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
ZS-9XoX20AV
ICLR.cc/2021/Conference
2021
GraphSAD: Learning Graph Representations with Structure-Attribute Disentanglement
["Minghao Xu", "Hang Wang", "Bingbing Ni", "Wenjun Zhang", "Jian Tang"]
Graph Neural Networks (GNNs) learn effective node/graph representations by aggregating the attributes of neighboring nodes, which commonly derives a single representation mixing the information of graph structure and node attributes. However, these two kinds of information might be semantically inconsistent and could b...
["Graph Representation Learning", "Disentangled Representation Learning"]
ABSTRACTGraph Neural Networks (GNNs) learn effective node/graph representations byaggregating the attributes of neighboring nodes, which commonly derives a sin-gle representation mixing the information of graph structure andnode attributes .However, these two kinds of information might be semantically inconsistent andc...
2KZlXB-KMXb
Feel confused on the motivation
3: Clear rejection
1. I am confused on the motivation part. Although the authors provide two examples (the prediction of a user’s social class mainly relies on his/her social network structure) to illustrate the benefits of disentangled representation for certain downstream tasks and I agree on the statement, it raises another question t...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title GraphSAD: Learning Graph Representations with Structure-Attribute Disentanglement ### Paper Abstract Graph Neural Networks (GNNs) learn effective node/graph representati...
dJbf5SqbFrM
ICLR.cc/2021/Conference
2021
Continuous Transfer Learning
["Jun Wu", "Jingrui He"]
Transfer learning has been successfully applied across many high-impact applications. However, most existing work focuses on the static transfer learning setting, and very little is devoted to modeling the time evolving target domain, such as the online reviews for movies. To bridge this gap, in this paper, we focus on...
["continuous transfer", "time", "target domain", "current target domain", "transfer learning", "many", "applications", "work focuses", "static transfer", "setting"]
ABSTRACTTransfer learning has been successfully applied across many high-impact applica-tions. However, most existing work focuses on the static transfer learning setting,and very little is devoted to modeling the time evolving target domain, such as theonline reviews for movies. To bridge this gap, in this paper, we f...
XdJtgahGN29
This paper focuses on the continuous transfer learning setting with a time evolving target domain. A major challenge is that the relationship between the source domain and the target domain changes over time. This paper proposes a new C-divergence to measure the discrepancy between domains. It could be utilized to inst...
6: Marginally above acceptance threshold
(1) This article is more detailed in its theoretical analysis, and the C-divergence has advantages over other divergence from the theoretical analysis. (2) This article proposes a new continuous transfer learning setting. It theoretically analyzes the upper bound of the target domain error, and utilizes the C-diverg...
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 Continuous Transfer Learning ### Paper Abstract Transfer learning has been successfully applied across many high-impact applications. However, most existing work focuses...
Rq31tXaqXq
ICLR.cc/2021/Conference
2021
VideoFlow: A Framework for Building Visual Analysis Pipelines
["Yue Wu", "Jianqiang Huang", "Jiangjie Zhen", "Guokun Wang", "Chen Shen", "Chang Zhou", "Xian-Sheng Hua"]
The past years have witnessed an explosion of deep learning frameworks like PyTorch and TensorFlow since the success of deep neural networks. These frameworks have significantly facilitated algorithm development in multimedia research and production. However, how to easily and efficiently build an end-to-end visual ana...
["Computation graph", "Resource", "Computer vision", "Deep learning", "Framework", "Software"]
ABSTRACTThe past years have witnessed an explosion of deep learning frameworks like Py-Torch and TensorFlow since the success of deep neural networks. These frame-works have significantly facilitated algorithm development in multimedia researchand production. However, how to easily and efficiently build an end-to-end v...
9f_CIX4_5l
An engineering introduction to a visual analysis system but not a scientific paper
3: Clear rejection
Summary ------- This paper introduce a framework for visual analysis (so-called VideoFlow) that will be open-sourced. VideoFlow is introduced with providing a flexible, efficient, extensible, and secure visual analysis framework for both academia and industry. It has been adopted in tens of intelligent visual analysis...
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 VideoFlow: A Framework for Building Visual Analysis Pipelines ### Paper Abstract The past years have witnessed an explosion of deep learning frameworks like PyTorch and ...
hga6dk7nxFB
ICLR.cc/2021/Conference
2021
Deep Learning Proteins using a Triplet-BERT network
["Mark Lennox", "Neil M. Robertson", "Barry Devereux"]
Modern sequencing technology has produced a vast quantity of proteomic data, which has been key to the development of various deep learning models within the field. However, there are still challenges to overcome with regards to modelling the properties of a protein, especially when labelled resources are scarce. Devel...
["Deep Learning", "Transformers", "Metric Learning", "Proteomics"]
ABSTRACTModern sequencing technology has produced a vast quantity of proteomic data,which has been key to the development of various deep learning models withinthe field. However, there are still challenges to overcome with regards to mod-elling the properties of a protein, especially when labelled resources are scarce...
hznyTH52q0
Official Blind Review #1
3: Clear rejection
########################################################################## Summary: The authors proposed the Triple-BERT approach for regression tasks in protein biology: plasma membrane localization, thermostability, peak absorption, and enantioselectivity. They adopted triplet loss to fine-tune pre-trained BERT repr...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Deep Learning Proteins using a Triplet-BERT network ### Paper Abstract Modern sequencing technology has produced a vast quantity of proteomic data, which has been key to...
HJxeGb5pTm
AKBC.ws/2019/Conference
2019
OPIEC: An Open Information Extraction Corpus
["Kiril Gashteovski", "Sebastian Wanner", "Sven Hertling", "Samuel Broscheit", "Rainer Gemulla"]
Open information extraction (OIE) systems extract relations and their arguments from natural language text in an unsupervised manner. The resulting extractions are a valuable resource for downstream tasks such as knowledge base construction, open question answering, or event schema induction. In this paper, we ...
["open information extraction", "text analytics"]
Automated Knowledge Base Construction (2019) Conference paperOPIEC: An Open Information Extraction CorpusKiril Gashteovski k.gashteovski@uni-mannheim.deSebastian Wanner swanner@mail.uni-mannheim.deSven Hertling sven@informatik.uni-mannheim.deSamuel Broscheit broscheit@informatik.uni-mannheim.deRainer Gemulla rgemulla@u...
B1eTN16XfN
OPIEC: An Open Information Extraction Corpus
7: Good paper, accept
In this paper, the authors build a new corpus for information extraction which is larger comparing to the prior public corpora and contains information not existing in current corpora. The dataset can be useful in other applications. The paper is well written and easy to follow. It also provides details about the corpu...
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 OPIEC: An Open Information Extraction Corpus ### Paper Abstract Open information extraction (OIE) systems extract relations and their arguments from natural language t...
hRHX6XW9_Gu
Interspeech.org/2023/Workshop/SSW
2023
Diffusion Transformer for Adaptive Text-to-Speech
["Haolin Chen", "Philip N. Garner"]
Given the success of diffusion in synthesizing realistic speech, we investigate how diffusion can be included in adaptive text-to-speech systems. Inspired by the adaptable layer norm modules for Transformer, we adapt a new backbone of diffusion models, Diffusion Transformer, for acoustic modeling. Specifically, the ada...
["speech synthesis", "adaptive TTS", "diffusion transformer", "adaptive layer norm"]
Diffusion Transformer for Adaptive Text-to-SpeechHaolin Chen1,2, Philip N. Garner11Idiap Research Institute, Martigny, Switzerland2 ́Ecole Polytechnique F ́ed ́erale de Lausanne, Switzerlandhchen@idiap.ch, pgarner@idiap.chAbstractGiven the success of diffusion in synthesizing realistic speech,we investigate how diffusi...
JbIhAZfQ7xd
Adaptive transformer fo TTS
7: Good paper, accept
This paper proposes to adapt a diffusion transformer (DiT) to the case of adaptive TTS. The DiT model is then used as the decoder providing interesting benefits in terms of quality and speed. First experiment compares the proposed DiT to a non-causal Wavenet and shows a similar output quality (naturalness) with a per...
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 Diffusion Transformer for Adaptive Text-to-Speech ### Paper Abstract Given the success of diffusion in synthesizing realistic speech, we investigate how diffusion can be...
0iBileNBBfi
ICML.cc/2021/Workshop/AML
2021
Combating Adversaries with Anti-Adversaries
["Motasem Alfarra", "Juan Camilo Perez", "Ali Thabet", "Adel Bibi", "Philip Torr", "Bernard Ghanem"]
Deep neural networks are vulnerable to small input perturbations known as adversarial attacks. Inspired by the fact that these adversaries are constructed by iteratively minimizing the confidence of a network for the true class label, we propose the anti-adversary layer, aimed at countering this effect. In particular, ...
["Adversarial Attacks", "Network Robustness"]
Combating Adversaries with Anti-AdversariesMotasem Alfarra1Juan C. P ́erez1Ali Thabet1Adel Bibi2Philip H.S. Torr2Bernard Ghanem1AbstractDeep neural networks are vulnerable to small in-put perturbations known as adversarial attacks.Inspired by the fact that these adversaries are con-structed by iteratively minimizing th...
RGJ_4_3RBc0
More experiments of white-box attack defense is required.
Accept
This paper introduces an anti-adversary layer in the network structure, which is equivalent to changing the network structure and pulling data points away from the classification decision boundary, The paper proves that it can effectively resist black box, gray box and adaptive attacks. Experiments are carried out on v...
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 Combating Adversaries with Anti-Adversaries ### Paper Abstract Deep neural networks are vulnerable to small input perturbations known as adversarial attacks. Inspired by...
rygnfn0qF7
ICLR.cc/2019/Conference
2019
Language Model Pre-training for Hierarchical Document Representations
["Ming-Wei Chang", "Kristina Toutanova", "Kenton Lee", "Jacob Devlin"]
Hierarchical neural architectures can efficiently capture long-distance dependencies and have been used for many document-level tasks such as summarization, document segmentation, and fine-grained sentiment analysis. However, effective usage of such a large context can difficult to learn, especially in the case where t...
["language model", "document segmentation", "algorithms", "hierarchical document representations", "representations", "dependencies", "many", "tasks"]
ABSTRACTHierarchical neural architectures can efficiently capture long-distance dependen-cies and have been used for many document-level tasks such as summarization,document segmentation, and sentiment analysis. However, effective usage of sucha large context can be difficult to learn, especially in the case where ther...
SyeacLV537
Good contribution with a few missing baselines and implementation details
6: Marginally above acceptance threshold
In this work, the authors explore different ways to pre-train contextualized word and sentence representations for use in other tasks. They propose two main methods: a straight-forward extension of the ElMO model for hierarchical uni-directional language models, and a de-noising auto-encoder type method which allows to...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Language Model Pre-training for Hierarchical Document Representations ### Paper Abstract Hierarchical neural architectures can efficiently capture long-distance dependen...
IfyIzP_GSpn
MICCAI.org/2022/Challenge/FLARE
2022
Abdominal Organ Segmentation via Self Training
["Hexin Dong", "Zifan Chen", "Jie Zhao", "Mingze Yuan", "Fei Yu", "Jiangdong Zhang", "Li Zhang"]
This paper proposes a semi-supervised abdominal organ segmentation approach based on self-training. Self-training(ST) helps improve the segmentation network’s decision boundary with unlabeled images and generated pseudo labels. Experiment results show that our method has significantly outperformed the non-ST baseline, ...
["Semantic Segmentation", "Self Training"]
Abdominal Organ Segmentation via Self TrainingHexin Dong1, Zifan Chen1, Jie Zhao3, Mingze Yuan1, Fei Yu1, JiangdongZhang1, and Li Zhang1,21Center for Data Science, Peking University, Beijing, China2Center for Data Science in Health and Medicine, Peking University, Beijing, China3National Engineering Laboratory for Big ...
ssAPKfh15E8
This paper lacks a description of the prediction efficiency. In addition, this paper is too short.
4: Ok but not good enough - rejection
In this paper, a semi-supervised method using nnUNet to generate pseudo labels is proposed to continuously improve the segmentation effect of the model. However, this paper does not mention the final prediction part of the network and the prediction efficiency of the network. If the original framework of nnUNet is used...
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 Abdominal Organ Segmentation via Self Training ### Paper Abstract This paper proposes a semi-supervised abdominal organ segmentation approach based on self-training. Sel...
rJf0BjAqYX
ICLR.cc/2019/Conference
2019
Like What You Like: Knowledge Distill via Neuron Selectivity Transfer
["Zehao Huang", "Naiyan Wang"]
Despite deep neural networks have demonstrated extraordinary power in various applications, their superior performances are at expense of high storage and computational costs. Consequently, the acceleration and compression of neural networks have attracted much attention recently. Knowledge Transfer (KT), which aims at...
["Knowledge Distill"]
ABSTRACTDespite deep neural networks have demonstrated extraordinary power in variousapplications, their superior performances are at expense of high storage and com-putational costs. Consequently, the acceleration and compression of neural net-works have attracted much attention recently. Knowledge Transfer (KT), whic...
BygynUFT3X
Need to improve clarity and interpretability
4: Ok but not good enough - rejection
This submission proposes a novel loss function, based on Maximum Mean Discrepancy (MMD), for knowledge transfer (distillation) from a teacher network to a student network, which matches the spatial distribution of neuron activations between the two. The proposed approach is interesting but there is significant room fo...
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 Like What You Like: Knowledge Distill via Neuron Selectivity Transfer ### Paper Abstract Despite deep neural networks have demonstrated extraordinary power in various ap...
lZr9s1x0mE
ICML.cc/2021/Workshop/AutoML
2021
Towards Explaining Hyperparameter Optimization via Partial Dependence Plots
["Julia Moosbauer", "Julia Herbinger", "Giuseppe Casalicchio", "Marius Lindauer", "Bernd Bischl"]
Automated hyperparameter optimization (HPO) can support practitioners to obtain peak performance in machine learning models. However, there is often a lack of valuable insights into the effects of different hyperparameters on the final model performance. This lack of comprehensibility and transparency makes it difficul...
["Automated machine learning", "interpretable machine learning", "partial dependence plots"]
Towards Explaining Hyperparameter Optimization viaPartial Dependence PlotsJulia Moosbauer;1julia.moosbauer@stat.uni-muenchen.daJulia Herbinger;1julia.herbinger@stat.uni-muenchen.deGiuseppe Casalicchio1giuseppe.casalicchio@stat.uni-muenchen.deMarius Lindauer2lindauer@tnt.uni-hannover.deBernd Bischl1bernd.bischl@stat.uni...
BhSkCmSRltd
A small but concrete contribution for boosting trust on hyperparameter auto-optimization
8: Top 50% of accepted papers, clear accept
The paper attempts to provide and enrich explanations on how values of hyperparameters affect the performance of a machine learning model, which can help boost confidence on automated methods for searching for the optimal hyperparameter values. The method is based on a well established method for explaining numerical ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Towards Explaining Hyperparameter Optimization via Partial Dependence Plots ### Paper Abstract Automated hyperparameter optimization (HPO) can support practitioners to o...
rJgJDAVKvB
ICLR.cc/2020/Conference
2020
Learning to Plan in High Dimensions via Neural Exploration-Exploitation Trees
["Binghong Chen", "Bo Dai", "Qinjie Lin", "Guo Ye", "Han Liu", "Le Song"]
We propose a meta path planning algorithm named \emph{Neural Exploration-Exploitation Trees~(NEXT)} for learning from prior experience for solving new path planning problems in high dimensional continuous state and action spaces. Compared to more classical sampling-based methods like RRT, our approach achieves much bet...
["learning to plan", "representation learning", "learning to design algorithm", "reinforcement learning", "meta learning"]
ABSTRACTWe propose a meta path planning algorithm named Neural Exploration-ExploitationTrees (NEXT) for learning from prior experience for solving new path planning problemsin high dimensional continuous state and action spaces. Compared to more classicalsampling-based methods like RRT, our approach achieves much bette...
Hylq3SQaYr
Official Blind Review #2
6: Weak Accept
This paper proposes an approach to learn how to plan in continuous spaces using neural nets to learn a value function and a policy for scoring and sampling next-step candidates in a stochastic tree search. The networks are updated as more planning tasks are executed, producing more data for the policy and value functi...
<|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 Plan in High Dimensions via Neural Exploration-Exploitation Trees ### Paper Abstract We propose a meta path planning algorithm named \emph{Neural Exploration...
