.g., binary) ... | EWkBW2PUhIN | An interesting idea on an important issue | 7: Good paper, accept | This paper explores to solve an often ignored issue in quantization: accumulation precision. As the bit-width of input scales down, the area/energy cost of the accumulator starts to dominate. The cyclic method proposed by the authors at the first glance is not intuitive. However, it's surprising that the surveyed mode... | 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
WrapNet: Neural Net Inference with Ultra-Low-Precision Arithmetic
### Paper Abstract
Low-precision neural networks represent both weights and activations with few bits,... |
S7KQa2XYpiQ | ICLR.cc/2021/Conference | 2021 | Improve Novel Class Generalization By Adaptive Feature Distribution for Few-Shot Learning | ["Ran Tao", "Marios Savvides"] | In this work, we focus on improving the novel class generalization of few-shot learning. By addressing the difference between feature distributions of base and novel classes, we propose the adaptive feature distribution method which is to finetune one scale vector using the support set of novel classes. The scale vecto... | ["Novel Class Generalization", "Finetuning One Scale Vector", "Adaptive Feature Distribution", "Cross-Domain"] | ABSTRACTIn this work, we focus on improving the novel class generalization of few-shotlearning. By addressing the difference between feature distributions of base andnovel classes, we propose the adaptive feature distribution method which is tofinetune one scale vector using the support set of novel classes. The scale ... | KPFFE34fbz3 | Proposes an interesting approach but suffers from clarity issues | 4: Ok but not good enough - rejection | Summary
========
This paper proposes a simple approach to few-shot classification which first pre-trains a feature extractor via the classification loss on all base classes, as is usually done. Then, to solve each test task, they propose to learn a task-specific scaling co-efficient for each (normalized) feature of th... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Improve Novel Class Generalization By Adaptive Feature Distribution for Few-Shot Learning
### Paper Abstract
In this work, we focus on improving the novel class generali... |
1uufzxsxfL | AKBC.ws/2020/Conference | 2020 | Revisiting Evaluation of Knowledge Base Completion Models | ["Pouya Pezeshkpour", "Yifan Tian", "Sameer Singh"] | Representing knowledge graphs (KGs) by learning embeddings for entities and relations has led to accurate models for existing KG completion benchmarks. However, due to the open-world assumption of existing KGs, evaluation of KG completion uses ranking metrics and triple classification with negative samples, and is thus... | ["Knowledge Graph Completion", "Link prediction", "Calibration", "Triple Classification"] | Automated Knowledge Base Construction (2020) Conference paperRevisiting Evaluation of Knowledge Base Completion ModelsPouya Pezeshkpour PEZESHKP @UCI.EDUUniversity of California, IrvineYifan Tian YIFANT @UCI.EDUUniversity of California, IrvineSameer Singh SAMEER @UCI.EDUUniversity of California, IrvineAbstractRepresent... | 18RI3Qm8S | A good analysis on popular KB-completion datasets plus a carefully labeled triple classification dataset | 7: Good paper, accept | This paper first analyzes several popular KB-completion datasets and their evaluation methods. Several issues have been highlighted and discussed, such as the assumptions made in the ranking metrics, skewed distributions on semi-inverse relations (in WN18RR & YAGO3-10), confidence scores by popular methods are not cal... | 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
Revisiting Evaluation of Knowledge Base Completion Models
### Paper Abstract
Representing knowledge graphs (KGs) by learning embeddings for entities and relations has le... |
SJgw51HFDr | ICLR.cc/2020/Conference | 2020 | Sparse Weight Activation Training | ["Md Aamir Raihan", "Tor M. Aamodt"] | Training convolutional neural networks (CNNs) is time consuming. Prior work has explored how to reduce the computational demands of training by eliminating gradients with relatively small magnitude. We show that eliminating small magnitude components has limited impact on the direction of high-dimensional vectors. Howe... | ["Sparsity", "Training", "Acceleration", "Pruning", "Compression"] | ABSTRACTTraining convolutional neural networks (CNNs) is time consuming. Prior workhas explored how to reduce the computational demands of training by eliminat-ing gradients with relatively small magnitude. We show that eliminating smallmagnitude components has limited impact on the direction of high-dimensionalvectors... | S1xw9KbyKH | Official Blind Review #3 | 6: Weak Accept | This paper studies training neural networks with sparse weights and sparse activations (SWAT training). By using sparse weights in forward passes as well as sparse weights and activations in backward passes, SWAT can reduce the computation overhead and also reduce the training memory footprint. 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
Sparse Weight Activation Training
### Paper Abstract
Training convolutional neural networks (CNNs) is time consuming. Prior work has explored how to reduce the computati... | |
XTPfeOoZD8 | NeurIPS.cc/2022/Workshop/SVRHM | 2022 | Measuring the Alignment of ANNs and Primate V1 on Luminance and Contrast Response Characteristics | ["Stephanie Olaiya", "Tiago Marques", "James J. DiCarlo"] | Some artificial neural networks (ANNs) are the current state-of-the-art in modeling the primate ventral stream and object recognition behavior. However, how well they align with luminance and contrast processing in early visual areas is not known. Here, we compared luminance and contrast processing in ANN models of V1 ... | ["Primary Visual Cortex", "Luminance", "Contrast", "Artificial Neural Networks"] | Measuring the Alignment of ANNs and Primate V1 onLuminance and Contrast Response CharacteristicsStephanie Olaiya1,∗, Tiago Marques2,3,4,5,6,∗,§, James J. DiCarlo2,3,41Department of Neuroscience, Brown University2Department of Brain and Cognitive Sciences, MIT, Cambridge, MA021393McGovern Institute for Brain Research, M... | 88g-mj48Tbk | Moving towards psychological plausiblity in addition to neural plausibility for CNNs | 6: Marginally above acceptance threshold | This paper attempts to make CNNs more psychologically plausible by investigating behaviour in addition to neural plausibility. Comparisons between single-neuron recordings of Non-human primate V1 spiking data and Deep Neural Networks neurons are made. Specifically, luminance and contrast response characteristics are in... | 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
Measuring the Alignment of ANNs and Primate V1 on Luminance and Contrast Response Characteristics
### Paper Abstract
Some artificial neural networks (ANNs) are the curre... |
Hyl8yANFDB | ICLR.cc/2020/Conference | 2020 | Assessing Generalization in TD methods for Deep Reinforcement Learning | ["Emmanuel Bengio", "Doina Precup", "Joelle Pineau"] | Current Deep Reinforcement Learning (DRL) methods can exhibit both data inefficiency and brittleness, which seem to indicate that they generalize poorly. In this work, we experimentally analyze this issue through the lens of memorization, and show that it can be observed directly during training. More precisely, we fi... | ["reinforcement learning", "deep learning", "generalization"] | ABSTRACTCurrent Deep Reinforcement Learning (DRL) methods can exhibit both data in-efficiency and brittleness, which seem to indicate that they generalize poorly. Inthis work, we experimentally analyze this issue through the lens of memorization ,and show that it can be observed directly during training. More precisely... | rJx71n4AtS | Official Blind Review #3 | 6: Weak Accept | The manuscript is analyzing the "generalization" in TD(lambda) methods. It includes supervised learning from trajectories, on-policy imitation learning, and basic RL setting. Moreover, memoization performance has also been measured. Main conclusion is the fact that TD(0) performs very similar to tabular learning failin... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Assessing Generalization in TD methods for Deep Reinforcement Learning
### Paper Abstract
Current Deep Reinforcement Learning (DRL) methods can exhibit both data ineffi... | |
SJNDWNOlg | ICLR.cc/2017/conference | 2017 | What Is the Best Practice for CNNs Applied to Visual Instance Retrieval? | ["Jiedong Hao", "Jing Dong", "Wei Wang", "Tieniu Tan"] | Previous work has shown that feature maps of deep convolutional neural networks (CNNs)
can be interpreted as feature representation of a particular image region. Features aggregated from
these feature maps have been exploited for image retrieval tasks and achieved state-of-the-art performances in
recent years. The key ... | ["Computer vision", "Deep learning"] | ABSTRACTPrevious work has shown that feature maps of deep convolutional neural networks(CNNs) can be interpreted as feature representation of a particular image region.Features aggregated from these feature maps have been exploited for image re-trieval tasks and achieved state-of-the-art performances in recent years. T... | By5gfegEg | An outdated method with misleading claims. | 3: Clear rejection | This paper explores different strategies for instance-level image retrieval with deep CNNs. The approach consists of extracting features from a network pre-trained for image classification (e.g. VGG), and post-process them for image retrieval. In other words, the network is off-the-shelf and solely acts as a feature ex... | 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
What Is the Best Practice for CNNs Applied to Visual Instance Retrieval?
