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
SJDaqqveg
ICLR.cc/2017/conference
2017
An Actor-Critic Algorithm for Sequence Prediction
["Dzmitry Bahdanau", "Philemon Brakel", "Kelvin Xu", "Anirudh Goyal", "Ryan Lowe", "Joelle Pineau", "Aaron Courville", "Yoshua Bengio"]
We present an approach to training neural networks to generate sequences using actor-critic methods from reinforcement learning (RL). Current log-likelihood training methods are limited by the discrepancy between their training and testing modes, as models must generate tokens conditioned on their previous guesses rath...
["Natural language processing", "Deep learning", "Reinforcement Learning", "Structured prediction"]
ABSTRACTWe present an approach to training neural networks to generate sequences usingactor-critic methods from reinforcement learning (RL). Current log-likelihoodtraining methods are limited by the discrepancy between their training and testingmodes, as models must generate tokens conditioned on their previous guesses...
SyLHhObEg
Review
8: Top 50% of accepted papers, clear accept
The paper presents a nice application of actor-critic method for conditional sequence prediction. The critic is trained conditional to target sequence output, while the actor is conditional on input sequence. The paper presents a number of interesting design decisions in order to tackle non-standard RL problem with act...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title An Actor-Critic Algorithm for Sequence Prediction ### Paper Abstract We present an approach to training neural networks to generate sequences using actor-critic methods ...
Yv_B743qacL
NeurIPS.cc/2022/Workshop/SyntheticData4ML
2022
Conditional Progressive Generative Adversarial Network for satellite image generation
["Renato Paulo Da Costa Cardoso", "Sofia Vallecorsa", "Edoardo Nemni"]
Image generation and image completion are rapidly evolving fields, thanks to machine learning algorithms that are able to realistically replace missing pixels. However, generating large high resolution images, with a large level of details, presents important computational challenges. In this work, we formulate the im...
["Machine Learning", "Generative Models", "Earth Observation", "GAN", "Auto Encoders"]
Conditional Progressive Generative AdversarialNetwork for satellite image generationRenato Cardoso∗CERN,1 Esplanade des Particules,Geneva, SwitzerlandSofia VallecorsaCERN,1 Esplanade des Particules,Geneva, SwitzerlandEdoardo NemniUnited Nations Satellite Centre (UNOSAT),7 bis, Avenue de la Paix,CH-1202 Geneva 2, Switze...
QLBLR1g9fbm
Needs more improvement but good application domain
6: Marginally above acceptance threshold
The authors propose a CGAN-based iterative image completion algorithm for satellite image generation. The results are preliminary training up to 256 x 256 Strengths: * First proof of concept of a methodology for the generation of large high-resolution images, such as those typically used in Earth Observation (EO) p...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Conditional Progressive Generative Adversarial Network for satellite image generation ### Paper Abstract Image generation and image completion are rapidly evolving field...
BkgOM1rKvr
ICLR.cc/2020/Conference
2020
The Surprising Behavior Of Graph Neural Networks
["Vivek Kothari", "Catherine Tong", "Nicholas Lane"]
We highlight a lack of understanding of the behaviour of Graph Neural Networks (GNNs) in various topological contexts. We present 4 experimental studies which counter-intuitively demonstrate that the performance of GNNs is weakly dependent on the topology, sensitive to structural noise and the modality (attributes or e...
["Graph Neural Networks", "Graph Toplogy", "Noise", "Attributed Networks"]
ABSTRACTWe highlight a lack of understanding of the behaviour of Convolutional Graph Neu-ral Networks (GNNs) in various topological contexts. We present 4 experimentalstudies which counter-intuitively demonstrate that the performance of GNNs isweakly dependent on the topology, sensitive to structural noise and the moda...
ryeDNdsitr
Official Blind Review #1
6: Weak Accept
This paper analyzes the properties of graph neural networks (GNNs). It shows that several hypothesis that one might intuitively make about the behavior of GNNs, do not actually hold. In fact, some observations are even contradictory, indicating that GNNs' performance is not robust, and care needs to be taken when using...
<|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 Surprising Behavior Of Graph Neural Networks ### Paper Abstract We highlight a lack of understanding of the behaviour of Graph Neural Networks (GNNs) in various topo...
Ifvv3qpPsqB
ICLR.cc/2021/Conference
2021
Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation
["Xiaogang Xu", "Hengshuang Zhao", "Jiaya Jia"]
Adversarial training is promising for improving the robustness of deep neural networks towards adversarial perturbations, especially on the classification task. The effect of this type of training on semantic segmentation, contrarily, just commences. We make the initial attempt to explore the defense strategy on semant...
["adversarial defense", "semantic segmentation", "robustness"]
ABSTRACTAdversarial training is promising for improving the robustness of deep neural net-works towards adversarial perturbations, especially on the classification task. Theeffect of this type of training on semantic segmentation, contrarily, just com-mences. We make the initial attempt to explore the defense strategy ...
a4NmNT8Uunz
Toward more adversarially robust semantic segmentation, but major weaknesses prevail
3: Clear rejection
Summary: The submission takes a closer look at robustness against adversarial attacks in semantic segmentation. In particular, it proposes a novel training framework termed DDC-AT (for ‘Dynamic Divide-and-Conquer’) where pixel samples from a set of paired adversarial and non-adversarial images are classified into two s...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation ### Paper Abstract Adversarial training is promising for improving the robustness of dee...
uRKqXoN-Ic9
ICLR.cc/2021/Conference
2021
Evaluating Robustness of Predictive Uncertainty Estimation: Are Dirichlet-based Models Reliable?
["Anna-Kathrin Kopetzki", "Bertrand Charpentier", "Daniel Z\u00fcgner", "Sandhya Giri", "Stephan G\u00fcnnemann"]
Robustness to adversarial perturbations and accurate uncertainty estimation are crucial for reliable application of deep learning in real world settings. Dirichlet-based uncertainty (DBU) models are a family of models that predict the parameters of a Dirichlet distribution (instead of a categorical one) and promise to ...
["dbu models", "robustness", "models", "robust", "uncertainty estimates", "able", "data", "predictive uncertainty estimation", "models reliable", "results"]
ABSTRACTRobustness to adversarial perturbations and accurate uncertainty estimation arecrucial for reliable application of deep learning in real world settings. Dirichlet-based uncertainty (DBU) models are a family of models that predict the parametersof a Dirichlet distribution (instead of a categorical one) and promi...
3LtHJe-a8E
A nice contribution for evaluation/quantification of uncertainty
7: Good paper, accept
In this work, the authors seek to evaluate the robustness and uncertainty of Dirichlet-Based Uncertainty models (DBUs). DBUs are models which rather than outputting a multinoulli distribution (as is the case with DNNs and BNNs) output the parameters of a Dirichet distribution. The authors clearly state and review the h...
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 Evaluating Robustness of Predictive Uncertainty Estimation: Are Dirichlet-based Models Reliable? ### Paper Abstract Robustness to adversarial perturbations and accurate ...
HJekvT4twr
ICLR.cc/2020/Conference
2020
RGTI:Response generation via templates integration for End to End dialog
["Yuxin Zhang", "Songyan Liu"]
End-to-end models have achieved considerable success in task-oriented dialogue area, but suffer from the challenges of (a) poor semantic control, and (b) little interaction with auxiliary information. In this paper, we propose a novel yet simple end-to-end model for response generation via mixed templates, which can ad...
["End-to-end dialogue systems", "transformer", "pointer-generate network"]
ABSTRACTEnd-to-end models have achieved considerable success in task-oriented dialoguearea, but suffer from the challenges of (a) poor semantic control, and (b) little in-teraction with auxiliary information. In this paper, we propose a novel yet simpleend-to-end model for response generation via mixed templates, which...
ByeBak8bqH
Official Blind Review #3
1: Reject
1. Summary: The authors proposed a deep neural network-based model to generate responses fro task-oriented dialogue systems. The model mainly contains two parts, the first part is to retrieve relevant responses based on question and encode them into templates, the second part is a decoder to generate the response based...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title RGTI:Response generation via templates integration for End to End dialog ### Paper Abstract End-to-end models have achieved considerable success in task-oriented dialogu...
r1gIwgSYwr
ICLR.cc/2020/Conference
2020
Localized Meta-Learning: A PAC-Bayes Analysis for Meta-Leanring Beyond Global Prior
["Chenghao Liu", "Tao Lu", "Doyen Sahoo", "Yuan Fang", "Steven C.H. Hoi."]
Meta-learning methods learn the meta-knowledge among various training tasks and aim to promote the learning of new tasks under the task similarity assumption. However, such meta-knowledge is often represented as a fixed distribution, which is too restrictive to capture various specific task information. In this work, w...
["localized meta-learning", "PAC-Bayes", "meta-learning"]
ABSTRACTMeta-learning methods learn the meta-knowledge among various training tasks 1and aim to promote the learning of new tasks under the task similarity assumption. 2However, such meta-knowledge is often represented as a fixed distribution, which 3is too restrictive to capture various specific task information. In t...
rkeqdjJOcS
Official Blind Review #2
8: Accept
This paper discusses a theoretical analysis of localized meta-learning. The authors are motivated from a PAC-Bayes perspective of meta-learning wherein the system learns a hyper-prior on the hypothesis space using the posteriors of the weak-learners learnt on their individual tasks. The first contribution 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 Localized Meta-Learning: A PAC-Bayes Analysis for Meta-Leanring Beyond Global Prior ### Paper Abstract Meta-learning methods learn the meta-knowledge among various train...
rJehNT4YPr
ICLR.cc/2020/Conference
2020
I Am Going MAD: Maximum Discrepancy Competition for Comparing Classifiers Adaptively
["Haotao Wang", "Tianlong Chen", "Zhangyang Wang", "Kede Ma"]
The learning of hierarchical representations for image classification has experienced an impressive series of successes due in part to the availability of large-scale labeled data for training. On the other hand, the trained classifiers have traditionally been evaluated on small and fixed sets of test images, which are...