B1xIj3VYvr
ICLR.cc/2020/Conference
2020
Weakly Supervised Clustering by Exploiting Unique Class Count
["Mustafa Umit Oner", "Hwee Kuan Lee", "Wing-Kin Sung"]
A weakly supervised learning based clustering framework is proposed in this paper. As the core of this framework, we introduce a novel multiple instance learning task based on a bag level label called unique class count (ucc), which is the number of unique classes among all instances inside the bag. In this task, no an...
["weakly supervised clustering", "weakly supervised learning", "multiple instance learning"]
ABSTRACTA weakly supervised learning based clustering framework is proposed in this pa-per. As the core of this framework, we introduce a novel multiple instance learningtask based on a bag level label called unique class count ( ucc), which is the numberof unique classes among all instances inside the bag. In this tas...
BygIZGO3YS
Official Blind Review #3
6: Weak Accept
The authors present a novel weakly-supervised multiple instance learning model based on a bag level label called unique class count(UCC). The core content of the method is to learn mapping between bags and their associated bag level ucc labels and then to predict the ucc labels of unseen bags. Positive: (1) The author...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Weakly Supervised Clustering by Exploiting Unique Class Count ### Paper Abstract A weakly supervised learning based clustering framework is proposed in this paper. As th...
EPUtNe7a9ta
logconference.io/LOG/2022/Conference
2022
Neighborhood-aware Scalable Temporal Network Representation Learning
["Yuhong Luo", "Pan Li"]
Temporal networks have been widely used to model real-world complex systems such as financial systems and e-commerce systems. In a temporal network, the joint neighborhood of a set of nodes often provides crucial structural information useful for predicting whether they may interact at a certain time. However, recent r...
["Temporal Network", "Representation Learning", "Graph Neural Networks"]
Neighborhood-aware Scalable Temporal NetworkRepresentation LearningYuhong Luo∗University of Massachusettsyuhongluo@umass.eduPan LiPurdue Universitypanli@purdue.eduAbstractTemporal networks have been widely used to model real-world complex systemssuch as financial systems and e-commerce systems. In a temporal network, t...
fwDoNPBWiAC
The paper proposes Neighborhood-Aware Temporal network model (NAT) for temporal network representation learning for link prediction tasks. The authors design a dictionary-type neighborhood representation and dedicated data structure N-cache (for fast parallel computing) to record down-sampled set of the neighboring nod...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Neighborhood-aware Scalable Temporal Network Representation Learning ### Paper Abstract Temporal networks have been widely used to model real-world complex systems such ...
Byex4T2XcV
icaps-conference.org/ICAPS/2019/Workshop/XAIP
2019
Model-Free Model Reconciliation
["Anonymous"]
Designing agents capable of explaining complex sequential decisions remain a significant open problem in automated decision-making. Recently, there has been a lot of interest in developing approaches for generating such explanations for various decision-making paradigms. One such approach has been the idea of explanati...
["model reconciliation", "user", "model", "agents capable", "complex sequential decisions", "significant open problem", "automated", "lot", "interest", "approaches"]
Model-Free Model ReconciliationSarath Sreedharan, Alberto Olmo Hernandez, Aditya Prasad Mishra andSubbarao KambhampatiSchool of Computing, Informatics, and Decision Systems Engineering,Arizona State University, Tempe, AZ 85281 USAfssreedh3, aolmoher, amishr28, rao g@ asu.eduAbstractDesigning agents capable of explainin...
S1eE7Wzw3V
Model Reconciliation with learned model
4: Top 50% of accepted papers, clear accept
This paper extends previous works from the same authors on model reconciliation. The novelty here is that they drop the assumption about having the explicit user model. In this work, the model is instead learned. This is clearly an interesting work, and it's worth presenting and discussing it at the workshop. Some com...
3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Model-Free Model Reconciliation ### Paper Abstract Designing agents capable of explaining complex sequential decisions remain a significant open problem in automated dec...
r1xlvi0qYm
ICLR.cc/2019/Conference
2019
Learning to Remember More with Less Memorization
["Hung Le", "Truyen Tran", "Svetha Venkatesh"]
Memory-augmented neural networks consisting of a neural controller and an external memory have shown potentials in long-term sequential learning. Current RAM-like memory models maintain memory accessing every timesteps, thus they do not effectively leverage the short-term memory held in the controller. We hypothesize t...
["memory-augmented neural networks", "writing optimization"]
Published as a conference paper at ICLR 2019Learning to Remember Morewith Less MemorizationHung Le, Truyen Tran and Svetha VenkateshApplied AI Institute, Deakin University, Geelong, Australia{lethai,truyen.tran,svetha.venkatesh}@deakin.edu.auAbstractMemory-augmented neural networks consisting of a neural controller and...
SygH6Dzjpm
Very interesting and well written paper on augmented memories
7: Good paper, accept
This paper looks at ways to improve memory-writing in memory augmented neural networks. Authors proposed two methods to compare against "regular writing" method as well as compare against each other, namely "uniform writing" and "cached uniform writing". Latter one attempts to utilize a small size memory efficiently by...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning to Remember More with Less Memorization ### Paper Abstract Memory-augmented neural networks consisting of a neural controller and an external memory have shown ...
WgYphAbJZS-
MICCAI.org/2022/Challenge/FLARE
2022
Coarse to Fine Automatic Segmentation of Abdominal Multiple Organs Based on Semi-Supervised Network
["Yi Lv", "Yu Ning", "Junchen Wang"]
Abdominal multi-organ segmentation is fast becoming a key instrument in preoperative diagnosis. Using the results of abdominal CT image segmentation for three-dimensional reconstruction is an intuitive and accurate method for surgical planning. In this paper, we propose a stable three-stage fast automatic segmentation ...
["Medical Image Segmentation", "Deep Learning", "Neural Network"]
Coarse to Fine Automatic Segmentation ofAbdominal Multiple OrgansYi Lv1,2[0000 −0002−3778−430X], Yu Ning2[0000 −0001−5319−3588], and JunchenWang2,3[0000 −0002−0916−9932]∗1North China Research Institute of Electro-optics, Beijing, 100015, China2School of Mechanical Engineering and Automation, Beihang University, Beijing...
MdCbCVeaGds
Review of Coarse to Fine Automatic Segmentation of Abdominal Multiple Organs Based on Semi-Supervised Network
7: Good paper, accept
Most training details are descripted, but the adopted semi-supervised learning method is missing.
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 Coarse to Fine Automatic Segmentation of Abdominal Multiple Organs Based on Semi-Supervised Network ### Paper Abstract Abdominal multi-organ segmentation is fast becomin...
HyevIJStwH
ICLR.cc/2020/Conference
2020
Understanding Why Neural Networks Generalize Well Through GSNR of Parameters
["Jinlong Liu", "Yunzhi Bai", "Guoqing Jiang", "Ting Chen", "Huayan Wang"]
As deep neural networks (DNNs) achieve tremendous success across many application domains, researchers tried to explore in many aspects on why they generalize well. In this paper, we provide a novel perspective on these issues using the gradient signal to noise ratio (GSNR) of parameters during training process of DNNs...
["DNN", "generalization", "GSNR", "gradient descent"]
ABSTRACTAs deep neural networks (DNNs) achieve tremendous success across many appli-cation domains, researchers tried to explore in many aspects on why they gener-alize well. In this paper, we provide a novel perspective on these issues usingthe gradient signal to noise ratio (GSNR) of parameters during training proces...
HkxvaNRhFB
Official Blind Review #1
6: Weak Accept
In this work, the authors suggest a new point of view on generalization through the lens of the distribution of the per-sample gradients. The authors consider the variance and mean of the per-sample gradients for each parameter of the model and define for each parameter the Gradient Signal to Noise ratio (GSNR). The GS...
<|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 Why Neural Networks Generalize Well Through GSNR of Parameters ### Paper Abstract As deep neural networks (DNNs) achieve tremendous success across many app...
dPWotG03R-h
MIDL.io/2023/Short_Paper_Track
2023
Investigate Sex Dimorphism of Cerebral Myelination Across Lifespan by Leveraging Conditional Variational Autoencoder
["Jinghang Li", "Linghang Wang", "Chang-le Chen", "Tamer Ibrahim", "Howard Aizenstein", "Minjie Wu"]
In this work we investigated the potential sex differences in white matter aging using conditional variational autoencoder (cVAE) on myelin content MR images. The cVAE model was trained along with a supervised brain age prediction model, which learns the representation of myelination aging process within a single end-t...
["Conditional VAE", "Brain age prediction", "Sex differences", "Myelin map"]
Medical Imaging with Deep Learning – Under Review 2023 Short Paper – MIDL 2023 submissionInvestigate Sex Dimorphism of Cerebral Myelination AcrossLifespan by Leveraging Conditional Variational AutoencoderJinghang Li1jil202@pitt.eduLinghai Wang1L.Wang@pitt.eduChang-le Chen1chc348@pitt.eduTamer Ibrahim1,2,tibrahim@pitt.e...
j3FQzsaJTB3
Interesting application but unclear methods/contribution attribution
5: Marginally below acceptance threshold
This paper proposes using a conditional VAE on myelin content MR images trained with brain age prediction to learn representation of myelination aging process in single model. Training was conducted using a single aging dataset and used to assess sex differences in myelination and aging. Strengths: + Interesting topic...
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 Investigate Sex Dimorphism of Cerebral Myelination Across Lifespan by Leveraging Conditional Variational Autoencoder ### Paper Abstract In this work we investigated the ...
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...
f6aO96Ue5Gd
This paper presents a pretty exciting and novel robot learning system that performs well in around 10 real-world robotics tasks and is able to handle unseen distractor objects and novel object configurations to some degree during test time. This work has a few noteworthy strengths. First of all, the algorithmic design...
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...
r1lQeoCVu4
ICLR.cc/2019/Workshop/LLD
2019
Data for free: Fewer-shot algorithm learning with parametricity data augmentation
["Owen Lewis", "Katherine Hermann"]
We address the problem of teaching an RNN to approximate list-processing algorithms given a small number of input-output training examples. Our approach is to generalize the idea of parametricity from programming language theory to formulate a semantic property that distinguishes common algorithms from arbitrary non-al...
["data augmentation", "algorithm learning", "RNN"]
ABSTRACTWe address the problem of teaching an RNN to approximate list-processing al-gorithms given a small number of input-output training examples. Our approachis to generalize the idea of parametricity from programming language theory toformulate a semantic property that distinguishes common algorithms from arbi-trar...
SJlrvdkW5V
Review
4: Top 50% of accepted papers, clear accept
The paper tackles the problem of teaching an RNNs to approximate list-processing algorithms. The authors argue that what distinguishes algorithms from arbitrary functions is that they commute with a family of element-wise changes to their inputs and propose a method that learns RNN functions to approximate such family ...
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 Data for free: Fewer-shot algorithm learning with parametricity data augmentation ### Paper Abstract We address the problem of teaching an RNN to approximate list-proces...
HkeMYJHYvS
ICLR.cc/2020/Conference
2020
High-Frequency guided Curriculum Learning for Class-specific Object Boundary Detection
["VSR Veeravasarapu", "Deepak Mittal", "Abhishek Goel", "Maneesh Singh"]
This work addresses class-specific object boundary extraction, i.e., retrieving boundary pixels that belong to a class of objects in the given image. Although recent ConvNet-based approaches demonstrate impressive results, we notice that they produce several false-alarms and misdetections when used in real-world applic...
["Computer Vision", "Object Contour Detection", "Curriculum Learning", "Wavelets", "Aerial Imagery"]
ABSTRACTThis work addresses class-specific object boundary extraction, i.e., retrievingboundary pixels that belong to a class of objects in the given image. Althoughrecent ConvNet -based approaches demonstrate impressive results, we notice thatthey produce several false-alarms and misdetections when used in real-world ...
r1grXbxL5H
Official Blind Review #3
3: Weak Reject
The paper shows the efficiency of curriculum learning and using problem specific features for contour detection. The authors consider a network trained for class-specific edge detection (e.g. outlining edges of roads in an image). They propose two problem domain tricks to improve the performance: - use curriculum lear...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title High-Frequency guided Curriculum Learning for Class-specific Object Boundary Detection ### Paper Abstract This work addresses class-specific object boundary extraction, ...
_i0cgACQ4mY
eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track
2021
Ontology Modelling for Materials Science Experiments
["Mehwish Alam", "Henk Birkholz", "Danilo Dessi'", "Christoph Eberl", "Heike Fliegl", "Peter Gumbsch", "Philipp von Hartrott", "Lutz M\u00e4dler", "Markus Niebel", "Harald Sack", "Akhil Thomas"]
Materials are either enabler or bottleneck for the vast majority of technological innovations. The digitization of materials and processes is mandatory to create live production environments which represent physical entities and their aggregations and thus allow to represent, share, and understand materials changes. ...
["Materials Science", "Ontology Design", "Data Curation"]
Ontology Modelling for Materials ScienceExperimentsMehwish Alam1;2, Henk Birkholz3, Danilo Dess 1;2, Christoph Eberl4, HeikeFliegl5, Peter Gumbsch4, Philipp von Hartrott4, Lutz M adler3, MarkusNiebel4, Harald Sack1;2, Akhil Thomas41FIZ Karlsruhe { Leibniz Institute for Information Infrastructure, Germany2Karlsruhe Inst...
ruO1l0nai55
Too early work; not yet interesting
4: Ok but not good enough - rejection
This poster submission argues for the need for an ontology describing concepts (primarily processes and materials) in materials science and engineering, and describes an under-development ontology filling this need. I do not believe this work is suitable for a poster presentation. It presents some early and rough idea...
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 Ontology Modelling for Materials Science Experiments ### Paper Abstract Materials are either enabler or bottleneck for the vast majority of technological innovations. T...
SJ1nzBeA-
ICLR.cc/2018/Conference
2018
Multi-Task Learning for Document Ranking and Query Suggestion
["Wasi Uddin Ahmad", "Kai-Wei Chang", "Hongning Wang"]
We propose a multi-task learning framework to jointly learn document ranking and query suggestion for web search. It consists of two major components, a document ranker, and a query recommender. Document ranker combines current query and session information and compares the combined representation with document represe...
["Multitask Learning", "Document Ranking", "Query Suggestion"]
ABSTRACTWe propose a multi-task learning framework to jointly learn document ranking andquery suggestion for web search. It consists of two major components, a documentranker and a query recommender. Document ranker combines current query andsession information and compares the combined representation with documentrepr...
SJe5w-Ylz
This work proposes a multi-task learning framework, M-NSRF, that can jointly learn document ranking and query suggestion. A session recurrence is introduced and modeled by an extra LSTM. The authors carry out extensive experiments to verify their algorithm. But I think this work is not of enough novelty and the experim...
4: Ok but not good enough - rejection
Novelty: It looks quite straightforward to combine document ranking and query suggestion. For the model architecture, it is a standard multi-task learning framework. For the “session encoder”, it is also proposed (at least, used) in (Sordoni et al., CIKM 2015). Therefore, I think the technical novelty of the work is l...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Multi-Task Learning for Document Ranking and Query Suggestion ### Paper Abstract We propose a multi-task learning framework to jointly learn document ranking and query s...
Hklcm0VYDS
ICLR.cc/2020/Conference
2020
How noise affects the Hessian spectrum in overparameterized neural networks
["Mingwei Wei", "David Schwab"]
Stochastic gradient descent (SGD) forms the core optimization method for deep neural networks. While some theoretical progress has been made, it still remains unclear why SGD leads the learning dynamics in overparameterized networks to solutions that generalize well. Here we show that for overparameterized networks wit...
["noise", "optimization", "loss landscape", "Hessian"]
ABSTRACTStochastic gradient descent (SGD) forms the core optimization method for deepneural networks. While some theoretical progress has been made, it still remainsunclear why SGD leads the learning dynamics in overparameterized networks tosolutions that generalize well. Here we show that for overparameterized network...
BkermuA2tS
Official Blind Review #3
6: Weak Accept
This paper considers the problem of how noise coming from the gradient-based update affects the geometry of the hessian matrix when training a neural network. The paper makes an interesting claim that around a local minimum, if the noise in SGD is aligned with the hessian matrix of the network, then doing SGD update ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title How noise affects the Hessian spectrum in overparameterized neural networks ### Paper Abstract Stochastic gradient descent (SGD) forms the core optimization method for d...
xboZWqM_ELA
ICLR.cc/2021/Conference
2021
Debiased Graph Neural Networks with Agnostic Label Selection Bias
["Shaohua Fan", "Xiao Wang", "Chuan Shi", "Kun Kuang", "Nian Liu", "Bai Wang"]
Most existing Graph Neural Networks (GNNs) are proposed without considering the selection bias in data, i.e., the inconsistent distribution between the training set with test set. In reality, the test data is not even available during the training process, making selection bias agnostic. Training GNNs with biased selec...