### Paper Abstract
Previous work has shown that feature maps of deep convolutional neural networ... |
9p2CltauWEY | ICLR.cc/2021/Conference | 2021 | On Size Generalization in Graph Neural Networks | ["Gilad Yehudai", "Ethan Fetaya", "Eli Meirom", "Gal Chechik", "Haggai Maron"] | Graph neural networks (GNNs) can process graphs of different sizes but their capacity to generalize across sizes is still not well understood. Size generalization is key to numerous GNN applications, from solving combinatorial optimization problems to learning in molecular biology. In such problems, obtaining labels an... | ["graph neural networks", "gnn", "generalization", "Weisfeiler-Lehman"] | ABSTRACTGraph neural networks (GNNs) can process graphs of different sizes but theircapacity to generalize across sizes is still not well understood. Size generalizationis key to numerous GNN applications, from solving combinatorial optimizationproblems to learning in molecular biology. In such problems, obtaining labe... | lEa9viabW5 | Novel and well-motivated investigation into GNN size generalization error | 7: Good paper, accept | Summary:
This paper investigates the issue of generalizability of GNNs when trained on small graphs and tested on larger graphs, a common setting in graph learning. The paper argues that the ability of constant-depth GNNs to generalize is not dependent on the size difference, but on the difference in distributions of n... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On Size Generalization in Graph Neural Networks
### Paper Abstract
Graph neural networks (GNNs) can process graphs of different sizes but their capacity to generalize ac... |
kBVJ2NtiY- | ICLR.cc/2021/Conference | 2021 | Learning What To Do by Simulating the Past | ["David Lindner", "Rohin Shah", "Pieter Abbeel", "Anca Dragan"] | Since reward functions are hard to specify, recent work has focused on learning policies from human feedback. However, such approaches are impeded by the expense of acquiring such feedback. Recent work proposed that agents have access to a source of information that is effectively free: in any environment that humans h... | ["imitation learning", "reward learning", "reinforcement learning"] | ABSTRACTSince reward functions are hard to specify, recent work has focused on learningpolicies from human feedback. However, such approaches are impeded by theexpense of acquiring such feedback. Recent work proposed that agents have accessto a source of information that is effectively free: in any environment that hum... | qYadRr6LzR0 | Cool idea but a bit limited in a number of ways. I did not find the empirical results fully convincing. | 5: Marginally below acceptance threshold | This paper introduces an algorithm, called deep reward learning by simulating the past (deep RLSP), that seeks to infer a reward function by looking at states in demonstration data. An example of this described in the paper is an environment with a vase: if demonstration data shows an intact vase in the presence of 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
Learning What To Do by Simulating the Past
### Paper Abstract
Since reward functions are hard to specify, recent work has focused on learning policies from human feedbac... |
Hyewf3AqYX | ICLR.cc/2019/Conference | 2019 | A Frank-Wolfe Framework for Efficient and Effective Adversarial Attacks | ["Jinghui Chen", "Jinfeng Yi", "Quanquan Gu"] | Depending on how much information an adversary can access to, adversarial attacks can be classified as white-box attack and black-box attack. In both cases, optimization-based attack algorithms can achieve relatively low distortions and high attack success rates. However, they usually suffer from poor time and query co... | ["efficient", "framework", "effective adversarial attacks", "attack", "attack algorithms", "algorithms", "much information", "adversary", "access", "adversarial attacks"] | ABSTRACTDepending on how much information an adversary can access to, adversarial at-tacks can be classified as white-box attack and black-box attack. In both cases,optimization-based attack algorithms can achieve relatively low distortions andhigh attack success rates. However, they usually suffer from poor time and q... | SyllU9Iq37 | A method to produce adversarial attack using a Frank-Wolfe inspired method | 5: Marginally below acceptance threshold | This paper provide a method to produce adversarial attack using a Frank-Wolfe inspired method.
I have some concerns about the motivation of this method:
- What are the motivations to use Frank-Wolfe ? Usually this algorithm is used when the constraints are to complicated to have a tractable projection (which is not... | 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 Frank-Wolfe Framework for Efficient and Effective Adversarial Attacks
### Paper Abstract
Depending on how much information an adversary can access to, adversarial atta... |
5zErZzsW2U1 | ICLR.cc/2021/Conference | 2021 | Category Disentangled Context: Turning Category-irrelevant Features Into Treasures | ["Keke Tang", "Guodong Wei", "Jie Zhu", "Yuexin Ma", "Runnan Chen", "Zhaoquan Gu", "Wenping Wang"] | Deep neural networks have achieved great success in computer vision, thanks to their ability in extracting category-relevant semantic features. On the contrary, irrelevant features (e.g., background and confusing parts) are usually considered to be harmful. In this paper, we bring a new perspective on the potential ben... | ["context", "irrelevant features", "cdc", "features", "category", "treasures category", "great success", "computer vision", "thanks"] | ABSTRACTDeep neural networks have achieved great success in computer vision, thanks totheir ability in extracting category-relevant semantic features. On the contrary,irrelevant features (e.g., background and confusing parts) are usually consideredto be harmful. In this paper, we bring a new perspective on the potentia... | sz0uV8GCxxd | Paper review from AnonReviewer4 | 4: Ok but not good enough - rejection | Summary:
The paper proposes an attention method for image classification by considering the category-irrelevant information. Specifically, the authors first pre-train a network to extract the attention map as the category irrelevant cue. Then the paper introduces a few designs (feature channel-wise operations and 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
Category Disentangled Context: Turning Category-irrelevant Features Into Treasures
### Paper Abstract
Deep neural networks have achieved great success in computer visio... |
B9UN92zQnRF | ML_Reproducibility_Challenge/2021/Fall | 2021 | Learn to Resolve Conversational Dependency: A Consistency Training Framework for Conversational Question Answering | ["Anonymous"] | Scope of Reproducibility
In this process, in order to evaluate the accuracy of the claims mentioned in the paper and also the reusability of the paper codes, the codes introduced in GitHub were used. For this purpose, it was carried out in accordance with the instructions mentioned in it. Due to severe hardware limitat... | ["authors", "google colab", "model", "article", "gpu", "learn", "conversational dependency", "consistency training framework", "conversational question", "process"] | Reproducibility report of Learn to Resolve ConversationalDependency: A Consistency Training Framework forConversational Question Answering paperAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2In this process, in order to evaluate the accuracy of the claims mentioned in the p... | r8QlGfvc8ec | Short but detailed evaluation | 7: Good paper, accept | The authors did what was expected. They reproduced the results given their hardware and matched the results. They did try it with different parameters and got similar results, which shows robustness. There was one case they weren't able to reproduce | 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
Learn to Resolve Conversational Dependency: A Consistency Training Framework for Conversational Question Answering
### Paper Abstract
Scope of Reproducibility In this pr... |
CXQpLhjrTwJ | NeurIPS.cc/2021/Workshop/SVRHM | 2021 | Signal Strength and Noise Drive Feature Preference in CNN Image Classifiers | ["Max Wolff", "Stuart Wolff"] | Feature preference in Convolutional Neural Network (CNN) image classifiers is integral to their decision making process, and while the topic has been well studied, it is still not understood at a fundamental level. We test a range of task relevant feature attributes (including shape, texture, and color) with varying de... | ["shape", "texture", "color", "signal strength", "cnn image classifiers", "preference", "convolutional neural network", "cnn"] | Signal Strength and Noise Drive Feature Preferencein CNN Image ClassifiersMax WolffWesleyan Universitymswolff@wesleyan.eduStuart Wolffs.wolff1621@gmail.comAbstractFeature preference in Convolutional Neural Network (CNN) image classifiers isintegral to their decision making process, and while the topic has been well stu... | WKGVrUf7gxb | Interesting research question, but the research and paper structure is unclear | 4: Ok but not good enough - rejection | The study presented in this paper aims to better understand which features of images play a more significant role in determining the CNN final classification decision. The research stems from a line of research from Geirhos and colleagues who showed that CNNs tend to have a bias towards the texture whereas humans tend ... | 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
Signal Strength and Noise Drive Feature Preference in CNN Image Classifiers
### Paper Abstract
Feature preference in Convolutional Neural Network (CNN) image classifiers... |
IKQYmATn893 | ijcai.org/IJCAI/2021/Workshop/NSNLI | 2021 | Exploring Multi-hop Reasoning Process in NLU from the View of Bayesian Probability | ["Yitian Li", "Jidong Tian", "Hao HE", "Yaohui Jin"] | Emerging pre-trained language models (PTLMs), such as BERT and RoBERTa, have already achieved great success on many natural language understanding (NLU) tasks, spurring widespread interest for their potential in scientific and social areas, with accompanying criticism on their ambiguousness of reasoning, especially mul... | ["Reasoning", "Bayesian probability"] | Exploring Multi-hop Reasoning Process in NLU from the View of BayesianProbabilityYitian Li1;2,Jidong Tian1;2,Hao He1;2andYaohui Jin1;21MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University2State Key Lab of Advanced Optical Communication System and Network, Shanghai Jiao TongUniversityfyiti... | WASpMuR-AZp | Review | 6: Marginally above acceptance threshold | This paper analyses whether reasoning capabilities of a pretrained language model can be compared to a Bayesian network for a probabilistic reasoning process.
An interesting apporach of the paper is to test this out via intermediate results in the k-hop reasoning environment, where the model's probability of predicting... | 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
Exploring Multi-hop Reasoning Process in NLU from the View of Bayesian Probability
### Paper Abstract
Emerging pre-trained language models (PTLMs), such as BERT and RoB... |
D5lK-IW_xS | MIDL.io/2020/Conference | 2020 | Model Averaging and Augmented Inference for Stable Echocardiography Segmentation using 2D ConvNets | ["Joshua V. Stough"] | The automatic segmentation of heart substructures in 2D echocardiography images is a goal common to both clinicians and researchers. Convolutional neural networks (CNNs) have recently shown the best average performance. However, on the rare occasions that a trained CNN fails, it can fail spectacularly. To mitigate thes... | ["Convolutional Neural Networks", "Echocardiography", "Segmentation", "Data Augmentation"] | Medical Imaging with Deep Learning { Under Review 2020 Short Paper { MIDL 2020 submissionModel Averaging and Augmented Inference for StableEchocardiography Segmentation using 2D ConvNetsAuthor(s) names withheld email(s) withheldAddress withheldEditors: Under Review for MIDL 2020AbstractThe automatic segmentation of hea... | b_hjgIKX6U | Interesting paper but rather incremental contribution with very brief experimental evaluation | 2: Weak reject | The paper proposes to improve results of echocardiography imagery segmentation using model averaging and augmented inference. These ideas are not particularly novel, but have proven to be valuable in multiple recent studies.