["model comparison"]
ABSTRACTThe learning of hierarchical representations for image classification has experi-enced an impressive series of successes due in part to the availability of large-scale labeled data for training. On the other hand, the trained classifiers havetraditionally been evaluated on small and fixed sets of test images, w...
HJxHV7JPjB
Official Blind Review #3
3: Weak Reject
This paper proposes a new method to compare existing classifiers, which does not use fixed test set and adaptively sample it from an arbitrarily large corpus of unlabeled images. The main idea seems similar to adopting active learning for the test set selection. One of the main advantage is that it can select a sampl...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title I Am Going MAD: Maximum Discrepancy Competition for Comparing Classifiers Adaptively ### Paper Abstract The learning of hierarchical representations for image classifica...
S1qI2FJDM
ICLR.cc/2018/Workshop
2018
Defending DNN Adversarial Attacks with Pruning and Logits Augmentation
["Shaokai Ye", "Siyue Wang", "Xiao Wang", "Bo Yuan", "Wujie Wen", "Xue Lin"]
Deep neural networks (DNNs) have been shown to be powerful models and perform extremely well on many complicated artificial intelligent tasks. However, recent research found that these powerful models are vulnerable to adversarial attacks, i.e., intentionally added imperceptible perturbations to DNN inputs can easily m...
["Adversarial Attacks", "Neural Network Security", "Weight Pruning", "Logits Augmentation"]
ABSTRACTDeep neural networks (DNNs) have been shown to be powerful models and per-form extremely well on many complicated artificial intelligent tasks. However,recent research found that these powerful models are vulnerable to adversarialattacks, i.e., intentionally added imperceptible perturbations to DNN inputs canea...
SyedPisdz
review
4: Ok but not good enough - rejection
The paper presents two approaches to makes models robust to adversarial attacks. The first one consider pruning model weights similarly to what was proposed in Han et al (2015). The second one consider multiplying the weights of the last layer by a large constant (they call this "logits augmentation"). I don't understa...
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 Defending DNN Adversarial Attacks with Pruning and Logits Augmentation ### Paper Abstract Deep neural networks (DNNs) have been shown to be powerful models and perform e...
SSSGs3M7nRY
ML_Reproducibility_Challenge/2021/Fall
2021
Reproducibility report: Hate Speech Detection based on Sentiment Knowledge Sharing
["Matteo Brivio", "Cagri Coltekin"]
This report summarises our efforts to reproduce the results presented in the ACL2021 paper Hate Speech Detection based on Sentiment Knowledge Sharing by Zhou et al. (2021), as part of the ML Reproducibility Challenge 2021. Scope of Reproducibility The main goal of this reproducibility attempt is to confirm the effecti...
["hate speech detection"]
Reproducibility report:Hate Speech Detection based on Sentiment Knowledge SharingAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1This report summarises our efforts to reproduce the results presented in the ACL2021 paper Hate Speech Detection 2based on Sentiment Knowledge Sharing by Zhou et al. (2021)...
rIcxqV2Uoxq
Reviewing "Reproducibility report: Hate Speech Detection based on Sentiment Knowledge Sharing"
6: Marginally above acceptance threshold
The paper addresses an important and timely topic of tooling hate speech detection on social media. It studies reproducing a 2021 ACL study by Zhou et al. (2021), building on its report, codebase, and communication with the original authors. In addition to trying to verify the three key claims made by Zhou et al (2021)...
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 Reproducibility report: Hate Speech Detection based on Sentiment Knowledge Sharing ### Paper Abstract This report summarises our efforts to reproduce the results present...
rF2kJgYAr3
ICLR.cc/2021/Conference
2021
NeuralLog: a Neural Logic Language
["Victor Guimar\u00e3es", "Vitor Santos Costa"]
Application domains that require considering relationships among objects which have real-valued attributes are becoming even more important. In this paper we propose NeuralLog, a first-order logic language that is compiled to a neural network. The main goal of NeuralLog is to bridge logic programming and deep learning,...
["Deep learning", "logic programming", "neural network", "relational learning"]
ABSTRACTApplication domains that require considering relationships among objects whichhave real-valued attributes are becoming even more important. In this paper wepropose NeuralLog, a first-order logic language that is compiled to a neural net-work. The main goal of NeuralLog is to bridge logic programming and deeplea...
BUZrvyutfp
Nice paper but needs another round of improvement.
5: Marginally below acceptance threshold
The key idea in this paper is to combine the power of logic programming and neural networks. Specifically this allows for neural networks to be defined as logic programs but handle numeric attributes and functions. The paper proposes NeuralLog, a FOL language that is compiled to a neural network. The key improvement ov...
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 NeuralLog: a Neural Logic Language ### Paper Abstract Application domains that require considering relationships among objects which have real-valued attributes are beco...
ZccHKNXFHV
ICLR.cc/2022/Track/Blog
2022
Diffusion Models Using a Single Equation
["Anonymous"]
In this work I present a novel viewpoint and a simplistic implementation and explanation of denoising diffusion models, and also the intuition that we force these models to sample from the data distribution by misleading them. The aim of this blogpost is to lower the barrier of entry to the field of diffusion models, b...
["generative modeling", "denoising diffusion", "DDIM", "DDPM"]
S8cxGrJg1J9
Interesting topic, but need to focus on making the material understandable by the target audience.
5: Marginally below acceptance threshold
**Overview**: The blog post aims to illustrate a simple interpretation to Diffusion Denoising Implicit Models, with an TensorFlow implementation embedded in. While I am interested in seeing a blog post on this topic, I feel that the current version can be a bit confusing to the target audience. **Strengths**: The bl...
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 Diffusion Models Using a Single Equation ### Paper Abstract In this work I present a novel viewpoint and a simplistic implementation and explanation of denoising diffusi...
U5zU7yDUhKx
graphicsinterface.org/Graphics_Interface/2023/Conference
2023
Temperature Illusions in Mixed Reality using Color and Dynamic Graphics
["Connor Wilson", "Daniel J. Rea", "Scott Bateman"]
Sensory illusions -- where a sensory stimulus causes people to perceive effects that are altered by a different sensory stimulus -- have the potential to enrich mixed-reality based interactions which typically lack the ability to affect many physical senses. The well-known color-temperature illusion is a sensory illusi...
["sensory illusion", "perception", "visual effects", "mixed reality", "augmented reality"]
ABSTRACTSensory illusions – where a sensory stimulus causes people to per-ceive effects that are altered by a different sensory stimulus – havethe potential to enrich mixed-reality based interactions. The well-known color-temperature illusion is a sensory illusion that causespeople to, somewhat counter intuitively, per...
xVdbv5HqQFA
interesting problem, potential for followup work
7: Good paper, accept
This paper presents an investigation into the color-temperature illusion in an augmented reality context. The experiment found weak support for the previously-reported illusion, where warm-colored objects are perceived as cooler than they are, while cool-colored objects are perceived as warmer. Further, and new to this...
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 Temperature Illusions in Mixed Reality using Color and Dynamic Graphics ### Paper Abstract Sensory illusions -- where a sensory stimulus causes people to perceive effect...
tmvULc0laNv
ICLR.cc/2021/Conference
2021
Gradient flow encoding with distance optimization adaptive step size
["Kyriakos Flouris", "Anna Volokitin", "Gustav Bredell", "Ender Konukoglu"]
The autoencoder model uses an encoder to map data samples to a lower dimensional latent space and then a decoder to map the latent space representations back to the data space. Implicitly, it relies on the encoder to approximate the inverse of the decoder network, so that samples can be mapped to and back from the late...
["gradient flow", "encoder", "decoder", "data samples", "latent space representations", "latent space", "ode", "adjoint", "model"]
ABSTRACTThe autoencoder model uses an encoder to map data samples to a lower dimen-sional latent space and then a decoder to map the latent space representations backto the data space. Implicitly, it relies on the encoder to approximate the inverse ofthe decoder network, so that samples can be mapped to and back from t...
IygxXKlGu7B
Iterative optimization of latent representation is an old idea
2: Strong rejection
Summary ------------- This paper proposes to get rid of the encoder of autoencoders by optimizing the latent representation directly. To find the optimal latent state the authors test gradient descent ("GFE"), Nesterov momentum ("GFE-2nd order"), and gradient descent with line search ("GFE-amd"). Empirical results are ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Gradient flow encoding with distance optimization adaptive step size ### Paper Abstract The autoencoder model uses an encoder to map data samples to a lower dimensional ...
K5YasWXZT3O
ICLR.cc/2021/Conference
2021
Tilted Empirical Risk Minimization
["Tian Li", "Ahmad Beirami", "Maziar Sanjabi", "Virginia Smith"]
Empirical risk minimization (ERM) is typically designed to perform well on the average loss, which can result in estimators that are sensitive to outliers, generalize poorly, or treat subgroups unfairly. While many methods aim to address these problems individually, in this work, we explore them through a unified frame...