["GRAPH NEURAL NETWORKS", "LABEL SELECTION BIAS"]
ABSTRACTMost existing Graph Neural Networks (GNNs) are proposed without consideringthe selection bias in data, i.e., the inconsistent distribution between the trainingset with test set. In reality, the test data is not even available during the trainingprocess, making selection bias agnostic. Training GNNs with biased ...
IDzagb30jvZ
Marginal improvement and inconclusive evaluation
5: Marginally below acceptance threshold
Summary: The authors propose two different regularization terms to help mitigate the effect of label selection bias. The regularizers are well motivated and can be applied to different GNN models. Reasons for score: Overall, I recommend a weak reject. While the theoretical analysis is interesting, the experimental ev...
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 Debiased Graph Neural Networks with Agnostic Label Selection Bias ### Paper Abstract Most existing Graph Neural Networks (GNNs) are proposed without considering the sele...
ryZElGZ0Z
ICLR.cc/2018/Conference
2018
Discovery of Predictive Representations With a Network of General Value Functions
["Matthew Schlegel", "Andrew Patterson", "Adam White", "Martha White"]
The ability of an agent to {\em discover} its own learning objectives has long been considered a key ingredient for artificial general intelligence. Breakthroughs in autonomous decision making and reinforcement learning have primarily been in domains where the agent's goal is outlined and clear: such as playing a game ...
["Reinforcement Learning", "General Value Functions", "Predictive Representations"]
ABSTRACTThe ability of an agent to discover its own learning objectives has long beenconsidered a key ingredient for artificial general intelligence. Breakthroughs inautonomous decision making and reinforcement learning have primarily been indomains where the agent’s goal is outlined and clear: such as playing a game t...
Bk0FsiXbM
Ultimately falls short of convincing
5: Marginally below acceptance threshold
This paper presents a suite of algorithmic ideas to learn a network of generalized value functions. The majority of the technical contribution is dedicated to a new algorithm, AdaGain, that adaptively selects a step size for an RL algorithm; the authors claim this algorithm has sparsifying properties. AdaGain is used...
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 Discovery of Predictive Representations With a Network of General Value Functions ### Paper Abstract The ability of an agent to {\em discover} its own learning objective...
BkeDGJBKvB
ICLR.cc/2020/Conference
2020
Multitask Soft Option Learning
["Maximilian Igl", "Andrew Gambardella", "Jinke He", "Nantas Nardelli", "N. Siddharth", "Wendelin B\u00f6hmer", "Shimon Whiteson"]
We present Multitask Soft Option Learning (MSOL), a hierarchical multi-task framework based on Planning-as-Inference. MSOL extends the concept of Options, using separate variational posteriors for each task, regularized by a shared prior. The learned soft-options are temporally extended, allowing a higher-level master ...
["Hierarchical Reinforcement Learning", "Reinforcement Learning", "Control as Inference", "Options", "Multitask Learning"]
ABSTRACTWe present Multitask Soft Option Learning ( MSOL ), a hierarchical multitaskframework based on Planning as Inference. MSOL extends the concept of options,using separate variational posteriors for each task, regularized by a shared prior.This allows fine-tuning of options for new tasks without forgetting their l...
H1li36yCYS
Official Blind Review #2
3: Weak Reject
Summary: The authors propose a method for learning hierarchical policies in a multi-task setting built on options and casts the concepts into the Planning as Inference framework. They claim that the method brings several advantages such as: -Stabilize end-to-end mtl, -Leads to coordination b/w master policies, -Allows ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Multitask Soft Option Learning ### Paper Abstract We present Multitask Soft Option Learning (MSOL), a hierarchical multi-task framework based on Planning-as-Inference. M...
rywUcQogx
ICLR.cc/2017/conference
2017
Differentiable Canonical Correlation Analysis
["Matthias Dorfer", "Jan Schl\u00fcter", "Gerhard Widmer"]
Canonical Correlation Analysis (CCA) computes maximally-correlated linear projections of two modalities. We propose Differentiable CCA, a formulation of CCA that can be cast as a layer within a multi-view neural network. Unlike Deep CCA, an earlier extension of CCA to nonlinear projections, our formulation enables ...
["Multi-modal learning"]
ABSTRACTCanonical Correlation Analysis (CCA) computes maximally-correlated linear pro-jections of two modalities. We propose Differentiable CCA, a formulation of CCAthat can be cast as a layer within a multi-view neural network. Unlike Deep CCA,an earlier extension of CCA to nonlinear projections, our formulation enabl...
ry-2Cn1Eg
Needs significant work before it can be publishable
3: Clear rejection
The authors propose to combine a CCA objective with a downstream loss. This is a really nice and natural idea. However, both the execution and presentation leave a lot to be desired in the current version of the paper. It is not clear what the overall objective is. This was asked in a pre-review question but the an...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Differentiable Canonical Correlation Analysis ### Paper Abstract Canonical Correlation Analysis (CCA) computes maximally-correlated linear projections of two modalities...
SylR-CEKDS
ICLR.cc/2020/Conference
2020
Modeling question asking using neural program generation
["Ziyun Wang", "Brenden M. Lake"]
People ask questions that are far richer, more informative, and more creative than current AI systems. We propose a neural program generation framework for modeling human question asking, which represents questions as formal programs and generates programs with an encoder-decoder based deep neural network. From extensi...
["question asking", "language generation", "program induction", "reinforcement learning", "density estimation", "cognitive science"]
ABSTRACTPeople ask questions that are far richer, more informative, and more creativethan current AI systems. We propose a neural program generation framework formodeling human question asking, which represents questions as formal programsand generates programs with an encoder-decoder based deep neural network. Fromext...
Hye7F-uXtB
Official Blind Review #1
1: Reject
The authors explore different ways to generate questions about the current state of a “Battleship” game. To do this, they introduce a neural network architecture with a convolutional encoder and a Transformer-based decoder and consider both supervised and reinforcement-learned training approaches. They evaluate the int...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Modeling question asking using neural program generation ### Paper Abstract People ask questions that are far richer, more informative, and more creative than current AI...
7t1FcJUWhi3
ICLR.cc/2021/Conference
2021
Neural Networks for Learning Counterfactual G-Invariances from Single Environments
["S Chandra Mouli", "Bruno Ribeiro"]
Despite —or maybe because of— their astonishing capacity to fit data, neural networks are believed to have difficulties extrapolating beyond training data distribution. This work shows that, for extrapolations based on finite transformation groups, a model’s inability to extrapolate is unrelated to its capacity. Rather...
["Extrapolation", "G-invariance regularization", "Counterfactual inference", "Invariant subspaces"]
ABSTRACTDespite —or maybe because of— their astonishing capacity to fit data, neural net-works are believed to have difficulties extrapolating beyond training data distribu-tion. This work shows that, for extrapolations based on finite transformation groups,a model’s inability to extrapolate is unrelated to its capacit...
O6jKFmLOo2
Seems like a good paper, although I found it hard to follow in places (updated)
7: Good paper, accept
POST REBUTTAL UPDATE: I am increasing my confidence in this paper from 2 to 3 - I still believe the paper can use some more clarity but enough points have been explained and updated in the draft for me to feel more confident in my evaluation. I think the ideas in this paper are quite interesting - for this reason I co...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Neural Networks for Learning Counterfactual G-Invariances from Single Environments ### Paper Abstract Despite —or maybe because of— their astonishing capacity to fit dat...
_MxHo0GHsH6
ICLR.cc/2021/Conference
2021
Once Quantized for All: Progressively Searching for Quantized Compact Models
["Mingzhu Shen", "Feng Liang", "Chuming Li", "Chen Lin", "Ming Sun", "Junjie Yan", "Wanli Ouyang"]
Automatic search of Quantized Neural Networks (QNN) has attracted a lot of attention. However, the existing quantization-aware Neural Architecture Search (NAS) approaches inherit a two-stage search-retrain schema, which is not only time-consuming but also adversely affected by the unreliable ranking of architectures du...
["quantized neural networks", "network architecture search", "compact models"]
ABSTRACTAutomatic search of Quantized Neural Networks (QNN) has attracted a lot ofattention. However, the existing quantization-aware Neural Architecture Search(NAS) approaches inherit a two-stage search-retrain schema, which is not onlytime-consuming but also adversely affected by the unreliable ranking of archi-tectu...
QeHKiPg6_mD
The paper shows promising results, but needs further polishing
4: Ok but not good enough - rejection
Summary: This paper performs a joint optimisation for DNN models, making the NAS scheme is aware of both the quantisation and architectural search spaces. The paper presented a large range of comparisons to different quantisation strategies and ran a lot of experiments to support their claims. However, the writing qua...
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 Once Quantized for All: Progressively Searching for Quantized Compact Models ### Paper Abstract Automatic search of Quantized Neural Networks (QNN) has attracted a lot o...
HkeZQJBKDB
ICLR.cc/2020/Conference
2020
Universal approximations of permutation invariant/equivariant functions by deep neural networks
["Akiyoshi Sannai", "Yuuki Takai", "Matthieu Cordonnier"]
In this paper, we develop a theory about the relationship between $G$-invariant/equivariant functions and deep neural networks for finite group $G$. Especially, for a given $G$-invariant/equivariant function, we construct its universal approximator by deep neural network whose layers equip $G$-actions and each affine t...
["finite group", "invariant", "equivariant", "neural networks"]
ABSTRACTIn this paper, we develop a theory about the relationship between G-invariant/equivariant functions and deep neural networks for finite group G. Es-pecially, for a given G-invariant/equivariant function, we construct its universalapproximator by deep neural network whose layers equip G-actions and each affinetr...
SJlwDP3aYS
Official Blind Review #3
3: Weak Reject
*Paper summary* The authors develop a universal approximation theorem for neural networks that are symmetric with respect to the symmetric group (permutations). They also formally show that the number of free parameters is to train an equivariant network is smaller that the number in a non-equivariant network, leadin...
<|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 approximations of permutation invariant/equivariant functions by deep neural networks ### Paper Abstract In this paper, we develop a theory about the relations...
S1gd7nCcF7
ICLR.cc/2019/Conference
2019
Self-Supervised Generalisation with Meta Auxiliary Learning
["Shikun Liu", "Edward Johns", "Andrew Davison"]
Auxiliary learning has been shown to improve the generalisation performance of a principal task. But typically, this requires manually-defined auxiliary tasks based on domain knowledge. In this paper, we consider that it may be possible to automatically learn these auxiliary tasks to best suit the principal task, towar...
["meta learning", "auxiliary learning", "multi-task learning", "self-supervised learning"]
ABSTRACTAuxiliary learning has been shown to improve the generalisation performanceof a principal task. But typically, this requires manually-defined auxiliary tasksbased on domain knowledge. In this paper, we consider that it may be possi-ble to automatically learn these auxiliary tasks to best suit the principal task...
rJl-0ufypm
The paper is an incremental contribution for an artificially sounding problem
4: Ok but not good enough - rejection
This paper proposes an algorithm for auxiliary learning. Given a target prediction task to be learned on training data, the auxiliary learning utilizes external training data to improve learning. The authors focus on a setup where both target and external training data come from the same distribution but differ in clas...
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 Self-Supervised Generalisation with Meta Auxiliary Learning ### Paper Abstract Auxiliary learning has been shown to improve the generalisation performance of a principal...
SklEhlHtPr
ICLR.cc/2020/Conference
2020
DeepPCM: Predicting Protein-Ligand Binding using Unsupervised Learned Representations
["Paul Kim", "Robin Winter", "Djork-Arn\u00e9 Clevert"]
In-silico protein-ligand binding prediction is an ongoing area of research in computational chemistry and machine learning based drug discovery, as an accurate predictive model could greatly reduce the time and resources necessary for the detection and prioritization of possible drug candidates. Proteochemometric model...
["Unsupervised Representation Learning", "Computational biology", "computational chemistry", "protein-ligand binding"]
ABSTRACTIn-silico protein-ligand binding prediction is an ongoing area of research in com-putational chemistry and machine learning based drug discovery, as an accuratepredictive model could greatly reduce the time and resources necessary for thedetection and prioritization of possible drug candidates. Proteochemometri...
BkxnneP9cS
Official Blind Review #4
3: Weak Reject
The authors present a model with state-of-the-art performance for predicting protein-ligand affinity and provide a thorough set of benchmarks to illustrate the superiority of combining learned low-dimensional embedding representations of both ligands and proteins. The authors then show that these learned representatio...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title DeepPCM: Predicting Protein-Ligand Binding using Unsupervised Learned Representations ### Paper Abstract In-silico protein-ligand binding prediction is an ongoing area o...
Hygm8jC9FQ
ICLR.cc/2019/Conference
2019
FAVAE: SEQUENCE DISENTANGLEMENT USING IN- FORMATION BOTTLENECK PRINCIPLE
["Masanori Yamada", "Kim Heecheol", "Kosuke Miyoshi", "Hiroshi Yamakawa"]
A state-of-the-art generative model, a ”factorized action variational autoencoder (FAVAE),” is presented for learning disentangled and interpretable representations from sequential data via the information bottleneck without supervision. The purpose of disentangled representation learning is to obtain interpretable and...
["disentangled representation learning"]
ABSTRACTWe propose the factorized action variational autoencoder (FA V AE), a state-of-the-art generative model for learning disentangled and interpretable representationsfrom sequential data via the information bottleneck without supervision. The pur-pose of disentangled representation learning is to obtain interpreta...
Syl0S_Iihm
Interesting and valuable model extension, but with rather early results
6: Marginally above acceptance threshold
This paper proposes an extension to VAE to model sequential datasets and extract disentangled representations of their evolution. It consists of a straight extension of CCI-VAE (Burgess et al 2018) to accept sequential data, combining it with a Ladder VAE architecture (VLAE, Zhao et al 2017). They show early results on...
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 FAVAE: SEQUENCE DISENTANGLEMENT USING IN- FORMATION BOTTLENECK PRINCIPLE ### Paper Abstract A state-of-the-art generative model, a ”factorized action variational autoenc...
S1eU0KxE_4
ICLR.cc/2019/Workshop/drlStructPred
2019
A Study of State Aliasing in Structured Prediction with RNNs
["Layla El Asri", "Adam Trischler"]
End-to-end reinforcement learning agents learn a state representation and a policy at the same time. Recurrent neural networks (RNNs) have been trained successfully as reinforcement learning agents in settings like dialogue that require structured prediction. In this paper, we investigate the representations learned by...
["deep reinforcement learning", "structured prediction", "dialogue"]
ABSTRACTEnd-to-end reinforcement learning agents learn a state representation and a pol-icy at the same time. Recurrent neural networks (RNNs) have been trained suc-cessfully as reinforcement learning agents in settings like dialogue that requirestructured prediction. In this paper, we investigate the representations l...
S1efcAsUK4
Questions regarding the simple maze
1: Strong rejection
The authors investigate state aliasing of the learned representations of RNNs trained with policy gradient methods. They start with a simple maze example and claim to show that state aliasing occurs when several states share the same optimal action and the agent is trained via policy gradient. They find that this does ...
2: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title A Study of State Aliasing in Structured Prediction with RNNs ### Paper Abstract End-to-end reinforcement learning agents learn a state representation and a policy at the...
BklpOo09tQ
ICLR.cc/2019/Conference
2019
EFFICIENT TWO-STEP ADVERSARIAL DEFENSE FOR DEEP NEURAL NETWORKS
["Ting-Jui Chang", "Yukun He", "Peng Li"]
In recent years, deep neural networks have demonstrated outstanding performancein many machine learning tasks. However, researchers have discovered that thesestate-of-the-art models are vulnerable to adversarial examples: legitimate examples added by small perturbations which are unnoticeable to human eyes. Adversaria...
["Adversarial Examples", "Adversarial Training", "FGSM", "IFGSM", "Robustness"]
ABSTRACTIn recent years, deep neural networks have demonstrated outstanding performancein many machine learning tasks. However, researchers have discovered that thesestate-of-the-art models are vulnerable to adversarial examples: legitimate exam-ples added by small perturbations which are unnoticeable to human eyes. Ad...
HkeoqNhshQ
Missing any theoretical justification, but encouraging empirical result
6: Marginally above acceptance threshold
The paper introduces a two-step adversarial defense method, to generate two adversarial examples per clean sample and include them in the actual training loop to achieve robustness. The main claim is that the proposed two-step scheme can outperform more expensive iterative methods such as IFGSM; hence achieving robustn...
