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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
-xjStp_F9o | logconference.io/LOG/2022/Conference | 2022 | Learning Graph Search Heuristics | ["Michal P\u00e1ndy", "Weikang Qiu", "Gabriele Corso", "Petar Veli\u010dkovi\u0107", "Zhitao Ying", "Jure Leskovec", "Pietro Lio"] | Searching for a path between two nodes in a graph is one of the most well-studied and fundamental problems in computer science. In numerous domains such as robotics, AI, or biology, practitioners develop search heuristics to accelerate their pathfinding algorithms. However, it is a laborious and complex process to hand... | ["graph search heuristics", "phil", "heuristic function", "path", "nodes", "graph", "fundamental problems", "computer science", "numerous domains", "robotics"] | Learning Graph Search HeuristicsMichal PándyUniversity of CambridgeWeikang QiuYale UniversityGabriele CorsoMITPetar Veli ˇckovi ́cDeepMindRex YingYale UniversityJure LeskovecStanford UniversityPietro LiòUniversity of CambridgeAbstractSearching for a path between two nodes in a graph is one of the most well-studiedand... | 3wvhFwMmGpl | Summary:
The paper studies the well-studied problem of finding a path between two nodes. While being a fundamental problem, this has a range of applications in various fields such as robotics and graphs. The paper proposes a training algorithm for discovering paths between two nodes based on graph representation learn... | <|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 Graph Search Heuristics
### Paper Abstract
Searching for a path between two nodes in a graph is one of the most well-studied and fundamental problems in compute... | |||
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... | O0oO0Ac7hDf | First convergence result for single-timescale actor-critic | 8: Top 50% of accepted papers, clear accept | Summary:
This paper proves that the proposed single-timescale actor-critic algorithm with KL regularization converges to a globally optimal policy when appropriate function approximation is used. The authors define a specific form of actor/critic updates, which are PPO update for the actor and the single application o... | 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
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... |
a6--BnpcdB | MIDL.io/2023/Short_Paper_Track | 2023 | Unsupervised Plaque Segmentation on Whole Slide Images | ["Johann Christopher Engster", "Nele Blum", "Tobias Reinberger", "Pascal Stagge", "Thorsten Buzug", "Jeanette Erdmann", "Zouhair Aherrahrou", "Maik Stille"] | Atherosclerosis is a multifactorial disease in which deposits of fat form in the arteries. These plaques can cause ischemic heart disease or other follow-up diseases. To investigate etiology and possible treatment options, mice were used as models and histological whole slide images (WSI) stained with Oil-Red-O (ORO) w... | ["Atherosclerosis", "Unsupervised Segmentation", "Plaque Segmentation", "WSI"] | Medical Imaging with Deep Learning – Under Review 2023 Short Paper – MIDL 2023 submissionUnsupervised Plaque Segmentation on Whole Slide ImagesJohann Engster1,2johann.christopher.engster@imte.fraunhofer.deNele Blum2blum@imt.uni-luebeck.deTobias Reinberger3,4,5tobias.reinberger@uni-luebeck.dePascal Stagge1,2pascal.stagg... | hdKk8zzvZ90 | Unsupervised Plaque Segmentation on Whole Slide Images | 6: Marginally above acceptance threshold | This paper presents an approach based on unsupervised segmentation to identify atherosclerotic plaques in histology of coronary arteries from a large cohort of mice.
The slides are stained with a special staining that highlights the plaque in red, compared to the rest of artery tissue.
The unsupervised approach is base... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Unsupervised Plaque Segmentation on Whole Slide Images
### Paper Abstract
Atherosclerosis is a multifactorial disease in which deposits of fat form in the arteries. Thes... |
Byxl-04KvH | ICLR.cc/2020/Conference | 2020 | NESTED LEARNING FOR MULTI-GRANULAR TASKS | ["Rapha\u00ebl Achddou", "J. Matias Di Martino", "Guillermo Sapiro"] | Standard deep neural networks (DNNs) used for classification are trained in an end-to-end fashion for very specific tasks - object recognition, face identification, character recognition, etc. This specificity often leads to overconfident models that generalize poorly to samples that are not from the original training ... | ["Nested learning"] | ABSTRACTStandard deep neural networks (DNNs) used for classification are trained in anend-to-end fashion for very specific tasks - object recognition, face identification,character recognition, etc. This specificity often leads to overconfident modelsthat generalize poorly to samples that are not from the original trai... | HygywE9K5B | Official Blind Review #4 | 3: Weak Reject | Summary: The problem addressed is how to train a DNN to learn a hierarchical representation of the input that has heterogeneously annotated labels at various levels of granularity. After training the network’s goal is to emit sequentially finer grained labels with corresponding confidence. The authors use information 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
NESTED LEARNING FOR MULTI-GRANULAR TASKS
### Paper Abstract
Standard deep neural networks (DNNs) used for classification are trained in an end-to-end fashion for very sp... | |
hP-pn8bKffe | ICLR.cc/2021/Conference | 2021 | Boltzman Tuning of Generative Models | ["Victor Berger", "Michele Sebag"] | The paper focuses on the a posteriori tuning of a generative model in order to favor the generation of good instances in the sense of some external differentiable criterion. The proposed approach, called Boltzmann Tuning of Generative Models (BTGM) applies to a wide range of applications. It covers conditional genera... | ["generative models", "variational inference"] | ABSTRACTThe paper focuses on the a posteriori tuning of a generative model in order tofavor the generation of good instances in the sense of some external differentiablecriterion. The proposed approach, called Boltzmann Tuning of Generative Models(BTGM ), applies to a wide range of applications. It covers conditional g... | Bkr8Trk-hMW | Review | 4: Ok but not good enough - rejection | The paper studies the problem of tuning a pretrained deep generative model p with an external criterion f, aiming to bias the model to produce samples that maximize f. To this end, a variational framework is presented, which formulates the problem as a constriant optimization problem. Specifically, a variational distri... | 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
Boltzman Tuning of Generative Models
### Paper Abstract
The paper focuses on the a posteriori tuning of a generative model in order to favor the generation of good insta... |
YWtLZvLmud7 | ICLR.cc/2021/Conference | 2021 | BERTology Meets Biology: Interpreting Attention in Protein Language Models | ["Jesse Vig", "Ali Madani", "Lav R. Varshney", "Caiming Xiong", "richard socher", "Nazneen Rajani"] | Transformer architectures have proven to learn useful representations for protein classification and generation tasks. However, these representations present challenges in interpretability. In this work, we demonstrate a set of methods for analyzing protein Transformer models through the lens of attention. We show that... | ["interpretability", "black box", "computational biology", "representation learning", "attention", "transformers", "visualization", "natural language processing"] | ABSTRACTTransformer architectures have proven to learn useful representations for pro-tein classification and generation tasks. However, these representations presentchallenges in interpretability. In this work, we demonstrate a set of methods foranalyzing protein Transformer models through the lens of attention. We sh... | YsoptjIClH_ | Systematic analysis of of interpretability of Transformer-based language attention models used for protein sequences | 6: Marginally above acceptance threshold | The authors of this manuscript conducted a comprehensive analysis of the interpretability of self-attention language models when learning from protein sequences. Specifically, five multi-head self-attention models from NLP were used to model protein sequences. The following tasks were conducted: (1) whether attention c... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
BERTology Meets Biology: Interpreting Attention in Protein Language Models
### Paper Abstract
Transformer architectures have proven to learn useful representations for p... |
HJYQLb-RW | ICLR.cc/2018/Conference | 2018 | On the limitations of first order approximation in GAN dynamics | ["Jerry Li", "Aleksander Madry", "John Peebles", "Ludwig Schmidt"] | Generative Adversarial Networks (GANs) have been proposed as an approach to learning generative models. While GANs have demonstrated promising performance on multiple vision tasks, their learning dynamics are not yet well understood, neither in theory nor in practice. In particular, the work in this domain has been foc... | ["GANs", "first order dynamics", "convergence", "mode collapse"] | ABSTRACTGenerative Adversarial Networks (GANs) have been proposed as an approach tolearning generative models. While GANs have demonstrated promising perfor-mance on multiple vision tasks, their learning dynamics are not yet well understood,neither in theory nor in practice. In particular, the work in this domain has b... | SyLbm9eWM | Good (but limited) step towards understanding dynamics of adversarial training in GANs | 7: Good paper, accept | Summary:
This paper studies the dynamics of adversarial training of GANs for Gaussian mixture model. The generator is a mixture of two Gaussians in one dimension. Discriminator is union of two intervals. Synthetic data is generated from a mixture of two Gaussians in one dimension. On this data, adversarial training is... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the limitations of first order approximation in GAN dynamics
### Paper Abstract
Generative Adversarial Networks (GANs) have been proposed as an approach to learning g... |
kvwWjYQtmw | logconference.io/LOG/2022/Conference | 2022 | GARNET: Reduced-Rank Topology Learning for Robust and Scalable Graph Neural Networks | ["Chenhui Deng", "Xiuyu Li", "Zhuo Feng", "Zhiru Zhang"] | Graph neural networks (GNNs) have been increasingly deployed in various applications that involve learning on non-Euclidean data. However, recent studies show that GNNs are vulnerable to graph adversarial attacks. Although there are several defense methods to improve GNN robustness by eliminating adversarial components... | ["graph neural networks", "adversarial robustness", "reduced-rank approximation", "probabilistic graphical model"] | GARNET: Reduced-Rank Topology Learning for Robust andScalable Graph Neural NetworksChenhui DengCornell Universitycd574@cornell.eduXiuyu LiUC Berkeleyxiuyu@berkeley.eduZhuo FengStevens Institute of Technologyzfeng12@stevens.eduZhiru ZhangCornell Universityzhiruz@cornell.eduAbstractGraph neural networks (GNNs) have been ... | Kr8QTBjlN4 | ## Summary
This paper proposed GARNET, a scalable GNN model with boosted robustness against adversarial structural attacks of GNNs by recovering the clean graph structure using similarity of low-rank spectral embeddings and edge pruning on probabilistic graphical model; the recovered graph has a slightly reduced rank ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
GARNET: Reduced-Rank Topology Learning for Robust and Scalable Graph Neural Networks
### Paper Abstract
Graph neural networks (GNNs) have been increasingly deployed in v... | |||
TGZ8P21_jHL | NeurIPS.cc/2022/Workshop/SyntheticData4ML | 2022 | Synthetic Clinical Trial Data while Preserving Subject-Level Privacy | ["Mandis Beigi", "Afrah Shafquat", "Jason Mezey", "Jacob W. Aptekar"] | Clinical trials capture high-quality data for millions of patients each year, yet these data are largely unavailable for research beyond the scope of any individual trial due to a combination of regulatory, intellectual property, and patient privacy barriers. Synthetic clinical trial data that captures the analytical p... | ["synthetic data", "clinical trials", "machine learning", "privacy"] | Synthetic Clinical Trial Data while PreservingSubject-Level PrivacyMandis BeigiMedidata Solutions, a Dassault Systèmes companyNew York, NY 10014Mandis.Beigi@3ds.comAfrah ShafquatMedidata Solutions, a Dassault Systèmes companyNew York, NY 10014Afrah.Shafquat@3ds.comJason G. MezeyDepartment of Computational Biology, Co... | d_WwetcdcH | Novel method for generating synthetic tabular data for clinical trials, some open questions about assessment of privacy | 8: Top 50% of accepted papers, clear accept | # Summary
The authors propose a novel method for generating synthetic clinical trial data, addressing the challenges that the small sample size of training data typically bring. The authors also carry out a benchmarking exercise to compare to existing methods.
# Discussion
Pro:
* The proposed method appears to work... | 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
Synthetic Clinical Trial Data while Preserving Subject-Level Privacy
### Paper Abstract
Clinical trials capture high-quality data for millions of patients each year, yet... |
rkX9Z_kwf | ICLR.cc/2018/Workshop | 2018 | Evaluating visual "common sense" using fine-grained classification and captioning tasks | ["Raghav Goyal", "Farzaneh Mahdisoltani", "Guillaume Berger", "Waseem Gharbieh", "Ingo Bax", "Roland Memisevic"] | We introduce the Something-something V2 dataset, which contains captions of finely-varying human-object interactions. We also discuss various baseline models, and show that neural networks show surprisingly strong performance on many of the very hard, detailed discrimination tasks associated with this dataset. | ["video dataset", "fine-grained understanding", "common sense", "video captioning"] | Workshop track - ICLR 2018EVALUATING VISUAL “COMMON SENSE ”USING FINE -GRAINED CLASSIFICATION AND CAPTIONING TASKSRaghav Goyal, Farzaneh Mahdisoltani, Guillaume Berger, Waseem Gharbieh,Ingo Bax, Roland MemisevicTwenty Billion Neurons Inc.ffirstname.lastname g@twentybn.com1 I NTRODUCTIONUnderstanding concepts in the wor... | Hy_wEn3uf | Further discussions would have made the paper stronger | 6: Marginally above acceptance threshold |
==========================================
Summary
==========================================
The paper proposes a new metric to measure the discriminative properties of a classification model beyond its top-k accuracy.
This metric is obtained by grouping annotated classes by different criteria, e.g. "pretending", ... | 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
Evaluating visual "common sense" using fine-grained classification and captioning tasks
### Paper Abstract
We introduce the Something-something V2 dataset, which contain... |
GXJPLbB5P-y | ICLR.cc/2021/Conference | 2021 | Simplifying Models with Unlabeled Output Data | ["Sang Michael Xie", "Tengyu Ma", "Percy Liang"] | We focus on prediction problems with high-dimensional outputs that are subject to output validity constraints, e.g. a pseudocode-to-code translation task where the code must compile. For these problems, labeled input-output pairs are expensive to obtain, but "unlabeled" outputs, i.e. outputs without corresponding input... | ["semi-supervised learning", "structured prediction"] | ABSTRACTWe focus on prediction problems with high-dimensional outputs that are subjectto output validity constraints, e.g. a pseudocode-to-code translation task where thecode must compile. For these problems, labeled input-output pairs are expensiveto obtain, but “unlabeled” outputs, i.e. outputs without corresponding ... | bUN-Pzesp6 | new interesting framework to leverage unlabelled output data | 6: Marginally above acceptance threshold | The paper introduces a “predict-and-denoise” model for structured prediction, specifically for tasks where the output has to adhere to some constraints e.g. natural language, code etc. This framework allows leveraging of unlabelled output data to train the denoiser, which consequently allows the base predictor to be of... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Simplifying Models with Unlabeled Output Data
### Paper Abstract
We focus on prediction problems with high-dimensional outputs that are subject to output validity constr... |
DtLOYYnnhiC | MLSys/2023/Workshop/RCLWN | 2023 | Efficient and Light-Weight Federated Learning via Asynchronous Distributed Dropout | ["Chen Dun", "Mirian Del Carmen Hipolito Garcia", "Christopher Jermaine", "Dimitrios Dimitriadis", "Anastasios Kyrillidis", "Anastasios Kyrillidis"] | Asynchronous learning protocols have regained attention lately, especially in the Federated Learning (FL) setup, where slower clients can severely impede the learning process. Herein, we propose \texttt{AsyncDrop}, a novel asynchronous FL framework that utilizes dropout regularization to handle device heterogeneity in ... | ["Efficient large scale neural network training", "federated learning"] | ABSTRACTAsynchronous learning protocols have regained attention lately, especially in the Federated Learning (FL) setup,where slower clients can severely impede the learning process. Herein, we propose AsyncDrop , a novel asyn-chronous FL framework that utilizes dropout regularization to handle device heterogeneity in ... | Ijz7fCJSUdK | The authors propose Hetero AsyncDrop, a novel algorithm for asynchronous federated learning which reduces the communication overhead of a FL session by allowing participating clients to asynchronously update only a fraction of the global model parameters. | 7: Good paper, accept | In standard federated learning (FL), multiple devices with heterogeneous datasets and computational capabilities train a global model without sharing their own private data. One issue in synchronous FL is that the time needed to compute a FL round is determined by the slowest node. Therefore, the authors present an alg... | 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
Efficient and Light-Weight Federated Learning via Asynchronous Distributed Dropout
### Paper Abstract
Asynchronous learning protocols have regained attention lately, esp... |
Skeh-xBYDH | ICLR.cc/2020/Conference | 2020 | On Symmetry and Initialization for Neural Networks | ["Ido Nachum", "Amir Yehudayoff"] | This work provides an additional step in the theoretical understanding of neural networks. We consider neural networks with one hidden layer and show that when learning symmetric functions, one can choose initial conditions so that standard SGD training efficiently produces generalization guarantees. We empirically ver... | ["Neural Network Theory", "Symmetry"] | ABSTRACTThis work provides an additional step in the theoretical understanding of neuralnetworks. We consider neural networks with one hidden layer and show that whenlearning symmetric functions, one can choose initial conditions so that standardSGD training efficiently produces generalization guarantees. We empiricall... | H1gUil1TYS | Official Blind Review #2 | 3: Weak Reject | The paper studies the problem of learning the class of symmetric boolean function, that is, functions that depend only on |x| = \sum_i x_i. The paper shows that with proper initialization, one-hidden layer over-parametrized networks can learn this class of functions. The main observation that the authors make is that 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
On Symmetry and Initialization for Neural Networks
### Paper Abstract
This work provides an additional step in the theoretical understanding of neural networks. We consi... | |
BylahsR9tX | ICLR.cc/2019/Conference | 2019 | Low-Rank Matrix Factorization of LSTM as Effective Model Compression | ["Genta Indra Winata", "Andrea Madotto", "Jamin Shin", "Elham J. Barezi"] | Large-scale Long Short-Term Memory (LSTM) cells are often the building blocks of many state-of-the-art algorithms for tasks in Natural Language Processing (NLP). However, LSTMs are known to be computationally inefficient because the memory capacity of the models depends on the number of parameters, and the inherent rec... | ["NLP", "LSTM", "Compression", "Low Rank", "Norm Analysis"] | ABSTRACTLarge-scale Long Short-Term Memory (LSTM) cells are often the building blocksof many state-of-the-art algorithms for tasks in Natural Language Processing(NLP). However, LSTMs are known to be computationally inefficient becausethe memory capacity of the models depends on the number of parameters, and theinherent... | HylTAeS8hm | Good technical report of applying decomposition methods to weights of LSTM. But lack of originality and significance. | 4: Ok but not good enough - rejection |
[PROS]
[quality]
The paper did extensive experimental studies on compressing weights of LSTM with two matrix decomposition methods (SVD and NMF). Their experiments cover three NLP downstream tasks and two benchmark datasets, and performed the decomposition methods on state-of-the-art language models. They explored e... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Low-Rank Matrix Factorization of LSTM as Effective Model Compression
### Paper Abstract
Large-scale Long Short-Term Memory (LSTM) cells are often the building blocks of ... |
o2ko2D_uvXJ | ICLR.cc/2021/Conference | 2021 | Group-Connected Multilayer Perceptron Networks | ["Mohammad Kachuee", "Sajad Darabi", "Shayan Fazeli", "Majid Sarrafzadeh"] | Despite the success of deep learning in domains such as image, voice, and graphs, there has been little progress in deep representation learning for domains without a known structure between features. For instance, a tabular dataset of different demographic and clinical factors where the feature interactions are not gi... | ["gmlp", "domains", "operations", "multilayer perceptron networks", "deep representation", "expressive feature combinations", "groups", "success", "deep learning", "image"] | ABSTRACTDespite the success of deep learning in domains such as image, voice, and graphs,there has been little progress in deep representation learning for domains withouta known structure between features. For instance, a tabular dataset of differentdemographic and clinical factors where the feature interactions are n... | i91vC_wKTNa | learning group features | 5: Marginally below acceptance threshold | The authors proposed a neural network architecture, Group-Connected Multilayer Perceptron (GMLP), which automatically groups the input features and extracts high level representations according to the groups. This paper focuses on classification problems.