["exponential tilting", "models of learning and generalization", "label noise robustness", "fairness"]
ABSTRACTEmpirical risk minimization (ERM) is typically designed to perform well on theaverage loss, which can result in estimators that are sensitive to outliers, generalizepoorly, or treat subgroups unfairly. While many methods aim to address theseproblems individually, in this work, we explore them through a unified ...
qLJgnLurpD
Review of "Tilted Empirical Risk Minimization"
8: Top 50% of accepted papers, clear accept
* Summary of the paper. The paper presents a modification of the empirical risk minimization (ERM) framework in order to obtain more robust or fairer results, using tilted objective. The paper exhibit several properties of the tilted empirical risk minimization (TERM) algorithm applied to classification and regr...
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 Tilted Empirical Risk Minimization ### Paper Abstract Empirical risk minimization (ERM) is typically designed to perform well on the average loss, which can result in es...
OQ08SN70M1V
ICLR.cc/2021/Conference
2021
Better Fine-Tuning by Reducing Representational Collapse
["Armen Aghajanyan", "Akshat Shrivastava", "Anchit Gupta", "Naman Goyal", "Luke Zettlemoyer", "Sonal Gupta"]
Although widely adopted, existing approaches for fine-tuning pre-trained language models have been shown to be unstable across hyper-parameter settings, motivating recent work on trust region methods. In this paper, we present a simplified and efficient method rooted in trust region theory that replaces previously used...
["finetuning", "nlp", "representational learning", "glue"]
ABSTRACTAlthough widely adopted, existing approaches for fine-tuning pre-trained lan-guage models have been shown to be unstable across hyper-parameter settings,motivating recent work on trust region methods. This paper presents a simpli-fied and efficient method rooted in trust region theory that replaces previouslyus...
DIKqZkFzVpe
A simple and effective method for preserving generalizability of a model when fine-tuning.
6: Marginally above acceptance threshold
Summary The paper proposes a method for finetuning pre-trained models that ensures the generalization ability of the representation is maintained. The key innovation is that the computationally expensive ascent step in the mirror descent method of SMART can be replaced by simply injecting noise. The results support th...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Better Fine-Tuning by Reducing Representational Collapse ### Paper Abstract Although widely adopted, existing approaches for fine-tuning pre-trained language models have...
wWK7yXkULyh
ICLR.cc/2021/Conference
2021
MONGOOSE: A Learnable LSH Framework for Efficient Neural Network Training
["Beidi Chen", "Zichang Liu", "Binghui Peng", "Zhaozhuo Xu", "Jonathan Lingjie Li", "Tri Dao", "Zhao Song", "Anshumali Shrivastava", "Christopher Re"]
Recent advances by practitioners in the deep learning community have breathed new life into Locality Sensitive Hashing (LSH), using it to reduce memory and time bottlenecks in neural network (NN) training. However, while LSH has sub-linear guarantees for approximate near-neighbor search in theory, it is known to have i...
["Large-scale Deep Learning", "Large-scale Machine Learning", "Efficient Training", "Randomized Algorithms"]
ABSTRACTRecent advances by practitioners in the deep learning community have breathednew life into Locality Sensitive Hashing (LSH), using it to reduce memory andtime bottlenecks in neural network (NN) training. However, while LSH has sub-linear guarantees for approximate near-neighbor search in theory, it is known toh...
1R222uVdrJG
A principled approach to updating LSH based ANNS for faster matrix multiplication in NN training
7: Good paper, accept
Some neural network runs involve layers with a large number of neurons. These require large matrix-vector or matrix-multiplication which can slow their training/inference. However, if the output of mat-vec/mul is dominated by a few neurons with which the activation has large inner product (a matmul can be thought of 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 MONGOOSE: A Learnable LSH Framework for Efficient Neural Network Training ### Paper Abstract Recent advances by practitioners in the deep learning community have breathe...
B1lKDlHtwS
ICLR.cc/2020/Conference
2020
Measuring causal influence with back-to-back regression: the linear case
["Jean-Remi King", "Francois Charton", "Maxime Oquab", "David Lopez-Paz"]
Identifying causes from observations can be particularly challenging when i) potential factors are difficult to manipulate individually and ii) observations are complex and multi-dimensional. To address this issue, we introduce “Back-to-Back” regression (B2B), a method designed to efficiently measure, from a set of co-...
["regression", "causal influence", "linear case", "observations", "causes", "potential factors", "difficult", "complex", "issue"]
ABSTRACTIdentifying causes from observations can be particularly challenging when i) poten-tial factors are difficult to manipulate individually and ii) observations are complexand multi-dimensional. To address this issue, we introduce “Back-to-Back” regres-sion (B2B), a method designed to efficiently measure, from a s...
Bkgo2vCzcr
Official Blind Review #2
6: Weak Accept
This paper proposes "Back-to-Back" regression for estimating the causal influence between X and Y in the linear model Y=(XE+N)F, where the E denotes a diagonal matrix of causal influences. Furthermore, this work theoretically shows the consistency of B2B and the experiments also verify the effectiveness of this approac...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Measuring causal influence with back-to-back regression: the linear case ### Paper Abstract Identifying causes from observations can be particularly challenging when i) ...
HJ0toxywG
ICLR.cc/2018/Workshop
2018
Share or Split : which is more efficient?
["Yinpeng Chen", "Zicheng Liu", "Lijuan Wang", "Zhengyou Zhang"]
In this paper, we investigate two different feature representations in convolutional neural networks (CNN): (a) Share - all classes share the same feature space, and a fully connected layer is used to decode class activation, and (b) Split - each class has its own feature space and a class is activated if its correspon...
["share", "split", "feature space", "efficient", "class", "data", "different feature representations", "convolutional neural networks", "cnn", "classes share"]
ABSTRACTIn this paper, we investigate two different feature representations in convolutionalneural networks (CNN): (a) Share - all classes share the same feature space, and afully connected layer is used to decode class activation, and (b) Split - each classhas its own feature space and a class is activated if its corr...
HkKcywJKz
hard to draw many conclusions
5: Marginally below acceptance threshold
This paper compares two approaches to final feature layer representation in a classifier: "Share", which uses the same feature vector for all classes, and "Split", which has a different feature vector for each class, and chooses the "winner" based on which one has largest magnitude. These two representations are 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 Share or Split : which is more efficient? ### Paper Abstract In this paper, we investigate two different feature representations in convolutional neural networks (CNN): ...
rJgfjjC9Ym
ICLR.cc/2019/Conference
2019
Backprop with Approximate Activations for Memory-efficient Network Training
["Ayan Chakrabarti", "Benjamin Moseley"]
With innovations in architecture design, deeper and wider neural network models deliver improved performance on a diverse variety of tasks. But the increased memory footprint of these models presents a challenge during training, when all intermediate layer activations need to be stored for back-propagation. Limited GPU...
["Back-propagation", "Memory Efficient Training", "Approximate Gradients", "Deep Learning"]
ABSTRACTWith innovations in architecture design, deeper and wider neural network mod-els deliver improved performance on a diverse variety of tasks. But the increasedmemory footprint of these models presents a challenge during training, when allintermediate layer activations need to be stored for back-propagation. Limi...
SJlbSuiI37
Borderline paper
5: Marginally below acceptance threshold
This paper proposes to use 8/4-bit approximation of activations to save the memory cost during gradient computation. The proposed technique is simple and straightforward. On the other hand, the proposed method only saves up to a constant cost of the storage. With the constant factor (4x, 8x) depending on whether fp16 ...
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 Backprop with Approximate Activations for Memory-efficient Network Training ### Paper Abstract With innovations in architecture design, deeper and wider neural network m...
HylKJhCcKm
ICLR.cc/2019/Conference
2019
Generalized Capsule Networks with Trainable Routing Procedure
["Zhenhua Chen", "Chuhua Wang", "Tiancong Zhao", "David Crandall"]
CapsNet (Capsule Network) was first proposed by Sabour et al. (2017) and lateranother version of CapsNet was proposed by Hinton et al. (2018). CapsNet hasbeen proved effective in modeling spatial features with much fewer parameters.However, the routing procedures (dynamic routing and EM routing) in both pa-pers are no...
["Capsule networks", "generalization", "scalability", "adversarial robustness"]
ABSTRACTCapsNet (Capsule Network) was first proposed by Sabour et al. (2017) and lateranother version of CapsNet was proposed by Hinton et al. (2018). CapsNet hasbeen proved effective in modeling spatial features with much fewer parameters.However, the routing procedures (dynamic routing and EM routing) in both pa-pers...
rJe2vOjb3m
The paper is poorly written, not clear about the contributions, evaluation is questionable
4: Ok but not good enough - rejection
Pros: The paper claims to make CapsuleNet's routing trainable. The proposed G-CapsNet have two variants (within feature map and across feature map). It presents evaluation of G-CapsNet in terms of robustness and generalization. It is interesting that Capsule networks are as bad as traditional CNNs for strong white b...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Generalized Capsule Networks with Trainable Routing Procedure ### Paper Abstract CapsNet (Capsule Network) was first proposed by Sabour et al. (2017) and lateranother ve...
ryGWhJBtDB
ICLR.cc/2020/Conference
2020
Hyperparameter Tuning and Implicit Regularization in Minibatch SGD
["Samuel L Smith", "Erich Elsen", "Soham De"]
This paper makes two contributions towards understanding how the hyperparameters of stochastic gradient descent affect the final training loss and test accuracy of neural networks. First, we argue that stochastic gradient descent exhibits two regimes with different behaviours; a noise dominated regime which typically a...