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 TWO-STEP ADVERSARIAL DEFENSE FOR DEEP NEURAL NETWORKS ### Paper Abstract In recent years, deep neural networks have demonstrated outstanding performancein many...
B1ePui0ctQ
ICLR.cc/2019/Conference
2019
SnapQuant: A Probabilistic and Nested Parameterization for Binary Networks
["Kuan Wang", "Hao Zhao", "Anbang Yao", "Aojun Zhou", "Dawei Sun", "Yurong Chen"]
In this paper, we study the problem of training real binary weight networks (without layer-wise or filter-wise scaling factors) from scratch under the Bayesian deep learning perspective, meaning that the final objective is to approximate the posterior distribution of binary weights rather than reach a point estimation....
["Binary weight networks", "neural network quantization", "reinforcement learning"]
ABSTRACTIn this paper, we study the problem of training real binary weight networks (withoutlayer-wise or filter-wise scaling factors) from scratch under the Bayesian deeplearning perspective, meaning that the final objective is to approximate the posteriordistribution of binary weights rather than reach a point estima...
BJxuov7q3X
Bayesian view on deep network binarization - interesting idea but lacks in clarity and experiments
4: Ok but not good enough - rejection
The paper treats network binarization as learning a bernoulli probability that each weight is 0 or 1. It uses a hierarchical approach to factor the probability in to parameters shared between weights/kernels/layers. The result is a network that can be used to sample a binary network. The main novelty is that full-preci...
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 SnapQuant: A Probabilistic and Nested Parameterization for Binary Networks ### Paper Abstract In this paper, we study the problem of training real binary weight networks...
r1CE9GWR-
ICLR.cc/2018/Conference
2018
Understanding GANs: the LQG Setting
["Soheil Feizi", "Changho Suh", "Fei Xia", "David Tse"]
Generative Adversarial Networks (GANs) have become a popular method to learn a probability model from data. Many GAN architectures with different optimization metrics have been introduced recently. Instead of proposing yet another architecture, this paper aims to provide an understanding of some of the basic issues su...
["Generative Adversarial Networks", "Wasserstein", "Generalization", "PCA"]
ABSTRACTGenerative Adversarial Networks (GANs) have become a popular method to learna probability model from data. Many GAN architectures with different optimiza-tion metrics have been introduced recently. Instead of proposing yet another ar-chitecture, this paper aims to provide an understanding of some of the basic i...
r1WP0c2bz
LQG setting reduces the GAN optimization to essentially PCA and I believe is too simple to give insights into more complex GANs.
5: Marginally below acceptance threshold
Summary: This paper studies GANs in the following LQG setting: Input data distribution (P_Y) is Gaussian with zero mean and Identity covariance. Loss function is quadratic. Generator is also considered to be a Gaussian distribution (linear function of the input Gaussian noise). The paper considers two settings for dis...
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 GANs: the LQG Setting ### Paper Abstract Generative Adversarial Networks (GANs) have become a popular method to learn a probability model from data. Many G...
rkgT3jRct7
ICLR.cc/2019/Conference
2019
Large-Scale Answerer in Questioner's Mind for Visual Dialog Question Generation
["Sang-Woo Lee", "Tong Gao", "Sohee Yang", "Jaejun Yoo", "Jung-Woo Ha"]
Answerer in Questioner's Mind (AQM) is an information-theoretic framework that has been recently proposed for task-oriented dialog systems. AQM benefits from asking a question that would maximize the information gain when it is asked. However, due to its intrinsic nature of explicitly calculating the information gain, ...
["questioner", "mind", "answerer", "aqm", "question", "information gain", "dialog", "error"]
ABSTRACTAnswerer in Questioner’s Mind (AQM) is an information-theoretic frameworkthat has been recently proposed for task-oriented dialog systems. AQM benefitsfrom asking a question that would maximize the information gain when it is asked.However, due to its intrinsic nature of explicitly calculating the information g...
rJg0MNRKhX
Contributions seem incremental and concerns regarding the formulated approach
6: Marginally above acceptance threshold
The paper proposes an improvement over the AQM approach for an information-theoretic framework for task-oriented dialog systems. Specifically, the paper tries to circumvent the problem of explicitly calculating the information gain while asking a question in the AQM setting. While the original AQM approach sweeps over ...
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 Large-Scale Answerer in Questioner's Mind for Visual Dialog Question Generation ### Paper Abstract Answerer in Questioner's Mind (AQM) is an information-theoretic framew...
PDLPdWHdp-h
ICLR.cc/2021/Conference
2021
Understanding Adversarial Attacks on Autoencoders
["Elsa Riachi", "Frank Rudzicz"]
Adversarial vulnerability is a fundamental limitation of deep neural networks which remains poorly understood. Recent work suggests that adversarial attacks on deep neural network classifiers exploit the fact that non-robust models rely on superficial statistics to form predictions. While such feat...
["adversarial attacks", "autoencoders", "predictive", "low reconstruction error", "adversarial", "pairs", "autoencoders adversarial vulnerability", "fundamental limitation", "deep neural networks", "understood"]
ABSTRACTAdversarial vulnerability is a fundamental limitation of deep neural networkswhich remains poorly understood. Recent work suggests that adversarial attackson deep neural network classifiers exploit the fact that non-robust models relyon superficial statistics to form predictions. While such features are semanti...
9VhPh_pl7rx
Interesting topic, hard to follow
5: Marginally below acceptance threshold
The paper studies the predictive characteristics of adversarial attack on autoencoders. It is shown that the perturbation is related to the sparse representation of input. While to conclude this proposal more future works are required, it opens up an interesting topic in this field. However, I have some comments: - The...
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 Understanding Adversarial Attacks on Autoencoders ### Paper Abstract Adversarial vulnerability is a fundamental limitation of deep neural networks which remains...
HHiiQKWsOcV
ICLR.cc/2021/Conference
2021
Explaining the Efficacy of Counterfactually Augmented Data
["Divyansh Kaushik", "Amrith Setlur", "Eduard H Hovy", "Zachary Chase Lipton"]
In attempts to produce machine learning models less reliant on spurious patterns in NLP datasets, researchers have recently proposed curating counterfactually augmented data (CAD) via a human-in-the-loop process in which given some documents and their (initial) labels, humans must revise the text to make a counterfactu...
["humans in the loop", "annotation artifacts", "text classification", "sentiment analysis", "natural language inference"]
ABSTRACTIn attempts to produce machine learning models less reliant on spurious patterns inNLP datasets, researchers have recently proposed curating counterfactually aug-mented data (CAD) via a human-in-the-loop process in which given some docu-ments and their (initial) labels, humans must revise the text to make a cou...
nffkXY8T7e6
Nice paper
7: Good paper, accept
This paper presents a perspective on counterfactually augmented data (CAD) and experiments providing evidence that CAD elicited from humans indeed bears resemblance to interventions on labels or variables that are causally related to the labels. This is already the stated intuition behind CAD and why CAD may help with...
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 Explaining the Efficacy of Counterfactually Augmented Data ### Paper Abstract In attempts to produce machine learning models less reliant on spurious patterns in NLP dat...
9-uyJRiORwh
eswc-conferences.org/ESWC/2021/Workshop/KGCW
2021
Integrating nested data into knowledge graphs with RML fields
["Thomas Delva", "Dylan Van Assche", "Pieter Heyvaert", "Ben De Meester", "Anastasia Dimou"]
To support business decisions or improve operational efficiency, heterogeneous data is often integrated into a knowledge graph. This inte- gration can be achieved with one of the existing declarative mapping lan- guages, which offer declarative data integration in the form of knowledge graphs. However, current mapping ...
["RML", "nested data", "KGC challenges"]
Integrating nested data into knowledge graphswith RML fieldsThomas Delva( )[0000000195212185], Dylan Van Assche[0000000271959935], PieterHeyvaert[0000000215835719], Ben De Meester[0000000302480987], and AnastasiaDimou( )[0000000321387972]IDLab, Department of Electronics and Information Systems, Ghent University - imecT...
jQucqquAnxp
Good and interesting proposal with some weak points coming from the lack of implementation
7: Good paper, accept
This article presents a proposal called RML fields in the form of a new "syntax" for RML. Fields and nested fields are presented as a way to solve specific problems when dealing with hierarchical data and, therefore, solve 5 of the open challenges in the W3C Knowledge Graph Construction Community Group. The strongest ...
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 Integrating nested data into knowledge graphs with RML fields ### Paper Abstract To support business decisions or improve operational efficiency, heterogeneous data is o...
HyEeMu_xx
ICLR.cc/2017/conference
2017
Progressive Attention Networks for Visual Attribute Prediction
["Paul Hongsuck Seo", "Zhe Lin", "Scott Cohen", "Xiaohui Shen", "Bohyung Han"]
We propose a novel attention model which can accurately attend to target objects of various scales and shapes in images. The model is trained to gradually suppress irrelevant regions in an input image via a progressive attentive process over multiple layers of a convolutional neural network. The attentive process in ea...
["Deep learning", "Computer vision", "Multi-modal learning"]
ABSTRACTWe propose a novel attention model which can accurately attend to target objectsof various scales and shapes in images. The model is trained to gradually suppressirrelevant regions in an input image via a progressive attentive process over multiplelayers of a convolutional neural network. The attentive process ...
HJKt06-Ng
Review
7: Good paper, accept
The paper presents an architecture to incrementally attend to image regions - at multiple layers of a deep CNN. In contrast to most other models, the model does not apply a weighted average pooling in the earlier layers of the network but only in the last layer. Instead, the features are reweighted in each layer with 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 Progressive Attention Networks for Visual Attribute Prediction ### Paper Abstract We propose a novel attention model which can accurately attend to target objects of var...
k2TyMLwuikx
ICLR.cc/2021/Conference
2021
Model-Free Energy Distance for Pruning DNNs
["Mohammadreza Soltani", "Suya Wu", "Yuerong Li", "Jie Ding", "Vahid Tarokh"]
We propose a novel method for compressing Deep Neural Networks (DNNs) with competitive performance to state-of-the-art methods. We measure a new model-free information between the feature maps and the output of the network. Model-freeness of our information measure guarantees that no parametric assumptions on the featu...
["Pruning", "Residual Networks", "Structured Method", "Energy Distance"]
ABSTRACTWe propose a novel method for compressing Deep Neural Networks (DNNs)with competitive performance to state-of-the-art methods. We measure a newmodel-free information between the feature maps and the output of the network.Model-freeness of our information measure guarantees that no parametric as-sumptions on the...
uliM4WRi0I
Model-Free Energy Distance for Pruning DNNs
5: Marginally below acceptance threshold
This paper introduces a metric called "energy distance" in DNN pruning to evaluate the relationship between feature maps and the output. However, the rationale behind this metric is not clear. The definition and discussions of this metric in Section 3.2 are inconsistent with the way it is used in Section 3.3, making 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 Model-Free Energy Distance for Pruning DNNs ### Paper Abstract We propose a novel method for compressing Deep Neural Networks (DNNs) with competitive performance to stat...
MLSvqIHRidA
ICLR.cc/2021/Conference
2021
Contrastive Divergence Learning is a Time Reversal Adversarial Game
["Omer Yair", "Tomer Michaeli"]
Contrastive divergence (CD) learning is a classical method for fitting unnormalized statistical models to data samples. Despite its wide-spread use, the convergence properties of this algorithm are still not well understood. The main source of difficulty is an unjustified approximation which has been used to derive the...
["Unsupervised learning", "energy based model", "adversarial learning", "contrastive divergence", "noise contrastive estimation"]
ABSTRACTContrastive divergence (CD) learning is a classical method for fitting unnormal-ized statistical models to data samples. Despite its wide-spread use, the conver-gence properties of this algorithm are still not well understood. The main sourceof difficulty is an unjustified approximation which has been used to d...
yNU4iv7jxr
Establishes an interesting theoretical link
7: Good paper, accept
Summary To implement the contrastive divergence (CD) algorithm in practice, an intractable term is typically omitted from the gradient. This leads to an approximation. This work shows that the resulting algorithm can in fact be viewed as an exact algorithm targeting a different, adversarial objective. The derivation 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 Contrastive Divergence Learning is a Time Reversal Adversarial Game ### Paper Abstract Contrastive divergence (CD) learning is a classical method for fitting unnormalize...
HygUOoC5KX
ICLR.cc/2019/Conference
2019
Are Generative Classifiers More Robust to Adversarial Attacks?
["Yingzhen Li", "John Bradshaw", "Yash Sharma"]
There is a rising interest in studying the robustness of deep neural network classifiers against adversaries, with both advanced attack and defence techniques being actively developed. However, most recent work focuses on discriminative classifiers, which only model the conditional distribution of the labels given the ...
["generative models", "adversarial attack", "defence", "detection", "Bayes' rule"]
ABSTRACTThere is a rising interest in studying the robustness of deep neural network classi-fiers against adversaries, with both advanced attack and defence techniques beingactively developed. However, most recent work focuses on discriminative classi-fiers, which only model the conditional distribution of the labels g...
B1ePMbbcnQ
Interesting direction but some important prior work is missing and evaluation does not yet convincingly support claims
4: Ok but not good enough - rejection
In this work the authors propose and analyse generative models as defences against adversarial examples. In addition, three detection methods are introduced and an extension to deep features is suggested. My main concerns are as follows (details below): * Important prior work is not mentioned. * Evaluation with direct...
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 Are Generative Classifiers More Robust to Adversarial Attacks? ### Paper Abstract There is a rising interest in studying the robustness of deep neural network classifier...
hsFN92eQEla
ICLR.cc/2021/Conference
2021
EVALUATION OF NEURAL ARCHITECTURES TRAINED WITH SQUARE LOSS VS CROSS-ENTROPY IN CLASSIFICATION TASKS
["Like Hui", "Mikhail Belkin"]
Modern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the square loss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of stand...
["large scale learning", "square loss vs cross-entropy", "classification", "experimental evaluation"]
ABSTRACTModern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the squareloss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of...
FM-XMSDc2rT
The paper shows that the square loss is usually as effective as the cross-entropy loss across classification tasks in a variety of domains and model architectures.
7: Good paper, accept
This paper questions the omnipresence of cross-entropy loss for classification tasks while showing that square loss also yields competitive results. The experimental section spans a wide variety of tasks and architectures covering NLP, ASR, and Vision. The authors find that the model performance is more stable w.r.t. t...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title EVALUATION OF NEURAL ARCHITECTURES TRAINED WITH SQUARE LOSS VS CROSS-ENTROPY IN CLASSIFICATION TASKS ### Paper Abstract Modern neural architectures for classification ta...
SygQvs0cFQ
ICLR.cc/2019/Conference
2019
Variational Smoothing in Recurrent Neural Network Language Models
["Lingpeng Kong", "Gabor Melis", "Wang Ling", "Lei Yu", "Dani Yogatama"]
We present a new theoretical perspective of data noising in recurrent neural network language models (Xie et al., 2017). We show that each variant of data noising is an instance of Bayesian recurrent neural networks with a particular variational distribution (i.e., a mixture of Gaussians whose weights depend on statis...
["data", "variational smoothing", "new theoretical perspective", "xie et", "variant", "instance", "particular variational distribution"]
ABSTRACTWe present a new theoretical perspective of data noising in recurrent neural networklanguage models (Xie et al., 2017). We show that each variant of data noising is aninstance of Bayesian recurrent neural networks with a particular variational distribu-tion (i.e., a mixture of Gaussians whose weights depend on ...
H1x17T9eTX
Interesting paper.
7: Good paper, accept
The paper presents a Bayesian/ variational interpretation of data noising in recurrent networks (Xie et al. 2017). Overall I found the paper interesting and well presented. The authors first review the work of Xie et al. 2017, that proposes data noisy for regularizing recurrent networks. This is done by randomly repla...
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 Variational Smoothing in Recurrent Neural Network Language Models ### Paper Abstract We present a new theoretical perspective of data noising in recurrent neural network...
ryeYpJSKwr
ICLR.cc/2020/Conference
2020
Meta-Learning Acquisition Functions for Transfer Learning in Bayesian Optimization
["Michael Volpp", "Lukas P. Fr\u00f6hlich", "Kirsten Fischer", "Andreas Doerr", "Stefan Falkner", "Frank Hutter", "Christian Daniel"]
Transferring knowledge across tasks to improve data-efficiency is one of the open key challenges in the field of global black-box optimization. Readily available algorithms are typically designed to be universal optimizers and, therefore, often suboptimal for specific tasks. We propose a novel transfer learning method ...