The architecture can be decomposed to three stages. The first s... | 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
Group-Connected Multilayer Perceptron Networks
### Paper Abstract
Despite the success of deep learning in domains such as image, voice, and graphs, there has been little... |
rJeO3aVKPB | ICLR.cc/2020/Conference | 2020 | Faster Neural Network Training with Data Echoing | ["Dami Choi", "Alexandre Passos", "Christopher J. Shallue", "George E. Dahl"] | In the twilight of Moore's law, GPUs and other specialized hardware accelerators have dramatically sped up neural network training. However, earlier stages of the training pipeline, such as disk I/O and data preprocessing, do not run on accelerators. As accelerators continue to improve, these earlier stages will increa... | ["systems", "faster training", "large scale"] | ABSTRACTIn the twilight of Moore’s law, GPUs and other specialized hardware acceleratorshave dramatically sped up neural network training. However, earlier stages of thetraining pipeline, such as disk I/O and data preprocessing, do not run on accelerators.As accelerators continue to improve, these earlier stages will i... | HklnaKqnYH | Official Blind Review #3 | 3: Weak Reject | The authors propose a simple method for avoiding bottlenecks during NN training, whereby training examples are utilized multiple times per read. The work focuses on cases where the cost of preparing a minibatch exceeds that of a training step (both a forward pass and, subsequently, a parameter update). Working within s... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Faster Neural Network Training with Data Echoing
### Paper Abstract
In the twilight of Moore's law, GPUs and other specialized hardware accelerators have dramatically sp... | |
Oj2hGyJwhwX | ICLR.cc/2021/Conference | 2021 | SelfNorm and CrossNorm for Out-of-Distribution Robustness | ["Zhiqiang Tang", "Yunhe Gao", "Yi Zhu", "Zhi Zhang", "Mu Li", "Dimitris N. Metaxas"] | Normalization techniques are crucial in stabilizing and accelerating the training of deep neural networks. However, they are mainly designed for the independent and identically distributed (IID) data, not satisfying many real-world out-of-distribution (OOD) situations. Unlike most previous works, this paper presents tw... | ["selfnorm", "crossnorm", "ood generalization", "statistics", "robustness selfnorm", "robustness normalization techniques", "crucial", "training", "deep neural networks", "independent"] | ABSTRACTNormalization techniques are crucial in stabilizing and accelerating the trainingof deep neural networks. However, they are mainly designed for the indepen-dent and identically distributed (IID) data, not satisfying many real-world out-of-distribution (OOD) situations. Unlike most previous works, this paper pre... | DaNjwapGtn | Not sure yet if the paper can be published. | 5: Marginally below acceptance threshold | **Summary**
The paper introduces two simple modules, SelfNorm (SN) and CrossNorm (CN), that are highly modular and can theoretically be attached to different parts of the CNNs to control the balance between style and content cues for their recognition. SN is a lightly parametrised module that gives freedom to the CNN 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
SelfNorm and CrossNorm for Out-of-Distribution Robustness
### Paper Abstract
Normalization techniques are crucial in stabilizing and accelerating the training of deep ne... |
BkxeSkO7ZB | MICCAI.org/2019/Workshop/COMPAY | 2019 | Self-Supervised Similarity Learning for Digital Pathology | ["Jacob Gildenblat", "Eldad Klaiman"] | Using features extracted from networks pretrained on ImageNet
is a common practice in applications of deep learning for digital
pathology. However it presents the downside of missing domain specifi c
image information. In digital pathology, supervised training data is expensive
and dicult to collect. We propose a self... | ["Deep Learning", "Self-Supervised", "Similarity Learning", "Digital Pathology"] | Self-Supervised Similarity Learning for DigitalPathologyJacob Gildenblat1;2and Eldad Klaiman31SagivTech Ltd.2DeePathology.ai3Pathology and Tissue Analytics, Pharma Research and Early Development, RocheInnovation Center Municheldad.klaiman@roche.comAbstract. Using features extracted from networks pretrained on Ima-geNet... | rkgsz1cuNH | Self-Supervised Similarity Learning for Digital Pathology | 5: Marginally below acceptance threshold | Two contributions of the paper are:
1) Applying Siamese Networks: Use of siamese networks to find similar and dissimilar patches. As the authors also mention, it is a known technique for similarity matching in computer vision applications. This paper contribution is the application (first time?, I am not familiar with... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Self-Supervised Similarity Learning for Digital Pathology
### Paper Abstract
Using features extracted from networks pretrained on ImageNet is a common practice in applic... |
rJlYsn4YwS | ICLR.cc/2020/Conference | 2020 | Gradient-free Neural Network Training by Multi-convex Alternating Optimization | ["Junxiang Wang", "Fuxun Yu", "Xiang Chen", "Liang Zhao"] | In recent years, stochastic gradient descent (SGD) and its variants have been the dominant optimization methods for training deep neural networks. However, SGD suffers from limitations such as the lack of theoretical guarantees, vanishing gradients, excessive sensitivity to input, and difficulties solving highly non-sm... | ["neural network", "alternating minimization", "global convergence"] | ABSTRACTIn recent years, stochastic gradient descent (SGD) and its variants have been thedominant optimization methods for training deep neural networks. However, SGDsuffers from limitations such as the lack of theoretical guarantees, vanishing gra-dients, excessive sensitivity to input, and difficulties solving highly... | HyxgOkVPcS | Official Blind Review #1 | 6: Weak Accept | The motivation for this paper is as follows:
Why SGD:
1. Easy
2. applied in online settings
However:
1. Convergence proof has been done by others, but the assumptions of their proofs cannot be applied to problems involving deep neural networks, which are highly nonsmooth and nonconvex
2. Suffer from gradient vanish... | <|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-free Neural Network Training by Multi-convex Alternating Optimization
### Paper Abstract
In recent years, stochastic gradient descent (SGD) and its variants hav... | |
rJlMBjAcYX | ICLR.cc/2019/Conference | 2019 | Optimizing for Generalization in Machine Learning with Cross-Validation Gradients | ["Barratt", "Shane", "Sharma", "Rishi"] | Cross-validation is the workhorse of modern applied statistics and machine learning, as it provides a principled framework for selecting the model that maximizes generalization performance. In this paper, we show that the cross-validation risk is differentiable with respect to the hyperparameters and training data for ... | ["generalization", "machine", "gradients", "risk", "workhorse", "modern", "statistics", "machine learning", "principled framework", "model"] | ABSTRACTCross-validation is the workhorse of modern applied statistics and machine learn-ing, as it provides a principled framework for selecting the model that maximizesgeneralization performance. In this paper, we show that the cross-validation riskis differentiable with respect to the hyperparameters and training da... | HkgzyGoV2Q | Interesting paper but with limited novelty and lacking convincing experiments | 5: Marginally below acceptance threshold | This paper proposes the so-called cross-validation gradient method (CVGM).
This is idea is to express the CV score as a differentiable function
of the hyperparameters and then to update hyperparameters with gradient
descent. Derivations are provided with Logistic regression and Elastic-Net
thanks to the sign splitting ... | 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
Optimizing for Generalization in Machine Learning with Cross-Validation Gradients
### Paper Abstract
Cross-validation is the workhorse of modern applied statistics and m... |
Io8oYQb4LRK | ICLR.cc/2021/Conference | 2021 | Non-greedy Gradient-based Hyperparameter Optimization Over Long Horizons | ["Paul Micaelli", "Amos Storkey"] | Gradient-based meta-learning has earned a widespread popularity in few-shot learning, but remains broadly impractical for tasks with long horizons (many gradient steps), due to memory scaling and gradient degradation issues. A common workaround is to learn meta-parameters online, but this introduces greediness which co... | ["Hyperparameter optimization", "Meta-learning"] | ABSTRACTGradient-based meta-learning has earned a widespread popularity in few-shotdeep learning, but remains broadly impractical for tasks with long horizons (manygradient steps), due to memory scaling and gradient degradation issues. A com-mon workaround is to learn meta-parameters online, but this introduces greedin... | 6HQMY4zt-_ | Reviewer 2 | 6: Marginally above acceptance threshold | The paper proposes an algorithm for tuning hyperparameters over long optimization problems to mitigate sub-optimal greedy solutions from short-horizon bias. This is done by:
(1) sharing hyperparameters across optimization steps which reduces the variance of hypergradients and computational cost
(2) A method for as... | 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
Non-greedy Gradient-based Hyperparameter Optimization Over Long Horizons
### Paper Abstract
Gradient-based meta-learning has earned a widespread popularity in few-shot l... |
VlRqY4sV9FO | ICLR.cc/2021/Conference | 2021 | Human-interpretable model explainability on high-dimensional data | ["Damien de Mijolla", "Christopher Frye", "Markus Kunesch", "John Mansir", "Ilya Feige"] | The importance of explainability in machine learning continues to grow, as both neural-network architectures and the data they model become increasingly complex. Unique challenges arise when a model's input features become high-dimensional: on one hand, principled model-agnostic approaches to explainability become too ... | ["data", "explainability", "model explainability", "latent features", "importance", "machine learning", "architectures", "become", "complex", "unique challenges"] | ABSTRACTThe importance of explainability in machine learning continues to grow, as bothneural-network architectures and the data they model become increasingly com-plex. Unique challenges arise when a model’s input features become high di-mensional: on one hand, principled model-agnostic approaches to explainabilitybec... | Fr_NKfSzAFh | Interesting method but no comparison to other concept-level explanations | 3: Clear rejection | Summary: I think that what is presented is a promising method for human interpretation of high-dimensional models. However, the experiments feel too much like toy examples, without rigorous attempts to validate the resulting feature attributions or to compare to other ways of getting concept-level attributions.
**Upda... | 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
Human-interpretable model explainability on high-dimensional data
### Paper Abstract
The importance of explainability in machine learning continues to grow, as both neur... |
HJxydABnbS | MICCAI.org/2019/Workshop/COMPAY | 2019 | NuClick: From Clicks in the Nuclei to Nuclear Boundaries | ["Mostafa Jahanifar", "Navid Alemi Koohbabnani", "Nasir Rajpoot"] | Best performing nuclear segmentation methods are based on deep learning algorithms that require a large amount of annotated data. However, collecting annotations for nuclear segmentation is a very labor-intensive and time-consuming task. Thereby, providing a tool that can facilitate and speed up this procedure is very ... | ["Interactive annotation", "nuclei segmentation", "instance segmentation and computational pathology"] | NuClick: From Clicks in the Nuclei to NuclearBoundariesMostafa Jahanifar?2, Navid Alemi Koohbanani?1,3, and Nasir Rajpoot1,31Department of Computer Science, University of Warwick, Coventry2Department of Research & Development, NRP Co., Tehran, Iran3Alan Turing Institute, London, UKAbstract. Best performing nuclear segm... | H1gtj9uCXH | Practical tool to facilitate data annotation for cell nuclei segmentation algorithms | 7: Good paper, accept | This paper focuses on facilitating the collection of annotated data to train deep-learning algorithms for cell nuclei segmentation. It presents a simple CNN-based approach named NuClick which is able to segment nuclei boundaries from a single point position inside each nucleus. The approach works by generating inclusio... | 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
NuClick: From Clicks in the Nuclei to Nuclear Boundaries
### Paper Abstract
Best performing nuclear segmentation methods are based on deep learning algorithms that requi... |
SkeZEhR5FQ | ICLR.cc/2019/Conference | 2019 | Learning Graph Decomposition | ["Jie Song", "Bjoern Andres", "Michael Black", "Otmar Hilliges", "Siyu Tang"] | We propose a novel end-to-end trainable framework for the graph decomposition problem. The minimum cost multicut problem is first converted to an unconstrained binary cubic formulation where cycle consistency constraints are incorporated into the objective function. The new optimization problem can be viewed as a Condi... | ["multicut graph decomposition", "optimization by learning", "pose estimation", "clustering"] | ABSTRACTWe propose a novel end-to-end trainable framework for the graph decompositionproblem. The minimum cost multicut problem is first converted to an uncon-strained binary cubic formulation where cycle consistency constraints are incor-porated into the objective function. The new optimization problem can be viewedas... | SyejICVyam | The authors proposed an end-to-end trainable framework for joint representation learning and graph decomposition. The proposed method achieved on-par performance consistently across classes. | 7: Good paper, accept | The paper is very well written and the motivation of the end-to-end trainable framework for graph decomposition problem is very clear and strong. The proposed method is also fairly novel, especially the decision to convert the discrete optimization problem to an unconstrained binary cubic problem.
Finally, the propose... | 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 Graph Decomposition
### Paper Abstract
We propose a novel end-to-end trainable framework for the graph decomposition problem. The minimum cost multicut problem ... |
HkwKQJgYg | ICLR.cc/2017/workshop | 2017 | Neural Style Representations of Fine Art | ["Jeremiah Johnson"] | The artistic style of a painting is a subtle aesthetic judgment used by art historians for grouping and classifying artwork. The neural style algorithm introduced by Gatys et. al. (2016) substantially succeeds in image style transfer, the task of merging the style of one image with the content of another. This work inv... | ["artistic style", "neural style algorithm", "neural style representations", "fine art", "painting", "subtle aesthetic judgment", "art historians", "artwork", "gatys et"] | ABSTRACTThe artistic style of a painting is a subtle aesthetic judgment used by art histori-ans for grouping and classifying artwork. The neural style algorithm introducedby Gatys et al. (2016) substantially succeeds in image style transfer, the task ofmerging the style of one image with the content of another. This wo... | SyL2y7eox | Interesting work but missing analyses. | 5: Marginally below acceptance threshold | This submission explores how the Neural Style representation by Gatys et al. can be used to classify the style of paintings.