["SGD", "momentum", "batch size", "learning rate", "noise", "temperature", "implicit regularization", "optimization", "generalization"]
ABSTRACTThis paper makes two contributions towards understanding how the hyperparame-ters of stochastic gradient descent affect the final training loss and test accuracyof neural networks. First, we argue that stochastic gradient descent exhibits tworegimes with different behaviours; a noise dominated regime which typi...
rJxkq6waYr
Official Blind Review #2
3: Weak Reject
This paper studies the properties of SGD as a function of batch size and learning rate. Authors argue that SGD has two regimes: a noise dominated regime (small batch size) and curvature dominated regime (large batch size). Authors conduct through numerical experiments highlighting how learning rate changes as a 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 Hyperparameter Tuning and Implicit Regularization in Minibatch SGD ### Paper Abstract This paper makes two contributions towards understanding how the hyperparameters of...
ByOExmWAb
ICLR.cc/2018/Conference
2018
MaskGAN: Better Text Generation via Filling in the _______
["William Fedus", "Ian Goodfellow", "Andrew M. Dai"]
Neural text generation models are often autoregressive language models or seq2seq models. Neural autoregressive and seq2seq models that generate text by sampling words sequentially, with each word conditioned on the previous model, are state-of-the-art for several machine translation and summarization benchmarks. These...
["Deep learning", "GAN"]
ABSTRACTNeural text generation models are often autoregressive language models or seq2seqmodels. These models generate text by sampling words sequentially, with eachword conditioned on the previous word, and are state-of-the-art for several machinetranslation and summarization benchmarks. These benchmarks are often def...
rkgAfEoeG
Very thorough empirical study
7: Good paper, accept
Generating high-quality sentences/paragraphs is an open research problem that is receiving a lot of attention. This text generation task is traditionally done using recurrent neural networks. This paper proposes to generate text using GANs. GANs are notorious for drawing images of high quality but they have a hard time...
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 MaskGAN: Better Text Generation via Filling in the _______ ### Paper Abstract Neural text generation models are often autoregressive language models or seq2seq models. N...
HkMhoDITb
ICLR.cc/2018/Conference
2018
Reinforcement Learning via Replica Stacking of Quantum Measurements for the Training of Quantum Boltzmann Machines
["Anna Levit", "\u0006 Daniel Crawford", "Navid Ghadermarzy", "Jaspreet S. Oberoi", "Ehsan Zahedinejad", "Pooya Ronagh"]
Recent theoretical and experimental results suggest the possibility of using current and near-future quantum hardware in challenging sampling tasks. In this paper, we introduce free-energy-based reinforcement learning (FERL) as an application of quantum hardware. We propose a method for processing a quantum annealer’s ...
["Quantum Annealing", "Reinforcement Learning", "Boltzmann Machines", "Markov Chain Monte Carlo"]
ABSTRACTRecent theoretical and experimental results suggest the possibility of using currentand near-future quantum hardware in challenging sampling tasks. In this paper,we introduce free-energy-based reinforcement learning (FERL) as an applicationof quantum hardware. We propose a method for processing a quantum anneal...
SJ22faFez
replica stacking for quantum annealers used for RL
4: Ok but not good enough - rejection
There is no scientific consensus on whether quantum annealers such as the D-Wave 2000Q that use the transverse-field Ising models yield any gains over classical methods (c.f. https://arxiv.org/abs/1703.00622). However, it is an exciting research area and this paper is an interesting demonstration of the feasibility of ...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Reinforcement Learning via Replica Stacking of Quantum Measurements for the Training of Quantum Boltzmann Machines ### Paper Abstract Recent theoretical and experimental...
ZPa2SyGcbwh
ICLR.cc/2021/Conference
2021
Learning with Feature-Dependent Label Noise: A Progressive Approach
["Yikai Zhang", "Songzhu Zheng", "Pengxiang Wu", "Mayank Goswami", "Chao Chen"]
Label noise is frequently observed in real-world large-scale datasets. The noise is introduced due to a variety of reasons; it is heterogeneous and feature-dependent. Most existing approaches to handling noisy labels fall into two categories: they either assume an ideal feature-independent noise, or remain heuristic wi...
["Noisy Label", "Deep Learning", "Classification"]
ABSTRACTLabel noise is frequently observed in real-world large-scale datasets. The noise isintroduced due to a variety of reasons; it is heterogeneous and feature-dependent.Most existing approaches to handling noisy labels fall into two categories: theyeither assume an ideal feature-independent noise, or remain heurist...
BvmU2V48tyB
This paper proposes a progressive label correction algorithm by correcting labels and refine the model iteratively.
7: Good paper, accept
Comments: Label noise is very frequently in many real world applications. However, the noise can be with different distributions. If we build the learning model under a certain distribution, it is difficult to capture the discriminative information. In this paper, without assuming that the noise is a certain distributi...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning with Feature-Dependent Label Noise: A Progressive Approach ### Paper Abstract Label noise is frequently observed in real-world large-scale datasets. The noise i...
SklgHoRqt7
ICLR.cc/2019/Conference
2019
Metric-Optimized Example Weights
["Sen Zhao", "Mahdi Milani Fard", "Maya Gupta"]
Real-world machine learning applications often have complex test metrics, and may have training and test data that follow different distributions. We propose addressing these issues by using a weighted loss function with a standard convex loss, but with weights on the training examples that are learned to optimize the...
["example weights", "training", "test metric", "machine learning applications", "complex test metrics", "test data", "different distributions", "issues", "weighted loss function", "standard convex loss"]
ABSTRACTReal-world machine learning applications often have complex test metrics, andmay have training and test data that follow different distributions. We proposeaddressing these issues by using a weighted loss function with a standard convexloss, but with weights on the training examples that are learned to optimize...
Byg37fo2hm
Interesting problem. I think more discussion and experiments to understand better potential overfitting issues are needed here. I'm intrigued that the approach works as well as it does on the examples given, so there may be something interesting here.
4: Ok but not good enough - rejection
Pros: - Addresses several interesting and important problems all at once: covariate shift, concept drift, mismatch between training loss and test loss. - Fairly simple and elegant solution. - Multiple examples of the method working. - Clearly written Cons: - Examples don't feel like full-fledged machine learning examp...
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 Metric-Optimized Example Weights ### Paper Abstract Real-world machine learning applications often have complex test metrics, and may have training and test data that fo...
NZu3TrXvCk
MIDL.io/2023/Short_Paper_Track
2023
Deep Learning-Based Segmentation of Locally Advanced Breast Cancer on MRI in Relation to Residual Cancer Burden: A Multi-Institutional Cohort Study
["Mark Janse", "Liselore Maria Janssen", "Bas H.M. van der Velden", "Maaike Moman", "Elian Wolters-van der Ben", "Marc Kock", "Max A. Viergever", "Paul van Diest", "Kenneth Gilhuijs"]
While several methods have been proposed for automated assessment of breast-cancer response to neoadjuvant chemotherapy on breast MRI, limited information is available about their performance across multiple institutions. In this paper, we assess the value and robustness of nnU-Net-derived volumes of locally advanc...
["Breast MRI", "response monitoring", "segmentation", "nnU-Net"]
Medical Imaging with Deep Learning 2023Deep Learning-Based Segmentation of Locally AdvancedBreast Cancer on MRI in Relation to Residual CancerBurden: A Multi-Institutional Cohort StudyMark Janse1m.h.a.janse-2@umcutrecht.nlLiselore Janssen1l.m.janssen-11@umcutrecht.nlBas van der Velden1B.H.M.vanderVelden-2@umcutrecht.nl...
Wo8-0DDLhNp
This paper evaluates the performance and robustness of nnU-Net, a deep learning model, in estimating changes in tumor load on breast MRI for assessing residual cancer burden after neoadjuvant chemotherapy. The study shows that nnU-Net is accurate in estimating tumor changes, associated with residual cancer burden, and ...
6: Marginally above acceptance threshold
In this paper, the authors trained an nnU-Net to segment locally advanced breast cancer (LABC) on MRI and used it to estimate tumor volume changes associated with residual cancer burden (RCB) after neoadjuvant chemotherapy. They found that nnU-Net accurately estimated changes in tumor load on DCE-MRI and that the metho...
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-Based Segmentation of Locally Advanced Breast Cancer on MRI in Relation to Residual Cancer Burden: A Multi-Institutional Cohort Study ### Paper Abstract ...
4T489T4yav
ICLR.cc/2021/Conference
2021
Differentiable Segmentation of Sequences
["Erik Scharw\u00e4chter", "Jonathan Lennartz", "Emmanuel M\u00fcller"]
Segmented models are widely used to describe non-stationary sequential data with discrete change points. Their estimation usually requires solving a mixed discrete-continuous optimization problem, where the segmentation is the discrete part and all other model parameters are continuous. A number of estimation algorithm...
["segmented models", "segmentation", "change point detection", "concept drift", "warping functions", "gradient descent"]
ABSTRACTSegmented models are widely used to describe non-stationary sequential data withdiscrete change points. Their estimation usually requires solving a mixed discrete-continuous optimization problem, where the segmentation is the discrete part andall other model parameters are continuous. A number of estimation alg...
dYdf4kqyFfQ
Official Blind Review #2
7: Good paper, accept
[summary] The paper proposes a relaxed way to solve the segmentation of sequence that can directly leverage the deep learning architectures. The relaxed model allows each segmentation parameter to be a linear interpolations between two consecutive parameters depends on a continuous warping function. The paper then prop...