In particular, the authors claim that averaging the predictions from multiple models improves ... | 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
Model Averaging and Augmented Inference for Stable Echocardiography Segmentation using 2D ConvNets
### Paper Abstract
The automatic segmentation of heart substructures i... |
BJesDsA9t7 | ICLR.cc/2019/Conference | 2019 | Better Accuracy with Quantified Privacy: Representations Learned via Reconstructive Adversarial Network | ["Sicong Liu", "Anshumali Shrivastava", "Junzhao Du", "Lin Zhong"] | The remarkable success of machine learning, especially deep learning, has produced a variety of cloud-based services for mobile users. Such services require an end user to send data to the service provider, which presents a serious challenge to end-user privacy. To address this concern, prior works either add noise to ... | ["end-user privacy", "utility", "feature learning", "adversarial training"] | ABSTRACTThe remarkable success of machine learning, especially deep learning, has pro-duced a variety of cloud-based services for mobile users. Such services require anend user to send data to the service provider, which presents a serious challenge toend-user privacy. To address this concern, prior works either add no... | B1xF4md62m | Nice idea. Need better experiments. | 4: Ok but not good enough - rejection | Privacy concerns arise when data is shared with third parties, a common occurrence. This paper proposes a privacy-preserving classification framework that consists of an encoder that extracts features from data, a classifier that performs the actual classification, and a decoder that tries to reconstruct the original d... | 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
Better Accuracy with Quantified Privacy: Representations Learned via Reconstructive Adversarial Network
### Paper Abstract
The remarkable success of machine learning, es... |
eZllW0F5aM_ | ICLR.cc/2021/Conference | 2021 | Don't stack layers in graph neural networks, wire them randomly | ["Diego Valsesia", "Giulia Fracastoro", "Enrico Magli"] | Graph neural networks have become a staple in problems addressing learning and analysis of data defined over graphs. However, several results suggest an inherent difficulty in extracting better performance by increasing the number of layers. Besides the classic vanishing gradient issues, recent works attribute this to ... | ["Graph neural networks", "random architectures"] | ABSTRACTGraph neural networks have become a staple in problems addressing learningand analysis of data defined over graphs. However, several results suggest aninherent difficulty in extracting better performance by increasing the number oflayers. Besides the classic vanishing gradient issues, recent works attribute thi... | 9q8J98wcmro | More details are needed. | 5: Marginally below acceptance threshold | This paper utilizes Randomly Wired architectures to boost deep GNNs. Theoretical analyses verify that randomly wired architectures behave like path ensemble and it enables adaptive receptive field. Experimental results on three non-popular datasets demonstrate the strength of the proposed model. Overall, the idea is in... | 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
Don't stack layers in graph neural networks, wire them randomly
### Paper Abstract
Graph neural networks have become a staple in problems addressing learning and analysi... |
rJljdh4KDH | ICLR.cc/2020/Conference | 2020 | Multi-Scale Representation Learning for Spatial Feature Distributions using Grid Cells | ["Gengchen Mai", "Krzysztof Janowicz", "Bo Yan", "Rui Zhu", "Ling Cai", "Ni Lao"] | Unsupervised text encoding models have recently fueled substantial progress in NLP. The key idea is to use neural networks to convert words in texts to vector space representations (embeddings) based on word positions in a sentence and their contexts, which are suitable for end-to-end training of downstream tasks. We s... | ["Grid cell", "space encoding", "spatially explicit model", "multi-scale periodic representation", "unsupervised learning"] | ABSTRACTUnsupervised text encoding models have recently fueled substantial progress inNatural Language Processing (NLP). The key idea is to use neural networks toconvert words in texts to vector space representations (embeddings) based on wordpositions in a sentence and their contexts, which are suitable for end-to-end... | Hye56-ChYr | Official Blind Review #3 | 6: Weak Accept | This paper presents a new method called "Space2Vec" to compute spatial embeddings of a pixel in a spatial data. The primary motivation of Space2Vec is to integrate representations of different spatial scales which could potentially make the spatial representations more informative and meaningful as features. Space2Vec ... | <|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-Scale Representation Learning for Spatial Feature Distributions using Grid Cells
### Paper Abstract
Unsupervised text encoding models have recently fueled substan... | |
BJepX2A9tX | ICLR.cc/2019/Conference | 2019 | Rotation Equivariant Networks via Conic Convolution and the DFT | ["Benjamin Chidester", "Minh N. Do", "Jian Ma"] | Performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Many image classification tasks, such as those related to cellular imaging, exhibit invariance to rotation. In particular, to aid convolutional neural networks in learning rotation invariance, we consider a s... | ["deep learning", "rotation equivariance", "bioimaging analysis"] | ABSTRACTPerformance of neural networks can be significantly improved by encoding knowninvariance for particular tasks. In particular, to aid convolutional neural networksin learning rotation invariance, we consider a simple, efficient conic convolutionalscheme that encodes rotational equivariance, along with a method f... | Hkgm-rB22m | Okay paper with limited novelty and lacking experimental evidence for main supposed advantages of proposed method | 4: Ok but not good enough - rejection | Disclaimer: I've already reviewed an earlier version of this manuscript for NIPS 2018.
# Summary
In the context of image classification, the paper proposes a convolutional neural network architecture with rotation-equivariant feature maps that are eventually made rotation-invariant by using the magnitude of the 2D di... | 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
Rotation Equivariant Networks via Conic Convolution and the DFT
### Paper Abstract
Performance of neural networks can be significantly improved by encoding known invaria... |
S1lVhxSYPH | ICLR.cc/2020/Conference | 2020 | Ternary MobileNets via Per-Layer Hybrid Filter Banks | ["Dibakar Gope", "Jesse G Beu", "Urmish Thakker", "Matthew Mattina"] | MobileNets family of computer vision neural networks have fueled tremendous progress in the design and organization of resource-efficient architectures in recent years. New applications with stringent real-time requirements in highly constrained devices require further compression of MobileNets-like already computeeffi... | ["Model compression", "ternary quantization", "energy-efficient models"] | ABSTRACTMobileNets family of computer vision neural networks have fueled tremendousprogress in the design and organization of resource-efficient architectures in re-cent years. New applications with stringent real-time requirements in highly con-strained devices require further compression of MobileNets-like already co... | ryxEfElTYB | Official Blind Review #2 | 3: Weak Reject | The authors focus on quantizing the MobileNets architecture to ternary values, resulting in less space and compute.
The space of making neural networks more energy efficient is vital towards their deployment in the real world.
I think the authors over-state their claims of no loss in accuracy, in Table 2 we see a cle... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Ternary MobileNets via Per-Layer Hybrid Filter Banks
### Paper Abstract
MobileNets family of computer vision neural networks have fueled tremendous progress in the desig... | |
MPO4oML_JC | ICLR.cc/2021/Conference | 2021 | Coordinated Multi-Agent Exploration Using Shared Goals | ["Iou-Jen Liu", "Unnat Jain", "Alex Schwing"] | Exploration is critical for good results of deep reinforcement learning algorithms and has drawn much attention. However, existing multi-agent deep reinforcement learning algorithms still use mostly noise-based techniques. It was recognized recently that noise-based exploration is suboptimal in multi-agent settings, an... | ["Multi-agent RL", "Deep RL", "Exploration"] | ABSTRACTExploration is critical for good results of deep reinforcement learning algorithmsand has attracted much attention. However, existing multi-agent deep reinforce-ment learning algorithms still use mostly noise-based techniques. It was recog-nized recently that noise-based exploration is suboptimal in multi-agent... | AFPeknCI_oT | Proposed technique has limited applicability | 5: Marginally below acceptance threshold | The paper proposes to improve the exponential sample complexity of finding a coordinated multi-agent strategy by learning an exploration policy for each agent that conditions on a shared goal. The exploration policy is mixed with the normal RL policy according to a parameter alpha, which is scaled down over time. The s... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Coordinated Multi-Agent Exploration Using Shared Goals
### Paper Abstract
Exploration is critical for good results of deep reinforcement learning algorithms and has draw... |
3IKKBxByalk | ICML.cc/2020/Workshop/SAS | 2020 | Adversarial representation learning for private speech generation | ["David Ericsson", "Adam \u00d6stberg", "Edvin Listo Zec", "John Martinsson", "Olof Mogren"] | As more data is collected in various settings across organizations, companies, and countries, there has been an increase in the demand of user privacy. Developing privacy preserving methods for data analytics is thus an important area of research. In this work we present a model based on generative adversarial networks... | ["adversarial networks", "representation learning", "speech", "audio", "privacy"] | Adversarial representation learning for private speech generationDavid Ericsson* 1 2Adam ̈Ostberg* 1 2Edvin Listo Zec2John Martinsson2Olof Mogren2AbstractAs more data is collected in various settingsacross organizations, companies, and countries,there has been an increase in the demand of userprivacy. Developing privac... | W-ULi23RUQL | Privacy in speech , applied to generation | 6: Marginally above acceptance threshold | Privacy in speech processing is a critical issue, both now and in the future. Because this paper sets out to do private speech generation, and manages to do so to some extent, I argue for acceptance despite the following critiques.
Given the stated goal of privacy preserving - isn't operating at either a higher level ... | 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
Adversarial representation learning for private speech generation
### Paper Abstract
As more data is collected in various settings across organizations, companies, and c... |
rJma2bZCW | ICLR.cc/2018/Conference | 2018 | Three factors influencing minima in SGD | ["Stanis\u0142aw Jastrz\u0119bski", "Zac Kenton", "Devansh Arpit", "Nicolas Ballas", "Asja Fischer", "Amos Storkey", "Yoshua Bengio"] | We study the statistical properties of the endpoint of stochastic gradient descent (SGD). We approximate SGD as a stochastic differential equation (SDE) and consider its Boltzmann Gibbs equilibrium distribution under the assumption of isotropic variance in loss gradients.. Through this analysis, we find that three fact... | ["SGD", "Deep Learning", "Generalization"] | ABSTRACTWe study the statistical properties of the endpoint of stochastic gradient descent(SGD). We approximate SGD as a stochastic differential equation (SDE) andconsider its Boltzmann Gibbs equilibrium distribution under the assumption ofisotropic variance in loss gradients.. Through this analysis, we find that three... | BkC-HgcxG | Interesting paper, however not convincing theoretical results | 3: Clear rejection | In this paper, the authors present an analysis of SGD within an SDE framework. The ideas and the presented results are interesting and are clearly of interest to the deep learning community. The paper is well-written overall.