The authors train a linear classifier which uses the neural style representation (gram matrices of a certain selection of layers) to predict the style. This approach is compared to a random fores... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Style Representations of Fine Art
### Paper Abstract
The artistic style of a painting is a subtle aesthetic judgment used by art historians for grouping and class... |
wiSgdeJ29ee | ICLR.cc/2021/Conference | 2021 | Fine-Tuning Offline Reinforcement Learning with Model-Based Policy Optimization | ["Adam Villaflor", "John Dolan", "Jeff Schneider"] | In offline reinforcement learning (RL), we attempt to learn a control policy from a fixed dataset of environment interactions. This setting has the potential benefit of allowing us to learn effective policies without needing to collect additional interactive data, which can be expensive or dangerous in real-world syste... | ["Offline Reinforcement Learning", "Model-Based Reinforcement Learning", "Off-policy Reinforcement Learning", "uncertainty estimation"] | ABSTRACTIn offline reinforcement learning (RL), we attempt to learn a control policy from afixed dataset of environment interactions. This setting has the potential benefit ofallowing us to learn effective policies without needing to collect additional inter-active data, which can be expensive or dangerous in real-worl... | EiXORgsKHPC | Review on the Paper "Fine-Tuning Offline Reinforcement Learning with Model-Based Policy Optimization" | 4: Ok but not good enough - rejection | 1. Summary
This paper proposes an improved offline RL (batch RL) algorithm combining the state-of-the-art behavior-regularization actor-critic method (Nair et al., 2020) with a model-based RL technique. N-trained probabilistic dynamics models generate fictitious trajectories with uncertainty-penalized rewards after pr... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Fine-Tuning Offline Reinforcement Learning with Model-Based Policy Optimization
### Paper Abstract
In offline reinforcement learning (RL), we attempt to learn a control ... |
B1edvs05Y7 | ICLR.cc/2019/Conference | 2019 | Sparse Binary Compression: Towards Distributed Deep Learning with minimal Communication | ["Felix Sattler", "Simon Wiedemann", "Klaus-Robert M\u00fcller", "Wojciech Samek"] | Currently, progressively larger deep neural networks are trained on ever growing data corpora. In result, distributed training schemes are becoming increasingly relevant. A major issue in distributed training is the limited communication bandwidth between contributing nodes or prohibitive communication cost in general.... | ["sbc", "sparse binary compression", "towards", "deep learning", "distributed training", "number", "experiments", "minimal communication"] | ABSTRACTCurrently, progressively larger deep neural networks are trained on ever growingdata corpora. In result, distributed training schemes are becoming increasingly rele-vant. A major issue in distributed training is the limited communication bandwidthbetween contributing nodes or prohibitive communication cost in g... | rJgh-xo53X | important problem, limited novelty, significance not clearly established because reporting focuses on means (bit/communication metric) not ends (optimization time) | 6: Marginally above acceptance threshold | Lowering costs for communicating weights between workers is an important intermediate goal for distributed optimization, since presumably it can limit the parallelization achievable once available bandwidth is saturated. This work reports reasonable approaches to try to overcome this through a mix of techniques, thoug... | 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
Sparse Binary Compression: Towards Distributed Deep Learning with minimal Communication
### Paper Abstract
Currently, progressively larger deep neural networks are train... |
ryserbZR- | ICLR.cc/2018/Conference | 2018 | Classification and Disease Localization in Histopathology Using Only Global Labels: A Weakly-Supervised Approach | ["Pierre Courtiol", "Eric W. Tramel", "Marc Sanselme", "Gilles Wainrib"] | Analysis of histopathology slides is a critical step for many diagnoses, and in particular in oncology where it defines the gold standard. In the case of digital histopathological analysis, highly trained pathologists must review vast whole-slide-images of extreme digital resolution (100,000^2 pixels) across multiple z... | ["Weakly Supervised Learning", "Medical Imaging", "Histopathology", "Deep Feature Extraction"] | ABSTRACTAnalysis of histopathology slides is a critical step for many diagnoses, and inparticular in oncology where it defines the gold standard. In the case of digitalhistopathological analysis, highly trained pathologists must review vast whole-slide-images of extreme digital resolution ( 100;0002pixels) across multi... | SkWQLvebf | Down-to-earth practical application of DL in a medico-clinical context | 6: Marginally above acceptance threshold | This paper proposes a deep learning (DL) approach (pre-trained CNNs) to the analysis of histopathological images for disease localization.
It correctly identifies the problem that DL usually requires large image databases to provide competitive results, while annotated histopathological data repositories are costly 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
Classification and Disease Localization in Histopathology Using Only Global Labels: A Weakly-Supervised Approach
### Paper Abstract
Analysis of histopathology slides is ... |
eEeyRrKVfbL | ICLR.cc/2021/Conference | 2021 | Balancing training time vs. performance with Bayesian Early Pruning | ["Mohit Rajpal", "Yehong Zhang", "Bryan Kian Hsiang Low"] | Pruning is an approach to alleviate overparameterization of deep neural networks (DNN) by zeroing out or pruning DNN elements with little to no efficacy at a given task. In contrast to related works that do pruning before or after training, this paper presents a novel method to perform early pruning of DNN elements (e.... | ["Efficient Training", "Multi-Output Gaussian Process", "Gaussian Process", "Bayesian", "Single-shot network pruning", "Dynamic Sparse Reparameterization", "Lottery Ticket Hypothesis"] | ABSTRACTPruning is an approach to alleviate overparameterization of deep neural network(DNN) by zeroing out or pruning DNN elements with little to no efficacy at a giventask. In contrast to related works that do pruning before or after training, this paperpresents a novel method to perform early pruning of DNN elements... | QQsaYZ8Icj | Well organized, interesting contribution, experiments could be improved | 7: Good paper, accept | This paper presents a training-time pruning method for deep neural network. The main idea is to include network elements only if they could improve the predictive mean of the saliency, where the saliency measures the efficiency of the network elements in terms of minimizing the loss function.
The paper presents a clea... | 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Balancing training time vs. performance with Bayesian Early Pruning
### Paper Abstract
Pruning is an approach to alleviate overparameterization of deep neural networks (... |
e1RgaHCQrT | Interspeech.org/2023/Workshop/SSW | 2023 | Voice Cloning: Training Speaker Selection with Limited Multi-Speaker Corpus | ["David Guennec", "Lily Wadoux", "Aghilas Sini", "Nelly Barbot", "Damien Lolive"] | Text-to-speech synthesis with few data is a challenging task, in particular when choosing the target speaker is not an option. Voice cloning is a popular method to alleviate these issues using only a few minutes of target speech. To do this, the model must first be trained on a large corpus of thousands of hours and hu... | ["Voice Cloning", "Speaker Adaptation", "Speaker Encoding", "Speaker Similarity", "Data Selection", "Neural Speech Synthesis"] | Voice Cloning: Training Speaker Selection in Limited Multi-Speaker CorpusDavid Guennec, Lily Wadoux, Aghilas Sini, Nelly Barbot, Damien LoliveUniv Rennes, CNRS, IRISA, France{firstname.secondname }@irisa.frAbstractText-To-Speech synthesis with few data is a challengingtask, in particular when choosing the target speake... | Gznp5OTtTIC | Experiments are reasonable, but technical novelty is limited | 6: Marginally above acceptance threshold | The aim of this paper is to investigate the performance of voice cloning approaches for text-to-speech (TTS) synthesis, i.e., speaker adaptation and speaker encoding, when only a small number of samples are available for the target speaker and the scale of multi-speaker corpora is limited. Specifically, to investigate ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Voice Cloning: Training Speaker Selection with Limited Multi-Speaker Corpus
### Paper Abstract
Text-to-speech synthesis with few data is a challenging task, in particula... |
HJxqMhC5YQ | ICLR.cc/2019/Conference | 2019 | End-to-End Multi-Lingual Multi-Speaker Speech Recognition | ["Hiroshi Seki", "Takaaki Hori", "Shinji Watanabe", "Jonathan Le Roux", "John R. Hershey"] | The expressive power of end-to-end automatic speech recognition (ASR) systems enables direct estimation of the character or word label sequence from a sequence of acoustic features. Direct optimization of the whole system is advantageous because it not only eliminates the internal linkage necessary for hybrid systems,... | ["end-to-end ASR", "multi-lingual ASR", "multi-speaker ASR", "code-switching", "encoder-decoder", "connectionist temporal classification"] | ABSTRACTThe expressive power of end-to-end automatic speech recognition (ASR) systemsenables direct estimation of the character or word label sequence from a sequenceof acoustic features. Direct optimization of the whole system is advantageous be-cause it not only eliminates the internal linkage necessary for hybrid sy... | HkeUS1ulpm | not enough novelty | 3: Clear rejection | This paper presents an end-to-end system that can recognize single-channel multiple-speaker speech with multiple languages.
Pros:
- The paper is well written.
- It shows the existing end-to-end multi-lingual ASR (Seki et al., 2018b) and end-to-end multi-speaker ASR (Seki et al., 2018a) techniques can be combined witho... | 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
End-to-End Multi-Lingual Multi-Speaker Speech Recognition
### Paper Abstract
The expressive power of end-to-end automatic speech recognition (ASR) systems enables direct... |
69y5fzvaAT | robot-learning.org/CoRL/2023/Conference | 2023 | RoboCook: Long-Horizon Elasto-Plastic Object Manipulation with Diverse Tools | ["Haochen Shi", "Huazhe Xu", "Samuel Clarke", "Yunzhu Li", "Jiajun Wu"] | Humans excel in complex long-horizon soft body manipulation tasks via flexible tool use: bread baking requires a knife to slice the dough and a rolling pin to flatten it. Often regarded as a hallmark of human cognition, tool use in autonomous robots remains limited due to challenges in understanding tool-object interac... | ["Deformable Object Manipulation", "Long-horizon Planning", "Model Learning", "Tool Usage"] | RoboCook: Long-Horizon Elasto-Plastic ObjectManipulation with Diverse ToolsHaochen Shi1*Huazhe Xu1*†Samuel Clarke1Yunzhu Li1,2Jiajun Wu11Stanford University2UIUC *Equal contributionhttps://hshi74.github.io/robocookAbstract: Humans excel in complex long-horizon soft body manipulation tasksvia flexible tool use: bread ba... | AMB0sJ6DGAP | Strength:
* Challenging long-horizon and deformable object manipulation task;
* Extensive studies with comparison to baselines and a generalisation task.
Weakness:
* Limited method novelty;
* Unclear that how large the datasets are for a sufficient coverage of state space;
* Lacking insights to interpret the performan... | 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
RoboCook: Long-Horizon Elasto-Plastic Object Manipulation with Diverse Tools
### Paper Abstract
Humans excel in complex long-horizon soft body manipulation tasks via fle... | ||
Y5TgO3J_Glc | ICLR.cc/2021/Conference | 2021 | Neurosymbolic Deep Generative Models for Sequence Data with Relational Constraints | ["Halley Young", "Maxwell Du", "Osbert Bastani"] | Recently, there has been significant progress designing deep generative models that generate realistic sequence data such as text or music. Nevertheless, it remains difficult to incorporate high-level structure to guide the generative process. We propose a novel approach for incorporating structure in the form of relat... | ["neurosymbolic", "sequence", "program synthesis", "generative", "constraint", "music", "poetry"] | Under review as a conference paper at ICLR 2020Neurosymbolic Deep Generative Models forSequence Data with Relational ConstraintsAnonymous authorsPaper under double-blind reviewAbstractRecently, there has been signicant progress designing deep generative mod-els that generate realistic sequence data such as text or musi... | CWS1rIhHkFY | Interesting method, but insufficient comparison to existing methods and analysis of results | 4: Ok but not good enough - rejection | **Summary**
This paper presents a novel method of incorporating structural and relational constraints into the generative process of sequences. This approach is explored in both the music and poetry domains using several generative approaches. An analysis of the resulting sequences finds that they successfully incorpo... | 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
Neurosymbolic Deep Generative Models for Sequence Data with Relational Constraints
### Paper Abstract
Recently, there has been significant progress designing deep genera... |
HygaikBKvS | ICLR.cc/2020/Conference | 2020 | Off-Policy Actor-Critic with Shared Experience Replay | ["Simon Schmitt", "Matteo Hessel", "Karen Simonyan"] | We investigate the combination of actor-critic reinforcement learning algorithms with uniform large-scale experience replay and propose solutions for two challenges: (a) efficient actor-critic learning with experience replay (b) stability of very off-policy learning. We employ those insights to accelerate hyper-paramet... | ["Reinforcement Learning", "Off-Policy Learning", "Experience Replay"] | ABSTRACTWe investigate the combination of actor-critic reinforcement learning algorithmswith uniform large-scale experience replay and propose solutions for two chal-lenges: (a) efficient actor-critic learning with experience replay (b) stability ofoff-policy learning where agents learn from other agents behaviour. We ... | HygTpk4Z9S | Official Blind Review #3 | 6: Weak Accept | This paper investigates off-policy actor critic (AC) learning with experience replay using V-trace. It shows that V-trace policy gradient is not guaranteed to converge to a local optimal solution. To mitigate the bias and variance problem of V-trace and importance sampling, a trust region approach is proposed to adapti... | <|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 Actor-Critic with Shared Experience Replay
### Paper Abstract
We investigate the combination of actor-critic reinforcement learning algorithms with uniform la... | |
7TBP8k7TLFA | ICLR.cc/2021/Conference | 2021 | Universal Approximation Theorem for Equivariant Maps by Group CNNs | ["Wataru Kumagai", "Akiyoshi Sannai"] | Group symmetry is inherent in a wide variety of data distributions. Data processing that preserves symmetry is described as an equivariant map and often effective in achieving high performance. Convolutional neural networks (CNNs) have been known as models with equivariance and shown to approximate equivariant maps for... | ["Universal Approximation Theorem", "CNN", "Deep Learning", "Symmetry"] | ABSTRACTGroup symmetry is inherent in a wide variety of data distributions. Data pro-cessing that preserves symmetry is described as an equivariant map and ofteneffective in achieving high performance. Convolutional neural networks (CNNs)have been known as models with equivariance and shown to approximate equivari-ant ... | kxROIr6WRkv | Important problem, interesting results but very hard to follow | 5: Marginally below acceptance threshold | After discussions:
I read the author's response and other reviews. The authors made a considerable effort to address my concerns. As a result, the paper has improved and I increased my score.