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 Differentiable Segmentation of Sequences ### Paper Abstract Segmented models are widely used to describe non-stationary sequential data with discrete change points. Thei...
bEoxzW_EXsa
ICLR.cc/2021/Conference
2021
Wasserstein-2 Generative Networks
["Alexander Korotin", "Vage Egiazarian", "Arip Asadulaev", "Alexander Safin", "Evgeny Burnaev"]
We propose a novel end-to-end non-minimax algorithm for training optimal transport mappings for the quadratic cost (Wasserstein-2 distance). The algorithm uses input convex neural networks and a cycle-consistency regularization to approximate Wasserstein-2 distance. In contrast to popular entropic and quadratic regular...
["wasserstein-2 distance", "optimal transport maps", "non-minimax optimization", "cycle-consistency regularization", "input-convex neural networks"]
ABSTRACTWe propose a novel end-to-end non-minimax algorithm for training optimal trans-port mappings for the quadratic cost (Wasserstein-2 distance). The algorithm usesinput convex neural networks and a cycle-consistency regularization to approximateWasserstein-2 distance. In contrast to popular entropic and quadratic ...
NrCemgimoKR
Reasonable approach but writing to be improved
6: Marginally above acceptance threshold
The paper proposes a new method for learning an optimal pushforward (for the quadratic cost) from a distribution to another distribution based on samples of both distributions. The optimal transport problem is first written equivalently as a minimax problem over set of convex functions, as in Makkuva et al. 19. Then, t...
2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Wasserstein-2 Generative Networks ### Paper Abstract We propose a novel end-to-end non-minimax algorithm for training optimal transport mappings for the quadratic cost (...
Bylnx209YX
ICLR.cc/2019/Conference
2019
Adversarial Attacks on Graph Neural Networks via Meta Learning
["Daniel Z\u00fcgner", "Stephan G\u00fcnnemann"]
Deep learning models for graphs have advanced the state of the art on many tasks. Despite their recent success, little is known about their robustness. We investigate training time attacks on graph neural networks for node classification that perturb the discrete graph structure. Our core principle is to use meta-grad...
["graph mining", "adversarial attacks", "meta learning", "graph neural networks", "node classification"]
ABSTRACTDeep learning models for graphs have advanced the state of the art on many tasks.Despite their recent success, little is known about their robustness. We investigatetraining time attacks on graph neural networks for node classification that perturbthe discrete graph structure. Our core principle is to use meta-...
H1lFhSrF3X
Good paper of using meta-learning to solve the bilevel optimization problem in graph attacking
7: Good paper, accept
This paper proposes an algorithm to alter the structure of a graph by adding/deleting edges so as to degrade the global performance of node classification. The main idea is to use the idea of meta-gradients from meta-learning to solve the bilevel optimization problem. The paper is clearly presented. The main contribu...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Adversarial Attacks on Graph Neural Networks via Meta Learning ### Paper Abstract Deep learning models for graphs have advanced the state of the art on many tasks. Despi...
HJgK0h4Ywr
ICLR.cc/2020/Conference
2020
Theory and Evaluation Metrics for Learning Disentangled Representations
["Kien Do", "Truyen Tran"]
We make two theoretical contributions to disentanglement learning by (a) defining precise semantics of disentangled representations, and (b) establishing robust metrics for evaluation. First, we characterize the concept “disentangled representations” used in supervised and unsupervised methods along three dimensions–in...
["disentanglement", "metrics"]
ABSTRACTWe make two theoretical contributions to disentanglement learning by (a) defin-ing precise semantics of disentangled representations, and (b) establishing robustmetrics for evaluation. First, we characterize the concept “disentangled represen-tations” used in supervised and unsupervised methods along three dime...
Bkg0XRg6KB
Official Blind Review #2
6: Weak Accept
This paper provides a mathematically-grounded set of axes to describe the effectiveness of latent representations: informativeness (as the mutual information between input and z), separability (I(x, zi, zj) = 0, i!=j), and interpretability. They accompany these with metrics to evaluate representations across those crit...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Theory and Evaluation Metrics for Learning Disentangled Representations ### Paper Abstract We make two theoretical contributions to disentanglement learning by (a) defin...
ByeLmn0qtX
ICLR.cc/2019/Conference
2019
Variational Domain Adaptation
["Hirono Okamoto", "Shohei Ohsawa", "Itto Higuchi", "Haruka Murakami", "Mizuki Sango", "Zhenghang Cui", "Masahiro Suzuki", "Hiroshi Kajino", "Yutaka Matsuo"]
This paper proposes variational domain adaptation, a unified, scalable, simple framework for learning multiple distributions through variational inference. Unlike the existing methods on domain transfer through deep generative models, such as StarGAN (Choi et al., 2017) and UFDN (Liu et al., 2018), the variational doma...
["domain adaptation", "variational inference", "multi-domain"]
ABSTRACTThis paper proposes variational domain adaptation , a unified, scalable, simpleframework for learning multiple distributions through variational inference. Un-like the existing methods on domain transfer through deep generative models, suchas StarGAN (Choi et al., 2017) and UFDN (Liu et al., 2018), the variatio...
HJlwXuL527
A preference learning generative model (in deep setting), with somewhat unintuitive setting and weak experimental evaluation
4: Ok but not good enough - rejection
1) Summary of the paper: The paper brings up a relatively new problem of learning a generative model for multiple domains. The domains, D1,...,Dn, may refer to person-specific preferred images, for instance, and they focus on how to build generative models P(x|Di), which represents a set of images preferred by subject...
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 Variational Domain Adaptation ### Paper Abstract This paper proposes variational domain adaptation, a unified, scalable, simple framework for learning multiple distribut...
NMgB4CVnMh
ICLR.cc/2021/Conference
2021
Acoustic Neighbor Embeddings
["Woojay Jeon"]
This paper proposes a novel acoustic word embedding called Acoustic Neighbor Embeddings where speech or text of arbitrary length are mapped to a vector space of fixed, reduced dimensions by adapting stochastic neighbor embedding (SNE) to sequential inputs. The Euclidean distance between coordinates in the embedding spa...
["text", "dimensions", "form", "embeddings", "novel acoustic word", "acoustic neighbor embeddings", "speech", "arbitrary length", "vector space"]
ABSTRACTThis paper proposes a novel acoustic word embedding called Acoustic Neigh-bor Embeddings where speech or text of arbitrary length are mapped to a vectorspace of fixed, reduced dimensions by adapting stochastic neighbor embedding(SNE) to sequential inputs. The Euclidean distance between coordinates in theembeddi...
b7dA9eXjf9M
Interesting idea, experimental validation falls short
6: Marginally above acceptance threshold
### Overview: This paper proposes a new acoustic word embedding approach, where acoustic and text embeddings are jointly learned. The text encoder takes either phonemes or characters as input. The novelty of the paper lies in a new loss, which is based on stochastic neighbour embeddings (SNE). The acoustic embedding n...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Acoustic Neighbor Embeddings ### Paper Abstract This paper proposes a novel acoustic word embedding called Acoustic Neighbor Embeddings where speech or text of arbitrary...
rkgTdkrtPH
ICLR.cc/2020/Conference
2020
NoiGAN: NOISE AWARE KNOWLEDGE GRAPH EMBEDDING WITH GAN
["Kewei Cheng", "Yikai Zhu", "Ming Zhang", "Yizhou Sun"]
Knowledge graph has gained increasing attention in recent years for its successful applications of numerous tasks. Despite the rapid growth of knowledge construction, knowledge graphs still suffer from severe incompletion and inevitably involve various kinds of errors. Several attempts have been made to complete knowle...
["Knowledge graph embedding", "Noise aware"]
ABSTRACTKnowledge graph has gained increasing attention in recent years for its success-ful applications of numerous tasks. Despite the rapid growth of knowledge con-struction, knowledge graphs still suffer from severe incompletion and inevitablyinvolve various kinds of errors. Several attempts have been made to comple...
HklhuRpxiH
Official Blind Review #4
3: Weak Reject
This paper presented a jointly learning framework based on GAN for tackling both knowledge graph completion and noise detection simultaneously. Existing works only deal with each of task independently and did not investigate the benefits of coping with both tasks together. The paper is well motivated. In order to achie...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title NoiGAN: NOISE AWARE KNOWLEDGE GRAPH EMBEDDING WITH GAN ### Paper Abstract Knowledge graph has gained increasing attention in recent years for its successful applications...
rJejta4KDS
ICLR.cc/2020/Conference
2020
SELF-KNOWLEDGE DISTILLATION ADVERSARIAL ATTACK
["Ma Xiaoxiong[1]", "Wang Renzhi[1]", "Tian Cong", "Dong Zeqian", "Duan Zhenhua"]
Neural networks show great vulnerability under the threat of adversarial examples. By adding small perturbation to a clean image, neural networks with high classification accuracy can be completely fooled. One intriguing property of the adversarial examples is transferability. This property allows adversarial ex...
["Adversarial Examples", "Transferability", "black-box targeted attack", "Distillation"]
ABSTRACTNeural networks show great vulnerability under the threat of adversarial exam-ples. By adding small perturbation to a clean image, neural networks with highclassification accuracy can be completely fooled. One intriguing property of theadversarial examples is transferability. This property allows adversarial ex...