["Transfer Learning", "Meta Learning", "Bayesian Optimization", "Reinforcement Learning"]
ABSTRACTTransferring knowledge across tasks to improve data-efficiency is one of the openkey challenges in the field of global black-box optimization. Readily availablealgorithms are typically designed to be universal optimizers and, therefore, oftensuboptimal for specific tasks. We propose a novel transfer learning me...
S1xux0YRKr
Official Blind Review #1
8: Accept
This paper proposes a framework for meta learning neural acquisition functions for the Bayesian optimization of various underivable functions. The neural acquisition functions are learned using proximal policy optimization in an outer loop on different problems on the same domain, and the learned acquisition function ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Meta-Learning Acquisition Functions for Transfer Learning in Bayesian Optimization ### Paper Abstract Transferring knowledge across tasks to improve data-efficiency is o...
B1nxTzbRZ
ICLR.cc/2018/Conference
2018
Forward Modeling for Partial Observation Strategy Games - A StarCraft Defogger
["Gabriel Synnaeve", "Zeming Lin", "Jonas Gehring", "Vasil Khalidov", "Nicolas Carion", "Nicolas Usunier"]
This paper we present a defogger, a model that learns to predict future hidden information from partial observations. We formulate this model in the context of forward modeling and leverage spatial and sequential constraints and correlations via convolutional neural networks and long short-term memory networks, respect...
["forward modeling", "partially observable", "deep learning", "strategy game", "real-time strategy"]
ABSTRACTThis paper we present a defogger , a model that learns to predict future hiddeninformation from partial observations. We formulate this model in the context offorward modeling and leverage spatial and sequential constraints and correlationsvia convolutional neural networks and long short-term memory networks, r...
Sy46KdXfM
Review
5: Marginally below acceptance threshold
# Summary This paper introduces a new prediction problem where the model should predict the hidden opponent's state as well as the agent's state. This paper presents a neural network architecture which takes the map information and several other features and reconstructs the unit occupancy and count information in 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 Forward Modeling for Partial Observation Strategy Games - A StarCraft Defogger ### Paper Abstract This paper we present a defogger, a model that learns to predict future...
qdBih7UIx3
MIDL.io/2020/Conference
2020
A hierarchical fusion framework integrating random projection-based classifiers: application in head and neck squamous carcinoma cancer
["Tianlan Mo", "Chao Zheng"]
Ensemble methods achieves better performance than single classifier model. Classifier diversity and fusion architecture are equally important for building a successful multi-classifier system. In this study, we introduced random projection to obtain the required classifier diversity and then proposed a hierarchical fra...
["ensemble method", "random projection", "fusion architecture", "ensemble diversity"]
A hierarchical fusion framework to integrate random projectionbased classifiers: application in head and neck squamous carcinoma cancer Abstract: Ensemble methods achieves better performance than single classifier model. Classifier diversity and fusion arc hitecture are equally important for building a successful multi...
esZ0u3ttc
Ensemble fusion approach for combining classifiers
1: Strong reject
- Introduction is a summary of ensemble learning, rather than what the problem is with existing methods that needs to be solved. This means the solution is not well motivated or clear as to the reasoning. - Weird formatting issues in the paper, unclear notation. - Not clear what the different ensemble methods are - Cli...
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 hierarchical fusion framework integrating random projection-based classifiers: application in head and neck squamous carcinoma cancer ### Paper Abstract Ensemble metho...
rJlHIo09KQ
ICLR.cc/2019/Conference
2019
Gradient-based Training of Slow Feature Analysis by Differentiable Approximate Whitening
["Merlin Sch\u00fcler", "Hlynur Dav\u00ed\u00f0 Hlynsson", "Laurenz Wiskott"]
We propose Power Slow Feature Analysis, a gradient-based method to extract temporally slow features from a high-dimensional input stream that varies on a faster time-scale, as a variant of Slow Feature Analysis (SFA). While displaying performance comparable to hierarchical extensions to the SFA algorithm, such as Hiera...
["Slow Feature Analysis", "Deep Learning", "Spectral Embedding", "Temporal Coherence"]
ABSTRACTWe propose Power Slow Feature Analysis, a gradient-based method to extracttemporally slow features from a high-dimensional input stream that varies on afaster time-scale, as a variant of Slow Feature Analysis (SFA). While displayingperformance comparable to hierarchical extensions to the SFA algorithm, such asH...
BJlEOF32n7
Gradient-based SFA
6: Marginally above acceptance threshold
In this paper the authors present a differentiable objective for slow feature analysis, to facilitate end-to-end training. I am not clear on the novelty of this formulation, as it appears to have been proposed in a similar form in previous works (e.g., A maximum-likelihood interpretation for slow feature analysis by ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Gradient-based Training of Slow Feature Analysis by Differentiable Approximate Whitening ### Paper Abstract We propose Power Slow Feature Analysis, a gradient-based meth...
rkltE0VKwH
ICLR.cc/2020/Conference
2020
Coordinated Exploration via Intrinsic Rewards for Multi-Agent Reinforcement Learning
["Shariq Iqbal", "Fei Sha"]
Solving tasks with sparse rewards is one of the most important challenges in reinforcement learning. In the single-agent setting, this challenge has been addressed by introducing intrinsic rewards that motivate agents to explore unseen regions of their state spaces. Applying these techniques naively to the multi-agent ...
["multi-agent reinforcement learning", "multi-agent", "exploration", "intrinsic motivation", "MARL", "coordinated exploration"]
ABSTRACTSolving tasks with sparse rewards is one of the most important challenges in rein-forcement learning. In the single-agent setting, this challenge has been addressedby introducing intrinsic rewards that motivate agents to explore unseen regionsof their state spaces. Applying these techniques naively to the multi...
rJeoQlwTKB
Official Blind Review #2
3: Weak Reject
Overall I like the approach in the paper. It proposes a nice 2 pronged method for exploiting exploration via intrinsic rewards for multi-agent systems. The parts that a bit lacking with the current version of the paper in this are the evaluation tasks are few and a bit simple and I think there needs to be more discussi...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Coordinated Exploration via Intrinsic Rewards for Multi-Agent Reinforcement Learning ### Paper Abstract Solving tasks with sparse rewards is one of the most important ch...
ryZERzWCZ
ICLR.cc/2018/Conference
2018
The Information-Autoencoding Family: A Lagrangian Perspective on Latent Variable Generative Modeling
["Shengjia Zhao", "Jiaming Song", "Stefano Ermon"]
A variety of learning objectives have been recently proposed for training generative models. We show that many of them, including InfoGAN, ALI/BiGAN, ALICE, CycleGAN, VAE, $\beta$-VAE, adversarial autoencoders, AVB, and InfoVAE, are Lagrangian duals of the same primal optimization problem. This generalization reveals t...
["Generative Models", "Variational Autoencoder", "Generative Adversarial Network"]
ABSTRACTA variety of learning objectives have been recently proposed for training gener-ative models. We show that many of them, including InfoGAN, ALI/BiGAN,ALICE, CycleGAN, V AE, -V AE, adversarial autoencoders, A VB, and InfoV AE,are Lagrangian duals of the same primal optimization problem. This generaliza-tion reve...
S1ufxZqlG
Not clear what specific insights exist or what problem this solves
4: Ok but not good enough - rejection
EDIT: I have read the authors' rebuttals and other reviews. My opinion has not been changed. I recommend the authors significantly revise their work, streamlining the narrative and making clear what problems and solutions they solve. While I enjoy the perspective of unifying various paths, it's unclear what insights co...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title The Information-Autoencoding Family: A Lagrangian Perspective on Latent Variable Generative Modeling ### Paper Abstract A variety of learning objectives have been recent...
ryxMX2R9YQ
ICLR.cc/2019/Conference
2019
CGNF: Conditional Graph Neural Fields
["Tengfei Ma", "Cao Xiao", "Junyuan Shang", "Jimeng Sun"]
Graph convolutional networks have achieved tremendous success in the tasks of graph node classification. These models could learn a better node representation through encoding the graph structure and node features. However, the correlation between the node labels are not considered. In this paper, we propose a novel ar...
["graph neural networks", "energy models", "conditional random fields", "label correlation"]
ABSTRACTGraph convolutional networks have achieved tremendous success in the tasks ofgraph node classification. These models could learn a better node representationthrough encoding the graph structure and node features. However, the correlationbetween the node labels are not considered. In this paper, we propose a nov...
Syx6o21R2Q
Novelty is incremental.
5: Marginally below acceptance threshold
This paper proposes a conditional graph neural network to explore the label correlation. In fact, this method combines graph convolutional neural network and CRF together. The novelty is incremental. No new insight is provided toward this topic. For experiments, how did you train DeepWalk (Node2vec)? By using all nod...
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 CGNF: Conditional Graph Neural Fields ### Paper Abstract Graph convolutional networks have achieved tremendous success in the tasks of graph node classification. These m...
A-3YCvm5Gl
NeurIPS.cc/2022/Workshop/Offline_RL
2022
Offline Reinforcement Learning for Customizable Visual Navigation
["Dhruv Shah", "Arjun Bhorkar", "Hrishit Leen", "Ilya Kostrikov", "Nicholas Rhinehart", "Sergey Levine"]
Robotic navigation often requires not only reaching a distant goal, but also satisfying intermediate user preferences on the path, such as obeying the rules of the road or preferring some surfaces over others. Our goal in this paper is to devise a robotic navigation system that can utilize previously collect data to le...
["robotics", "reinforcement learning"]
Offline Reinforcement Learning forCustomizable Visual NavigationDhruv Shah†, Arjun Bhorkar†, Hrishit Leen, Ilya Kostrikov, Nick Rhinehart, Sergey LevineUC BerkeleyAbstract: Robotic navigation often requires not only reaching a distant goal, butalso satisfying intermediate user preferences on the path, such as obeying t...
JU-7Rz8xgL
Official Review
6: Marginally above acceptance threshold
The authors propose a combination of offline, goal-based Reinforcement Learning and topological planning for learning visual navigation of robots in the real world. Given some utility functions expressing human navigation preferences (e.g. staying in sunlight or in the grass), different reward functions are constructed...
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 Offline Reinforcement Learning for Customizable Visual Navigation ### Paper Abstract Robotic navigation often requires not only reaching a distant goal, but also satisfy...
7AQUzh5ntX_
ICLR.cc/2021/Conference
2021
Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in First-person Simulated 3D Environments
["Wilka Torrico Carvalho", "Anthony Liang", "Kimin Lee", "Sungryull Sohn", "Honglak Lee", "Richard Lewis", "Satinder Singh"]
First-person object-interaction tasks in high-fidelity, 3D, simulated environments such as the AI2Thor virtual home-environment pose significant sample-efficiency challenges for reinforcement learning (RL) agents learning from sparse task rewards. To alleviate these challenges, prior work has provided extensive supervi...
["object-centric", "representation learning", "reinforcement learning", "sparse reward"]
ABSTRACTFirst-person object-interaction tasks in high-fidelity, 3D, simulated environmentssuch as the AI2Thor virtual home-environment pose significant sample-efficiencychallenges for reinforcement learning (RL) agents learning from sparse task re-wards. To alleviate these challenges, prior work has provided extensive ...
_BA1JiSslW
Paper achieves good results on the chosen tasks. Why were these tasks the ones chosen?
6: Marginally above acceptance threshold
Summary ========================================== This paper describes a way learn an attention mechanism for objects in a 3D environment without explicit supervision. This attention mechanism allows the proposed network, ROMA, to outperform existing approaches without having access to the ground truth state of the ob...
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 Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in First-person Simulated 3D Environments ### Paper Abstract First-person object-interaction tasks in h...
By4ryDOUf
ICLR.cc/2018/Workshop
2018
Evaluation of Deep Reinforcement Learning Methods for Modular Robots
["Risto Kojcev", "Nora Etxezarreta", "Alejandro Hern\u00e1ndez", "V\u00edctor Mayoral"]
We propose a novel framework for Deep Reinforcement Learning (DRL) in modular robotics using traditional robotic tools that extend state-of-the-art DRL implementations and provide an end-to-end approach which trains a robot directly from joint states. Moreover, we present a novel technique to transfer these DLR methods...
["Robotics", "modular robots", "simulation", "reinforcement learning", "simulation framework"]
ABSTRACTWe propose a novel framework for Deep Reinforcement Learning (DRL) in mod-ular robotics using traditional robotic tools that extend state-of-the-art DRL imple-mentations and provide an end-to-end approach which trains a robot directly fromjoint states. Moreover, we present a novel technique to transfer these DL...
B1iILbbYf
Promise of comparison between DRL methods, but nowhere to be found
3: Clear rejection
- The task in this paper appears extremely simple (reach a fixed point). This begs the question if the generalization to modular robots with more modules (3 DOF to 4 DOF) is valid? It seems that the task is just too simple and can be solved with both setups. It would be more interesting to see a task for which the perf...
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 Evaluation of Deep Reinforcement Learning Methods for Modular Robots ### Paper Abstract We propose a novel framework for Deep Reinforcement Learning (DRL) in modular rob...
SyOK1Sg0W
ICLR.cc/2018/Conference
2018
Adaptive Quantization of Neural Networks
["Soroosh Khoram", "Jing Li"]
Despite the state-of-the-art accuracy of Deep Neural Networks (DNN) in various classification problems, their deployment onto resource constrained edge computing devices remains challenging due to their large size and complexity. Several recent studies have reported remarkable results in reducing this complexity throug...
["Deep Neural Networks", "Model Quantization", "Model Compression"]
ABSTRACTDespite the state-of-the-art accuracy of Deep Neural Networks (DNN) in variousclassification problems, their deployment onto resource constrained edge comput-ing devices remains challenging due to their large size and complexity. Severalrecent studies have reported remarkable results in reducing this complexity...
rkIpwEslz
The paper proposes a method for quantizing neural networks that allows weights to be quantized with different precision depending on their importance
6: Marginally above acceptance threshold
I have read the responses to the concerns raised by all reviewers. I find the clarifications and modifications satisfying, therefore I keep my rating of the paper to above acceptance threshold. ----------------- ORIGINAL REVIEW: The paper proposes a method for quantizing neural networks that allows weights to be quan...
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 Adaptive Quantization of Neural Networks ### Paper Abstract Despite the state-of-the-art accuracy of Deep Neural Networks (DNN) in various classification problems, their...
SkhU2fcll
ICLR.cc/2017/conference
2017
Deep Multi-task Representation Learning: A Tensor Factorisation Approach
["Yongxin Yang", "Timothy M. Hospedales"]
Most contemporary multi-task learning methods assume linear models. This setting is considered shallow in the era of deep learning. In this paper, we present a new deep multi-task representation learning framework that learns cross-task sharing structure at every layer in a deep network. Our approach is based on genera...
["deep", "representation learning", "tensor factorisation", "representation", "contemporary", "methods", "linear models", "setting", "shallow", "era"]
ABSTRACTMost contemporary multi-task learning methods assume linear models. This set-ting is considered shallow in the era of deep learning. In this paper, we presenta new deep multi-task representation learning framework that learns cross-tasksharing structure at every layer in a deep network . Our approach is based o...
B17a61-Vg
8: Top 50% of accepted papers, clear accept
This paper proposed a deep multi-task representation learning framework that learns cross-task sharing structure at every layer in a deep network with tensor factorization and end-to-end knowledge sharing. This approach removed the requirement of a user-defined multi-task sharing strategy in conventional approach. Their...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Deep Multi-task Representation Learning: A Tensor Factorisation Approach ### Paper Abstract Most contemporary multi-task learning methods assume linear models. This sett...
S1xI_TEtwS
ICLR.cc/2020/Conference
2020
Amata: An Annealing Mechanism for Adversarial Training Acceleration
["Nanyang Ye", "Qianxiao Li", "Zhanxing Zhu"]
Despite of the empirical success in various domains, it has been revealed that deep neural networks are vulnerable to maliciously perturbed input data that much degrade their performance. This is known as adversarial attacks. To counter adversarial attacks, adversarial training formulated as a form of robust optimizati...
["mechanism", "adversarial attacks", "adversarial training", "pgd", "amata", "adversarial training acceleration", "empirical success", "various domains", "deep neural networks"]
ABSTRACTDespite of the empirical success in various domains, it has been revealed that deepneural networks are vulnerable to maliciously perturbed input data that much de-grade their performance. This is known as adversarial attacks. To counter ad-versarial attacks, adversarial training formulated as a form of robust o...