Having said that, I am still not sure that the paper is ready for publication in its current form. My main concern is still th... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Universal Approximation Theorem for Equivariant Maps by Group CNNs
### Paper Abstract
Group symmetry is inherent in a wide variety of data distributions. Data processing... |
mQPBmvyAuk | ICLR.cc/2021/Conference | 2021 | BREEDS: Benchmarks for Subpopulation Shift | ["Shibani Santurkar", "Dimitris Tsipras", "Aleksander Madry"] | We develop a methodology for assessing the robustness of models to subpopulation shift---specifically, their ability to generalize to novel data subpopulations that were not observed during training. Our approach leverages the class structure underlying existing datasets to control the data subpopulations that comprise... | ["benchmarks", "distribution shift", "hierarchy", "robustness"] | ABSTRACTWe develop a methodology for assessing the robustness of models to subpopulationshift—specifically, their ability to generalize to novel data subpopulations that werenot observed during training. Our approach leverages the class structure underlyingexisting datasets to control the data subpopulations that compr... | JFo4ZbXHLly | Straightforward approach for evaluating model robustness to subpopulation shift | 6: Marginally above acceptance threshold | This paper addresses the problem of model robustness to subpopulation shift. Authors propose building large-scale subpopulation shift benchmarks wherein the data subpopulations present during model training and evaluation differ. In this regard, their approach is based on leveraging existing dataset labels and use them... | 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
BREEDS: Benchmarks for Subpopulation Shift
### Paper Abstract
We develop a methodology for assessing the robustness of models to subpopulation shift---specifically, thei... |
2Qrd-Yw4YmF | robot-learning.org/CoRL/2023/Conference | 2023 | Sequential Dexterity: Chaining Dexterous Policies for Long-Horizon Manipulation | ["Yuanpei Chen", "Chen Wang", "Li Fei-Fei", "Karen Liu"] | Many real-world manipulation tasks consist of a series of subtasks that are significantly different from one another. Such long-horizon, complex tasks highlight the potential of dexterous hands, which possess adaptability and versatility, capable of seamlessly transitioning between different modes of functionality with... | ["Dexterous Manipulation", "Reinforcement Learning", "Long-Horizon Manipulation"] | Sequential Dexterity: Chaining Dexterous Policiesfor Long-Horizon ManipulationYuanpei Chen∗, Chen Wang∗, Li Fei-Fei, C. Karen LiuStanford UniversityAbstract: Many real-world manipulation tasks consist of a series of subtasks thatare significantly different from one another. Such long-horizon, complex taskshighlight the... | -vHfDu0ikV | Strengths:
- To my knowledge, this paper presents a novel approach for long horizon dexterous manipulation
- I think the paper tackles a very difficult and important problem in an interesting and novel way
- The transition feasibility function is very useful for even more general purpose settings
- I also think the b... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Sequential Dexterity: Chaining Dexterous Policies for Long-Horizon Manipulation
### Paper Abstract
Many real-world manipulation tasks consist of a series of subtasks tha... | ||
I6QHpMdZD5k | ICLR.cc/2021/Conference | 2021 | Learning to Solve Nonlinear Partial Differential Equation Systems To Accelerate MOSFET Simulation | ["Seungcheol Han", "Jonghyun Choi", "Sung-Min Hong"] | Semiconductor device simulation uses numerical analysis, where a set of coupled nonlinear partial differential equations is solved with the iterative Newton-Raphson method. Since an appropriate initial guess to start the Newton-Raphson method is not available, a solution of practical importance with desired boundary co... | ["Partial differential equation", "nonlinear equation", "Newton-Raphson method", "convolutional neural network"] | ABSTRACTSemiconductor device simulation uses numerical analysis, where a set of cou-pled nonlinear partial differential equations is solved with the iterative Newton-Raphson method. Since an appropriate initial guess to start the Newton-Raphsonmethod is not available, a solution of practical importance with desired bou... | ZHg9PaoU_YW | Mehtod is not new; good engineering application | 5: Marginally below acceptance threshold | The paper proposed to use a CNN as a low-accuracy solver to output a good initial guess from the device parameters and then the solution is refined by an expensive numerical solver to get a good accuracy. The experiment on the MOSFET problem shows that, compared with the conventional numerical solver, this hybrid strat... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Solve Nonlinear Partial Differential Equation Systems To Accelerate MOSFET Simulation
### Paper Abstract
Semiconductor device simulation uses numerical analy... |
SJLlmG-AZ | ICLR.cc/2018/Conference | 2018 | Understanding image motion with group representations | ["Andrew Jaegle", "Stephen Phillips", "Daphne Ippolito", "Kostas Daniilidis"] | Motion is an important signal for agents in dynamic environments, but learning to represent motion from unlabeled video is a difficult and underconstrained problem. We propose a model of motion based on elementary group properties of transformations and use it to train a representation of image motion. While most metho... | ["vision", "motion", "recurrent neural networks", "self-supervised learning", "unsupervised learning", "group theory"] | ABSTRACTMotion is an important signal for agents in dynamic environments, but learning torepresent motion from unlabeled video is a difficult and underconstrained problem.We propose a model of motion based on elementary group properties of transfor-mations and use it to train a representation of image motion. While mos... | SkG-iVtlz | Interesting idea, but more detailed experimental validation is needed. | 5: Marginally below acceptance threshold | Paper proposes an approach for learning video motion features in an unsupervised manner. A number of constraints are used to optimize the neural network that consists of CNN + RNN (LSTM). Constraints stem from group structure of sequences and include associativity and inevitability. For example, forward-backward motion... | 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
Understanding image motion with group representations
### Paper Abstract
Motion is an important signal for agents in dynamic environments, but learning to represent mot... |
BJgd7m0xRZ | ICLR.cc/2018/Conference | 2018 | Unsupervised Adversarial Anomaly Detection using One-Class Support Vector Machines | ["Prameesha Sandamal Weerasinghe", "Tansu Alpcan", "Sarah Monazam Erfani", "Christopher Leckie"] | Anomaly detection discovers regular patterns in unlabeled data and identifies the non-conforming data points, which in some cases are the result of malicious attacks by adversaries. Learners such as One-Class Support Vector Machines (OCSVMs) have been successfully in anomaly detection, yet their performance may degrade... | ["anomaly detection", "one class support vector machine", "adversarial learning"] | ABSTRACTAnomaly detection discovers regular patterns in unlabeled data and identifies thenon-conforming data points, which in some cases are the result of maliciousattacks by adversaries. Learners such as One-Class Support Vector Machines(OCSVMs) have been successfully used in anomaly detection, yet their perfor-mance ... | BJLMRY2gG | Paper needs a clearer explanation and more thorough experiments | 4: Ok but not good enough - rejection | In this paper, the authors explore how using random projections can be used to make OCSVM robust to adversarially perturbed training data. While the intuition is nice and interesting, the paper is not very clear in describing the attack and the experiments do not appropriately test whether this method actually provide... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Unsupervised Adversarial Anomaly Detection using One-Class Support Vector Machines
### Paper Abstract
Anomaly detection discovers regular patterns in unlabeled data and... |
HMqNjkBEqP4 | ICLR.cc/2021/Conference | 2021 | Bayesian Meta-Learning for Few-Shot 3D Shape Completion | ["Masanori Koyama", "Toshiki Nakanishi", "Shin-ichi Maeda", "Vitor Campagnolo Guizilini", "Adrien Gaidon"] | Estimating the 3D shape of real-world objects is a key perceptual challenge. It requires going from partial observations, which are often too sparse and incomprehensible for the human eye, to detailed shape representations that vary significantly across categories and instances. We propose to cast shape completion as a... | ["shape completion", "Meta-learning", "Few-shot", "3D reconstruction"] | ABSTRACTEstimating the 3D shape of real-world objects is a key perceptual challenge. Itrequires going from partial observations, which are often too sparse and incompre-hensible for the human eye, to detailed shape representations that vary significantlyacross categories and instances. We propose to cast shape completi... | b1Vd19hVzjU | Interesting idea and impressive results, but lack of literature review. | 7: Good paper, accept | This paper introduces a meta-learning approach for the neural implicit representation of 3D shapes. The main idea, in my understanding, is to consider the points in the input point cloud as few-shot examples of the object so that each of them can be encoded in a way to best approximate the entire object information. 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
Bayesian Meta-Learning for Few-Shot 3D Shape Completion
### Paper Abstract
Estimating the 3D shape of real-world objects is a key perceptual challenge. It requires goin... |
fV8CfTjnXgH | NeurIPS.cc/2022/Workshop/SVRHM | 2022 | Fast temporal decoding from large-scale neural recordings in monkey visual cortex | ["Jerome Hadorn", "Zuowen Wang", "Bodo Rueckauer", "Xing Chen", "Pieter R. Roelfsema", "Shih-Chii Liu"] | With new developments in electrode and nanoscale technology, a large-scale multi-electrode cortical neural prosthesis with thousands of stimulation and recording electrodes is becoming viable. Such a system will be useful as both a neuroscience tool and a neuroprosthesis.
In the context of a visual neuroprosthesis, a ... | ["rapid neural decoding", "high-channel count vision neuroprosthesis", "informative electrode selection", "deep learning methods"] | Fast temporal decoding from large-scale neuralrecordings in monkey visual cortexJerome Hadorn1∗, Zuowen Wang1∗, Bodo Rueckauer2,Xing Chen3, Pieter R. Roelfsema4, Shih-Chii Liu11Sensors Group, Institute of Neuroinformatics, University of Zurich and ETH Zurich2Radboud University Nijmegen3University of Pittsburgh School o... | xWwg1wrJei | Clear decoding project on unique dataset with strong results | 8: Top 50% of accepted papers, clear accept | The project uses a unique monkey physiology dataset based on Utah array recordings, aimed at developing visual prostheses. It demonstrates and investigates DNN-based decoding of letter characters from early visual areas V1 and V4, presented during a visual discrimination task.
Overall this decoding project is quite 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
Fast temporal decoding from large-scale neural recordings in monkey visual cortex
### Paper Abstract
With new developments in electrode and nanoscale technology, a large... |
rke2HRVYvH | ICLR.cc/2020/Conference | 2020 | Stochastic Prototype Embeddings | ["Tyler R. Scott", "Karl Ridgeway", "Michael C. Mozer"] | Supervised deep-embedding methods project inputs of a domain to a representational space in which same-class instances lie near one another and different-class instances lie far apart. We propose a probabilistic method that treats embeddings as random variables. Extending a state-of-the-art deterministic method, Protot... | ["deep embeddings", "stochastic embeddings", "probabilistic embeddings", "deep metric learning", "few-shot learning"] | ABSTRACTSupervised deep-embedding methods project inputs of a domain to a representa-tional space in which same-class instances lie near one another and different-classinstances lie far apart. We propose a probabilistic method that treats embeddingsas random variables. Extending a state-of-the-art deterministic method,... | S1xNTO8ptS | Official Blind Review #1 | 3: Weak Reject | Authors propose Stochastic Prototype Embeddings, as a probabilistic extension of Prototype Networks (snell et al, 2017) for few-shot classification. Authors claim their method leads to better few-shot results in the case of corrupted data, and support their claims with a fair number of results over N-Mnist, with 'Hedge... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Stochastic Prototype Embeddings
### Paper Abstract
Supervised deep-embedding methods project inputs of a domain to a representational space in which same-class instances... | |
rJlDnoA5Y7 | ICLR.cc/2019/Conference | 2019 | Von Mises-Fisher Loss for Training Sequence to Sequence Models with Continuous Outputs | ["Sachin Kumar", "Yulia Tsvetkov"] | The Softmax function is used in the final layer of nearly all existing sequence-to-sequence models for language generation. However, it is usually the slowest layer to compute which limits the vocabulary size to a subset of most frequent types; and it has a large memory footprint. We propose a general technique for rep... | ["Language Generation", "Regression", "Word Embeddings", "Machine Translation"] | ABSTRACTThe Softmax function is used in the final layer of nearly all existing sequence-to-sequence models for language generation. However, it is usually the slowest layerto compute which limits the vocabulary size to a subset of most frequent types;and it has a large memory footprint. We propose a general technique f... | HylClIbC2m | cool new approach with some limitations | 6: Marginally above acceptance threshold | This paper proposes to replace the softmax over the vocab in the decoder with a single embedding layer using the Von Mises-Fisher distribution, which speeds up training 2.5x compared to a standard softmax+cross entropy decoder. The goal is admirable, as the softmax during training is a huge time sink (the proposed appr... | 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
Von Mises-Fisher Loss for Training Sequence to Sequence Models with Continuous Outputs
### Paper Abstract
The Softmax function is used in the final layer of nearly all e... |
oW4rUGjbMYg | ACM.org/ICMI/2023/Workshop/GENEA_Challenge | 2023 | The KCL-SAIR team's entry to the GENEA Challenge 2023 Exploring Role-based Gesture Generation in Dyadic Interactions: Listener vs. Speaker | ["Viktor Schmuck", "Nguyen Tan Viet Tuyen", "Oya Celiktutan"] | This paper presents the KCL-SAIR team's contribution to the GENEA Challenge 2023. As this year's challenge addressed gesture generation in a dyadic context instead of a monadic one, our aim was to investigate how the previous state-of-the-art approach can be improved to be more applicable for the generation of both spe... | ["datasets", "Tacotron2", "gesture generation", "dyadic interaction"] | ABSTRACTThis paper presents the KCL-SAIR team’s contribution to the GE-NEA Challenge 2023. As this year’s challenge addressed gesturegeneration in a dyadic context instead of a monadic one, our aimwas to investigate how the previous state-of-the-art approach can beimproved to be more applicable for the generation of bo... | CJFqLJPVVc | I like the idea of having speaker model and listener model | 10: Top 5% of accepted papers, seminal paper | The paper proposed to have two models (a speaker model and a listener model) for gesture generation in dyadic situation.
The authors split dataset into a speaker dataset and a listener dataset (recognized by speaker ID) and separately trained a Tactron2-based (for text-to-audio generation) seq2seq model. For generatio... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The KCL-SAIR team's entry to the GENEA Challenge 2023 Exploring Role-based Gesture Generation in Dyadic Interactions: Listener vs. Speaker
### Paper Abstract
This paper ... |
rkxwVZPpPV | icaps-conference.org/ICAPS/2019/Workshop/HSDIP | 2019 | Beyond Cost-to-go Estimates in Situated Temporal Planning | ["Anonymous"] | Heuristic search research often deals with finding algorithms for offline planning which aim to minimize the number of expanded nodes or planning time. In online planning, algorithms for real-time search or deadline-aware search have been considered before. However, in this paper, we are interested in the problem of {\... | ["heuristic search", "automated planning", "temporal planning", "real time search"] | Beyond Cost-to-go Estimates in Situated Temporal PlanningAndrew Coles,2Shahaf S. Shperberg,1Erez Karpas,4Solomon E. Shimony1Wheeler Ruml,31Ben-Gurion University, Israel;2King’s College London, UK;3University of New Hampshire, USA;4Technion, Israel;andrew.coles@kcl.ac.uk, shperbsh@post.bgu.ac.il, karpase@technion.ac.ils... | SylYe8R4t4 | Interesting experimental observation, but significance and contribution not clear | 5: Marginally below acceptance threshold | Very interesting topic, but not clear exactly what is the contribution or the message the authors want to convey.
Key details of the AE2 model are not defined in the paper. The authors say that it is an MDP but basic facts about it are not present in the manuscript. This is a major omission, and made quite difficult ... | 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
Beyond Cost-to-go Estimates in Situated Temporal Planning
### Paper Abstract
Heuristic search research often deals with finding algorithms for offline planning which aim... |
SJlRUkrFPS | ICLR.cc/2020/Conference | 2020 | Learning transport cost from subset correspondence | ["Ruishan Liu", "Akshay Balsubramani", "James Zou"] | Learning to align multiple datasets is an important problem with many applications, and it is especially useful when we need to integrate multiple experiments or correct for confounding. Optimal transport (OT) is a principled approach to align datasets, but a key challenge in applying OT is that we need to specify a c... | ["correspondence", "datasets", "side information", "transport cost", "cost function", "available", "images", "multiple datasets", "important problem"] | ABSTRACTLearning to align multiple datasets is an important problem with many applications,and it is especially useful when we need to integrate multiple experiments orcorrect for confounding. Optimal transport (OT) is a principled approach toalign datasets, but a key challenge in applying OT is that we need to specify... | SklG_8hatS | Official Blind Review #1 | 6: Weak Accept | # Summary
This paper proposes a new way to learn the optimal transport (OT) cost between two datasets that can utilize subset correspondence information. The main idea is to use a neural network to define a transport cost between two examples instead of using the popular Euclidean distance and have a L2 regularizer tha... | <|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 transport cost from subset correspondence
### Paper Abstract
Learning to align multiple datasets is an important problem with many applications, and it is espec... | |
HJeOekHKwr | ICLR.cc/2020/Conference | 2020 | Smoothness and Stability in GANs | ["Casey Chu", "Kentaro Minami", "Kenji Fukumizu"] | Generative adversarial networks, or GANs, commonly display unstable behavior during training. In this work, we develop a principled theoretical framework for understanding the stability of various types of GANs. In particular, we derive conditions that guarantee eventual stationarity of the generator when it is trained... | ["generative adversarial networks", "stability", "smoothness", "convex conjugate"] | ABSTRACTGenerative adversarial networks, or GANs, commonly display unstable behaviorduring training. In this work, we develop a principled theoretical framework forunderstanding the stability of various types of GANs. In particular, we deriveconditions that guarantee eventual stationarity of the generator when it is tr... | Skl_j_8FKB | Official Blind Review #2 | 6: Weak Accept | The work studies the relationship between the stability and the smoothness of GANs based on the proposition which was proposed by Bertsekas . It explains many nontrivial empirical observations when one is training GANs, including both of the necessities of the spectral normalization and the gradient penalty, in a theor... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Smoothness and Stability in GANs
### Paper Abstract
Generative adversarial networks, or GANs, commonly display unstable behavior during training. In this work, we develo... | |
BJg4NgBKvH | ICLR.cc/2020/Conference | 2020 | Training binary neural networks with real-to-binary convolutions | ["Brais Martinez", "Jing Yang", "Adrian Bulat", "Georgios Tzimiropoulos"] | This paper shows how to train binary networks to within a few percent points (~3-5%) of the full precision counterpart. We first show how to build a strong baseline, which already achieves state-of-the-art accuracy, by combining recently proposed advances and carefully adjusting the optimization procedure. Secondly, we... | ["binary networks"] | ABSTRACTThis paper shows how to train binary networks to within a few percent points(35%) of the full precision counterpart. We first show how to build a strongbaseline, which already achieves state-of-the-art accuracy, by combining recentlyproposed advances and carefully adjusting the optimization procedure. Secondly,... | B1lneTXwcH | Official Blind Review #4 | 6: Weak Accept | A. Summary
Problem:
Binary NNs promise to make neural networks compatible with devices that only have access to limited computational resources (fundamental to embed NNs in mobiles or IoT devices). However, the loss of computational accuracy comes with a great loss of performances. Designing the right learning algorit... | <|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 binary neural networks with real-to-binary convolutions
### Paper Abstract
This paper shows how to train binary networks to within a few percent points (~3-5%) ... | |
HJgLZR4KvH | ICLR.cc/2020/Conference | 2020 | Dynamics-Aware Unsupervised Discovery of Skills | ["Archit Sharma", "Shixiang Gu", "Sergey Levine", "Vikash Kumar", "Karol Hausman"] | Conventionally, model-based reinforcement learning (MBRL) aims to learn a global model for the dynamics of the environment. A good model can potentially enable planning algorithms to generate a large variety of behaviors and solve diverse tasks. However, learning an accurate model for complex dynamical systems is diffi... | ["reinforcement learning", "unsupervised learning", "model-based learning", "deep learning", "hierarchical reinforcement learning"] | ABSTRACTConventionally, model-based reinforcement learning (MBRL) aims to learn aglobal model for the dynamics of the environment. A good model can poten-tially enable planning algorithms to generate a large variety of behaviors andsolve diverse tasks. However, learning an accurate model for complex dynami-cal systems ... | SJg5q-L5Yr | Official Blind Review #1 | 8: Accept | This paper proposes a novel approach to learn a continuous set of skills (where a skill is associated with a latent vector and the skill policy network takes that vector as an extra input) by pure unsupervised exploration using as intrinsic reward a proxy for the mutual information between next states and the skill (g... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Dynamics-Aware Unsupervised Discovery of Skills
### Paper Abstract
Conventionally, model-based reinforcement learning (MBRL) aims to learn a global model for the dynamic... | |
aeA3LLDDQe2 | ICLR.cc/2021/Conference | 2021 | Fairness guarantee in analysis of incomplete data | ["Yiliang Zhang", "Qi Long"] | Missing data are prevalent and present daunting challenges in real data analysis. While there is a growing body of literature on fairness in analysis of fully observed data, there has been little work on investigating fairness in analysis of incomplete data when the goal is to develop a fair algorithm in the complete d... | ["Algorithmic fairness", "missing data analysis", "domain adaptation"] | ABSTRACTMissing data are prevalent and present daunting challenges in real data analysis.While there is a growing body of literature on fairness in analysis of fully observeddata, there has been little work on investigating fairness in analysis of incompletedata when the goal is to develop a fair algorithm in the compl... | 43QFda_Y4z | Important Topic that Merits Further Study. Rigorous Theoretical Analysis with Unclear Value in Practice. | 4: Ok but not good enough - rejection | # Summary
This paper presents a theoretical analysis of the estimation error of group fairness metrics in settings with missing data. The authors consider standard mechanisms that result in missing data and derive an upper bound and a lower bound on the estimation error for classification and regression problems. 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
Fairness guarantee in analysis of incomplete data
### Paper Abstract
Missing data are prevalent and present daunting challenges in real data analysis. While there is a g... |
V-SLFW_asA | graphicsinterface.org/Graphics_Interface/2023/Conference | 2023 | Immersive Haptic Authoring: Creating Haptic Experiences On-Demand within VR through Direct Manipulation | ["Anonymous"] | This paper presents immersive haptic authoring, a novel way to author haptic experiences within VR. Designing haptics to create rich VR experiences is becoming increasingly common, yet designers lack the tools to iteratively design, experience, and test such experiences easily. Designing haptics using desktop tools is ... | ["virtual reality", "haptics", "design process"] | ABSTRACTThis paper presents immersive haptic authoring , a novel way toauthor haptic experiences within VR. Designing haptics to createrich VR experiences is becoming increasingly common, yet design-ers lack the tools to iteratively design, experience, and test suchexperiences easily. Designing haptics using desktop to... | a0KM7OaX2j | Useful direction but work is in early stages | 3: Clear rejection | This paper presents an immersive authoring tool for creating haptic feedback in VR experiences.