ByeNLz1sFH
Official Blind Review #3
1: Reject
This paper proposes an attack method to improve the transferability of targeted adversarial examples. The proposed method uses a temperature T to convert the logit of the network, and calculates the gradient based on the new logit, yielding the distillation-based attack method. It has been integrated into FGSM and MI-F...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title SELF-KNOWLEDGE DISTILLATION ADVERSARIAL ATTACK ### Paper Abstract Neural networks show great vulnerability under the threat of adversarial examples. By adding small p...
YtgRjBw-7GJ
NeurIPS.cc/2022/Challenge/CellSeg
2022
Uncertainty-Aware Contour Proposal Networks for Cell Segmentation in Multi-Modality High-Resolution Microscopy Images
["Eric Upschulte", "Stefan Harmeling", "Katrin Amunts", "Timo Dickscheid"]
We present a simple framework for cell segmentation, based on uncertainty-aware Contour Proposal Networks (CPNs). It is designed to provide high segmentation accuracy while remaining computationally efficient, which makes it an ideal solution for high throughput microscopy applications. Each predicted cell is provided ...
["cell", "segmentation", "detection", "celldetection", "microscopy", "cpn", "deep", "learning", "resnet", "resnext", "contour", "proposal", "network"]
Uncertainty-Aware Contour Proposal Networks forCell Segmentation in Multi-Modality High-ResolutionMicroscopy ImagesEric Upschulte1,2, Stefan Harmeling3, Katrin Amunts1,4, Timo Dickscheid1,2,51Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich, Germany2Helmholtz AI, Research Centre Jülich, Germany3...
ZMGkkkjf2h
Clean and unified solution as well as strong results
9: Top 15% of accepted papers, strong accept
Summary: The authors present a novel framework for multi-modal cell segmentation. The fundamental segmentation structure is a contour proposal network. Besides, the authors propose an uncertainty estimation and new contour supervision to improve the segmentation accuracy. Finally, the framework achieves great segment...
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 Uncertainty-Aware Contour Proposal Networks for Cell Segmentation in Multi-Modality High-Resolution Microscopy Images ### Paper Abstract We present a simple framework fo...
SJUX_MWCZ
ICLR.cc/2018/Conference
2018
Predict Responsibly: Increasing Fairness by Learning to Defer
["David Madras", "Toniann Pitassi", "Richard Zemel"]
When machine learning models are used for high-stakes decisions, they should predict accurately, fairly, and responsibly. To fulfill these three requirements, a model must be able to output a reject option (i.e. say "``I Don't Know") when it is not qualified to make a prediction. In this work, we propose learning to de...
["Fairness", "IDK", "Calibration", "Automated decision-making", "Transparency", "Accountability"]
ABSTRACTWhen machine learning models are used for high-stakes decisions, they shouldpredict accurately, fairly, and responsibly. To fulfill these three requirements, amodel must be able to output a reject option (i.e. say “I Don’t Know”) when itis not qualified to make a prediction. In this work, we propose learning to...
HJMCQAFgf
not in scope of call for papers
4: Ok but not good enough - rejection
I like the direction this paper is going in by combining fairness objectives with deferment criteria and learning. However, I do not believe that this paper is in the scope defined by the ICLR call for papers, as there is nothing related to learned representations in it. I find it quite interesting that the authors g...
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 Predict Responsibly: Increasing Fairness by Learning to Defer ### Paper Abstract When machine learning models are used for high-stakes decisions, they should predict acc...
S1g7tpEYDS
ICLR.cc/2020/Conference
2020
From Variational to Deterministic Autoencoders
["Partha Ghosh", "Mehdi S. M. Sajjadi", "Antonio Vergari", "Michael Black", "Bernhard Scholkopf"]
Variational Autoencoders (VAEs) provide a theoretically-backed and popular framework for deep generative models. However, learning a VAE from data poses still unanswered theoretical questions and considerable practical challenges. In this work, we propose an alternative framework for generative modeling that is simple...
["Unsupervised learning", "Generative Models", "Variational Autoencoders", "Regularization"]
ABSTRACTVariational Autoencoders (V AEs) provide a theoretically-backed and popularframework for deep generative models. However, learning a V AE from data posesstill unanswered theoretical questions and considerable practical challenges. Inthis work, we propose an alternative framework for generative modeling that iss...
rkxPP-2itH
Official Blind Review #2
6: Weak Accept
This paper propose an extension to deterministic autoencoders. Motivated from VAEs, the authors propose RAEs, which replace the noise injection in the encoders of VAEs with an explicit regularization term on the latent representations. As a result, the model becomes a deterministic autoencoder with a L_2 regularization...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title From Variational to Deterministic Autoencoders ### Paper Abstract Variational Autoencoders (VAEs) provide a theoretically-backed and popular framework for deep generati...
uxpzitPEooJ
ICLR.cc/2021/Conference
2021
Graph Coarsening with Neural Networks
["Chen Cai", "Dingkang Wang", "Yusu Wang"]
As large scale-graphs become increasingly more prevalent, it poses significant computational challenges to process, extract and analyze large graph data. Graph coarsening is one popular technique to reduce the size of a graph while maintaining essential properties. Despite rich graph coarsening literature, there is onl...
["graph coarsening", "graph neural network", "Doubly-weighted Laplace operator"]
ABSTRACTAs large-scale graphs become increasingly more prevalent, it poses significantcomputational challenges to process, extract and analyze large graph data. Graphcoarsening is one popular technique to reduce the size of a graph while main-taining essential properties. Despite rich graph coarsening literature, there...
V70WqmsgR1p
Solid paper with theoretical and experimental results.
6: Marginally above acceptance threshold
The paper studied the problem of graph coarsening in the context of data-driven deep generative models. The authors studied a family of Laplacian operators and differentiable losses in order to construct high-quality coarse graphs. The paper is well-motivated by providing extensive theoretical analysis to support the r...
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 Graph Coarsening with Neural Networks ### Paper Abstract As large scale-graphs become increasingly more prevalent, it poses significant computational challenges to proce...
r1eVMnA9K7
ICLR.cc/2019/Conference
2019
Unsupervised Control Through Non-Parametric Discriminative Rewards
["David Warde-Farley", "Tom Van de Wiele", "Tejas Kulkarni", "Catalin Ionescu", "Steven Hansen", "Volodymyr Mnih"]
Learning to control an environment without hand-crafted rewards or expert data remains challenging and is at the frontier of reinforcement learning research. We present an unsupervised learning algorithm to train agents to achieve perceptually-specified goals using only a stream of observations and actions. Our agent s...
["deep reinforcement learning", "goals", "UVFA", "mutual information"]
ABSTRACTLearning to control an environment without hand-crafted rewards or expert dataremains challenging and is at the frontier of reinforcement learning research. Wepresent an unsupervised learning algorithm to train agents to achieve perceptually-specified goals using only a stream of observations and actions. Our a...
SygRw182s7
A strong paper with innovative ideas, but somewhat unclear methods and results
7: Good paper, accept
In this paper, the authors address the problem of learning to achieve perceptually specified goals in a fully unsupervised way. For doing so, they simultaneously learn a goal-conditioned policy and a goal achievement reward function based on the mutual information between goals sampled from an a priori distribution an...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Unsupervised Control Through Non-Parametric Discriminative Rewards ### Paper Abstract Learning to control an environment without hand-crafted rewards or expert data rema...
EyDgK7q5vwJ
ICLR.cc/2021/Conference
2021
Streamlining EM into Auto-Encoder Networks
["Yuangang Pan", "Ivor Tsang"]
We present a new deep neural network architecture, named EDGaM, for deep clustering. This architecture can seamlessly learn deep auto-encoders and capture common group features of complex inputs in the encoded latent space. The key idea is to introduce a differentiable Gaussian mixture neural network between an encoder...
["Deep Clustering", "Differentiable EM"]
ABSTRACTWe present a new deep neural network architecture, named EDGaM, for deepclustering. This architecture can seamlessly learn deep auto-encoders and capturecommon group features of complex inputs in the encoded latent space. Thekey idea is to introduce a differentiable Gaussian mixture network between anencoder an...
NlpXob7sqbS
Deep clustering with streaming EM
5: Marginally below acceptance threshold
This paper proposed a streaming EM algorithm to learn latent representations and cluster in the latent space simultaneously, through a deep GMM model. The EM algorithm was unfolded into network layers as the forward pass, and the global parameters (GMM, encoder, decoder) was optimized through the backward propagation. ...
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 Streamlining EM into Auto-Encoder Networks ### Paper Abstract We present a new deep neural network architecture, named EDGaM, for deep clustering. This architecture can ...
3YQAVD9_Dz3
ICLR.cc/2021/Conference
2021
NOSE Augment: Fast and Effective Data Augmentation Without Searching
["Qingrui Li", "Song Xie", "An\u0131l Oymagil", "Mustafa Furkan Eseoglu", "Ziyin Zhang", "CM Lee"]
Data augmentation has been widely used for enhancing the diversity of training data and model generalization. Different from traditional handcrafted methods, recent research introduced automated search for optimal data augmentation policies and achieved state-of-the-art results on image classification tasks. However, t...
["data augmentation", "stochastic policy", "multi-stage augmentation"]
ABSTRACTData augmentation has been widely used for enhancing the diversity of train-ing data and model generalization. Different from traditional handcrafted meth-ods, recent research introduced automated search for optimal data augmentationpolicies and achieved state-of-the-art results on image classification tasks. H...