However, the paper has important problems.
1) The analysis is widely based on the recent pa... | 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
Three factors influencing minima in SGD
### Paper Abstract
We study the statistical properties of the endpoint of stochastic gradient descent (SGD). We approximate SGD a... |
SkfNU2e0Z | ICLR.cc/2018/Conference | 2018 | Statestream: A toolbox to explore layerwise-parallel deep neural networks | ["Volker Fischer"] | Building deep neural networks to control autonomous agents which have to interact in real-time with the physical world, such as robots or automotive vehicles, requires a seamless integration of time into a network’s architecture. The central question of this work is, how the temporal nature of reality should be reflect... | ["model-parallel", "parallelization", "software platform"] | ABSTRACTBuilding deep neural networks to control autonomous agents which have to in-teract in real-time with the physical world, such as robots or automotive vehi-cles, requires a seamless integration of time into a network’s architecture. Thecentral question of this work is, how the temporal nature of reality should b... | B1KY-MqgG | Review of "STATESTREAM: A TOOLBOX TO EXPLORE LAYERWISE-PARALLEL DEEP NEURAL NETWORKS" | 5: Marginally below acceptance threshold | In this paper, the authors present an open-source toolbox to explore layerwise-parallel deep neural networks. They offer an interesting and detailed comparison of the temporal progression of layerwise-parallel and layerwise-sequential networks, and differences that can emerge in the results of these two computation str... | 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
Statestream: A toolbox to explore layerwise-parallel deep neural networks
### Paper Abstract
Building deep neural networks to control autonomous agents which have to int... |
48goXfYCVFX | ICLR.cc/2021/Conference | 2021 | Interpretable Relational Representations for Food Ingredient Recommendation Systems | ["Kana Maruyama", "Michael Spranger"] | Supporting chefs with ingredient recommender systems to create new recipes is challenging, as good ingredient combinations depend on many factors like taste, smell, cuisine style, texture among others. There have been few attempts to address these issues using machine learning. Importantly, useful models do obvious... | ["Metric Learning", "Gastronomy", "Memory Network", "Knowledge Graph", "Interpretable"] | ABSTRACTSupporting chefs with ingredient recommender systems to create new recipes ischallenging, as good ingredient combinations depend on many factors like taste,smell, cuisine style, texture among others. There have been few attempts to ad-dress these issues using machine learning. Useful Machine Learning models doo... | nnMaIcPjDZ | The paper is built upon modification to prior work related to memory network-based recommendations. The main contribution aspect of the paper is to apply this work in a completely new domain of food science. Overall the problem the authors are trying to solve is well defined and their approach is solid. | 7: Good paper, accept | Overall, I vote for accepting. The authors propose a novel approach to support chefs with creating/experimenting with new recipes to overcome the challenging combinations of taste/texture etc. that can result from addition of new ingredients. Their idea of adding interpretability to their results from a Food Knowledge ... | 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
Interpretable Relational Representations for Food Ingredient Recommendation Systems
### Paper Abstract
Supporting chefs with ingredient recommender systems to create new... |
9az9VKjOx00 | ICLR.cc/2021/Conference | 2021 | TopoTER: Unsupervised Learning of Topology Transformation Equivariant Representations | ["Xiang Gao", "Wei Hu", "Guo-Jun Qi"] | We present the Topology Transformation Equivariant Representation (TopoTER) learning, a general paradigm of unsupervised learning of node representations of graph data for the wide applicability to Graph Convolutional Neural Networks (GCNNs). We formalize the TopoTER from an information-theoretic perspective, by maximi... | ["Unsupervised learning", "node representations", "mutual information"] | ABSTRACTWe present the Topology Transformation Equivariant Representation (TopoTER)learning, a general paradigm of unsupervised learning of node representations ofgraph data for the wide applicability to Graph Convolutional Neural Networks(GCNNs). We formalize the TopoTER from an information-theoretic perspec-tive, by ... | 0AsbYbsOHe8 | A method for for self-training GNN | 7: Good paper, accept | The paper propose an unsupervised method for self-training of graph-neural-networks (GNNs). The authors provide information-theoretic justification to their method using maximization of the lower bound of the mutual information on their objective. Their approach is based on maximizing the mutual information between a 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
TopoTER: Unsupervised Learning of Topology Transformation Equivariant Representations
### Paper Abstract
We present the Topology Transformation Equivariant Representatio... |
BJlBNZDaP4 | icaps-conference.org/ICAPS/2019/Workshop/HSDIP | 2019 | A∗ Search and Bound-Sensitive Heuristics for Oversubscription Planning | ["Anonymous"] | Oversubscription planning (OSP) is the problem of finding plans that maximize the utility value of their end state while staying within a specified cost bound. Recently, it has been shown that OSP problems can be reformulated as classical planning problems with multiple cost functions but no utilities. Here we take ad... | ["oversubscription planning", "heuristic search", "A*", "bound-sensitive heuristics"] | ASearch and Bound-Sensitive Heuristics for Oversubscription PlanningMichael Katz1,Emil Keyder21IBM Research, Yorktown Heights, NY , USA2Invitae Corporation, San Francisco, CA, USAmichael.katz1@ibm.com, emilkeyder@gmail.comAbstractOversubscription planning (OSP) is the problemof finding plans that maximize the utility v... | SkxFLKli_V | missing related work - weak accept | 6: Marginally above acceptance threshold | The paper proposes modifications to admissible heuristics to make them better informed in a multi-criteria setting where one cost function is the minimization objective and one or more secondary cost functions are constrained by bounds. The modified heuristics are applied to a reformulation of oversubscription planning... | 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∗ Search and Bound-Sensitive Heuristics for Oversubscription Planning
### Paper Abstract
Oversubscription planning (OSP) is the problem of finding plans that maximize t... |
rywDjg-RW | ICLR.cc/2018/Conference | 2018 | Neural-Guided Deductive Search for Real-Time Program Synthesis from Examples | ["Ashwin Kalyan", "Abhishek Mohta", "Oleksandr Polozov", "Dhruv Batra", "Prateek Jain", "Sumit Gulwani"] | Synthesizing user-intended programs from a small number of input-output exam-
ples is a challenging problem with several important applications like spreadsheet
manipulation, data wrangling and code refactoring. Existing synthesis systems
either completely rely on deductive logic techniques that are extensively hand-
e... | ["Program synthesis", "deductive search", "deep learning", "program induction", "recurrent neural networks"] | ABSTRACTSynthesizing user-intended programs from a small number of input-output exam-ples is a challenging problem with several important applications like spreadsheetmanipulation, data wrangling and code refactoring. Existing synthesis systemseither completely rely on deductive logic techniques that are extensively ha... | SkPNib9ez | Incremental paper but well-written | 6: Marginally above acceptance threshold | This paper extends and speeds up PROSE, a programming by example system, by posing the selection of the next production rule in the grammar as a supervised learning problem.
This paper requires a large amount of background knowledge as it depends on understanding program synthesis as it is done in the programming lang... | 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
Neural-Guided Deductive Search for Real-Time Program Synthesis from Examples
### Paper Abstract
Synthesizing user-intended programs from a small number of input-output e... |
D5lK-IW_xS | MIDL.io/2020/Conference | 2020 | Model Averaging and Augmented Inference for Stable Echocardiography Segmentation using 2D ConvNets | ["Joshua V. Stough"] | The automatic segmentation of heart substructures in 2D echocardiography images is a goal common to both clinicians and researchers. Convolutional neural networks (CNNs) have recently shown the best average performance. However, on the rare occasions that a trained CNN fails, it can fail spectacularly. To mitigate thes... | ["Convolutional Neural Networks", "Echocardiography", "Segmentation", "Data Augmentation"] | Medical Imaging with Deep Learning { Under Review 2020 Short Paper { MIDL 2020 submissionModel Averaging and Augmented Inference for StableEchocardiography Segmentation using 2D ConvNetsAuthor(s) names withheld email(s) withheldAddress withheldEditors: Under Review for MIDL 2020AbstractThe automatic segmentation of hea... | yY-pqA1VlE2 | Model for improved performance on outliers on echocardiography segmentation using test-time augmentation and ensembling | 2: Weak reject | - The paper is very clearly written and the methods clearly described. The method involves ensembling 8 U-net models, trained on different overlapping folds of the echocardiography data with on-the-fly augmentations, and then applying test time augmentation by introducing 200 rotation variations and averaging the (unro... | 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
Model Averaging and Augmented Inference for Stable Echocardiography Segmentation using 2D ConvNets
### Paper Abstract
The automatic segmentation of heart substructures i... |
Sx-mvOvnmJj | ICLR.cc/2021/Conference | 2021 | Improving Calibration for Long-Tailed Recognition | ["Zhisheng Zhong", "Jiequan Cui", "Shu Liu", "Jiaya Jia"] | Deep neural networks often perform poorly when training datasets are heavily class-imbalanced. Recently, two-stage methods greatly improve the performances by decoupling representation learning and classifier learning. In this paper, we discover that networks trained on long-tailed datasets are more prone to miscalibra... | ["long tailed recognition", "network calibration", "label-aware smoothing", "mixup", "dataset bias"] | ABSTRACTDeep neural networks often perform poorly when training datasets are heavily class-imbalanced. Recently, two-stage methods greatly improve the performances bydecoupling representation learning and classifier learning. In this paper, we discoverthat networks trained on long-tailed datasets are more prone to misc... | QNzzYbirhI_ | Interesting paper with some promising results | 6: Marginally above acceptance threshold | Summary:
This paper addresses two major issues in previous works of long-tailed recognition: 1) models are over-confident on “head” classes. 2) the domain shift (dataset bias) is ignored in two-stage models. Accordingly, they propose two novel solutions including label-aware smoothing and shift learning on batch norma... | 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
Improving Calibration for Long-Tailed Recognition
### Paper Abstract
Deep neural networks often perform poorly when training datasets are heavily class-imbalanced. Recen... |
HksioDcxl | ICLR.cc/2017/conference | 2017 | Joint Training of Ratings and Reviews with Recurrent Recommender Networks | ["Chao-Yuan Wu", "Amr Ahmed", "Alex Beutel", "Alexander J. Smola"] | Accurate modeling of ratings and text reviews is at the core of successful recommender systems. While neural networks have been remarkably successful in modeling images and natural language, they have been largely unexplored in recommender system research. In this paper, we provide a neural network model that combines... | ["ratings", "reviews", "text reviews", "joint training", "recurrent recommender networks", "modeling", "core", "successful recommender systems", "neural networks"] | ABSTRACTAccurate modeling of ratings and text reviews is at the core of successful rec-ommender systems. While neural networks have been remarkably successful inmodeling images and natural language, they have been largely unexplored in rec-ommender system research. In this paper, we provide a neural network modelthat c... | BJ9p-XmEx | 6: Marginally above acceptance threshold | This paper proposed a joint model for rate prediction and text generation. The author compared the methods on a more realistic time based split setting, which requires “predict into the future.”