This research takes a useful and relevant direction toward designing haptic feedback for VR. Overall, the work as presented is fairly preliminary and is more like a poster contribution than a full research paper.
I would n... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Immersive Haptic Authoring: Creating Haptic Experiences On-Demand within VR through Direct Manipulation
### Paper Abstract
This paper presents immersive haptic authoring... |
rkzcvoA9YX | ICLR.cc/2019/Conference | 2019 | Few-Shot Learning by Exploiting Object Relation | ["Liangqu Long", "Wei Wang", "Jun Wen", "Meihui Zhang", "Qian Lin"] |
Few-shot learning trains image classifiers over datasets with few examples per category.
It poses challenges for the optimization algorithms, which typically require many examples to fine-tune the model parameters for new categories.
Distance-learning-based approaches avoid the optimization issue by embedding the im... | ["few-shot learning", "relation learning"] | ABSTRACTFew-shot learning trains image classifiers over datasets with few examples percategory. It poses challenges for the optimization algorithms, which typicallyrequire many examples to fine-tune the model parameters for new categories.Metric-learning-based approaches avoid the optimization issue by embedding theima... | H1xgGJa9hX | Minor extension of the existing method | 4: Ok but not good enough - rejection | - In evaluation of omniglot, ("a small improvement over the state-of-the-art should be considered as significant") since all the results are already quite high, the small difference are not much beneficial in practice. Further, to justify the significance of the proposed method, a statistical test should be conducted a... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Few-Shot Learning by Exploiting Object Relation
### Paper Abstract
Few-shot learning trains image classifiers over datasets with few examples per category. It poses ch... |
BygPq6VFvS | ICLR.cc/2020/Conference | 2020 | Enhancing Attention with Explicit Phrasal Alignments | ["Xuan-Phi Nguyen", "Shafiq Joty", "Thanh-Tung Nguyen"] | The attention mechanism is an indispensable component of any state-of-the-art neural machine translation system. However, existing attention methods are often token-based and ignore the importance of phrasal alignments, which are the backbone of phrase-based statistical machine translation. We propose a novel phrase-ba... | ["NMT", "Phrasal Attention", "Machine Translation", "Language Modeling"] | ABSTRACTThe attention mechanism is an indispensable component of any state-of-the-artneural machine translation system. However, existing attention methods are oftentoken-based and ignore the importance of phrasal alignments, which are the back-bone of phrase-based statistical machine translation. We propose a novel ph... | Byl2Sx7aYH | Official Blind Review #3 | 6: Weak Accept | This paper proposes an extension of the attention module that explicitly incorporates phrase information. Using convolution, attention scores are obtained independently for each n-gram type, and then combined. Transformer models with the proposed phrase attention are evaluated on multiple translation tasks, as well as ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Enhancing Attention with Explicit Phrasal Alignments
### Paper Abstract
The attention mechanism is an indispensable component of any state-of-the-art neural machine tran... | |
O1GEH9X8848 | ICLR.cc/2021/Conference | 2021 | A Point Cloud Generative Model Based on Nonequilibrium Thermodynamics | ["Shitong Luo", "Wei Hu"] | We present a probabilistic model for point cloud generation, which is critical for various 3D vision tasks such as shape completion, upsampling, synthesis and data augmentation. Inspired by the diffusion process in non-equilibrium thermodynamics, we view points in point clouds as particles in a thermodynamic system in ... | ["Point cloud", "Generation", "Generative model"] | ABSTRACTWe present a probabilistic model for point cloud generation, which is critical forvarious 3D vision tasks such as shape completion, upsampling, synthesis and dataaugmentation. Inspired by the diffusion process in non-equilibrium thermody-namics, we view points in point clouds as particles in a thermodynamic sys... | U2bnPNXwTvy | This paper focuses on the point cloud generation task inspired by non-equilibrium thermodynamics. To cope with various shapes, the latent code is applied to construct a conditioned Markov chain based framework to model rich point distributions. This is a very early work that deduces the point cloud generative task by t... | 6: Marginally above acceptance threshold | This paper focuses on the point cloud generation task inspired by non-equilibrium thermodynamics. Transforming noises to a desired shape is considered as the reverse diffusion process, where points are viewed as particles. To cope with various shapes, the latent code is applied to construct a conditioned Markov chain b... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Point Cloud Generative Model Based on Nonequilibrium Thermodynamics
### Paper Abstract
We present a probabilistic model for point cloud generation, which is critical f... |
ryl3ygHYDB | ICLR.cc/2020/Conference | 2020 | Lookahead: A Far-sighted Alternative of Magnitude-based Pruning | ["Sejun Park*", "Jaeho Lee*", "Sangwoo Mo", "Jinwoo Shin"] | Magnitude-based pruning is one of the simplest methods for pruning neural networks. Despite its simplicity, magnitude-based pruning and its variants demonstrated remarkable performances for pruning modern architectures. Based on the observation that magnitude-based pruning indeed minimizes the Frobenius distortion of a... | ["network magnitude-based pruning"] | ABSTRACTMagnitude-based pruning is one of the simplest methods for pruning neural net-works. Despite its simplicity, magnitude-based pruning and its variants demon-strated remarkable performances for pruning modern architectures. Based on theobservation that magnitude-based pruning indeed minimizes the Frobenius distor... | Bkgyu_mq5r | Official Blind Review #2 | 6: Weak Accept | *Summary*
The paper proposes a multi-layer alternative to magnitude-based pruning. The operations entailed in the previous, current, and subsequent layers are treated as linear operations (by omitting any nonlinearities), weights are selected for pruning to minimize the "Frobenius distortion", the Frobenius norm of 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
Lookahead: A Far-sighted Alternative of Magnitude-based Pruning
### Paper Abstract
Magnitude-based pruning is one of the simplest methods for pruning neural networks. De... | |
Hc8M6hGQhAF | ML_Reproducibility_Challenge/2021/Fall | 2021 | Reproducibility Study - SCOUTER: Slot Attention-based Classifier for Explainable Image Recognition | ["Anonymous"] |
Scope of Reproducibility
The experiments presented by Li et al. on SCOUTER were replicated to verify their claims on the properties of the
model. For the task of explainable image recognition, the authors state their main claims: (1) SCOUTER produces state-of-the-art performance values on positive and negative explan... | ["MLRC", "SCOUTER", "Explainable AI", "Image Recognition"] | Reproducibility Study - SCOUTER: Slot Attention-basedClassifier for Explainable Image RecognitionXXX XXXX XXX XXXXXXX XXXX XXX XXXXReproducibility SummaryScope of ReproducibilityThe experiments presented by Li et al. on SCOUTER were replicated to verify their claims on the properties of themodel. For the task of explai... | SclvSu7rb5 | Well written approached report | 8: Top 50% of accepted papers, clear accept | Authors provided very thorough detail on their reproducibility efforts. Their report has been well written and concepts have been well articulated. They managed to reproduce the results of the original papers with a very good precision, albeit with some deviations.
However, they tried to contact the authors to get some... | 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
Reproducibility Study - SCOUTER: Slot Attention-based Classifier for Explainable Image Recognition
### Paper Abstract
Scope of Reproducibility The experiments presente... |
BJg1fgBYwH | ICLR.cc/2020/Conference | 2020 | SAFE-DNN: A Deep Neural Network with Spike Assisted Feature Extraction for Noise Robust Inference | ["Xueyuan She", "Priyabrata Saha", "Daehyun Kim", "Yun Long", "Saibal Mukhopadhyay"] | We present a Deep Neural Network with Spike Assisted Feature Extraction (SAFE-DNN) to improve robustness of classification under stochastic perturbation of inputs. The proposed network augments a DNN with unsupervised learning of low-level features using spiking neuron network (SNN) with Spike-Time-Dependent-Plasticity... | ["Noise robust", "deep learning", "DNN", "image classification"] | ABSTRACTWe present a Deep Neural Network with Spike Assisted Feature Extraction (SAFE-DNN) to improve robustness of classification under stochastic perturbation ofinputs. The proposed network augments a DNN with unsupervised learning of low-level features using spiking neuron network (SNN) with Spike-Time-Dependent-Pla... | rygtIBsQ5S | Official Blind Review #1 | 6: Weak Accept | The paper shows that replacing feature extraction layers by spiking convolution network can improve the performance under random noise. The algorithm itself is simple since it's just a combiniation of STDP and standard CNN. The results shows improved performance under some random noise. Although the idea is cute, I fee... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
SAFE-DNN: A Deep Neural Network with Spike Assisted Feature Extraction for Noise Robust Inference
### Paper Abstract
We present a Deep Neural Network with Spike Assisted... | |
S1lk61BtvB | ICLR.cc/2020/Conference | 2020 | ``"Best-of-Many-Samples" Distribution Matching | ["Apratim Bhattacharyya", "Mario Fritz", "Bernt Schiele"] | Generative Adversarial Networks (GANs) can achieve state-of-the-art sample quality in generative modelling tasks but suffer from the mode collapse problem. Variational Autoencoders (VAE) on the other hand explicitly maximize a reconstruction-based data log-likelihood forcing it to cover all modes, but suffer from poore... | ["Distribution Matching", "Generative Adversarial Networks", "Variational Autoencoders"] | ABSTRACTGenerative Adversarial Networks (GANs) can achieve state-of-the-art samplequality in generative modelling tasks but suffer from the mode collapse prob-lem. Variational Autoencoders (V AE) on the other hand explicitly maximize areconstruction-based data log-likelihood forcing it to cover all modes, but suf-fer f... | r1ejzBAX5r | Official Blind Review #3 | 6: Weak Accept | This paper presents a new objective function for hybrid VAE-GANs. To overcome a number of known issues with VAE-GANs, this work uses multiple samples from the generator network to achieve a high data log-likelihood and low divergence to the latent prior.
In the experimental section, the ``"Best-of-Many-Samples" approac... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
``"Best-of-Many-Samples" Distribution Matching
### Paper Abstract
Generative Adversarial Networks (GANs) can achieve state-of-the-art sample quality in generative modell... | |
hga0T0Qcli5 | ICLR.cc/2021/Conference | 2021 | QuatRE: Relation-Aware Quaternions for Knowledge Graph Embeddings | ["Dai Quoc Nguyen", "Thanh Vu", "Tu Dinh Nguyen", "Dinh Phung"] | We propose an effective embedding model, named QuatRE, to learn quaternion embeddings for entities and relations in knowledge graphs. QuatRE aims to enhance correlations between head and tail entities given a relation within the Quaternion space with Hamilton product. QuatRE achieves this goal by further associating ea... | ["Knowledge graph embeddings", "Quaternion", "Hamilton product"] | QuatRE: Relation-Aware Quaternions for KnowledgeGraph EmbeddingsAbstractWe propose an effective embedding model, named QuatRE, to learn quaternionembeddings for entities and relations in knowledge graphs. QuatRE aims toenhance correlations between head and tail entities given a relation within theQuaternion space with ... | 8R55jU61J5I | KG completion model based on quaternions that requires improved evaluation | 4: Ok but not good enough - rejection | This work proposes a new model for KG completion, based on embedding entities and relations into quaternion space, achieving promising performance on standard KG completion datasets.
Although the proposed model achieves promising results, the comparison to most existing KG completion models is unfair given the differe... | 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
QuatRE: Relation-Aware Quaternions for Knowledge Graph Embeddings
### Paper Abstract
We propose an effective embedding model, named QuatRE, to learn quaternion embedding... |
H1gZsJBYwH | ICLR.cc/2020/Conference | 2020 | Hybrid Weight Representation: A Quantization Method Represented with Ternary and Sparse-Large Weights | ["Jinbae Park", "Sung-Ho Bae"] | Previous ternarizations such as the trained ternary quantization (TTQ), which quantized weights to three values (e.g., {−Wn, 0,+Wp}), achieved the small model size and efficient inference process. However, the extreme limit on the number of quantization steps causes some degradation in accuracy. To solve this problem... | ["quantized neural networks", "centralized quantization", "hybrid weight representation", "weighted ridge", "ternary weight"] | ABSTRACTPrevious ternarizations such as the trained ternary quantization (TTQ), whichquantized weights to three values (e.g., fWn, 0,+Wpg), achieved the smallmodel size and efficient inference process. However, the extreme limit on thenumber of quantization steps causes some degradation in accuracy. To solve thisproble... | SJlv2o_Jqr | Official Blind Review #2 | 6: Weak Accept | This paper is about quantization, and how to represent values as the finite number of states in a low bit width, using discretization. Particularly, they propose an approach to tackle the problems associated with previous ternarization which quantize weights to three values. Their approach is a hybrid weight representa... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Hybrid Weight Representation: A Quantization Method Represented with Ternary and Sparse-Large Weights
### Paper Abstract
Previous ternarizations such as the trained tern... | |
H1gB4RVKvB | ICLR.cc/2020/Conference | 2020 | Recurrent neural circuits for contour detection | ["Drew Linsley*", "Junkyung Kim*", "Alekh Ashok", "Thomas Serre"] | We introduce a deep recurrent neural network architecture that approximates visual cortical circuits (Mély et al., 2018). We show that this architecture, which we refer to as the 𝜸-net, learns to solve contour detection tasks with better sample efficiency than state-of-the-art feedforward networks, while also exhibiti... | ["Contextual illusions", "visual cortex", "recurrent feedback", "neural circuits"] | ABSTRACTWe introduce a deep recurrent neural network architecture that approximates knownvisual cortical circuits (Mély et al., 2018). We show that this architecture, whichwe refer to as the -Net, learns to solve contour detection tasks with better sampleefficiency than state-of-the-art feedforward networks, while als... | r1ee0w3k9S | Official Blind Review #4 | 6: Weak Accept | The authors propose a new artificial neural network architecture that is derived from a human visual model (Mély et al., 2018). The original (human vision) model can explain some of the human visual illusions, specifically contextual ones. While the adaptation from this human visual model to artificial neural networks ... | <|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 neural circuits for contour detection
### Paper Abstract
We introduce a deep recurrent neural network architecture that approximates visual cortical circuits (... | |
BJlFzkSFwr | ICLR.cc/2020/Conference | 2020 | Joint text classification on multiple levels with multiple labels | ["Miruna P\u00eeslar", "Marek Rei"] | Natural language uses words in an associative way to construct sentences: it is not words in isolation, but the appropriate use of hierarchical structures that makes communication successful. We propose a deep learning framework for explicitly tying together the representations between single words and full sentences, ... | ["multi-head attention", "zero-shot learning", "multi-task learning", "text classification", "sequence labeling"] | ABSTRACTNatural language uses words in an associative way to construct sentences: it is notwords in isolation, but the appropriate use of hierarchical structures that makescommunication successful. We propose a deep learning framework for explicitlytying together the representations between single words and full senten... | HygEMj8CKB | Official Blind Review #3 | 1: Reject | This paper addresses the problem of text classification by incorporating classification at both token and sentence levels, with the aim of better learning composition. The paper finds that introducing losses at multiple levels, along with additional auxiliary objectives, generally improves performance over simpler base... | <|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 text classification on multiple levels with multiple labels
### Paper Abstract
Natural language uses words in an associative way to construct sentences: it is not ... | |
ryV9NFyvM | ICLR.cc/2018/Workshop | 2018 | Exploring Single-Cell Data with Deep Multitasking Neural Networks | ["Matthew Amodio", "Krishnan Srinivasan", "David van Dijk", "Hussein Mohsen", "Kristina Yim", "Rebecca Muhle", "Kevin R. Moon", "Susan Kaech", "Ryan Sowell", "Ruth Montgomery", "James Noonan", "Guy Wolf", "Smita Krishnaswamy"] | Handling the vast amounts of single-cell RNA-sequencing and CyTOF data, which are now being generated in patient cohorts, presents a computational challenge due to the noise, complexity, sparsity and batch effects present. Here, we propose a unified deep neural network-based approach to automatically process and extrac... | ["autoencoder", "clustering", "unsupervised", "visualization", "computational biology", "multitask"] | ABSTRACTHandling the vast amounts of single-cell RNA-sequencing and CyTOF data,which are now being generated in patient cohorts, presents a computational chal-lenge due to the noise, complexity, sparsity and batch effects present. Here, wepropose a unified deep neural network-based approach to automatically processand ... | rJaLr_Vdz | Unclear presentation. | 3: Clear rejection | The authors propose a deep auto-encoder with "information dimension" regularisation on the bottleneck layer and an additional "maximal mean discrepancy" regularisation.