BJgaBtsAKH
Official Blind Review #3
6: Weak Accept
This paper proposed an annealing mechanism for PGD adversarial training in Madry et al. This mechanism gradually reduces the step size and increases the number of iterations of PGD maximization. The authors claim that the proposed method can achieve comparable performance as the original PGD adversarial training with l...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Amata: An Annealing Mechanism for Adversarial Training Acceleration ### Paper Abstract Despite of the empirical success in various domains, it has been revealed that dee...
Ud3DSz72nYR
ICLR.cc/2021/Conference
2021
Contrastive Explanations for Reinforcement Learning via Embedded Self Predictions
["Zhengxian Lin", "Kin-Ho Lam", "Alan Fern"]
We investigate a deep reinforcement learning (RL) architecture that supports explaining why a learned agent prefers one action over another. The key idea is to learn action-values that are directly represented via human-understandable properties of expected futures. This is realized via the embedded self-prediction (ES...
["Explainable AI", "Deep Reinforcement Learning"]
ABSTRACTWe investigate a deep reinforcement learning (RL) architecture that supports ex-plaining why a learned agent prefers one action over another. The key idea is tolearn action-values that are directly represented via human-understandable proper-ties of expected futures. This is realized via the embedded self-predi...
NNbbsJSpEf-
ESP appears to rely on careful GVF feature engineering; further information/evaluation needed
7: Good paper, accept
Edit: I have read the authors' response as well as the other reviews. Based on the additional results and added feature selection details, I now agree that ESP is generally applicable. Summary: The authors present ESP, an RL system that can then explain action choices in terms of future feature values. Generalize...
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 Contrastive Explanations for Reinforcement Learning via Embedded Self Predictions ### Paper Abstract We investigate a deep reinforcement learning (RL) architecture that ...
Sklw_kHtPH
ICLR.cc/2020/Conference
2020
AdamT: A Stochastic Optimization with Trend Correction Scheme
["Bingxin Zhou", "Xuebin Zheng", "Junbin Gao"]
Adam-typed optimizers, as a class of adaptive moment estimation methods with the exponential moving average scheme, have been successfully used in many applications of deep learning. Such methods are appealing for capability on large-scale sparse datasets. On top of that, they are computationally efficient and insensit...
["Optimization", "ADAM", "Stochastic Gradient Descent", "Deep Learning"]
ABSTRACTAdam-typed optimizers, as a class of adaptive moment estimation methods withthe exponential moving average scheme, have been successfully used in manyapplications of deep learning. Such methods are appealing for capability on large-scale sparse datasets. On top of that, they are computationally efficient and in...
r1lP-mqOsH
Official Blind Review #4
3: Weak Reject
Overall I think there is a room for this paper to improve. I expect to see more comprehensive experiments for comparing Adam and AdamT. The authors applied AdamT to four different tasks and compared training and testing loss difference. Are the models used in these tasks state-of-the-art? I did not see a discussion abo...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title AdamT: A Stochastic Optimization with Trend Correction Scheme ### Paper Abstract Adam-typed optimizers, as a class of adaptive moment estimation methods with the exponen...
YgrdmztE4OY
ICLR.cc/2021/Conference
2021
Federated Mixture of Experts
["Matthias Reisser", "Christos Louizos", "Efstratios Gavves", "Max Welling"]
Federated learning (FL) has emerged as the predominant approach for collaborative training of neural network models across multiple users, without the need to gather the data at a central location. One of the important challenges in this setting is data heterogeneity; different users have different data characteristics...
["Federated Learning", "personalized models", "non-i.i.d data"]
ABSTRACTFederated learning (FL) has emerged as the predominant approach for collabora-tive training of neural network models across multiple users, without the need togather the data at a central location. One of the important challenges in this set-ting is data heterogeneity; different users have different data charac...
u6K_L3Z-e0D
Interesting approach, but subpar results and needs more clarification
4: Ok but not good enough - rejection
-- Summary -- The paper proposes a method for federated learning of a mixture of experts model (FedMix). The approach allows training an ensemble of models each of which specializes to a subset of clients with similar data characteristics. The authors argue that this way of training an ensemble reduces the gradient div...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Federated Mixture of Experts ### Paper Abstract Federated learning (FL) has emerged as the predominant approach for collaborative training of neural network models acros...
SJTB5GZCb
ICLR.cc/2018/Conference
2018
Extending the Framework of Equilibrium Propagation to General Dynamics
["Benjamin Scellier", "Anirudh Goyal", "Jonathan Binas", "Thomas Mesnard", "Yoshua Bengio"]
The biological plausibility of the backpropagation algorithm has long been doubted by neuroscientists. Two major reasons are that neurons would need to send two different types of signal in the forward and backward phases, and that pairs of neurons would need to communicate through symmetric bidirectional connections. ...
["Deep Learning", "Backpropagation", "Fixed Point Recurrent Neural Network", "Biologically Plausible Learning", "Feedback Alignment", "Dynamical System", "Gradient-Free Optimization"]
ABSTRACTThe biological plausibility of the backpropagation algorithm has long beendoubted by neuroscientists. Two major reasons are that neurons would need tosend two different types of signal in the forward and backward phases, and thatpairs of neurons would need to communicate through symmetric bidirectionalconnectio...
HJqm1ilzG
Review
6: Marginally above acceptance threshold
The paper proposes a new learning algorithm for learning neural networks that may be biologically plausible. The paper builds upon the paper by Scellier & Bengio but doesn't assume symmetric weights. I couldn't judge how solid the biological plausibility argument but my understanding is that there is no universal agree...
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 Extending the Framework of Equilibrium Propagation to General Dynamics ### Paper Abstract The biological plausibility of the backpropagation algorithm has long been doub...
4hA23Eld-HU
ICLR.cc/2021/Conference
2021
Learning to Control on the Fly
["Zhanzhan Zhao"]
This paper proposes an algorithm which learns to control on the fly. The proposed algorithm has no access to the transition law of the environment, which is actually linear with bounded random noise, and learns to make decisions directly online without training phases or sub-optimal policies as the initial input. Neith...
["online decision making", "convergence", "regret bound", "bounded random noise"]
ABSTRACTThis paper proposes an algorithm which learns to control on the fly. The pro-posed algorithm has no access to the transition law of the environment, which isactually linear with bounded random noise, and learns to make decisions directlyonline without training phases or sub-optimal policies as the initial input...
cTV-fU4vXrX
An ellipsoidal approach to adaptive control of linear systems; but the results and the presentation are premature
4: Ok but not good enough - rejection
**General overview:** The paper studies fully observable stochastic linear systems with unknown state dynamics (but known input dynamics) and bounded noises. The goal is to drive the system to a small neighborhood of a target state, which is assumed to be the zero vector. The paper applies a standard quadratic loss fun...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning to Control on the Fly ### Paper Abstract This paper proposes an algorithm which learns to control on the fly. The proposed algorithm has no access to the transi...
BJxTjcLk5V
icaps-conference.org/ICAPS/2019/Workshop/WIPC
2019
Cost-Optimal Planning in IPC 2018: Symbolic Search and Pattern Databases vs. Portfolio Planning
["Anonymous"]
The optimal track is one the most exiting events in the International Planning Competition (IPC). In this position paper we argue that —despite of not winning the competition— symbolic search and pattern databases were likely the most influential planning approaches in the latest IPC in 2018, and, in continuation to t...
["Symbolic Search", "Planning Pattern Database", "Portfolio Planning"]
Cost-Optimal Planning in the IPC 2018:Symbolic Search and Planning Pattern Databases vs. Portfolio PlanningStefan Edelkamp andIonut MoraruKing’s College Londonfstefan.edelkamp,ionut.moraru g@kcl.ac.ukAbstractThe optimal track is one the most exiting events in the Inter-national Planning Competition (IPC).In this positi...
SJlaWAs6q4
Poorly executed analysis of the state-of-the-art in optimal classical planning following IPC 2018
3: Clear rejection
This paper states that it "a) provides a clearer picture on what is the currently leading technology according to IPC 2018 and b) discusses whether or not portfolios help to push or blur the outcome of a competition". I will discuss both of these points separately in the following. a) The paper claims that symbolic se...
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 Cost-Optimal Planning in IPC 2018: Symbolic Search and Pattern Databases vs. Portfolio Planning ### Paper Abstract The optimal track is one the most exiting events in th...
YCXrx6rRCXO
ICLR.cc/2021/Conference
2021
Faster Binary Embeddings for Preserving Euclidean Distances
["Jinjie Zhang", "Rayan Saab"]
We propose a fast, distance-preserving, binary embedding algorithm to transform a high-dimensional dataset $\mathcal{T}\subseteq\mathbb{R}^n$ into binary sequences in the cube $\{\pm 1\}^m$. When $\mathcal{T}$ consists of well-spread (i.e., non-sparse) vectors, our embedding method applies a stable noise-shaping quanti...
["Binary Embeddings", "Johnson-Lindenstrauss Transforms", "Sigma Delta Quantization"]
ABSTRACTWe propose a fast, distance-preserving, binary embedding algorithm to transforma high-dimensional dataset T Rninto binary sequences in the cube f1gm.WhenTconsists of well-spread (i.e., non-sparse) vectors, our embedding methodapplies a stable noise-shaping quantization scheme to Axwhere A2Rmnisa sparse Gaussian...
BTs5EfXjjdx
A solid, but fairly incremental contribution
7: Good paper, accept
This paper proposes a scheme for fast distance-preserving binary embeddings of R^n into {-1/+1}^m. Guarantees are given for embedding l2-distance in the host space. Under certain assumptions about the input data both encoding and decoding can be done in O(m) time/space, which improves substantially upon the O(n log 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 Faster Binary Embeddings for Preserving Euclidean Distances ### Paper Abstract We propose a fast, distance-preserving, binary embedding algorithm to transform a high-dim...
SklcyJBtvB
ICLR.cc/2020/Conference
2020
Off-policy Bandits with Deficient Support
["Noveen Sachdeva", "Yi Su", "Thorsten Joachims"]
Off-policy training of contextual-bandit policies is attractive in online systems (e.g. search, recommendation, ad placement), since it enables the reuse of large amounts of log data from the production system. State-of-the-art methods for off-policy learning, however, are based on inverse propensity score (IPS) weight...
["Recommender System", "Search Engine", "Counterfactual Learning"]
ABSTRACTOff-policy training of contextual-bandit policies is attractive in online systems(e.g. search, recommendation, ad placement), since it enables the reuse of largeamounts of log data. State-of-the-art methods for off-policy learning, however, arebased on inverse propensity score (IPS) weighting, which requires th...
BygT4d5oYr
Official Blind Review #3
3: Weak Reject
This paper talks about the problem of off-policy or batch learning in the contextual bandit setting without the complete support assumption. This problem setting is very realistic and encountered in most problems, especially in temporally extended settings, such as reinforcement learning. They compare three approaches ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Off-policy Bandits with Deficient Support ### Paper Abstract Off-policy training of contextual-bandit policies is attractive in online systems (e.g. search, recommendati...
OtAnbr1OQAW
ICLR.cc/2021/Conference
2021
Diverse Exploration via InfoMax Options
["Yuji Kanagawa", "Tomoyuki Kaneko"]
In this paper, we study the problem of autonomously discovering temporally abstracted actions, or options, for exploration in reinforcement learning. For learning diverse options suitable for exploration, we introduce the infomax termination objective defined as the mutual information between options and their correspo...
["Reinforcement Learning", "Hierachical Reinforcement Learning", "Exploration"]
ABSTRACTIn this paper, we study the problem of autonomously discovering temporally ab-stracted actions, or options, for exploration in reinforcement learning. For learningdiverse options suitable for exploration, we introduce the infomax termination ob-jective defined as the mutual information between options and their...
OvFyNpef_y
Review
5: Marginally below acceptance threshold
Review for "Diverse Exploration via Infomax Options" Thank you for your feedback. I found that the analysis of the method in the revision is informative. However, the comparison with the baselines is still lacking, and the experiments are only performed in simple environments. For these reasons, I keep my rating unch...
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 Diverse Exploration via InfoMax Options ### Paper Abstract In this paper, we study the problem of autonomously discovering temporally abstracted actions, or options, for...
qdBih7UIx3
MIDL.io/2020/Conference
2020
A hierarchical fusion framework integrating random projection-based classifiers: application in head and neck squamous carcinoma cancer
["Tianlan Mo", "Chao Zheng"]
Ensemble methods achieves better performance than single classifier model. Classifier diversity and fusion architecture are equally important for building a successful multi-classifier system. In this study, we introduced random projection to obtain the required classifier diversity and then proposed a hierarchical fra...
["ensemble method", "random projection", "fusion architecture", "ensemble diversity"]
A hierarchical fusion framework to integrate random projectionbased classifiers: application in head and neck squamous carcinoma cancer Abstract: Ensemble methods achieves better performance than single classifier model. Classifier diversity and fusion arc hitecture are equally important for building a successful multi...
Nu0maTYCh
The paper presents and ensemble method for carcinoma cancer classification. Valid paper. Does not pass threshold.
2: Weak reject
The idea is valid. And random projections are one method for stratified bootstrap sampling of datasets. Technically correct bu this paper falls significantly below the threshold of this conference. It is a passable workshop paper. The significance is rather weak. The evaluations reported in the table are the same 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 A hierarchical fusion framework integrating random projection-based classifiers: application in head and neck squamous carcinoma cancer ### Paper Abstract Ensemble metho...
HJxaC1rKDS
ICLR.cc/2020/Conference
2020
Imbalanced Classification via Adversarial Minority Over-sampling
["Jaehyung Kim", "Jongheon Jeong", "Jinwoo Shin"]
In most real-world scenarios, training datasets are highly class-imbalanced, where deep neural networks suffer from generalizing to a balanced testing criterion. In this paper, we explore a novel yet simple way to alleviate this issue via synthesizing less-frequent classes with adversarial examples of other classes. Su...
["imbalanced classification", "adversarial examples"]
ABSTRACTIn most real-world scenarios, training datasets are highly class-imbalanced, wheredeep neural networks suffer from generalizing to a balanced testing criterion. Inthis paper, we explore a novel yet simple way to alleviate this issue via synthe-sizing less-frequent classes with adversarial examples of other clas...
BkgN1MWT9B
Official Blind Review #3
3: Weak Reject
The authors propose to tackle imbalance classification using re-sampling methods. The idea is to generate adversarial examples of a model trained on the imbalanced dataset, in the least frequent classes. To the best of my understanding the rational behind this is that models usually overfit the least frequent class. ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Imbalanced Classification via Adversarial Minority Over-sampling ### Paper Abstract In most real-world scenarios, training datasets are highly class-imbalanced, where de...
Bk7wvW-C-
ICLR.cc/2018/Conference
2018
Exploring Asymmetric Encoder-Decoder Structure for Context-based Sentence Representation Learning
["Shuai Tang", "Hailin Jin", "Chen Fang", "Zhaowen Wang", "Virginia R. de Sa"]
Context information plays an important role in human language understanding, and it is also useful for machines to learn vector representations of language. In this paper, we explore an asymmetric encoder-decoder structure for unsupervised context-based sentence representation learning. As a result, we build an encoder...
["asymmetric structure", "RNN-CNN", "fast", "unsupervised", "representation", "sentence"]
ABSTRACTContext information plays an important role in human language understanding,and it is also useful for machines to learn vector representations of language. Inthis paper, we explore an asymmetric encoder-decoder structure for unsupervisedcontext-based sentence representation learning. As a result, we build an en...
Skrfq_Jlz
Good results, but tons of minor issues
6: Marginally above acceptance threshold
Update: I'm going to change my review to a 6 to acknowledge the substantial improvements you've made—I no longer fear that there are major errors in the paper, but this paper is still solidly borderline, and I'm not completely convinced that any new claim is true. The evidence presented for the main claim—that you can...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Exploring Asymmetric Encoder-Decoder Structure for Context-based Sentence Representation Learning ### Paper Abstract Context information plays an important role in human...
e6hMkY6MFcU
ICLR.cc/2021/Conference
2021
WordsWorth Scores for Attacking CNNs and LSTMs for Text Classification
["Nimrah Shakeel"]
Black box attacks on traditional deep learning models trained for text classifica- tion target important words in a piece of text, in order to change model prediction. Current approaches towards highlighting important features are time consuming and require large number of model queries. We present a simple yet novel m...