One major flaw of the paper is that it does not address the impact of BOW vs the RNN based text model, specifically RRN(rati... | 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
Joint Training of Ratings and Reviews with Recurrent Recommender Networks
### Paper Abstract
Accurate modeling of ratings and text reviews is at the core of successful r... | |
ryiAv2xAZ | ICLR.cc/2018/Conference | 2018 | Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples | ["Kimin Lee", "Honglak Lee", "Kibok Lee", "Jinwoo Shin"] | The problem of detecting whether a test sample is from in-distribution (i.e., training distribution by a classifier) or out-of-distribution sufficiently different from it arises in many real-world machine learning applications. However, the state-of-art deep neural networks are known to be highly overconfident in their... | ["classifiers", "samples", "classifier", "problem", "test sample", "distribution", "different", "many", "machine learning applications", "deep neural networks"] | ABSTRACTThe problem of detecting whether a test sample is from in-distribution (i.e., train-ing distribution by a classifier) or out-of-distribution sufficiently different from itarises in many real-world machine learning applications. However, the state-of-artdeep neural networks are known to be highly overconfident i... | B1klq-5lG | interesting idea for robust classification | 7: Good paper, accept | The manuscript proposes a generative approach to detect which samples are within vs. out of the sample space of the training distribution. This distribution is used to adjust the classifier so it makes confident predictions within sample, and less confident predictions out of sample, where presumably it is prone to mis... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples
### Paper Abstract
The problem of detecting whether a test sample is from in-distrib... |
Byxv9aioz | MIDL.amsterdam/2018/Conference | 2018 | Test-time augmentation with uncertainty estimation for deep learning-based medical image segmentation | ["Guotai Wang", "Wenqi Li", "Michael Aertsen", "Jan Deprest", "Sebastien Ourselin", "Tom Vercauteren"] | Data augmentation has been widely used for training deep learning systems for medical image segmentation and plays an important role in obtaining robust and transformation-invariant predictions. However, it has seldom been used at test time for segmentation and not been formulated in a consistent mathematical framework... | ["data augmentation", "uncertainty estimation", "image segmentation", "deep learning"] | Test-time augmentation with uncertainty estimationfor deep learning-based medical image segmentationGuotai WangUniversity College Londonguotai.wang.14@ucl.ac.ukWenqi LiUniversity College Londonwenqi.li@ucl.ac.ukMichael AertsenyKU Leuvenmichael.aertsen@uzleuven.beJan DeprestyKU Leuvenjan.deprest@uzleuven.beSébastien Ou... | B1WRAm4AG | Good topic to delve into, but we do not learn too much from this study | 2: Marginally below acceptance threshold | This paper investigates a heuristic trick that has been used a lot in deep learning system design known as test time augmentation.
Pro:
- Interesting to analyze this aspect/trick of optimizing the performance of deep networks (and possibly other methods too) in more detail
- Well written
Comments:
- The fetal experim... | 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
Test-time augmentation with uncertainty estimation for deep learning-based medical image segmentation
### Paper Abstract
Data augmentation has been widely used for train... |
ryllpA21vB | NeurIPS.cc/2019/Workshop/Program_Transformations | 2019 | Differentiation of High-Level Language Semantics | ["Michael Innes"] | Though analytic differentiation (AD) is a program transformation, AD tools have typically supported only very limited program representations, consisting of primitive mathematical operations and basic structured control flow. Zygote, an AD for the Julia language, instead operates on Julia code. This presents an interes... | ["zygote", "differentiation", "language semantics differentiation", "language semantics", "analytic differentiation", "program transformation", "ad tools", "limited program representations", "primitive mathematical operations"] | Differentiation of High-Level Language SemanticsMichael InnesJulia Computing, Inc.Edinburgh, UKmike.j.innes@gmail.comAbstractThough analytic differentiation (AD) is a program transformation, AD tools havetypically supported only very limited program representations, consisting of prim-itive mathematical operations and ... | Bke0oRWkOH | Solid implementation of AD in Julia | 7: Good paper, accept | The article is a good overview of all the AD features that have been made to work in Zygote, an implementation of AD in Julia. The whole language (including recursion and higher-level functions), excluding macros themselves, is supported.
Additional features have also been implemented, such as checkpointing, hooks, and... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Differentiation of High-Level Language Semantics
### Paper Abstract
Though analytic differentiation (AD) is a program transformation, AD tools have typically supported o... |
thhdrl4IdMm | ICLR.cc/2021/Conference | 2021 | A Chain Graph Interpretation of Real-World Neural Networks | ["Yuesong Shen", "Daniel Cremers"] | The last decade has witnessed a boom of deep learning research and applications achieving state-of-the-art results in various domains. However, most advances have been established empirically, and their theoretical analysis remains lacking. One major issue is that our current interpretation of neural networks (NNs) as ... | ["neural network interpretation", "chain graph", "deep learning theory", "probabilistic graphical model"] | ABSTRACTThe last decade has witnessed a boom of deep learning research and applicationsachieving state-of-the-art results in various domains. However, most advanceshave been established empirically, and their theoretical analysis remains lacking.One major issue is that our current interpretation of neural networks (NNs... | 9A3BfoaoUYc | alternative interpretation based on chain graph | 4: Ok but not good enough - rejection | This paper tries to interpret neural networks with chain graphs that provides theoretical analysis on various neural network components. Furthermore, this chain graph interpretation has been used to propose a new approach (architecture), which is a partially collapsed feed-forward. A layered chain graph representation ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Chain Graph Interpretation of Real-World Neural Networks
### Paper Abstract
The last decade has witnessed a boom of deep learning research and applications achieving s... |
rkeMj3VtvB | ICLR.cc/2020/Conference | 2020 | Multi-Task Adapters for On-Device Audio Inference | ["M. Tagliasacchi", "F. de Chaumont Quitry", "D. Roblek"] | The deployment of deep networks on mobile devices requires to efficiently use
the scarce computational resources, expressed as either available memory or
computing cost. When addressing multiple tasks simultaneously, it is extremely
important to share resources across tasks, especially when they all consume the
same in... | ["Audio", "multi-task learning"] | ABSTRACTThe deployment of deep networks on mobile devices requires to efficiently use thescarce computational resources, expressed as either available memory or computingcost. When addressing multiple tasks simultaneously, it is extremely importantto share resources across tasks, especially when they all consume the sa... | BJgWXCrpFS | Official Blind Review #2 | 3: Weak Reject | The authors present a method for the audio classification problem, using a pre-trained network (shared encoder) and creating adapters for each new task. This is an interesting alternative to transfer learning, and it would be great if minimal adapters can be designed to work with a baseline network.