It is unfortunately not clear from the paper how exactly the proposed technique works. The authors are telling what is done, but not how it is done. I... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Exploring Single-Cell Data with Deep Multitasking Neural Networks
### Paper Abstract
Handling the vast amounts of single-cell RNA-sequencing and CyTOF data, which are no... |
7Pkzm2FgUmq | robot-learning.org/CoRL/2023/Conference | 2023 | SLAP: Spatial-Language Attention Policies | ["Priyam Parashar", "Vidhi Jain", "Xiaohan Zhang", "Jay Vakil", "Sam Powers", "Yonatan Bisk", "Chris Paxton"] | Despite great strides in language-guided manipulation, existing work has been constrained to table-top settings. Table-tops allow for perfect and consistent camera angles, properties are that do not hold in mobile manipulation. Task plans that involve moving around the environment must be robust to egocentric views and... | ["learning from demonstration", "language-based robotics"] | SLAP: Spatial-Language Attention PoliciesPriyam Parashar1Vidhi Jain2Xiaohan Zhang1,3Jay Vakil1Sam Powers1,2Yonatan Bisk1,2Chris Paxton11Meta2Carnegie Mellon3SUNY BinghamtonAbstract: Despite great strides in language-guided manipulation, existing workhas been constrained to table-top settings. Table-tops allow for perfe... | 5CDrMrgSao | Strengths:
- The proposed hierarchical design is reasonable. Empirically, it demonstrates to outperform the original PerAct in 11/16 testing tasks (if counting each instruction with seen and unseen objects to be two tasks).
- The proposed method is clearly motivated and described in the paper. The technical part of ... | 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
SLAP: Spatial-Language Attention Policies
### Paper Abstract
Despite great strides in language-guided manipulation, existing work has been constrained to table-top setti... | ||
hR1D7pliQ2a | MICCAI.org/2022/Challenge/FLARE | 2022 | Semi-supervised Context-aware Multi-Organ Segmentation | ["jiacheng wang", "Yongkang Ma"] | The automatic segmentation of organs at risk is extremely dedicated to the clinical assistance that can significantly reduce the clinical resource cost. However, training such a good enough model usually requires a large amount of labeled data, or the model performance is
likely to meet heavy drop. Semi-supervised trai... | ["segmentation", "data", "automatic segmentation", "organs", "risk", "clinical assistance", "clinical resource cost", "good enough model", "large amount", "model performance"] | Semi-supervised Context-aware Multi-OrganSegmentationJiacheng Wang1and Yongkang Ma1Manteia Technologies Co.,Ltd, Xiamen, China{wangjiacheng, yongkangma}@manteiatech.comAbstract. The automatic segmentation of organs at risk is extremelydedicated to the clinical assistance that can significantly reduce the clin-ical reso... | veTBqAosVlU | Coarse-to-fine U-Net for multi-organ segmentation | 6: Marginally above acceptance threshold | 1. This work proposes a pseudo method for semi-supervised multi-organ segmentation. It first uses a 3D U-Net for coarse segmentation and then uses the same structure for refining the cropped prediction. The proposed method can use unlabeled data to achieve better performance.
2. The author describes the background 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
Semi-supervised Context-aware Multi-Organ Segmentation
### Paper Abstract
The automatic segmentation of organs at risk is extremely dedicated to the clinical assistance ... |
m08OHhXxl-5 | ICLR.cc/2021/Conference | 2021 | Privacy Preserving Recalibration under Domain Shift | ["Rachel Luo", "Shengjia Zhao", "Jiaming Song", "Jonathan Kuck", "Stefano Ermon", "Silvio Savarese"] | Classifiers deployed in high-stakes applications must output calibrated confidence scores, i.e. their predicted probabilities should reflect empirical frequencies. Typically this is achieved with recalibration algorithms that adjust probability estimates based on the real-world data; however, existing algorithms are no... | ["uncertainty calibration", "differential privacy"] | ABSTRACTClassifiers deployed in high-stakes applications must output calibrated confidencescores, i.e. their predicted probabilities should reflect empirical frequencies. Typi-cally this is achieved with recalibration algorithms that adjust probability estimatesbased on real-world data; however, existing algorithms are... | TgGh-KYSNi7 | This paper studies the problem of recalibrating a classifier under the presence of domain shift and the constraints of differential privacy. They show how to adapt several algorithms for dealing with domain shift to the paradigm of differential privacy, giving mechanisms that achieve both goals. | 6: Marginally above acceptance threshold | Summary:
This paper studies the problem of recalibrating a classifier under the presence of domain shift and the constraints of differential privacy. They show how to adapt several algorithms for dealing with domain shift to the paradigm of differential privacy, giving mechanisms that achieve both goals.
Strong Poin... | 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
Privacy Preserving Recalibration under Domain Shift
### Paper Abstract
Classifiers deployed in high-stakes applications must output calibrated confidence scores, i.e. th... |
rked_6NFwH | ICLR.cc/2020/Conference | 2020 | Path Space for Recurrent Neural Networks with ReLU Activations | ["Yue Wang", "Qi Meng", "Wei Chen", "Yuting Liu", "Zhi-Ming Ma", "Tie-Yan Liu"] | It is well known that neural networks with rectified linear units (ReLU) activation functions are positively scale-invariant (i.e., the neural network is invariant to positive rescaling of weights). Optimization algorithms like stochastic gradient descent that optimize the neural networks in the vector space of weights... | ["optimization", "neural network", "positively scale-invariant", "path space", "deep learning", "RNN"] | ABSTRACTIt is well known that neural networks with rectified linear units (ReLU) activationfunctions are positively scale-invariant (i.e., the neural network is invariant topositive rescaling of weights). Optimization algorithms like stochastic gradientdescent that optimize the neural networks in the vector space of we... | BJgnP_s15B | Official Blind Review #1 | 3: Weak Reject | Motivation: The authors motivate their work with the observation that neural networks with ReLU activations are positively scale invariant. Recent works have proposed parameter space called path space to leverage this insight for feedforward as well as CNNs, but this is not really the case for Recurrent Neural Networks... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Path Space for Recurrent Neural Networks with ReLU Activations
### Paper Abstract
It is well known that neural networks with rectified linear units (ReLU) activation fun... | |
BJlOcR4KwS | ICLR.cc/2020/Conference | 2020 | Channel Equilibrium Networks | ["Wenqi Shao", "Shitao Tang", "Xingang Pan", "Ping Tan", "Xiaogang Wang", "Ping Luo"] | Convolutional Neural Networks (CNNs) typically treat normalization methods such as batch normalization (BN) and rectified linear function (ReLU) as building blocks. Previous work showed that this basic block would lead to channel-level sparsity (i.e. channel of zero values), reducing computational complexity of CNNs. H... | ["Deep learning", "convolutional neural networks", "building block design"] | ABSTRACTConvolutional Neural Networks (CNNs) typically treat normalization methodssuch as batch normalization (BN) and rectified linear function-like activations ( e.g.ReLU) as building blocks. Previous work pointed out that learning feature chan-nels with equal magnitudes is important for a CNN to achieve good general... | HygbkYmJ9r | Official Blind Review #1 | 6: Weak Accept | This paper proposes a new block that gives diminishing
Pros)
(+) This paper is well-written and the idea looks interesting.
(+) The provided experiments look quite extensive. I like the experiments with both heavy and light network architectures to show the effectiveness of the proposed method.
(+) The authors provid... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Channel Equilibrium Networks
### Paper Abstract
Convolutional Neural Networks (CNNs) typically treat normalization methods such as batch normalization (BN) and rectified... | |
Byx91R4twB | ICLR.cc/2020/Conference | 2020 | Adversarial Video Generation on Complex Datasets | ["Aidan Clark", "Jeff Donahue", "Karen Simonyan"] | Generative models of natural images have progressed towards high fidelity samples by the strong leveraging of scale. We attempt to carry this success to the field of video modeling by showing that large Generative Adversarial Networks trained on the complex Kinetics-600 dataset are able to produce video samples of subs... | ["GAN", "generative model", "generative adversarial network", "video prediction"] | ABSTRACTGenerative models of natural images have progressed towards high fidelity samplesby the strong leveraging of scale. We attempt to carry this success to the field ofvideo modeling by showing that large Generative Adversarial Networks trained onthe complex Kinetics-600 dataset are able to produce video samples of... | HkeZSl0aKS | Official Blind Review #3 | 3: Weak Reject | Summary:
This paper tackles the problem of efficient video generation. The authors present a Dual-Video-Discriminator Generative Adversarial Network (DVD-GAN) composed of an image-level spatial discriminator and video-level temporal discriminator. DVD-GAN achieves state-of-the-art results when benchmarked against the F... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adversarial Video Generation on Complex Datasets
### Paper Abstract
Generative models of natural images have progressed towards high fidelity samples by the strong lever... | |
Kz42iQirPJI | ICLR.cc/2021/Conference | 2021 | Towards Learning to Remember in Meta Learning of Sequential Domains | ["Zhenyi Wang", "Tiehang Duan", "Donglin Zhan", "Changyou Chen"] | Meta-learning has made rapid progress in past years, with recent extensions made to avoid catastrophic forgetting in the learning process, namely continual meta learning. It is desirable to generalize the meta learner’s ability to continuously learn in sequential domains, which is largely unexplored to-date. We found t... | ["Meta learning", "Continual Learning", "Sequential Domain Learning"] | ABSTRACTMeta-learning has made rapid progress in past years, with recent extensionsmade to avoid catastrophic forgetting in the learning process, namely contin-ual meta learning. It is desirable to generalize the meta learner’s ability tocontinuously learn in sequential domains , which is largely unexplored to-date.We ... | PmeRfM3lkOx | Needs some clarity | 5: Marginally below acceptance threshold | This work focuses on sequential adaptation of a model without forgetting. Their goal is to minimize the catastrophic interference of the model while learning a new few-shot task coming from a different domain. To that end, they introduce a problem setup where the model receives a sequence of few-shot tasks from differe... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Learning to Remember in Meta Learning of Sequential Domains
### Paper Abstract
Meta-learning has made rapid progress in past years, with recent extensions made t... |
cI8WHuiW3x | MICCAI.org/2022/Challenge/FLARE | 2022 | A Simple Self-labeling Method for Semi-supervised Medical Image Segmentation | ["Ye Zhu"] | Leveraging a few labeled images and a large number of unlabeled images is crucial for medical image segmentation since labeling the medical data can be very expensive and time-consumed. Therefore we introduce a naïve but simple method to utilize the massive unlabeled
medical images for better training. We first use all... | ["Semi-supervise learning Pseudo labels"] | A Simple Self-labeling Method forSemi-supervised Medical Image SegmentationYe Zhu1and Hanlin Tian1The Chinese University of Hong Kong (Shenzhen), Chinazhuye1@cuhk.edu.cnAbstract. Leveraging a few labeled images and a large number of un-labeled images is crucial for medical image segmentation since labelingthe medical d... | I_O4I_b50F1 | This paper has sufficient sections, these beginning sections are good but the result section does not have enough detailed analysis | 4: Ok but not good enough - rejection | Authors adopt Swin U-Net as 2D segmentation problem, where they first train on labeled data, then use that trained model to generate pseudo-labels, then retrain on all of them.
This paper has sufficient sections, those beginning sections are good whereas the result section does not have enough detailed analysis; moreo... | 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 Simple Self-labeling Method for Semi-supervised Medical Image Segmentation
### Paper Abstract
Leveraging a few labeled images and a large number of unlabeled images is... |
HyljzgHtwS | ICLR.cc/2020/Conference | 2020 | Regularly varying representation for sentence embedding | ["Hamid Jalalzai", "Pierre Colombo", "Chlo\u00e9 Clavel", "Eric Gaussier", "Giovanna Varni", "Emmanuel Vignon", "Anne Sabourin"] | The dominant approaches to sentence representation in natural language rely on learning embeddings on massive corpuses. The obtained embeddings have desirable properties such as compositionality and distance preservation (sentences with similar meanings have similar representations). In this paper, we develop a novel m... | ["extreme value theory", "classification", "supvervised learning", "data augmentation", "representation learning"] | ABSTRACTThe dominant approaches to sentence representation in natural language relyon learning embeddings on massive corpuses. The obtained embeddings havedesirable properties such as compositionality and distance preservation (sentenceswith similar meanings have similar representations). In this paper, we develop anov... | S1evAPSTtS | Official Blind Review #1 | 1: Reject | The paper explores learning dilation-invariant sentence representations, with a goal of improving downstream task performance on rare events. A pre-trained embedding is encoded as a latent variable Z, which is constrained to be multi-variate heavy tailed. Separate classifiers are trained on the head and tail of the dis... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Regularly varying representation for sentence embedding
### Paper Abstract
The dominant approaches to sentence representation in natural language rely on learning embedd... | |
S1ANxQW0b | ICLR.cc/2018/Conference | 2018 | Maximum a Posteriori Policy Optimisation | ["Abbas Abdolmaleki", "Jost Tobias Springenberg", "Yuval Tassa", "Remi Munos", "Nicolas Heess", "Martin Riedmiller"] | We introduce a new algorithm for reinforcement learning called Maximum a-posteriori Policy Optimisation (MPO) based on coordinate ascent on a relative-entropy objective. We show that several existing methods can directly be related to our derivation. We develop two off-policy algorithms and demonstrate that they are co... | ["Reinforcement Learning", "Variational Inference", "Control"] | ABSTRACTWe introduce a new algorithm for reinforcement learning called Maximum a-posteriori Policy Optimisation (MPO) based on coordinate ascent on a relative-entropy objective. We show that several existing methods can directly be relatedto our derivation. We develop two off-policy algorithms and demonstrate thatthey ... | rypb6tngM | The paper presents an interesting new algorithm for deep reinforcement learning which outperforms state of the art methods. | 7: Good paper, accept | The paper presents a new algorithm for inference-based reinforcement learning for deep RL. The algorithm decomposes the policy update in two steps, an E and an M-step. In the E-step, the algorithm estimates a variational distribution q which is subsequentially used for the M-step to obtain a new policy. Two versions 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
Maximum a Posteriori Policy Optimisation
### Paper Abstract
We introduce a new algorithm for reinforcement learning called Maximum a-posteriori Policy Optimisation (MPO)... |
SJ-uGHcee | ICLR.cc/2017/conference | 2017 | Efficient iterative policy optimization | ["Nicolas Le Roux"] | We tackle the issue of finding a good policy when the number of policy updates is limited. This is done by approximating the expected policy reward as a sequence of concave lower bounds which can be efficiently maximized, drastically reducing the number of policy updates required to achieve good performance. We also ex... | ["number", "policy updates", "issue", "good policy", "limited", "expected policy reward", "sequence", "concave lower bounds", "good performance"] | ABSTRACTWe tackle the issue of finding a good policy when the number of policy updates islimited. This is done by approximating the expected policy reward as a sequenceof concave lower bounds which can be efficiently maximized, drastically reducingthe number of policy updates required to achieve good performance. We al... | rJ6N1xMVl | Review | 7: Good paper, accept | The paper considers the problem of reinforcement learning where the number of policy updates is required to be low. The problem is well motivated and the author provides an interesting modification to the PoWER algorithm, along with variational bounds on the value function (lemmas 3,4) which are interesting in themselv... | 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
Efficient iterative policy optimization
### Paper Abstract
We tackle the issue of finding a good policy when the number of policy updates is limited. This is done by app... |
C0qJUx5dxFb | ICLR.cc/2021/Conference | 2021 | Neural networks with late-phase weights | ["Johannes Von Oswald", "Seijin Kobayashi", "Joao Sacramento", "Alexander Meulemans", "Christian Henning", "Benjamin F Grewe"] | The largely successful method of training neural networks is to learn their weights using some variant of stochastic gradient descent (SGD). Here, we show that the solutions found by SGD can be further improved by ensembling a subset of the weights in late stages of learning. At the end of learning, we obtain back a si... | ["weights", "neural networks", "sgd", "learning", "weights neural networks", "successful", "variant", "stochastic gradient descent", "solutions", "subset"] | ABSTRACTThe largely successful method of training neural networks is to learn their weightsusing some variant of stochastic gradient descent (SGD). Here, we show that thesolutions found by SGD can be further improved by ensembling a subset of theweights in late stages of learning. At the end of learning, we obtain back... | 2G9emdue6Pa | Official review | 7: Good paper, accept |
### Summary
The authors propose late-phase weights, a method of updating the weights near the end of training via a splitting and ensembling mechanism. They analyze the benefits in the noisy quadratic setting. The method improves validation performance on a range of image recognition tasks and on enwiki8.