MDDfeoi4NYB
Review of NOSE Augment
3: Clear rejection
This paper aims to provide an effective augmentation strategy without the need for a separate search. The resulting method is called NOSE Augment, which is presented as a substitute for the previous AutoAugment type methods (e.g. Fast AutoAugment, Population Based Augmentation, RandAugment, Adversarial AutoAugment etc....
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 NOSE Augment: Fast and Effective Data Augmentation Without Searching ### Paper Abstract Data augmentation has been widely used for enhancing the diversity of training da...
OruB0mPF1Ef
NeurIPS.cc/2022/Workshop/SVRHM
2022
Joint rotational invariance and adversarial training of a dual-stream Transformer yields state of the art Brain-Score for Area V4
["William Berrios", "Arturo Deza"]
Modern high-scoring models of vision in the brain score competition do not stem from Vision Transformers. However, in this paper, we provide evidence against the unexpected trend of Vision Transformers (ViT) being not perceptually aligned with human visual representations by showing how a dual-stream Transformer, a Cro...
["Vision Transformer", "Brain-Score competition", "adversarial training", "rotation invariance."]
Joint rotational invariance and adversarial training ofa dual-stream Transformer yields state of the artBrain-Score for Area V4Anonymous Author(s)AffiliationAddressemailAbstractModern high-scoring models of vision in the brain score competition do not stem 1from Vision Transformers. However, in this paper, we provide e...
kyyk9Lw3lBl
It is a work that fits a bio-implausible machine learning architecture to vision. By switching from the kind of cost functions from which transformers arise, to those that can induce higher brain-score, the authors showed that transformers can be tuned to behave quite like IT and decently like V1. The result seems to s...
6: Marginally above acceptance threshold
Quality: good Clarity: very good Originality: fitting a biologically implausible artificial neural network to a brain is an innovation in the sense that, while obvious, most people won't take this route. Adversarial training is recently known, rotational invariance was well known, training using those I regard as incr...
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 Joint rotational invariance and adversarial training of a dual-stream Transformer yields state of the art Brain-Score for Area V4 ### Paper Abstract Modern high-scoring ...
Hkl_sAVtwr
ICLR.cc/2020/Conference
2020
Compressed Sensing with Deep Image Prior and Learned Regularization
["Dave Van Veen", "Ajil Jalal", "Mahdi Soltanolkotabi", "Eric Price", "Sriram Vishwanath", "Alexandros G. Dimakis"]
We propose a novel method for compressed sensing recovery using untrained deep generative models. Our method is based on the recently proposed Deep Image Prior (DIP), wherein the convolutional weights of the network are optimized to match the observed measurements. We show that this approach can be applied to solve any...
["compressed sensing", "sparsity", "inverse problems"]
ABSTRACTWe propose a novel method for compressed sensing recovery using untrained deepgenerative models. Our method is based on the recently proposed Deep ImagePrior (DIP), wherein the convolutional weights of the network are optimized tomatch the observed measurements. We show that this approach can be applied tosolve...
HkecZhUFcr
Official Blind Review #5
6: Weak Accept
This paper proposes a deep learning based compressed sensing method, CS-DIP which employs deep image prior algorithm to recovering signals especially for medical images from noisy measurements without pre-learning over large dataset. For further optimization, a learned regularization is leveraged over small training da...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Compressed Sensing with Deep Image Prior and Learned Regularization ### Paper Abstract We propose a novel method for compressed sensing recovery using untrained deep gen...
rybAWfx0b
ICLR.cc/2018/Conference
2018
COLD FUSION: TRAINING SEQ2SEQ MODELS TOGETHER WITH LANGUAGE MODELS
["Anuroop Sriram", "Heewoo Jun", "Sanjeev Satheesh", "Adam Coates"]
Sequence-to-sequence (Seq2Seq) models with attention have excelled at tasks which involve generating natural language sentences such as machine translation, image captioning and speech recognition. Performance has further been improved by leveraging unlabeled data, often in the form of a language model. In this work, w...
["Sequence-to-Sequence Models", "Speech Recognition", "Language Models"]
ABSTRACTSequence-to-sequence (Seq2Seq) models with attention have excelled at taskswhich involve generating natural language sentences such as machine translation,image captioning and speech recognition. Performance has further been improvedby leveraging unlabeled data, often in the form of a language model. In this wo...
Sy0xMaHlG
review
5: Marginally below acceptance threshold
The paper proposes a novel approach to integrate a language model (LM) to a seq2seq based speech recognition system (ASR). The LM is pretrained on separate data (presumably larger, potentially not the same exact distribution). It has a similar flavor as DeepFusion (DF), a previous work which also integrated an LM to a ...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title COLD FUSION: TRAINING SEQ2SEQ MODELS TOGETHER WITH LANGUAGE MODELS ### Paper Abstract Sequence-to-sequence (Seq2Seq) models with attention have excelled at tasks which i...
HJeu43ActQ
ICLR.cc/2019/Conference
2019
NOODL: Provable Online Dictionary Learning and Sparse Coding
["Sirisha Rambhatla", "Xingguo Li", "Jarvis Haupt"]
We consider the dictionary learning problem, where the aim is to model the given data as a linear combination of a few columns of a matrix known as a dictionary, where the sparse weights forming the linear combination are known as coefficients. Since the dictionary and coefficients, parameterizing the linear model are ...
["dictionary learning", "provable dictionary learning", "online dictionary learning", "sparse coding", "support recovery", "iterative hard thresholding", "matrix factorization", "neural architectures", "neural networks", "noodl"]
ABSTRACTWe consider the dictionary learning problem, where the aim is to model the givendata as a linear combination of a few columns of a matrix known as a dictionary ,where the sparse weights forming the linear combination are known as coeffi-cients . Since the dictionary and coefficients, parameterizing the linear m...
B1x81eSE2m
We found this work interesting. We think that some issues need to be further addressed by the authors.
7: Good paper, accept
The main contributions of this work are essentially on the theoretical aspects. It seems that the proposed algorithm is not very original because its two parts, namely prediction (coefficient estimation) and learning (dictionary update) have been widely used in the literature, using respectively a IHT and a gradient de...
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 NOODL: Provable Online Dictionary Learning and Sparse Coding ### Paper Abstract We consider the dictionary learning problem, where the aim is to model the given data as ...
HygW26VYwS
ICLR.cc/2020/Conference
2020
Attention Privileged Reinforcement Learning for Domain Transfer
["Sasha Salter", "Dushyant Rao", "Markus Wulfmeier", "Raia Hadsell", "Ingmar Posner"]
Applying reinforcement learning (RL) to physical systems presents notable challenges, given requirements regarding sample efficiency, safety, and physical constraints compared to simulated environments. To enable transfer of policies trained in simulation, randomising simulation parameters leads to more robust policies...
["sim-to-real", "domain randomisation", "attention", "transfer learning", "reinforcement learning"]
ABSTRACTApplying reinforcement learning (RL) to physical systems presents notable chal-lenges, given requirements regarding sample efficiency, safety, and physical con-straints compared to simulated environments. To enable transfer of policiestrained in simulation, randomising simulation parameters leads to more robust...
B1gwk6-RFH
Official Blind Review #2
3: Weak Reject
Overall the method is interesting but I am not overly convinced the method provides a significant improvement over simpler methods that are not compared to the work here. Also, the generalization and extrapolation experiments need more description. As well, the work talks about how this method can be used to accelerate...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Attention Privileged Reinforcement Learning for Domain Transfer ### Paper Abstract Applying reinforcement learning (RL) to physical systems presents notable challenges, ...
H139Q_gAW
ICLR.cc/2018/Conference
2018
Learning Graph Convolution Filters from Data Manifold
["Guokun Lai", "Hanxiao Liu", "Yiming Yang"]
Convolution Neural Network (CNN) has gained tremendous success in computer vision tasks with its outstanding ability to capture the local latent features. Recently, there has been an increasing interest in extending CNNs to the general spatial domain. Although various types of graph convolution and geometric convolutio...
["Label Propagation", "Depthwise separable convolution", "Graph and geometric convolution"]
ABSTRACTConvolution Neural Network (CNN) has gained tremendous success in computervision tasks with its outstanding ability to capture the local latent features. Re-cently, there has been an increasing interest in extending CNNs to the general spa-tial domain. Although various types of graph convolution and geometric c...
S1rTmy9xG
Extension of Depth-Wise-Convolution (Chollet et al. 2016) with improved performance.
6: Marginally above acceptance threshold
The paper presents an extension of the Xception network of (Chollet et al. 2016) 2D grids to generic graphs. The Xception network decouples the spatial correlations from depth channels correlations by having separate weights for each depth channel. The weights within a depth channel is shared thus maintaining the stati...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning Graph Convolution Filters from Data Manifold ### Paper Abstract Convolution Neural Network (CNN) has gained tremendous success in computer vision tasks with its...
HJG7m2AcF7
ICLR.cc/2019/Conference
2019
Context Mover's Distance & Barycenters: Optimal transport of contexts for building representations
["Sidak Pal Singh", "Andreas Hug", "Aymeric Dieuleveut", "Martin Jaggi"]
We propose a unified framework for building unsupervised representations of entities and their compositions, by viewing each entity as a histogram (or distribution) over its contexts. This enables us to take advantage of optimal transport and construct representations that effectively harness the geometry of the underl...