The adapter networ... | <|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-Task Adapters for On-Device Audio Inference
### Paper Abstract
The deployment of deep networks on mobile devices requires to efficiently use the scarce computation... | |
kLyLW3RRqU | ICLR.cc/2021/Conference | 2021 | Representation Quality Of Neural Networks Links To Adversarial Attacks and Defences | ["Shashank Kotyan", "Moe Matsuki", "Danilo Vasconcellos Vargas"] | Neural networks have been shown vulnerable to a variety of adversarial algorithms. A crucial step for understanding the rationale behind this lack of robustness is to assess the potential of the neural networks' representation to encode the existing features. Here, we propose a method to understand the representation q... | ["Understanding Neural Networks", "Representation Metrics", "Adversarial Machine Learning", "Adversarial Attacks", "Adversarial Defences"] | ABSTRACTNeural networks have been shown vulnerable to a variety of adversarial algorithms.A crucial step for understanding the rationale behind this lack of robustness is toassess the potential of the neural networks’ representation to encode the existingfeatures. Here, we propose a method to understand the representat... | EF-GGQnDpu2 | Further development needed? | 4: Ok but not good enough - rejection | This paper is concerned with the assessment of the vulnerability of neural networks to adversarial attack, according to its ability to interpret new classes using previously-learned features. The underlying assumption made is that combinations of features from a sufficiently rich learned collection should be able to he... | 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
Representation Quality Of Neural Networks Links To Adversarial Attacks and Defences
### Paper Abstract
Neural networks have been shown vulnerable to a variety of adversa... |
wta_8Hx2KD | ICLR.cc/2021/Conference | 2021 | Incorporating Symmetry into Deep Dynamics Models for Improved Generalization | ["Rui Wang", "Robin Walters", "Rose Yu"] | Recent work has shown deep learning can accelerate the prediction of physical dynamics relative to numerical solvers. However, limited physical accuracy and an inability to generalize under distributional shift limit its applicability to the real world. We propose to improve accuracy and generalization by incorporating... | ["deep sequence model", "equivariant neural network", "physics-guided deep learning", "AI for earth science"] | ABSTRACTRecent work has shown deep learning can accelerate the prediction of physi-cal dynamics relative to numerical solvers. However, limited physical accu-racy and an inability to generalize under distributional shift limits its applica-bility to the real world. We propose to improve accuracy and generalizationby in... | lO8ZHqy_NFp | Novelty in modeling physical dynamics with symmetry | 6: Marginally above acceptance threshold | This paper studies improving the modeling of physical dynamics with equivariant neural networks. In particular, this paper focuses on a new type of data governed by physical models. Several special symmetry groups are considered to better characterize the system, including uniform motion equivariance, resolution-indepe... | 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
Incorporating Symmetry into Deep Dynamics Models for Improved Generalization
### Paper Abstract
Recent work has shown deep learning can accelerate the prediction of phys... |
pdsec2YIOCx | ICLR.cc/2021/Conference | 2021 | Untangle: Critiquing Disentangled Recommendations | ["Preksha Nema", "Alexandros Karatzoglou", "Filip Radlinski"] | The core principle behind most collaborative filtering methods is to embed users and items in latent spaces, where individual dimensions are learned independently of any particular item attributes. It is thus difficult for users to control their recommendations based on particular aspects (critiquing). In this work, we... | ["Disentangling", "Recommender Systems", "VAE", "Critiquing", "Explainability"] | Under review as a conference paper at ICLR 2021Untangle :Critiquing Disentangled RecommendationsAnonymous authorsPaper under double-blind reviewAbstractThe core principle behind most collaborative filtering methods is to embedusers and items in latent spaces, where individual dimensions are learnedindependently of any ... | rrGBiv4NUZN | An interesting paper but below the bar | 4: Ok but not good enough - rejection | Summary:
The paper proposes a framework to learn disentangled representations for collaborative filtering systems. To model the user-item interactions, the authors adopt the likelihood model proposed by \beta-Multi-VAE. The auxiliary task of predicting item labels is considered to increase the disentanglement and the ... | 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
Untangle: Critiquing Disentangled Recommendations
### Paper Abstract
The core principle behind most collaborative filtering methods is to embed users and items in latent... |
R7aFOrR0b2 | ICLR.cc/2021/Conference | 2021 | Dataset Curation Beyond Accuracy | ["Johan Bjorck", "Carla P Gomes"] | Neural networks are known to be data-hungry, and collecting large labeled datasets is often a crucial step in deep learning deployment. Researchers have studied dataset aspects such as distributional shift and labeling cost, primarily using downstream prediction accuracy for evaluation. In sensitive real-world applicat... | ["crowd-sourcing", "calibration", "dataset", "uncertainty"] | ABSTRACTNeural networks are known to be data-hungry, and collecting large labeleddatasets is often a crucial step in deep learning deployment. Researchers havestudied dataset aspects such as distributional shift and labeling cost, primarily us-ing downstream prediction accuracy for evaluation. In sensitive real-world a... | X9QI5g_7BqY | If we were reporting accuracy in the paper experiments, how different would the conclusions be? | 4: Ok but not good enough - rejection | The paper is an empirical study looking at how different dataset properties affect model calibration in the context of vision tasks. All experiments use a specific well-known vision model (ResNet 50).
In particular, the dataset properties that are investigated are:
- Balanced/Unbalanced classes.
- Label quality.
- D... | 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
Dataset Curation Beyond Accuracy
### Paper Abstract
Neural networks are known to be data-hungry, and collecting large labeled datasets is often a crucial step in deep le... |
ByUEelW0- | ICLR.cc/2018/Conference | 2018 | Modifying memories in a Recurrent Neural Network Unit | ["Vlad Velici", "Adam Pr\u00fcgel-Bennett"] | Long Short-Term Memory (LSTM) units have the ability to memorise and use long-term dependencies between inputs to generate predictions on time series data. We introduce the concept of modifying the cell state (memory) of LSTMs using rotation matrices parametrised by a new set of trainable weights. This addition shows s... | ["LSTM", "RNN", "rotation matrix", "long-term memory", "natural language processing"] | ABSTRACTLong Short-Term Memory (LSTM) units have the ability to memorise and uselong-term dependencies between inputs to generate predictions on time seriesdata. We introduce the concept of modifying the cell state (memory) of LSTMsusing rotation matrices parametrised by a new set of trainable weights. This ad-dition s... | SkYqiWteM | Insufficient Justification and Comparison | 4: Ok but not good enough - rejection | The paper proposes to add a rotation operation in long short-term memory (LSTM) cells. It performs experiments on bAbI tasks and showed that the results are better than the simple baselines with original LSTM cells. There are a few problems with the paper.
Firstly, the title and abstract discuss "modifying memories", ... | 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
Modifying memories in a Recurrent Neural Network Unit
### Paper Abstract
Long Short-Term Memory (LSTM) units have the ability to memorise and use long-term dependencies ... |
ryxUMREYPr | ICLR.cc/2020/Conference | 2020 | Is There Mode Collapse? A Case Study on Face Generation and Its Black-box Calibration | ["Zhenyu Wu", "Ye Yuan", "Zhaowen Wang", "Jianming Zhang", "Zhangyang Wang", "Hailin Jin"] | Generative adversarial networks (GANs) nowadays are capable of producing im-ages of incredible realism. One concern raised is whether the state-of-the-artGAN’s learned distribution still suffers from mode collapse. Existing evaluation metrics for image synthesis focus on low-level perceptual quality. Diversity tests ... | ["Generative Adversarial Networks", "Mode Collapse", "Calibration"] | ABSTRACTGenerative adversarial networks (GANs) nowadays are capable of producing im-ages of incredible realism. One concern raised is whether the state-of-the-artGAN’s learned distribution still suffers from mode collapse. Existing evaluationmetrics for image synthesis focus on low-level perceptual quality. Diversity t... | HJlStebC5r | Official Blind Review #3 | 1: Reject | This work addresses the important problem of generation bias and a lack of diversity in generative models, which is often called model collapse. It proposed a new metric to measure the diversity of the generative model's "worst" outputs based on the sample clustering patterns. Furthermore, it proposed two blackbox appr... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Is There Mode Collapse? A Case Study on Face Generation and Its Black-box Calibration
### Paper Abstract
Generative adversarial networks (GANs) nowadays are capable of p... | |
HkxSOAEFDB | ICLR.cc/2020/Conference | 2020 | Octave Graph Convolutional Network | ["Heng Chang", "Yu Rong", "Somayeh Sojoudi", "Junzhou Huang", "Wenwu Zhu"] | Many variants of Graph Convolutional Networks (GCNs) for representation learning have been proposed recently and have achieved fruitful results in various domains. Among them, spectral-based GCNs are constructed via convolution theorem upon theoretical foundation from the perspective of Graph Signal Processing (GSP). H... | ["Graph Convolutional Networks", "Octave Convolution", "Graph Mining"] | ABSTRACTMany variants of Graph Convolutional Networks (GCNs) for representation learn-ing have been proposed recently and have achieved fruitful results in various do-mains. Among them, spectral-based GCNs are constructed via convolution theo-rem upon a theoretical foundation from the perspective of Graph Signal Proces... | rygxsZzmqH | Official Blind Review #3 | 3: Weak Reject | This paper proposes to use octave convolution to learn a representation of a graph.
Typically, a learning on a graph is done either in a spatial domain or in a spectral domain.