### Comment... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural networks with late-phase weights
### Paper Abstract
The largely successful method of training neural networks is to learn their weights using some variant of stoc... |
SJeFNlHtPS | ICLR.cc/2020/Conference | 2020 | Hidden incentives for self-induced distributional shift | ["David Scott Krueger", "Tegan Maharaj", "Shane Legg", "Jan Leike"] | Decisions made by machine learning systems have increasing influence on the world. Yet it is common for machine learning algorithms to assume that no such influence exists. An example is the use of the i.i.d. assumption in online learning for applications such as content recommendation, where the (choice of) content di... | ["distributional shift", "safety", "incentives", "specification", "content recommendation", "reinforcement learning", "online learning", "ethics"] | ABSTRACTDecisions made by machine learning systems have increasing influence on theworld. Yet it is common for machine learning algorithms to assume that no suchinfluence exists. An example is the use of the i.i.d. assumption in online learningfor applications such as content recommendation, where the (choice of) conte... | rJxhKoSaFB | Official Blind Review #2 | 6: Weak Accept | The authors study the phenomena of self-introduced distributional shift. They define the term along with the term hidden incentives for distributional shift. The latter describes factors that motivate the learner to change the distribution in order to achieve a higher performance. The authors study both phenomena in tw... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Hidden incentives for self-induced distributional shift
### Paper Abstract
Decisions made by machine learning systems have increasing influence on the world. Yet it is c... | |
L4v_5Qtshj7 | ICLR.cc/2021/Conference | 2021 | Goal-Driven Imitation Learning from Observation by Inferring Goal Proximity | ["Andrew Szot", "Youngwoon Lee", "Shao-Hua Sun", "Joseph J Lim"] | Humans can effectively learn to estimate how close they are to completing a desired task simply by watching others fulfill the task. To solve the task, they can then take actions towards states with higher estimated proximity to the goal. From this intuition, we propose a simple yet effective method for imitation learn... | ["Imitation Learning", "Learning from Observation"] | ABSTRACTHumans can effectively learn to estimate how close they are to completing a desiredtask simply by watching others fulfill the task. To solve the task, they can thentake actions towards states with higher estimated proximity to the goal. From thisintuition, we propose a simple yet effective method for imitation ... | IsW3Rh015v9 | Insufficient evaluation for a paper that comes with no theoratical insights | 5: Marginally below acceptance threshold | Summary
--------
The paper proposes a method for imitation learning for goal-directed tasks that uses a learned proximity function for computing rewards.
An ensemble of proximity functions is trained in supervised way to predict the time step (rescaled to the range $[0,1]$) for the expert's states. The ensemble is impr... | 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
Goal-Driven Imitation Learning from Observation by Inferring Goal Proximity
### Paper Abstract
Humans can effectively learn to estimate how close they are to completing ... |
guetrIHLFGI | ICLR.cc/2021/Conference | 2021 | The Deep Bootstrap Framework: Good Online Learners are Good Offline Generalizers | ["Preetum Nakkiran", "Behnam Neyshabur", "Hanie Sedghi"] | We propose a new framework for reasoning about generalization in deep learning.
The core idea is to couple the Real World, where optimizers take stochastic gradient steps on the empirical loss, to an Ideal World, where optimizers take steps on the population loss. This leads to an alternate decomposition of test error... | ["generalization", "optimization", "online learning", "understanding deep learning", "empirical investigation"] | ABSTRACTWe propose a new framework for reasoning about generalization in deep learning.The core idea is to couple the Real World, where optimizers take stochastic gradi-ent steps on the empirical loss, to an Ideal World, where optimizers take steps onthe population loss. This leads to an alternate decomposition of test... | xJ21Na-pLTH | Interesting framework but there appear to be some caveats | 6: Marginally above acceptance threshold | ######################################################################
1. Paper Summary
The authors propose a bootstrap framework for understanding generalization in deep learning. In particular, instead of the usual decomposition of test error as training error plus the generalization gap, the bootstrap framework... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The Deep Bootstrap Framework: Good Online Learners are Good Offline Generalizers
### Paper Abstract
We propose a new framework for reasoning about generalization in deep... |
SyeZIkrKwS | ICLR.cc/2020/Conference | 2020 | DyNet: Dynamic Convolution for Accelerating Convolution Neural Networks | ["Kane Zhang", "Jian Zhang", "Qiang Wang", "Zhao Zhong"] | Convolution operator is the core of convolutional neural networks (CNNs) and occupies the most computation cost. To make CNNs more efficient, many methods have been proposed to either design lightweight networks or compress models. Although some efficient network structures have been proposed, such as MobileNet or Shuf... | ["CNNs", "dynamic convolution kernel"] | ABSTRACTConvolution operator is the core of convolutional neural networks (CNNs) andoccupies the most computation cost. To make CNNs more efficient, many methodshave been proposed to either design lightweight networks or compress models.Although some efficient network structures have been proposed, such as MobileNetor ... | B1lC6hMCYS | Official Blind Review #1 | 6: Weak Accept | === Summary ===
The authors propose to use dynamic convolutional kernels as a means to reduce the computation cost in static CNNs while maintaining their performance. The dynamic kernels are obtained by a linear combination of static kernels where the weights of the linear combination are input-dependent (they are obta... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
DyNet: Dynamic Convolution for Accelerating Convolution Neural Networks
### Paper Abstract
Convolution operator is the core of convolutional neural networks (CNNs) and o... | |
YiR1NIojU2q | icaps-conference.org/ICAPS/2021/Workshop/XAIP | 2021 | Counterfactual Explanations for Optimization-Based Decisions in the Context of the GDPR | ["Anton Korikov", "Alexander Shleyfman", "Chris Beck"] | The General Data Protection Regulations (GDPR) entitle individuals to explanations for automated decisions. The form, comprehensibility, and even existence of such explanations remain open problems, investigated as part of explainable AI. We adopt the approach of counterfactual explanations and apply it to decisions ma... | ["Explainability", "Scheduling", "Constraint Optimization"] | Counterfactual Explanations for Optimization-Based Decisionsin the Context of the GDPRAnton Korikov ,Alexander Shleyfman and J. Christopher BeckDepartment of Mechanical & Industrial Engineering, University of Toronto, Toronto, Canadafkorikov, jcbg@mie.utoronto.ca, shleyfman.alexander@gmail.comAbstractThe General Data P... | 919pQ1Dq9n | A Preliminary Work but Is Definitely of Interest to the Workshop | 7: Good paper, accept | The paper looks at the problem of generating counterfactual explanations for decisions generated from optimization problems. The paper argues that one could view such explanations through the lens of inverse optimization (or through an even more general framework of NCE). While the general problem of inverse optimizati... | 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
Counterfactual Explanations for Optimization-Based Decisions in the Context of the GDPR
### Paper Abstract
The General Data Protection Regulations (GDPR) entitle individ... |
NND5KhQGYct | EMNLP/2020/Workshop/NLP-COVID | 2020 | Interactive Visualization and Simplified Pattern Discovery in the COVID-19 Open Research Dataset(CORD-19) | ["Wuraola Fisayo Oyewusi"] | This work explores the use of Natural Language Processing based algorithms for Large Text Mining and Interactive Visualization for the COVID-19 Open Research Dataset (CORD-19) Dataset. We developed a series of easy to use online interactive text visualization based on different percentages of mined text data of disease... | ["scattertext", "scispaCy", "Text Visualization", "NER", "NLP", "CORD-19"] | 1000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044045046047048049050051052053054055056057058059060061062063064065066067068069070071072073074075076077078079080081082083084085086087088089090091092093094095096097098099ACL 2020 Submission... | lyJXxUgHFzn | Work-in-progress, not ready for publication | 3: Clear rejection | I found this article reports work-in-progress and I think it is still not ready for publication. Even though a visualization tool could certainly help medical professionals, no URL is provided to test the system (only screenshots). The author does not discuss its use case (beyond mentioning "the study of patterns") nor... | 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
Interactive Visualization and Simplified Pattern Discovery in the COVID-19 Open Research Dataset(CORD-19)
### Paper Abstract
This work explores the use of Natural Langu... |
B1lCn64tvS | ICLR.cc/2020/Conference | 2020 | Improving SAT Solver Heuristics with Graph Networks and Reinforcement Learning | ["Vitaly Kurin", "Saad Godil", "Shimon Whiteson", "Bryan Catanzaro"] | We present GQSAT, a branching heuristic in a Boolean SAT solver trained with value-based reinforcement learning (RL) using Graph Neural Networks for function approximation. Solvers using GQSAT are complete SAT solvers that either provide a satisfying assignment or a proof of unsatisfiability, which is required for many... | ["SAT", "reinforcement learning", "graph neural networks", "heuristics", "DQN", "boolean satisfiability"] | ABSTRACTWe present GQSAT, a branching heuristic in a Boolean SAT solver trained withvalue-based reinforcement learning (RL) using Graph Neural Networks for func-tion approximation. Solvers using GQSAT are complete SAT solvers that eitherprovide a satisfying assignment or a proof of unsatisfiability, which is required f... | BkeMdpNkcr | Official Blind Review #2 | 3: Weak Reject | The paper proposes learning a branching heuristic to be used inside the SAT solver MiniSat using reinforcement learning. The state is represented as a graph representation of the Boolean formula as in previous works, and the policy is parameterized as a graph neural network. At each step of an episode the policy select... | <|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 SAT Solver Heuristics with Graph Networks and Reinforcement Learning
### Paper Abstract
We present GQSAT, a branching heuristic in a Boolean SAT solver trained... | |
Bke02gHYwB | ICLR.cc/2020/Conference | 2020 | Learn Interpretable Word Embeddings Efficiently with von Mises-Fisher Distribution | ["Minghong Yao", "Liansheng Zhuang", "Houqiang Li", "Jian Yang", "Shafei Wang"] | Word embedding plays a key role in various tasks of natural language processing. However, the dominant word embedding models don't explain what information is carried with the resulting embeddings. To generate interpretable word embeddings we intend to replace the word vector with a probability density distribution. Th... | ["word embedding", "natural language processing"] | ABSTRACTWord embedding plays a key role in various tasks of natural language processing. 1However, the dominant word embedding model don’t explain what information 2is carried with the resulting embeddings. To generate interpretable word embed- 3dings we intend to replace the word vector with a probability density dist... | Byg0bMQ5YS | Official Blind Review #3 | 1: Reject | Summary: This paper proposed a variational word embedding method, by using vMF distribution as prior and adding an entropy regularization term.
Strengths:
[+] In the experiment parts, the authors describe the experiment settings in detail.
[+] Detailed descriptions for each equation is given.
Weaknesses:
[-] Template... | <|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 Interpretable Word Embeddings Efficiently with von Mises-Fisher Distribution
### Paper Abstract
Word embedding plays a key role in various tasks of natural languag... | |
ShMlIzKgOW9 | aclweb.org/ACL/2022/Workshop/CSRR | 2022 | Knowledge-Augmented Language Models for Cause-Effect Relation Classification | ["Pedram Hosseini", "David A. Broniatowski", "Mona T. Diab"] | Previous studies have shown the efficacy of knowledge augmentation methods in pretrained language models. However, these methods behave differently across domains and downstream tasks. In this work, we investigate the augmentation of pretrained language models with knowledge graph data in the cause-effect relation clas... | ["commonsense causal reasoning", "cause-effect relation classification", "pretrained language models"] | Knowledge-Augmented Language Models for Cause-Effect RelationClassificationPedram Hosseini1David A. Broniatowski1Mona Diab1,21The George Washington University2Meta AIphosseini@gwu.eduAbstractPrevious studies have shown the efficacy ofknowledge augmentation methods in pretrainedlanguage models. However, these methodsbeh... | B6PeGTiJuGc | Unique limited contribution of cause-effect models created by knowledge-augmented LM pretraining | 5: Marginally below acceptance threshold | This paper proposes a method for knowledge-augmented LM pretraining with cause-effect information. The method is targetted towards causal reasoning benchmarks (TCR, COPA, and BCOPA-CE). The method performs better than vanilla and existing system baselines on TCR, and below some baselines on the COPA/BCOPA-CE tasks.
Th... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Knowledge-Augmented Language Models for Cause-Effect Relation Classification
### Paper Abstract
Previous studies have shown the efficacy of knowledge augmentation method... |
pzpytjk3Xb2 | ICLR.cc/2021/Conference | 2021 | Policy-Driven Attack: Learning to Query for Hard-label Black-box Adversarial Examples | ["Ziang Yan", "Yiwen Guo", "Jian Liang", "Changshui Zhang"] | To craft black-box adversarial examples, adversaries need to query the victim model and take proper advantage of its feedback. Existing black-box attacks generally suffer from high query complexity, especially when only the top-1 decision (i.e., the hard-label prediction) of the victim model is available. In this pape... | ["hard-label attack", "black-box attack", "adversarial attack", "reinforcement learning"] | ABSTRACTTo craft black-box adversarial examples, adversaries need to query the victimmodel and take proper advantage of its feedback. Existing black-box attacks gen-erally suffer from high query complexity, especially when only the top-1 decision(i.e., the hard-label prediction) of the victim model is available. In thi... | iCjCYRo9xeI | Official Blind Review #2 | 7: Good paper, accept | This paper proposes a new hard-label black-box adversarial attack method based on reinforcement learning. The general idea is to improve the adversarial noise compressing efficiency taking advantage of past queries and the policy network. Experiments are conducted on MNIST, CIFAR-10 and ImageNet and achieved superior p... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Policy-Driven Attack: Learning to Query for Hard-label Black-box Adversarial Examples
### Paper Abstract
To craft black-box adversarial examples, adversaries need to que... |
r1eiqi09K7 | ICLR.cc/2019/Conference | 2019 | Riemannian Adaptive Optimization Methods | ["Gary Becigneul", "Octavian-Eugen Ganea"] | Several first order stochastic optimization methods commonly used in the Euclidean domain such as stochastic gradient descent (SGD), accelerated gradient descent or variance reduced methods have already been adapted to certain Riemannian settings. However, some of the most popular of these optimization tools - namely A... | ["Riemannian optimization", "adaptive", "hyperbolic", "curvature", "manifold", "adam", "amsgrad", "adagrad", "rsgd", "convergence"] | ABSTRACTSeveral first order stochastic optimization methods commonly used in the Euclideandomain such as stochastic gradient descent ( SGD), accelerated gradient descentor variance reduced methods have already been adapted to certain Riemanniansettings. However, some of the most popular of these optimization tools name... | SylzCu-S3Q | Riemannian ADAM | 7: Good paper, accept | I have enjoyed reading this paper. The paper is accessible in most cases and provides a novel optimization technique. Having said this, I have a few concerns here,
- I am not sure why the notion of product manifolds is required in developing the technique. To me, all the arguments follow without that. Even if the aut... | 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
Riemannian Adaptive Optimization Methods
### Paper Abstract
Several first order stochastic optimization methods commonly used in the Euclidean domain such as stochastic ... |
BJxwPJHFwS | ICLR.cc/2020/Conference | 2020 | Robustness Verification for Transformers | ["Zhouxing Shi", "Huan Zhang", "Kai-Wei Chang", "Minlie Huang", "Cho-Jui Hsieh"] | Robustness verification that aims to formally certify the prediction behavior of neural networks has become an important tool for understanding model behavior and obtaining safety guarantees. However, previous methods can usually only handle neural networks with relatively simple architectures. In this paper, we consid... | ["Robustness", "Verification", "Transformers"] | ABSTRACTRobustness verification that aims to formally certify the prediction behavior ofneural networks has become an important tool for understanding model behav-ior and obtaining safety guarantees. However, previous methods can usually onlyhandle neural networks with relatively simple architectures. In this paper, we... | S1elFelaKS | Official Blind Review #2 | 3: Weak Reject | Overview:
This paper is dedicated to developing robustness verification techniques. They claim their methods can deal with verification problems for Transformers which includes cross-nonlinearity and cross-position dependency. The paper solves these key challenges which are not traceable for previous methods. Moreover... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robustness Verification for Transformers
### Paper Abstract
Robustness verification that aims to formally certify the prediction behavior of neural networks has become a... | |
q-cnWaaoUTH | ICLR.cc/2021/Conference | 2021 | Conformation-Guided Molecular Representation with Hamiltonian Neural Networks | ["Ziyao Li", "Shuwen Yang", "Guojie Song", "Lingsheng Cai"] | Well-designed molecular representations (fingerprints) are vital to combine medical chemistry and deep learning. Whereas incorporating 3D geometry of molecules (i.e. conformations) in their representations seems beneficial, current 3D algorithms are still in infancy. In this paper, we propose a novel molecular represen... | ["Molecular Representation", "Neural Physics Engines", "Molecular Dynamics", "Graph Neural Networks"] | ABSTRACTWell-designed molecular representations (fingerprints) are vital to combine med-ical chemistry and deep learning. Whereas incorporating 3D geometry ofmolecules (i.e. conformations) in their representations seems beneficial, current3D algorithms are still in infancy. In this paper, we propose a novel molecularre... | H1-zhj8otJi | Interesting approach for incorporate 3d inforamtion in molecular fingerprints, but lacking results | 5: Marginally below acceptance threshold | This paper proposes to use 3d conformations for learning molecular fingerprints by 1) training a generative model to predict the 3d coordinates and 2) use those to train a "fingerprint generator" to obtain fingerprints by learning to predict molecular properties. The paper is well structured and clearly written. The 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
Conformation-Guided Molecular Representation with Hamiltonian Neural Networks
### Paper Abstract
Well-designed molecular representations (fingerprints) are vital to comb... |
VvRbhkiAwR | aclweb.org/ACL/2020/Workshop/NLP-COVID | 2020 | Cross-language sentiment analysis of European Twitter messages during the COVID-19 pandemic | ["Anna Kruspe", "Matthias Haeberle", "Iona Kuhn", "Xiao Xiang Zhu"] | Social media data can be a very salient source of information during crises. User-generated messages provide a window into people’s minds during such times, allowing us insights about their moods and opinions. Due to the vast amounts of such messages, a large-scale analysis of population-wide developments becomes possi... | ["sentiment analysis", "covid-19", "twitter", "social media", "sentence embeddings", "europe"] | Cross-language sentiment analysis of European Twitter messages duringthe COVID-19 pandemicAnna KruspeGerman Aerospace Center (DLR)Institute of Data ScienceJena, Germanyanna.kruspe@dlr.deMatthias H ̈aberleTechnical University of Munich (TUM)Signal Processing in Earth Observation (SiPEO)Munich, Germanymatthias.haeberle@t... | 1cp_MEsz_cI | Review of "Cross-language sentiment analysis of European Twitter messages" -- interesting trends analysis but some more approach comparisons and tables for the data would be good. | 6: Marginally above acceptance threshold | The authors present an interesting, important and relevant trend analysis of sentiment across languages in several locales during the Covid-19 pandemic, using geo-tagged European Twitter data and pre-trained cross-lingual embeddings within a neural model.