["cnns", "lstms", "scores", "word importance", "wordsworth scores", "text", "single words"]
ABSTRACTBlack box attacks on traditional deep learning models trained for text classifica-tion target important words in a piece of text, in order to change model prediction.Current approaches towards highlighting important features are time consumingand require large number of model queries. We present a simple yet no...
mg4XZRnLvW
The paper should be improved before submission
2: Strong rejection
The paper is poorly written. This is especially the case for the proposed algorithm (the core contribution). This section is very difficult to understand, and notations are awkward. Everything is a bit messy. However, the experimental results are quite well presented (to be compared with the beginning of the paper).
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title WordsWorth Scores for Attacking CNNs and LSTMs for Text Classification ### Paper Abstract Black box attacks on traditional deep learning models trained for text classifi...
L-88RyVtXGr
ICLR.cc/2021/Conference
2021
Learning Deeply Shared Filter Bases for Efficient ConvNets
["Woochul Kang", "Daeyeon Kim"]
Recently, inspired by repetitive block structure of modern ConvNets, such as ResNets, parameter-sharing among repetitive convolution layers has been proposed to reduce the size of parameters. However, naive sharing of convolution filters poses many challenges such as overfitting and vanishing/exploding gradients, res...
["Deep learning", "ConvNets", "parameter sharing", "model compression", "convolutional neural networks", "recursive networks"]
ABSTRACTRecently, inspired by repetitive block structure of modern ConvNets, such asResNets, parameter-sharing among repetitive convolution layers has been proposedto reduce the size of parameters. However, naive sharing of convolution filtersposes many challenges such as overfitting and vanishing/exploding gradients,r...
awBMMQLPaCE
Nice solution to decrease computation complexity in CNNs. Needs more work on results section (writing + experiments)
6: Marginally above acceptance threshold
[Summary] This method proposes to decompose convolutional filters using a low rank filter basis where the convolutional operation in a layer consists of shareable filter basis and non-shareable layer coefficients. This is developped to save computational costs whilst maintaining performance. To regularise against vanis...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning Deeply Shared Filter Bases for Efficient ConvNets ### Paper Abstract Recently, inspired by repetitive block structure of modern ConvNets, such as ResNets, par...
SJlEs1HKDr
ICLR.cc/2020/Conference
2020
Attentive Sequential Neural Processes
["Jaesik Yoon", "Gautam Singh", "Sungjin Ahn"]
Sequential Neural Processes (SNP) is a new class of models that can meta-learn a temporal stochastic process of stochastic processes by modeling temporal transition between Neural Processes. As Neural Processes (NP) suffers from underfitting, SNP is also prone to the same problem, even more severely due to its temporal...
["meta-learning", "neural processes", "attention", "sequential modeling"]
ABSTRACTSequential Neural Processes (SNP) is a new class of models that temporally ex-tends Neural Processes (NP) and can meta-learn a sequence of stochastic pro-cesses. This learned function however under-fits the provided contexts as is alsothe case in NP. Applying attention to the contexts resolves this in NP but si...
rkgjX7waFr
Official Blind Review #1
1: Reject
This paper combines ideas from attentive and sequential neural processes to incorporate an attention mechanism to the existing sequential neural process, which results in an attentive sequential neural processes framework. While the idea is somewhat interesting, I think this paper is technically vague and not well-mot...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Attentive Sequential Neural Processes ### Paper Abstract Sequential Neural Processes (SNP) is a new class of models that can meta-learn a temporal stochastic process of ...
SylGpT4FPS
ICLR.cc/2020/Conference
2020
Last-iterate convergence rates for min-max optimization
["Jacob Abernethy", "Kevin A. Lai", "Andre Wibisono"]
While classic work in convex-concave min-max optimization relies on average-iterate convergence results, the emergence of nonconvex applications such as training Generative Adversarial Networks has led to renewed interest in last-iterate convergence guarantees. Proving last-iterate convergence is challenging because ma...
["min-max optimization", "zero-sum game", "saddle point", "last-iterate convergence", "non-asymptotic convergence", "global rates", "Hamiltonian", "sufficiently bilinear"]
ABSTRACTWhile classic work in convex-concave min-max optimization relies on average-iterate convergence results, the emergence of nonconvex applications such astraining Generative Adversarial Networks has led to renewed interest in last-iterateconvergence guarantees. Proving last-iterate convergence is challenging beca...
H1evEoR6YH
Official Blind Review #3
6: Weak Accept
Summary: The paper, considers methods for solving smooth unconstrained min-max optimization problems. In particular, the authors prove that the Hamiltonian Gradient Descent (HGD) algorithm converges with linear convergence rate to the min-max solution. One of the main contributions of this work is that the proposed a...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Last-iterate convergence rates for min-max optimization ### Paper Abstract While classic work in convex-concave min-max optimization relies on average-iterate convergenc...
Hkl9JlBYvr
ICLR.cc/2020/Conference
2020
VariBAD: A Very Good Method for Bayes-Adaptive Deep RL via Meta-Learning
["Luisa Zintgraf", "Kyriacos Shiarlis", "Maximilian Igl", "Sebastian Schulze", "Yarin Gal", "Katja Hofmann", "Shimon Whiteson"]
Trading off exploration and exploitation in an unknown environment is key to maximising expected return during learning. A Bayes-optimal policy, which does so optimally, conditions its actions not only on the environment state but on the agent’s uncertainty about the environment. Computing a Bayes-optimal policy is how...
["Meta-Learning", "Bayesian Reinforcement Learning", "BAMDPs", "Deep Reinforcement Learning"]
ABSTRACTTrading off exploration and exploitation in an unknown environment is key tomaximising expected return during learning. A Bayes-optimal policy, which doesso optimally, conditions its actions not only on the environment state but on theagent’s uncertainty about the environment. Computing a Bayes-optimal policyis...
SyxKPjfroH
Official Blind Review #4
8: Accept
*EDIT: Score increased after discussion with authors clarified many concerns raised below* Summary: This paper presents an algorithmic approach toward learning Bayes Optimal policies under the uncertainty of the environment. Leveraging meta-learning, the proposed variBAD approximate inference procedure is capable of a...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title VariBAD: A Very Good Method for Bayes-Adaptive Deep RL via Meta-Learning ### Paper Abstract Trading off exploration and exploitation in an unknown environment is key to ...
SkaxnKEYg
ICLR.cc/2017/workshop
2017
Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols
["Serhii Havrylov", "Ivan Titov"]
Learning to communicate through interaction, rather than relying on explicit supervision, is often considered a prerequisite for developing a general AI. We study a setting where two agents engage in playing a referential game and, from scratch, develop a communication protocol necessary to succeed in this game. We req...
["Natural language processing", "Deep learning", "Multi-modal learning", "Games"]
ABSTRACTLearning to communicate through interaction, rather than relying on explicit su-pervision, is often considered a prerequisite for developing a general AI. We studya setting where two agents engage in playing a referential game and, from scratch,develop a communication protocol necessary to succeed in this game....
SyHbXgGig
7: Good paper, accept
Applying the straight-through Gumbel-softmax estimator for end to end differentiable sequence generation is likely an early instance of a trick we will see used more broadly. This has also not been applied to this task before with previous approaches either being continuous or RL. That the process is the same during tr...
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 Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols ### Paper Abstract Learning to communicate through interaction, rather th...
Hke4l2AcKQ
ICLR.cc/2019/Conference
2019
MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders
["Xuezhe Ma", "Chunting Zhou", "Eduard Hovy"]
Variational Autoencoder (VAE), a simple and effective deep generative model, has led to a number of impressive empirical successes and spawned many advanced variants and theoretical investigations. However, recent studies demonstrate that, when equipped with expressive generative distributions (aka. decoders), VAE suff...
["VAE", "regularization", "auto-regressive"]
ABSTRACTVariational Autoencoder (V AE), a simple and effective deep generative model, hasled to a number of impressive empirical successes and spawned many advancedvariants and theoretical investigations. However, recent studies demonstrate that,when equipped with expressive generative distributions (aka. decoders), V ...
SJec11dj3X
Interesting paper with marginal results
7: Good paper, accept
This paper proposes changes to the ELBO loss used to train VAEs, to avoid posterior collapse. They motivate their additional components rather differently than what has been done in the literature so far, which I found quite interesting. They compare against appropriate baselines, on MNIST and OMNIGLOT, in a complete w...
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 MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders ### Paper Abstract Variational Autoencoder (VAE), a simple and effective deep generative mod...
ckeT8cMz_A
robot-learning.org/CoRL/2023/Conference
2023
REBOOT: Reuse Data for Bootstrapping Efficient Real-World Dexterous Manipulation
["Zheyuan Hu", "Aaron Rovinsky", "Jianlan Luo", "Vikash Kumar", "Abhishek Gupta", "Sergey Levine"]
Dexterous manipulation tasks involving contact-rich interactions pose a significant challenge for both model-based control systems and imitation learning algorithms. The complexity arises from the need for multi-fingered robotic hands to dynamically establish and break contacts, balance forces on the non-prehensile obj...
["Dexterous Manipulation", "Reinforcement Learning", "Sample-Efficient RL"]
REBOOT: Reuse Data for Bootstrapping EfficientReal-World Dexterous ManipulationZheyuan Hu1∗, Aaron Rovinsky1∗, Jianlan Luo1, Vikash Kumar2, Abhishek Gupta3, Sergey Levine11UC Berkeley2Meta AI Research3University of WashingtonFigure 1: REBOOT achieves 2Xsample efficiency boost on learning a variety of contact-rich real-...
hUqu6xGeY5
This work presents an interesting and practical exploration, confirming that leveraging prior non-task specific data can significantly enhance skill-learning efficiency in real-world dexterous manipulation tasks. One notable limitation is its confinement to image-based domains, where different tasks share the same obs...
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 REBOOT: Reuse Data for Bootstrapping Efficient Real-World Dexterous Manipulation ### Paper Abstract Dexterous manipulation tasks involving contact-rich interactions pose...
H1-IBSgMz
ICLR.cc/2018/Conference
2018
A Matrix Approximation View of NCE that Justifies Self-Normalization
["Jacob Goldberger", "Oren Melamud"]
Self-normalizing discriminative models approximate the normalized probability of a class without having to compute the partition function. This property is useful to computationally-intensive neural network classifiers, as the cost of computing the partition function grows linearly with the number of classes and may be...
["language modeling", "NCE", "self-normalization"]
ABSTRACTSelf-normalizing discriminative models approximate the normalized probability ofa class without having to compute the partition function. This property is useful tocomputationally-intensive neural network classifiers, as the cost of computing thepartition function grows linearly with the number of classes and m...
Hkfimc-Mf
Self-normalization results from NCE being a low rank approximation
6: Marginally above acceptance threshold
The paper presents a proof of the self normalization of NCE as a result of being a low-rank matrix approximation of low-rank approximation of the normalized conditional probabilities matrix. However, it seems that in equation 4, the authors assume that the noise distribution is a unigram model over words. However, one...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title A Matrix Approximation View of NCE that Justifies Self-Normalization ### Paper Abstract Self-normalizing discriminative models approximate the normalized probability of...
rcQdycl0zyk
ICLR.cc/2021/Conference
2021
Beyond Fully-Connected Layers with Quaternions: Parameterization of Hypercomplex Multiplications with $1/n$ Parameters
["Aston Zhang", "Yi Tay", "SHUAI Zhang", "Alvin Chan", "Anh Tuan Luu", "Siu Hui", "Jie Fu"]
Recent works have demonstrated reasonable success of representation learning in hypercomplex space. Specifically, “fully-connected layers with quaternions” (quaternions are 4D hypercomplex numbers), which replace real-valued matrix multiplications in fully-connected layers with Hamilton products of quaternions, both en...
["hypercomplex representation learning"]
ABSTRACTRecent works have demonstrated reasonable success of representation learningin hypercomplex space. Specifically, “fully-connected layers with quaternions”(quaternions are 4D hypercomplex numbers), which replace real-valued matrixmultiplications in fully-connected layers with Hamilton products of quaternions,bot...
cYxBqAPaOuI
Interesting idea with practical benefits
8: Top 50% of accepted papers, clear accept
The authors propose a novel way of parametrizing hypercomplex multiplications. The proposed parametrization helps with: (a) Generalizing the multiplication to arbitrary dimensions, and (b) Reducing the number of parameters. Building on this, the authors propose a parameterized hypercomplex multiplication (PHM) layer...
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 Fully-Connected Layers with Quaternions: Parameterization of Hypercomplex Multiplications with $1/n$ Parameters ### Paper Abstract Recent works have demonstrated ...
r1lZgyBYwS
ICLR.cc/2020/Conference
2020
HiLLoC: lossless image compression with hierarchical latent variable models
["James Townsend", "Thomas Bird", "Julius Kunze", "David Barber"]
We make the following striking observation: fully convolutional VAE models trained on 32x32 ImageNet can generalize well, not just to 64x64 but also to far larger photographs, with no changes to the model. We use this property, applying fully convolutional models to lossless compression, demonstrating a method to scale...
["compression", "variational inference", "lossless compression", "deep latent variable models"]
ABSTRACTWe make the following striking observation: fully convolutional V AE modelstrained on 3232 ImageNet can generalize well, not just to 64 64 but also to farlarger photographs, with no changes to the model. We use this property, applyingfully convolutional models to lossless compression, demonstrating a method tos...
BygJ9qoy9S
Official Blind Review #1
6: Weak Accept
The authors propose a method for lossless image compression based on using fully convolutional VAE models. These models are shown to generalize well when they are trained on small images (e.g. 32x32 and 64x64) and then applied to much larger images. The method is based on a fully vectorized implementation of bits back ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title HiLLoC: lossless image compression with hierarchical latent variable models ### Paper Abstract We make the following striking observation: fully convolutional VAE models...
Cs2DZT6icMd
NeurIPS.cc/2022/Workshop/Federated_Learning
2022
Federated Continual Learning to Detect Accounting Anomalies in Financial Auditing
["Marco Schreyer", "Hamed Hemati", "Damian Borth", "Miklos A. Vasarhelyi"]
The International Standards on Auditing (ISA) require auditors to collect reasonable assurance that financial statements are free of material misstatement. At the same time, a central objective of Continuous Assurance is the real-time assessment of digital accounting journal entries. Recently, driven by the advances in...
["federated learning", "continual learning", "tabular data", "financial auditing", "anomaly detection"]
Federated Continual Learning toDetect Accounting Anomalies in Financial AuditingMarco Schreyer1,2Hamed Hemati1Damian Borth1Miklos A. Vasarhelyi21University of St.Gallen, Switzerland2Rutgers, The State University of New Jersey, USA{firstname.lastname}@unisg.ch ,{firstname.lastname}@rutgers.eduAbstractThe International S...
tAAuR7ReRC6
The paper addresses a relevant topic. However, the descriptions of the proposed method and the structure of proposed method are not well introduced.
This paper proposes an FCL method to detect anomalies in financial auditing. The paper addresses a relevant topic. And the case studies are well presented. However, the descriptions of the proposed method and the structure of proposed method are not well introduced. Besides, detailed contributions in the paper are less...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Federated Continual Learning to Detect Accounting Anomalies in Financial Auditing ### Paper Abstract The International Standards on Auditing (ISA) require auditors to co...
v7zNWgyirXX
NeurIPS.cc/2021/Workshop/DLDE
2021
Fitting Regularized Population Dynamics with Neural Differential Equations
["David Calhas", "Rui Henriques"]
Neural differential equations (neural DEs) are yet to see success in its application as interpretable autoencoders/descriptors, where they directly model a population of signals with the learned vector field. In this manuscript, we show that there is a threshold to which these models capture the dynamics of a populatio...
["Neural differential equations", "Representational regularization", "Population dynamics", "Regression"]
Fitting Regularized Population Dynamics withNeural Differential EquationsDavid CalhasInstituto Superior TécnicoINESC-IDRui HenriquesInstituto Superior TécnicoINESC-IDAbstractNeural differential equations (neural DEs) are yet to see success in its applicationas interpretable autoencoders/descriptors, where they direct...
ADksf8lzhU-
Interesting work, but a lot more explanation needed
This work proposes and auto-encoder + neural ODE in latent space in order to represent population dynamics. The encoder consists of an attention mechanism with 2 heads, and the decoder is fully connected. There is an additional cost function in latent space which is aimed controlling what the authors call the dyna...
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 Fitting Regularized Population Dynamics with Neural Differential Equations ### Paper Abstract Neural differential equations (neural DEs) are yet to see success in its ap...