A spectral domain based approach uses a eigenvalue decomposition form of a graph Laplacian (a symmetric matrix) and learning a filter that acts... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Octave Graph Convolutional Network
### Paper Abstract
Many variants of Graph Convolutional Networks (GCNs) for representation learning have been proposed recently and ha... | |
BJlo91BYPr | ICLR.cc/2020/Conference | 2020 | Irrationality can help reward inference | ["Lawrence Chan", "Andrew Critch", "Anca Dragan"] | Specifying reward functions is difficult, which motivates the area of reward inference: learning rewards from human behavior. The starting assumption in the area is that human behavior is optimal given the desired reward function, but in reality people have many different forms of irrationality, from noise to myopia to... | ["preference inference", "inverse reinforcement learning", "reward inference", "irrationality"] | ABSTRACTSpecifying reward functions is difficult, which motivates the area of reward in-ference: learning rewards from human behavior. The starting assumption in thearea is that human behavior is optimal given the desired reward function, but inreality people have many different forms of irrationality, from noise to my... | S1e54b8AFH | Official Blind Review #1 | 3: Weak Reject | This paper studies reward inference from demonstrations of irrational experts. More specifically, a set of semantically meaningful experts' behaviors is considered which is derived from modifying the Bellman update. The quality of reward inference from these different experts is measured by two different scores and ana... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Irrationality can help reward inference
### Paper Abstract
Specifying reward functions is difficult, which motivates the area of reward inference: learning rewards from ... | |
Sy6iJDqlx | ICLR.cc/2017/conference | 2017 | Attend, Adapt and Transfer: Attentive Deep Architecture for Adaptive Transfer from multiple sources in the same domain | ["Janarthanan Rajendran", "Aravind Lakshminarayanan", "Mitesh M. Khapra", "Prasanna P", "Balaraman Ravindran"] | Transferring knowledge from prior source tasks in solving a new target task can be useful in several learning applications. The application of transfer poses two serious challenges which have not been adequately addressed. First, the agent should be able to avoid negative transfer, which happens when the transfer hampe... | ["Deep learning", "Reinforcement Learning", "Transfer Learning"] | ABSTRACTTransferring knowledge from prior source tasks in solving a new target task canbe useful in several learning applications. The application of transfer poses twoserious challenges which have not been adequately addressed. First, the agentshould be able to avoid negative transfer, which happens when the transfer ... | Sy-SiOZNe | Review | 7: Good paper, accept | The paper tackles important problems in multi-task reinforcement learning: avoid negative transfer and allow finer selective transfer. The method is based on soft attention mechanism, very general, and demonstrated to be applicable in both policy gradient and value iteration methods. The introduction of base network al... | 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
Attend, Adapt and Transfer: Attentive Deep Architecture for Adaptive Transfer from multiple sources in the same domain
### Paper Abstract
Transferring knowledge from pri... |
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... | Hkt131ZRf | Interesting work but authors need to highlight the novel aspects | 2: Marginally below acceptance threshold | This paper presents a recurrent inference machines (RIM) framework to accelerate the reconstruction of magnetic resonance imaging (MRI). The RIM model is a combination of alternating convolutional layers and gated recurrent units (GRU). A ReLU activation function is used after the convolution layers. This study is cond... | 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 ... |
HkLXCE9lx | ICLR.cc/2017/conference | 2017 | RL^2: Fast Reinforcement Learning via Slow Reinforcement Learning | ["Yan Duan", "John Schulman", "Xi Chen", "Peter L. Bartlett", "Ilya Sutskever", "Pieter Abbeel"] | Deep reinforcement learning (deep RL) has been successful in learning sophisticated behaviors automatically; however, the learning process requires a huge number of trials. In contrast, animals can learn new tasks in just a few trials, benefiting from their prior knowledge about the world. This paper seeks to bridge th... | ["Reinforcement Learning", "Deep learning"] | ABSTRACTDeep reinforcement learning (deep RL) has been successful in learning sophis-ticated behaviors automatically; however, the learning process requires a hugenumber of trials. In contrast, animals can learn new tasks in just a few trials, bene-fiting from their prior knowledge about the world. This paper seeks to ... | SJwrBiWVl | Review | 3: Clear rejection | The paper proposes to use RL methods on sequences of episodes instead of single episodes. The underlying idea is the problem of 'learning to learn', and the experimental protocol proposed here allows one to understand how a neural network-based RL model can keep memory of past episodes in order to improve its ability t... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
RL^2: Fast Reinforcement Learning via Slow Reinforcement Learning
### Paper Abstract
Deep reinforcement learning (deep RL) has been successful in learning sophisticated ... |
-5W5OBfFlwX | ICLR.cc/2021/Conference | 2021 | Regret Bounds and Reinforcement Learning Exploration of EXP-based Algorithms | ["Mengfan Xu", "Diego Klabjan"] | EXP-based algorithms are often used for exploration in multi-armed bandit. We revisit the EXP3.P algorithm and establish both the lower and upper bounds of regret in the Gaussian multi-armed bandit setting, as well as a more general distribution option. The analyses do not require bounded rewards compared to classical ... | ["exploration", "algorithms", "bounds", "reinforcement", "bandit", "algorithm", "lower", "upper bounds", "regret", "gaussian"] | ABSTRACTEXP-based algorithms are often used for exploration in multi-armed bandit. Werevisit the EXP3.P algorithm and establish both the lower and upper bounds ofregret in the Gaussian multi-armed bandit setting, as well as a more generaldistribution option. The analyses do not require bounded rewards compared toclassi... | mr6VACkYmZ9 | Contributions appear marginal, and some doubts about the regret lower bound. | 4: Ok but not good enough - rejection | The authors consider analyzing the EXP3.P algorithm for the case of unbounded reward functions, in the sense that the rewards are governed by a Gaussian distribution. The authors first demonstrate a regret lower bound result on the Gaussian MABs when the time horizon is bounded from above. Then, the authors proceed to ... | 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
Regret Bounds and Reinforcement Learning Exploration of EXP-based Algorithms
### Paper Abstract
EXP-based algorithms are often used for exploration in multi-armed bandit... |
bM4Iqfg8M2k | ICLR.cc/2021/Conference | 2021 | Graph Information Bottleneck for Subgraph Recognition | ["Junchi Yu", "Tingyang Xu", "Yu Rong", "Yatao Bian", "Junzhou Huang", "Ran He"] | Given the input graph and its label/property, several key problems of graph learning, such as finding interpretable subgraphs, graph denoising and graph compression, can be attributed to the fundamental problem of recognizing a subgraph of the original one. This subgraph shall be as informative as possible, yet con... | ["irregular graph data", "gib objective", "graph information bottleneck", "graph denoising", "subgraph", "informative", "framework", "subgraph recognition", "input graph"] | ABSTRACTGiven the input graph and its label/property, several key problems of graph learn-ing, such as finding interpretable subgraphs, graph denoising and graph compres-sion, can be attributed to the fundamental problem of recognizing a subgraph ofthe original one. This subgraph shall be as informative as possible, ye... | 0J7VemrR9ze | A good paper with a clear theoretical contribution and rigorous empirical evaluation - clear accept recommendation with medium reviewer confidence | 8: Top 50% of accepted papers, clear accept | # Summary
The paper introduces the Graph Information Bottleneck (GIB) which aims to learn the most-informative compressed representation $Z$ given graph $G$ with associated label $Y$. Further, it defines GIB-Subgraph which aims to learn the compressed representation as the subgraph $G_{sub}$ which maximizes the mutual... | 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
Graph Information Bottleneck for Subgraph Recognition
### Paper Abstract
Given the input graph and its label/property, several key problems of graph learning, such as f... |
B1xGGTEtDH | ICLR.cc/2020/Conference | 2020 | Universal Approximation with Deep Narrow Networks | ["Patrick Kidger", "Terry Lyons"] | The classical Universal Approximation Theorem certifies that the universal approximation property holds for the class of neural networks of arbitrary width. Here we consider the natural `dual' theorem for width-bounded networks of arbitrary depth. Precisely, let $n$ be the number of inputs neurons, $m$ be the number of... | ["deep learning", "universal approximation", "deep narrow networks"] | ABSTRACTThe classical Universal Approximation Theorem certifies that the universal ap-proximation property holds for the class of neural networks of arbitrary width.Here we consider the natural ‘dual’ theorem for width-bounded networks of arbi-trary depth. Precisely, let nbe the number of inputs neurons, mbe the number... | ByevCWg15H | Official Blind Review #1 | 6: Weak Accept | - This paper complement of the fundamental Universal approximation theorem variants
- Based of the Register model, that the authors seem to have developed themselves from the scratch, elegant, although non-obvious
- Proof is straightforward, although the pictural description can be enhanced. In reality the ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Universal Approximation with Deep Narrow Networks
### Paper Abstract
The classical Universal Approximation Theorem certifies that the universal approximation property ho... | |
ByxPYjC5KQ | ICLR.cc/2019/Conference | 2019 | Improving Generalization and Stability of Generative Adversarial Networks | ["Hoang Thanh-Tung", "Truyen Tran", "Svetha Venkatesh"] | Generative Adversarial Networks (GANs) are one of the most popular tools for learning complex high dimensional distributions. However, generalization properties of GANs have not been well understood. In this paper, we analyze the generalization of GANs in practical settings. We show that discriminators trained on discr... | ["GAN", "generalization", "gradient penalty", "zero centered", "convergence"] | ABSTRACTGenerative Adversarial Networks (GANs) are one of the most popular tools forlearning complex high dimensional distributions. However, generalization prop-erties of GANs have not been well understood. In this paper, we analyze thegeneralization of GANs in practical settings. We show that discriminators trainedon... | Bkgv8HHx3Q | An interesting read on the convergence of GANs with gradient penalties, lacking comparisons to WGAN-GP | 6: Marginally above acceptance threshold | Summary:
The paper proposes to add to the original GAN (2014) loss a zero-centered gradient penalty as the one defined in the WGAN-GP paper. It also provides an analysis on the mode collapse and lack of stability of classical GANs. The authors compare results using their penalty on a few synthetic examples and on imag... | 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
Improving Generalization and Stability of Generative Adversarial Networks
### Paper Abstract
Generative Adversarial Networks (GANs) are one of the most popular tools for... |
Sk0pHeZAW | ICLR.cc/2018/Conference | 2018 | Sparse Regularized Deep Neural Networks For Efficient Embedded Learning | ["Jia Bi"] | Deep learning is becoming more widespread in its application due to its power in solving complex classification problems. However, deep learning models often require large memory and energy consumption, which may prevent them from being deployed effectively on embedded platforms, limiting their applications. This work ... | ["Sparse representation", "Compression Deep Learning Models", "L1 regularisation", "Optimisation."] | Under review as a conference paper at ICLR 2018Sparse Regularized Deep Neural NetworksFor Efficient Embedded LearningAnonymous authorsPaper under double-blind reviewAbstractDeep learning is becoming more widespread in its application due to itspower in solving complex classification problems. However, deep learningmode... | Hku4bLqgM | The authors use l-1 regularized SVRG to promotes sparsity in the trained model. However, the paper lacks comparisons with some key literature, and experimentally the benefit of SVRG over SGD does not seem substantial. | 4: Ok but not good enough - rejection | The authors present an l-1 regularized SVRG based training algorithm that is able to force many weights of the network to be 0, hence leading to good compression of the model. The motivation for l-1 regularization is clear as it promotes sparse models, which lead to lower storage overheads during inference. The use of... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Sparse Regularized Deep Neural Networks For Efficient Embedded Learning
### Paper Abstract
Deep learning is becoming more widespread in its application due to its power ... |
oGvpaRgaAd | NoDaLiDa/2023/Conference | 2023 | Question Answering and Question Generation for Finnish | ["Ilmari Kylli\u00e4inen", "Roman Yangarber"] | Recent advances in the field of language modeling have improved the state-of-the-art in question answering (QA) and question generation (QG). However, the development of modern neural models, their benchmarks, and datasets for training them has mainly focused on English. Finnish, like many other languages, faces a sho... | ["computational linguistics", "question answering", "question generation", "deep learning", "transformer models"] | Question Answering and Question Generation for FinnishIlmari Kylliäinen and Roman YangarberUniversity of Helsinki, FinlandDepartment of Digital Humanitiesfirst.last@helsinki.fiAbstractRecent advances in the field of languagemodeling have improved the state-of-the-art in question answering (QA) and ques-tion generation... | Z8GgzfQZuy- | Question Answering and Question Generation for Finnish | 8: Top 50% of accepted papers, clear accept | This paper presents (mostly automatically translated) Finnish question answering and question generation datasets and experiments to finetune Finnish LM-s to the task. Creating datasets and benchmarks for new languages is very important and the most significant contribution of this work. In general, I found the work we... | 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
Question Answering and Question Generation for Finnish
### Paper Abstract
Recent advances in the field of language modeling have improved the state-of-the-art in questio... |
0MjC3uMthAb | ICLR.cc/2021/Conference | 2021 | Learning Flexible Classifiers with Shot-CONditional Episodic (SCONE) Training | ["Eleni Triantafillou", "Vincent Dumoulin", "Hugo Larochelle", "Richard Zemel"] | Early few-shot classification work advocates for episodic training, i.e. training over learning episodes each posing a few-shot classification task. However, the role of this training regime remains poorly understood, and its usefulness is still debated. Standard classification training methods (``pre-training'') follo... | ["few-shot classification", "few-shot learning", "episodic training", "meta-learning"] | ABSTRACTEarly few-shot classification work advocates for episodic training, i.e. training overlearning episodes each posing a few-shot classification task. However, the role ofthis training regime remains poorly understood, and its usefulness is still debated.Standard classification training methods (“pre-training”) fo... | mzfkA51ASJ_ | Not familiar with this area, but assigned to review | 5: Marginally below acceptance threshold | This paper aims to understand the role of this episodic fine-tuning phase and discovers that fine-tuning on episodes of a particular shot can specialize the pre-trained model to solving episodes of that shot at the expense of performance on other shots. It proposes a shot-conditional form of episodic fine-tuning.