The main contributions of the paper are: 1) the geo-tagged Euro... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Cross-language sentiment analysis of European Twitter messages during the COVID-19 pandemic
### Paper Abstract
Social media data can be a very salient source of informat... |
i_xiyGq6FNT | NeurIPS.cc/2021/Workshop/SVRHM | 2021 | What can 5.17 billion regression fits tell us about artificial models of the human visual system? | ["Colin Conwell", "Jacob S. Prince", "George A. Alvarez", "Talia Konkle"] | Rapid simultaneous advances in machine vision and cognitive neuroimaging present an unparalleled opportunity to assess the current state of artificial models of the human visual system. Here, we perform a large-scale benchmarking analysis of 72 modern deep neural network models to characterize with robust statistical p... | ["neuro_ai", "deep neural networks", "human vision", "model benchmarking", "big data", "fMRI"] | What can 5.17 billion regression fits tell us aboutartificial models of the human visual system?Colin Conwell∗, Jacob S. Prince, George A. Alvarez, Talia KonkleDepartment of Psychology, Harvard UniversityAbstractRapid simultaneous advances in machine vision and cognitive neuroimagingpresent an unparalleled opportunity ... | 1U9WQE0XL0 | A well written paper with somewhat unclear implications | 6: Marginally above acceptance threshold | The authors of the paper benchmark 72 neural networks to understand if they can be used to improve our understanding of the human visual system. They find two key results. First, that neural net architecture does not well correlate with the predictive power of the network. Secondly, different visual tasks lead to diffe... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
What can 5.17 billion regression fits tell us about artificial models of the human visual system?
### Paper Abstract
Rapid simultaneous advances in machine vision and co... |
rnVEiCJXuAF | uoft.ai/University_of_Toronto/2021/ProjectX | 2021 | Attention to Bayes: Self-Attention and Bayesian Methods for Predicting 3D Chromatin Folding From DNA Sequences | ["Anonymous"] | We present a novel model and approach, combining Transformers and VAE-like model that builds upon the established Akita/Basenji CNN architecture to enable exciting new possibilities of predicting 3D chromatin folding from DNA sequences. We establish the motivation for using self-attention and a simple yet feasible Baye... | ["transformer", "bioinformatics"] | Attention to Bayes: Self-Attention and BayesianMethods for Predicting 3D Chromatin Folding FromDNA SequencesAnonymous AuthorsAbstractWe present a novel model and approach, combining Transformers and V AE-likemodel that builds upon the established Akita/Basenji CNN architecture to enableexciting new possibilities of pre... | rEng3M3gXJc | A review of Attention to Bayes: Self-Attention and Bayesian Methods for Predicting 3D Chromatin Folding From DNA Sequences | 5: Marginally below acceptance threshold | The authors present an extension to a work in a well established biomedical field: predicting epigenetic markers directly from sequence. In this case, the task is to predict 3D chromatin-mediated contact within 1mb segments of the genome.
The authors do a good job covering relevant literature, especially Fudenberg e... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Attention to Bayes: Self-Attention and Bayesian Methods for Predicting 3D Chromatin Folding From DNA Sequences
### Paper Abstract
We present a novel model and approach, ... |
PTz0aXJp7A | logconference.io/LOG/2022/Conference | 2022 | Similarity-based Link Prediction from Modular Compression of Network Flows | ["Christopher Bl\u00f6cker", "Jelena Smiljani\u0107", "Ingo Scholtes", "Martin Rosvall"] | Node similarity scores are a foundation for machine learning in graphs for clustering, node classification, anomaly detection, and link prediction with applications in biological systems, information networks, and recommender systems. Recent works on link prediction use vector space embeddings to calculate node similar... | ["map equation", "community-based", "node similarity", "representation learning", "compression", "link prediction"] | Similarity-based Link Prediction from Modular Compressionof Network FlowsChristopher BlöckerIntegrated Science Lab, Department of PhysicsUmeå Universitychristopher.blocker@umu.seJelena Smiljani ́cIntegrated Science Lab, Department of PhysicsUmeå Universityjelena.smiljanic@umu.seIngo ScholtesCenter for Artificial Int... | g58WQQjPF5 | Summary:
The paper introduces MapSim, a method to estimate node similarities based on the Map Equation. The method has several advantages including an ability to compute asymmetric similarities for directed graphs and an average performance greater than competing methods such as node2vec. The method is novel as far... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Similarity-based Link Prediction from Modular Compression of Network Flows
### Paper Abstract
Node similarity scores are a foundation for machine learning in graphs for ... | |||
6lH8nkwKRXV | ICLR.cc/2021/Conference | 2021 | Graph Structural Aggregation for Explainable Learning | ["Alexis Galland", "marc lelarge"] | Graph neural networks have proven to be very efficient to solve several tasks in graphs such as node classification or link prediction. These algorithms that operate by propagating information from vertices to their neighbors allow one to build node embeddings that contain local information. In order to use graph neura... | ["graph", "deep", "learning"] | ABSTRACTGraph neural networks have proven to be very efficient to solve several tasks ingraphs such as node classification or link prediction. These algorithms that operateby propagating information from vertices to their neighbors allow one to build nodeembeddings that contain local information. In order to use graph ... | A4DqtLXz4- | Official Blind Review #3 | 7: Good paper, accept | This paper focuses on deriving explainable features for use in graph classification. To that end, they propose StructAgg that is essentially an aggregation process based on the structural roles of nodes that is then used in an end-to-end model. Experiments demonstrate the effectiveness of the proposed approach as it pr... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Graph Structural Aggregation for Explainable Learning
### Paper Abstract
Graph neural networks have proven to be very efficient to solve several tasks in graphs such as ... |
SPyxaz_h9Nd | ICLR.cc/2021/Conference | 2021 | PareCO: Pareto-aware Channel Optimization for Slimmable Neural Networks | ["Rudy Chin", "Ari S. Morcos", "Diana Marculescu"] | Slimmable neural networks provide a flexible trade-off front between prediction error and computational cost (such as the number of floating-point operations or FLOPs) with the same storage cost as a single model. They have been proposed recently for resource-constrained settings such as mobile devices. However, curren... | ["Adaptive Neural Network", "Slimmable Neural Network", "Channel Optimization", "Neural Architecture Search", "Convolutional Neural Network", "Image Classification"] | ABSTRACTSlimmable neural networks provide a flexible trade-off front between predictionerror and computational cost (such as the number of floating-point operations orFLOPs) with the same storage cost as a single model. They have been proposedrecently for resource-constrained settings such as mobile devices. However, c... | UAYJIj9XqBM | Good work, successfully extends previous work with proper techniques | 6: Marginally above acceptance threshold | The paper directly extends the “Slimmable network” by using per-layer width multipliers to allow more flexible network configurations. PareCO mainly aims to optimize both width multipliers and shared weights while considering Pareto-optimal. (Accuracy & Speed) Because the problem is intractable, the authors adopt the B... | 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
PareCO: Pareto-aware Channel Optimization for Slimmable Neural Networks
### Paper Abstract
Slimmable neural networks provide a flexible trade-off front between predictio... |
HJgyAoRqFQ | ICLR.cc/2019/Conference | 2019 | State-Denoised Recurrent Neural Networks | ["Michael C. Mozer", "Denis Kazakov", "Robert V. Lindsey"] | Recurrent neural networks (RNNs) are difficult to train on sequence processing tasks, not only because input noise may be amplified through feedback, but also because any inaccuracy in the weights has similar consequences as input noise. We describe a method for denoising the hidden state during training to achieve mor... | ["recurrent nets", "attractor nets", "denoising", "sequence processing"] | ABSTRACTRecurrent neural networks (RNNs) are difficult to train on sequence processingtasks, not only because input noise may be amplified through feedback, but alsobecause any inaccuracy in the weights has similar consequences as input noise. Wedescribe a method for denoising the hidden state during training to achiev... | S1gsJSsy2m | Review for State-Denoised Recurrent Neural Networks | 5: Marginally below acceptance threshold | In this paper the authors develop the clever idea to use attractor networks, inspired by Hopfield nets, to “denoise” a recurrent neural network. The idea is that for every normal step of an RNN, one induces an additional "dimension" of recurrency in order to create attractor dynamics around that particular hidden stat... | 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
State-Denoised Recurrent Neural Networks
### Paper Abstract
Recurrent neural networks (RNNs) are difficult to train on sequence processing tasks, not only because input ... |
Skx2iCNFwB | ICLR.cc/2020/Conference | 2020 | Stabilizing Neural ODE Networks with Stochasticity | ["Xuanqing Liu", "Tesi Xiao", "Si Si", "Qin Cao", "Sanjiv Kumar", "Cho-Jui Hsieh"] | Neural Ordinary Differential Equation (Neural ODE) has been proposed as a continuous approximation to the ResNet architecture. Some commonly used regularization mechanisms in discrete neural networks (e.g. dropout, Gaussian noise) are missing in current Neural ODE networks. In this paper, we propose a new continuous ne... | ["neural ode networks", "neural ode", "regularization mechanisms", "dropout", "stochasticity", "continuous approximation", "resnet architecture", "discrete neural networks", "gaussian noise"] | ABSTRACTNeural Ordinary Differential Equation (Neural ODE) has been proposed as acontinuous approximation to the ResNet architecture. Some commonly usedregularization mechanisms in discrete neural networks (e.g. dropout, Gaussiannoise) are missing in current Neural ODE networks. In this paper, we propose anew continuou... | S1emwR1y5H | Official Blind Review #2 | 6: Weak Accept | The paper introduces Neural Stochastic Differential Equations models which is a follow-up of last year paper on Neural Ordinary Differential Equations. The stochasticity is obtained by adding an injection noise term.
Pros:
As there is still no published paper that introduces Neural SDE, this paper would be well pla... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Stabilizing Neural ODE Networks with Stochasticity
### Paper Abstract
Neural Ordinary Differential Equation (Neural ODE) has been proposed as a continuous approximation ... | |
ByquB-WC- | ICLR.cc/2018/Conference | 2018 | Finding ReMO (Related Memory Object): A Simple neural architecture for Text based Reasoning | ["Jihyung Moon", "Hyochang Yang", "Sungzoon Cho"] | Memory Network based models have shown a remarkable progress on the task of relational reasoning.
Recently, a simpler yet powerful neural network module called Relation Network (RN) has been introduced.
Despite its architectural simplicity, the time complexity of relation network grows quadratically with data, hence l... | ["Natural Language Processing", "Deep Learning", "Reasoning"] | ABSTRACTTo solve the text-based question and answering task that requires relational rea-soning, it is necessary to memorize a large amount of information and find out thequestion relevant information from the memory. Most approaches were based onexternal memory and four components proposed by Memory Network. The dis-t... | rk-hlXcez | Finding ReMO review | 4: Ok but not good enough - rejection | This paper introduces Related Memory Network (RMN), an improvement over Relationship Networks (RN). RMN avoids growing the relationship time complexity as suffered by RN (Santoro et. Al 2017). RMN reduces the complexity to linear time for the bAbi dataset. RN constructs pair-wise interactions between objects in RN to 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
Finding ReMO (Related Memory Object): A Simple neural architecture for Text based Reasoning
### Paper Abstract
Memory Network based models have shown a remarkable progre... |
BBrgJZF4pfc | graphicsinterface.org/Graphics_Interface/2022/Conference | 2022 | Context and Minimalism: User Evaluations on Two Interpersonal Telepresence Systems | ["Anonymous"] | We present a field study with 10 pairs of adults to employ two prototypes of an interpersonal telepresence system: one catered to the Viewer, the individual experiencing a remote environment, and one catered to the Streamer, the individual sharing the remote experience to the Viewer through technological means. Based o... | ["Telepresence", "Live-streaming", "Remote experiences", "interpersonal telepresence"] | ABSTRACTWe present a field study with 10 pairs of adults to employ twoprototypes of an interpersonal telepresence system: one cateredto the Viewer, the individual experiencing a remote environment,and one catered to the Streamer, the individual sharing the remoteexperience to the Viewer through technological means. Bas... | PdYQgO3j-dK | A study that explicitly compares handheld telepresence system user preferences for both the remote and local viewer's needs. | 7: Good paper, accept | The authors conduct an exploratory comparison of two telepresence system, each trying to cater to either the remote or telepresent viewer. The key difference was a 360 degree camera that allowed a viewer to see and pan around in the environment. They found different preferences depending on role: the remote person pref... | 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
Context and Minimalism: User Evaluations on Two Interpersonal Telepresence Systems
### Paper Abstract
We present a field study with 10 pairs of adults to employ two prot... |
yNFwsrcEtO0 | ICLR.cc/2021/Conference | 2021 | Demon: Momentum Decay for Improved Neural Network Training | ["John Chen", "Cameron Wolfe", "Zhao Li", "Anastasios Kyrillidis"] | Momentum is a popular technique in deep learning for gradient-based optimizers. We propose a decaying momentum (Demon) rule, motivated by decaying the total contribution of a gradient to all future updates. Applying Demon to Adam leads to significantly improved training, notably competitive to momentum SGD with learnin... | ["deep learning", "large scale learning", "neural networks", "sgd"] | ABSTRACTMomentum is a popular technique in deep learning for gradient-based optimizers.We propose a decaying momentum (D EMON ) rule, motivated by decaying the to-tal contribution of a gradient to all future updates. Applying D EMON to Adamleads to significantly improved training, notably competitive to momentum SGDwit... | JHc86VIvKO9 | The topic of this paper is interesting | 5: Marginally below acceptance threshold | Overall, the paper is well-organized and easy to understand. The authors proposed a new momentum decay strategy for SGDM and Adam methods, experimental results shown the effectiveness of the propsoed methods. I have the following comments/questions.
1. The motivation of using $\beta_t$ is still not clear to me. Could... | 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
Demon: Momentum Decay for Improved Neural Network Training
### Paper Abstract
Momentum is a popular technique in deep learning for gradient-based optimizers. We propose ... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.