SkgQBn0cF7
ICLR.cc/2019/Conference
2019
Modeling the Long Term Future in Model-Based Reinforcement Learning
["Nan Rosemary Ke", "Amanpreet Singh", "Ahmed Touati", "Anirudh Goyal", "Yoshua Bengio", "Devi Parikh", "Dhruv Batra"]
In model-based reinforcement learning, the agent interleaves between model learning and planning. These two components are inextricably intertwined. If the model is not able to provide sensible long-term prediction, the executed planer would exploit model flaws, which can yield catastrophic failures. This paper focus...
["model-based reinforcement learning", "variation inference"]
ABSTRACTIn model-based reinforcement learning, the agent interleaves between modellearning and planning. These two components are inextricably intertwined. Ifthe model is not able to provide sensible long-term prediction, the executed plan-ner would exploit model flaws, which can yield catastrophic failures. This paper...
Skg9apeK2X
Interesting approach; not sure if really scales to long horizon problems
6: Marginally above acceptance threshold
The paper introduces an interesting approach to model learning for imitation and RL. Given the problem of maintaining multi-step predictions in the context of sequential decision making process, and deficiencies faced during planning with one-step models [1][2], it’s imperative to explore approaches that do multi-step ...
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 Modeling the Long Term Future in Model-Based Reinforcement Learning ### Paper Abstract In model-based reinforcement learning, the agent interleaves between model learnin...
BJgbzhC5Ym
ICLR.cc/2019/Conference
2019
NECST: Neural Joint Source-Channel Coding
["Kristy Choi", "Kedar Tatwawadi", "Tsachy Weissman", "Stefano Ermon"]
For reliable transmission across a noisy communication channel, classical results from information theory show that it is asymptotically optimal to separate out the source and channel coding processes. However, this decomposition can fall short in the finite bit-length regime, as it requires non-trivial tuning of hand-...
["joint source-channel coding", "deep generative models", "unsupervised learning"]
ABSTRACTFor reliable transmission across a noisy communication channel, classical resultsfrom information theory show that it is asymptotically optimal to separate out thesource and channel coding processes. However, this decomposition can fall shortin the finite bit-length regime, as it requires non-trivial tuning of ...
HkgDn_1T3X
Review
4: Ok but not good enough - rejection
Summary of paper: For the finite-bit case of the noisy communication channel model, it is suboptimal to optimize source coding (compression of input) and error correction (fault tolerance for inherent noise in the channel) separately. The authors propose a neural network model (NECST) that is very similar to the standa...
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 NECST: Neural Joint Source-Channel Coding ### Paper Abstract For reliable transmission across a noisy communication channel, classical results from information theory sh...
yT7-k6Q6gda
ICLR.cc/2021/Conference
2021
Catastrophic Fisher Explosion: Early Phase Fisher Matrix Impacts Generalization
["Stanislaw Kamil Jastrzebski", "Devansh Arpit", "Oliver \u00c5strand", "Giancarlo Kerg", "Huan Wang", "Caiming Xiong", "richard socher", "Kyunghyun Cho", "Krzysztof J. Geras"]
The early phase of training has been shown to be important in two ways for deep neural networks. First, the degree of regularization in this phase significantly impacts the final generalization. Second, it is accompanied by a rapid change in the local loss curvature influenced by regularization choices. Connecting thes...
["early phase of training", "implicit regularization", "SGD", "learning rate", "batch size", "Hessian", "Fisher Information Matrix", "curvature", "gradient norm"]
ABSTRACTThe early phase of training has been shown to be important in two ways for deepneural networks. First, the degree of regularization in this phase significantlyimpacts the final generalization. Second, it is accompanied by a rapid change inthe local loss curvature influenced by regularization choices. Connecting...
p4w7XJw71W2
Misleading and Even Wrong Writing
5: Marginally below acceptance threshold
# Summary This work claims SGD regularizes the model through penalizing the trace of Fisher in the early phase. This claim is supported by the similar generalization behavior of SGD (with optimal learning rate) and Fisher penalization (for SGD with small learning rate). A series experiments are conducted to verify the ...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Catastrophic Fisher Explosion: Early Phase Fisher Matrix Impacts Generalization ### Paper Abstract The early phase of training has been shown to be important in two ways...
r1gIdySFPH
ICLR.cc/2020/Conference
2020
Skew-Fit: State-Covering Self-Supervised Reinforcement Learning
["Vitchyr H. Pong", "Murtaza Dalal", "Steven Lin", "Ashvin Nair", "Shikhar Bahl", "Sergey Levine"]
Autonomous agents that must exhibit flexible and broad capabilities will need to be equipped with large repertoires of skills. Defining each skill with a manually-designed reward function limits this repertoire and imposes a manual engineering burden. Self-supervised agents that set their own goals can automate this pr...
["deep reinforcement learning", "goal space", "goal conditioned reinforcement learning", "self-supervised reinforcement learning", "goal sampling", "reinforcement learning"]
ABSTRACTAutonomous agents that must exhibit flexible and broad capabilities will need tobe equipped with large repertoires of skills. Defining each skill with a manually-designed reward function limits this repertoire and imposes a manual engineeringburden. Self-supervised agents that set their own goals can automate t...
BylJ3YZntB
Official Blind Review #2
6: Weak Accept
This paper introduced a very interesting idea to facilitate exploration in goal-conditioned reinforcement learning. The key idea is to learn a generative model of goal distribution to match the weighted empirical distribution, where the rare states receive larger weights. This encourages the model to generate more dive...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Skew-Fit: State-Covering Self-Supervised Reinforcement Learning ### Paper Abstract Autonomous agents that must exhibit flexible and broad capabilities will need to be eq...
hpZXQ6VVFxo
CUHK.edu.hk/2021/Course/IERG5350
2021
Deploying Reinforcement Learning in Water Transport
["Donald P. H. Wong", "Tsz Kui Chow"]
In this project, we deployed various types of reinforcement learning algorithms to resolves the rewards maximization problem in water sailing by reaching the destination with the highest priority using the smallest number of steps. We have implemented our own environment for this project and trained different agents us...
["steps", "reinforcement learning", "water transport", "project", "destination", "environment", "policy iteration", "value iteration", "dqn", "various types"]
Deploying Reinforcement Learning in WaterTransportPak Hei WongDepartment of Information EngineeringThe Chinese University of Hong KongShatin, Hong Kongwph019@ie.cuhk.edu.hkTsz Kui ChowDepartment of Information EngineeringThe Chinese University of Hong KongShatin, Hong Kong1155093841@link.cuhk.edu.hkAbstractIn this proj...
DCKY5YP13c
Deploying Reinforcement Learning in Water Transport
8: Top 50% of accepted papers, clear accept
summary: This paper deploys different kinds of reinforcement learning algorithms to solve the water sailing problem. Clarity: Clear problem definition and environment description. pos: 1. open-source implementation code. 2. provide an intuitive environment framework. 3. use different kinds of methods to solve the pro...
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 Deploying Reinforcement Learning in Water Transport ### Paper Abstract In this project, we deployed various types of reinforcement learning algorithms to resolves the re...
rJTKKKqeg
ICLR.cc/2017/conference
2017
Tracking the World State with Recurrent Entity Networks
["Mikael Henaff", "Jason Weston", "Arthur Szlam", "Antoine Bordes", "Yann LeCun"]
We introduce a new model, the Recurrent Entity Network (EntNet). It is equipped with a dynamic long-term memory which allows it to maintain and update a rep- resentation of the state of the world as it receives new data. For language under- standing tasks, it can reason on-the-fly as it reads text, not just when it is ...
["Natural language processing", "Deep learning"]
ABSTRACTWe introduce a new model, the Recurrent Entity Network (EntNet). It is equippedwith a dynamic long-term memory which allows it to maintain and update a rep-resentation of the state of the world as it receives new data. For language under-standing tasks, it can reason on-the-fly as it reads text, not just when i...
Sy9rRTHEg
7: Good paper, accept
The work proposes a variant of a recurrent neural network that can selectively update a fixed number of multiple memory slots to update entity states. The architecture is well motivated, especially with the motivating example, and the operation is shown to validate the intuition as shown in visualizations. Experiment...
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 Tracking the World State with Recurrent Entity Networks ### Paper Abstract We introduce a new model, the Recurrent Entity Network (EntNet). It is equipped with a dynamic...
Sk9QRnCIM
ICLR.cc/2018/Workshop
2018
Node Centralities and Classification Performance for Characterizing Node Embedding Algorithms
["Kento Nozawa", "Masanari Kimura", "Atsunori Kanemura"]
Embedding graph nodes into a vector space can allow the use of machine learning to e.g. predict node classes, but the study of node embedding algorithms is immature compared to the natural language processing field because of a diverse nature of graphs. We examine the performance of node embedding algorithms with respe...
["node", "algorithms", "classification performance", "node centralities", "algorithms node centralities", "graph nodes", "vector space", "use", "machine learning", "predict node classes"]
Workshop track - ICLR 2018N/o.sc/d.sc/e.sc C/e.sc/n.sc/t.sc/r.sc/a.sc/l.sc/i.sc/t.sc/i.sc/e.sc/s.sc /a.sc/n.sc/d.sc C/l.sc/a.sc/s.sc/s.sc/i.sc/f.sc/i.sc/c.sc/a.sc/t.sc/i.sc/o.sc/n.sc P/e.sc/r.sc/f.sc/o.sc/r.sc/m.sc/a.sc/n.sc/c.sc/e.sc/f.sc/o.sc/r.sc C/h.sc/a.sc/r.sc/a.sc/c.sc/t.sc/e.sc/r.sc/i.sc/z.sc/i.sc/n.sc/g.sc N/o...
ByjS8SZYM
Interesting but not quite there yet
4: Ok but not good enough - rejection
The idea overall is interesting: whether graph embeddings for a simple node classification task correlate with graph-wide metrics (such as centrality). For an empirical workshop paper seeking to compare embedding methods, you would expect to find a lot more methods in the list. After asking the authors some questions,...
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 Node Centralities and Classification Performance for Characterizing Node Embedding Algorithms ### Paper Abstract Embedding graph nodes into a vector space can allow the ...
wl0Kr_jqM2a
ICLR.cc/2021/Conference
2021
Testing Robustness Against Unforeseen Adversaries
["Daniel Kang", "Yi Sun", "Dan Hendrycks", "Tom B Brown", "Jacob Steinhardt"]
Most existing adversarial defenses only measure robustness to $L_p$ adversarial attacks. Not only are adversaries unlikely to exclusively create small $L_p$ perturbations, adversaries are unlikely to remain fixed. Adversaries adapt and evolve their attacks; hence adversarial defenses must be robust to a broad range of ...
["adversarial examples", "adversarial training", "adversarial attacks"]
ABSTRACTMost existing adversarial defenses only measure robustness to Lpadversarialattacks. Not only are adversaries unlikely to exclusively create small Lpperturba-tions, adversaries are unlikely to remain fixed. Adversaries adapt and evolve theirattacks; hence adversarial defenses must be robust to a broad range of u...
Mxm69kh1WQl
Good motivation but not unconvincing enough
5: Marginally below acceptance threshold
This paper proposes four novel efficient adversarial attack methods beyond Lp threat models. Together with other two existing attack methods, these six attack methods combine as a framework to evaluate robustness of defenses against unforeseen attacks. In this framework, the novel measure is normalized with the perform...
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 Testing Robustness Against Unforeseen Adversaries ### Paper Abstract Most existing adversarial defenses only measure robustness to $L_p$ adversarial attacks. Not only ar...
r1e44bPpP4
icaps-conference.org/ICAPS/2019/Workshop/HSDIP
2019
Learning How to Ground a Plan - Partial Grounding in Classical Planning
["Anonymous"]
Current classical planners are very successful in finding (non-optimal) plans, even for large planning instances. To do so, most planners rely on a preprocessing stage that computes a grounded representation of the task. Whenever the grounded task is too big to be generated (i.e., whenever this preprocess fails) the in...
["classical planning", "grounding", "heuristic search", "machine learning", "relational learning", "inductive logic programming"]
Learning How to Ground a Plan – Partial Grounding in Classical PlanningDaniel Gnad, ́Alvaro TorralbaSaarland UniversitySaarland Informatics CampusSaarbr ̈ucken, Germanyfgnad,torralbag@cs.uni-saarland.deMart ́ın Dom ́ınguez, Carlos Areces, Facundo BustosUniversidad Nacional de C ́ordobaC ́ordoba, Argentinafmardom75, car...
H1lHYmRStV
Very well-written paper tackling an interesting problem
9: Top 15% of accepted papers, strong accept
This paper presents an approach for only partially grounding classical planning tasks which are too large to be fully grounded by common grounding algorithms. The approach uses machine learning techniques to estimate the probability of operators belonging to a plan of the task, using information from small instances 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 Learning How to Ground a Plan - Partial Grounding in Classical Planning ### Paper Abstract Current classical planners are very successful in finding (non-optimal) plans,...
ZW0yXJyNmoG
ICLR.cc/2021/Conference
2021
Taming GANs with Lookahead-Minmax
["Tatjana Chavdarova", "Matteo Pagliardini", "Sebastian U Stich", "Fran\u00e7ois Fleuret", "Martin Jaggi"]
Generative Adversarial Networks are notoriously challenging to train. The underlying minmax optimization is highly susceptible to the variance of the stochastic gradient and the rotational component of the associated game vector field. To tackle these challenges, we propose the Lookahead algorithm for minmax optimizati...
["Minmax", "Generative Adversarial Networks"]
ABSTRACTGenerative Adversarial Networks are notoriously challenging to train. The under-lying minmax optimization is highly susceptible to the variance of the stochasticgradient and the rotational component of the associated game vector field. To tacklethese challenges, we propose the Lookahead algorithm for minmax opt...
ujcNKpKJ5Js
Official Blind Review #1
6: Marginally above acceptance threshold
##### Summary: This paper proposes a lookahead-minmax algorithm for optimizing minmax problems such as GANs, which updates the parameters (of both the generator and the discriminator) with the extrapolation. With a bilinear example, the authors show that the use of lookahead-minimax allows for convergence in cases whe...
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 Taming GANs with Lookahead-Minmax ### Paper Abstract Generative Adversarial Networks are notoriously challenging to train. The underlying minmax optimization is highly s...
rkxmPgrKwB
ICLR.cc/2020/Conference
2020
Weight-space symmetry in neural network loss landscapes revisited
["Berfin Simsek", "Johanni Brea", "Bernd Illing", "Wulfram Gerstner"]
Neural network training depends on the structure of the underlying loss landscape, i.e. local minima, saddle points, flat plateaus, and loss barriers. In relation to the structure of the landscape, we study the permutation symmetry of neurons in each layer of a deep neural network, which gives rise not only to multiple...
["Weight-space symmetry", "neural network landscapes"]
ABSTRACTNeural network training depends on the structure of the underlying loss landscape,i.e. local minima, saddle points, flat plateaus, and loss barriers. In relation to thestructure of the landscape, we study the permutation symmetry of neurons in eachlayer of a deep neural network, which gives rise not only to mul...
HklMZ14z5H
Official Blind Review #1
3: Weak Reject
This paper studies a special type of weight symmetry in neural networks. I think studying the geometry of neural-nets is an interesting and important direction for understanding neural-nets, and along this direction, weight-space symmetry is an important subject. However, it seems to me this paper does not make enough ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Weight-space symmetry in neural network loss landscapes revisited ### Paper Abstract Neural network training depends on the structure of the underlying loss landscape, i...
rJxt0JHKvS
ICLR.cc/2020/Conference
2020
Coloring graph neural networks for node disambiguation
["George Dasoulas", "Ludovic Dos Santos", "Kevin Scaman", "Aladin Virmaux"]
In this paper, we show that a simple coloring scheme can improve, both theoretically and empirically, the expressive power of Message Passing Neural Networks (MPNNs). More specifically, we introduce a graph neural network called Colored Local Iterative Procedure (CLIP) that uses colors to disambiguate identical node at...
["Graph neural networks", "separability", "node disambiguation", "universal approximation", "representation learning"]
ABSTRACTIn this paper, we show that a simple coloring scheme can improve, both theoreti-cally and empirically, the expressive power of Message Passing Neural Networks(MPNNs). More specifically, we introduce a graph neural network called ColoredLocal Iterative Procedure (CLIP) that uses colors to disambiguate identical ...
Bye8QHt0tH
Official Blind Review #1
3: Weak Reject
This paper proposes an extension of MPNN which leverages the random color augmentation to improve the representation power of MPNN. Authors also prove that two variants of the proposed method have universal representation power (one is exact and the other holds in expectation) from the separability perspective. Experim...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Coloring graph neural networks for node disambiguation ### Paper Abstract In this paper, we show that a simple coloring scheme can improve, both theoretically and empiri...