The ... | 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
Learning Flexible Classifiers with Shot-CONditional Episodic (SCONE) Training
### Paper Abstract
Early few-shot classification work advocates for episodic training, i.e.... |
BJlxm30cKm | ICLR.cc/2019/Conference | 2019 | An Empirical Study of Example Forgetting during Deep Neural Network Learning | ["Mariya Toneva*", "Alessandro Sordoni*", "Remi Tachet des Combes*", "Adam Trischler", "Yoshua Bengio", "Geoffrey J. Gordon"] | Inspired by the phenomenon of catastrophic forgetting, we investigate the learning dynamics of neural networks as they train on single classification tasks. Our goal is to understand whether a related phenomenon occurs when data does not undergo a clear distributional shift. We define a ``forgetting event'' to have occ... | ["catastrophic forgetting", "sample weighting", "deep generalization"] | ABSTRACTInspired by the phenomenon of catastrophic forgetting, we investigate the learningdynamics of neural networks as they train on single classification tasks. Our goalis to understand whether a related phenomenon occurs when data does not un-dergo a clear distributional shift. We define a “forgetting event” to hav... | SyxkXUThiQ | Thorough experiments which prove there exist "support examples" in neural network training. | 7: Good paper, accept | This paper studies the forgetting behavior of the training examples during SGD. Empirically it shows there are forgettable and unforgettable examples, unforgettable examples are like "support examples", one can achieve similar performance by training only on these "support examples". The paper also shows this phenomeno... | 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 Study of Example Forgetting during Deep Neural Network Learning
### Paper Abstract
Inspired by the phenomenon of catastrophic forgetting, we investigate the... |
B8mxkTzX2RY | ML_Reproducibility_Challenge/2021/Fall | 2021 | Replicating and Improving GAN2Shape Through Novel Shape Priors and Training Steps | ["Alfred Nilsson", "Alessio Galatolo"] | SCOPE OF REPRODUCIBILITY
Pan et al. propose an unsupervised method named GAN2Shape that purportedly is able to recover 3D information stored in the weights of a pre-trained StyleGAN2 model, to produce 3D shapes from 2D images. We aim to reproduce the 3D shape recovery and identify its strengths and weaknesses.
METHOD... | ["Unsupervised Learning", "Deep Learning", "Generative Modeling", "Computer Vision", "3D Shape"] | Replicating and Improving GAN2Shape Through Novel ShapePriors and Training StepsAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2Pan et al. [2021] propose an unsupervised method named GAN2Shape that purportedly is able to recover 3D information 3stored in the weights of a pre... | rMzxjieWUW9 | Good reproducibility study, writing and structure needs to be improved | 6: Marginally above acceptance threshold | This paper aims to reproduce "Do 2D GANs Know 3D Shape? Unsupervised 3D Shape Reconstruction from 2D Image GANs" by Pan et al.
The authors provide a complete and clear summary of the work that has been done. They also state clearly which of the research claims made by Pan et al. are subject to this reproducibility st... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Replicating and Improving GAN2Shape Through Novel Shape Priors and Training Steps
### Paper Abstract
SCOPE OF REPRODUCIBILITY Pan et al. propose an unsupervised method ... |
2VXyy9mIyU3 | ICLR.cc/2021/Conference | 2021 | Learning with Instance-Dependent Label Noise: A Sample Sieve Approach | ["Hao Cheng", "Zhaowei Zhu", "Xingyu Li", "Yifei Gong", "Xing Sun", "Yang Liu"] | Human-annotated labels are often prone to noise, and the presence of such noise will degrade the performance of the resulting deep neural network (DNN) models. Much of the literature (with several recent exceptions) of learning with noisy labels focuses on the case when the label noise is independent of features. Pract... | ["Learning with noisy labels", "instance-based label noise", "deep neural networks."] | ABSTRACTHuman-annotated labels are often prone to noise, and the presence of such noisewill degrade the performance of the resulting deep neural network (DNN) mod-els. Much of the literature (with several recent exceptions) of learning with noisylabels focuses on the case when the label noise is independent of features... | UQ4bxPhGPDK | Review for "Learning with Instance-Dependent Label Noise: A Sample Sieve Approach" | 6: Marginally above acceptance threshold | The authors of the paper propose a new method, the CORES (COnfidence REgularized Sample Sieve), to tackle the important problem of learning under instance dependent label noise. The proposed method, in essence, involves the use of a confidence regularization term that encourages more confident predictions and a sieving... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning with Instance-Dependent Label Noise: A Sample Sieve Approach
### Paper Abstract
Human-annotated labels are often prone to noise, and the presence of such noise ... |
B1gZV1HYvS | ICLR.cc/2020/Conference | 2020 | Multi-Agent Interactions Modeling with Correlated Policies | ["Minghuan Liu", "Ming Zhou", "Weinan Zhang", "Yuzheng Zhuang", "Jun Wang", "Wulong Liu", "Yong Yu"] | In multi-agent systems, complex interacting behaviors arise due to the high correlations among agents. However, previous work on modeling multi-agent interactions from demonstrations is primarily constrained by assuming the independence among policies and their reward structures.
In this paper, we cast the multi-agent... | ["Multi-agent reinforcement learning", "Imitation learning"] | ABSTRACTIn multi-agent systems, complex interacting behaviors arise due to the high cor-relations among agents. However, previous work on modeling multi-agent inter-actions from demonstrations is primarily constrained by assuming the indepen-dence among policies and their reward structures. In this paper, we cast the m... | B1gXtGEb5S | Official Blind Review #3 | 6: Weak Accept | The authors propose a decentralized adversarial imitation learning algorithm with correlated policies, which recovers each agent’s policy through approximating opponents action using opponent modeling. Extensive experimental results showed that the proposed framework, CoDAIL, better fits scenarios with correlated multi... | <|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 Interactions Modeling with Correlated Policies
### Paper Abstract
In multi-agent systems, complex interacting behaviors arise due to the high correlations am... | |
pqZV_srUVmK | ICLR.cc/2021/Conference | 2021 | Single-Timescale Actor-Critic Provably Finds Globally Optimal Policy | ["Zuyue Fu", "Zhuoran Yang", "Zhaoran Wang"] | We study the global convergence and global optimality of actor-critic, one of the most popular families of reinforcement learning algorithms. While most existing works on actor-critic employ bi-level or two-timescale updates, we focus on the more practical single-timescale setting, where the actor and critic are update... | ["optimal policy", "actor", "critic", "provably", "global optimality", "rate", "first time", "global convergence", "popular families", "reinforcement"] | ABSTRACTWe study the global convergence and global optimality of actor-critic, one of themost popular families of reinforcement learning algorithms. While most exist-ing works on actor-critic employ bi-level or two-timescale updates, we focus onthe more practical single-timescale setting, where the actor and critic are... | AljIbLyWOjd | Theory for an one-timescale AC | 8: Top 50% of accepted papers, clear accept | This is a theoretical paper. The paper studies convergence and optimality of the actor-critic with the function approximation in a single-timescale setting. The proposed single-timescale AC applies PPO for the actor and updates the critic by applying the Bellman operator to the critic one time. Despite the actor-critic... | 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
Single-Timescale Actor-Critic Provably Finds Globally Optimal Policy
### Paper Abstract
We study the global convergence and global optimality of actor-critic, one of the... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.