["representation learning", "wasserstein distance", "wasserstein barycenter", "entailment"]
ABSTRACTWe propose a unified framework for building unsupervised representations of enti-ties and their compositions, by viewing each entity as a histogram (ordistribution )over its contexts. This enables us to take advantage of optimal transport and con-struct representations that effectively harness the geometry of t...
SJgryzvchX
Interesting approach, proposes representation augmentation as opposed to representation learning and the proposed distance not used for training.
6: Marginally above acceptance threshold
The paper proposes a method to augment representation of an entity (such as a word) from standard "point in a vector space" to a histogram with bins located at some points in that vector space. In this model, the bins correspond the context objects, the location of which are the standard point embedding of those object...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Context Mover's Distance & Barycenters: Optimal transport of contexts for building representations ### Paper Abstract We propose a unified framework for building unsuper...
LpSGtq6F5xN
ICLR.cc/2021/Conference
2021
A Mixture of Variational Autoencoders for Deep Clustering
["Avi Caciularu", "Jacob Goldberger"]
In this study, we propose a deep clustering algorithm that utilizes a variational autoencoder (VAE) framework with a multi encoder-decoder neural architecture. This setup enforces a complementary structure that guides the learned latent representations towards a more meaningful space arrangement. It differs from previo...
["deep clustering", "variational auto encoder", "VAE"]
ABSTRACTIn this study, we propose a deep clustering algorithm that utilizes a variationalautoencoder (V AE) framework with a multi encoder-decoder neural architecture.This setup enforces a complementary structure that guides the learned latent rep-resentations towards a more meaningful space arrangement. It differs fro...
TPtLdLIeoB6
The authors propose a variant of variational deep clustering to include K encoder-decoder neural architectures to improve the quality of the data generations and the clustering performance. The results are rather convincing and the idea is worth exploring, however the paper contains many weaknesses that should be addre...
5: Marginally below acceptance threshold
1) Quality of the writing. The paper is rather difficult to read because of many typos and confusing statements. The authors should spend more time in refining the clarity of the presentation. Some suggestions: run a spell checker, avoid sloppy statements, be more meticulous in the description of related works and imp...
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 Mixture of Variational Autoencoders for Deep Clustering ### Paper Abstract In this study, we propose a deep clustering algorithm that utilizes a variational autoencode...
rkx9_gHtvS
ICLR.cc/2020/Conference
2020
Interactive Classification by Asking Informative Questions
["Lili Yu", "Howard Chen", "Sida I. Wang", "Yoav Artzi and Tao Lei"]
We propose an interactive classification approach for natural language queries. Instead of classifying given the natural language query only, we ask the user for additional information using a sequence of binary and multiple-choice questions. At each turn, we use a policy controller to decide if to present a ques...
["NLP", "interactive classification", "interactive system", "text classification", "data collection"]
ABSTRACTNatural language systems often rely on a single, potentially ambiguous input tomake one final prediction, which may simplify the problem but degrade end userexperience. Instead of making predictions with the natural language query only,we ask the user for additional information using a small number of binary an...
Hye5eDzGKH
Official Blind Review #2
6: Weak Accept
The authors show that a query classification system can be improved by providing additional information to the system. In particular, additional ‘tags’ are converted to question/answer pairs and the system can choose to ‘ask’ these questions to obtain the ‘answers’ before finalizing its decision. They use a determinist...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Interactive Classification by Asking Informative Questions ### Paper Abstract We propose an interactive classification approach for natural language queries. Inste...
whNntrHtB8D
ICLR.cc/2021/Conference
2021
Gradient Based Memory Editing for Task-Free Continual Learning
["Xisen Jin", "Junyi Du", "Xiang Ren"]
Prior work on continual learning often operate in a “task-aware” manner, by assuming that the task boundaries and identifies of the data examples are known at all times. While in practice, it is rarely the case that such information are exposed to the methods (i.e., thus called “task-free”)–a setting that is relatively...
["Continual learning", "task-free continual learning"]
ABSTRACTContinual learning often assumes a knowledge of (strict) task boundaries andidentities for the instances in a data stream— i.e., a “task-aware” setting. However,in practice it is rarely the case that practitioners can expose task information tothe model; thus needing “task-free” continual learning methods. Rece...
RM3qGt9pr8
Intriguing Empirical Results, But Lacking Theoretical Justification
6: Marginally above acceptance threshold
## Summary This paper proposes a task-free continual learning method called GMED that extends experience replay. The key idea is to modify the individual data points in the replay memory to maximize the one-step forgetting at the current time step. --- ## Pros - The idea of editing the data in a replay buffer with ...
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 Memory Editing for Task-Free Continual Learning ### Paper Abstract Prior work on continual learning often operate in a “task-aware” manner, by assuming th...
ByW2Avqgg
ICLR.cc/2017/conference
2017
Neural Causal Regularization under the Independence of Mechanisms Assumption
["Mohammad Taha Bahadori", "Krzysztof Chalupka", "Edward Choi", "Robert Chen", "Walter F. Stewart", "Jimeng Sun"]
Neural networks provide a powerful framework for learning the association between input and response variables and making accurate predictions. However, in many applications such as healthcare, it is important to identify causal relationships between the inputs and the response variables to be able to change the respon...
["Deep learning", "Applications"]
ABSTRACTNeural networks provide a powerful framework for learning the association betweeninput and response variables and making accurate predictions and offer promise inusing the rapidly growing volume of health care data to surface causal relationshipsthat cannot necessarily be tested in randomized clinical trials. I...
HJAeeLH4x
Possibly practical method, but methodology and analysis fall short
5: Marginally below acceptance threshold
This paper proposes to use a causality score to weight a sparsity regularizer. In that way, selected variables trade off between being causal and discriminative. The framework is primarily evaluated on a proprietary health dataset. While the dataset does give a good motivation to the problem setting, the paper falls...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Neural Causal Regularization under the Independence of Mechanisms Assumption ### Paper Abstract Neural networks provide a powerful framework for learning the association...
r1Aab85gg
ICLR.cc/2017/conference
2017
Offline bilingual word vectors, orthogonal transformations and the inverted softmax
["Samuel L. Smith", "David H. P. Turban", "Steven Hamblin", "Nils Y. Hammerla"]
Usually bilingual word vectors are trained "online''. Mikolov et al. showed they can also be found "offline"; whereby two pre-trained embeddings are aligned with a linear transformation, using dictionaries compiled from expert knowledge. In this work, we prove that the linear transformation between two spaces should be...
["Natural language processing", "Transfer Learning", "Applications"]
ABSTRACTUsually bilingual word vectors are trained “online”. Mikolov et al. (2013a)showed they can also be found “offline”; whereby two pre-trained embeddingsare aligned with a linear transformation, using dictionaries compiled from expertknowledge. In this work, we prove that the linear transformation between twospace...
rkM8HK44l
Review
6: Marginally above acceptance threshold
This paper extends preceding works to create a mapping between the word embedding space of two languages. The word embeddings had been independently trained on monolingual data only, and various forms of bilingual information is used to learn the mapping. This mapping is then used to measure the precision of translatio...
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 bilingual word vectors, orthogonal transformations and the inverted softmax ### Paper Abstract Usually bilingual word vectors are trained "online''. Mikolov et a...
qf6Nmm-_6Z
ICLR.cc/2021/Conference
2021
VECoDeR - Variational Embeddings for Community Detection and Node Representation
["Rayyan Ahmad Khan", "Muhammad Umer Anwaar", "Omran Kaddah", "Martin Kleinsteuber"]
In this paper, we study how to simultaneously learn two highly correlated tasks of graph analysis, i.e., community detection and node representation learning. We propose an efficient generative model called VECoDeR for jointly learning Variational Embeddings for COmmunity DEtection and node Representation. VECoDeR assu...
["Generative Models", "Node representation", "VECoDeR", "Graph Neural Networks", "Variational Embeddings"]
ABSTRACTIn this paper, we study how to simultaneously learn two highly correlated tasksof graph analysis, i.e., community detection and node representation learning.We propose an efficient generative model called VEC ODERfor jointly learn-ingVariational Embeddings for Community Detection and node Representation.VEC ODE...
CmDoD71gNBL
Interesting problem, but limited novelty
5: Marginally below acceptance threshold
This paper proposes a generative model for community detection and node representation in a unified framework. The underlying assumption of the work is that connected nodes in a graph should have similar node embeddings and similar community assignments. Experimental evaluations show improvement both on node classifica...
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 VECoDeR - Variational Embeddings for Community Detection and Node Representation ### Paper Abstract In this paper, we study how to simultaneously learn two highly correl...
BBxeo2Vuvbq
ICLR.cc/2022/Workshop/DGM4HSD
2022
Conditional Generative Quantile Networks via Optimal Transport
["Jesse Sun", "Dihong Jiang", "Yaoliang Yu"]
Quantile regression has a natural extension to generative modelling by leveraging a stronger pointwise convergence than in distribution. While the pinball quantile loss works well in the scalar case, it cannot be readily extended to the vector case. In this work, we propose a multivariate quantile approach for generati...
["Optimal Transport", "Generative Models", "Quantile Functions", "Time-Series Forecasting"]
ABSTRACTQuantile regression has a natural extension to generative modelling by leveraginga stronger pointwise convergence than in distribution. While the pinball quantileloss works well in the scalar case, it cannot be readily extended to the vector case.In this work, we propose a multivariate quantile approach for gen...
rTvlbXmz9G9
Quantile multivariate quantile generative model with optimal transport
6: Marginally above acceptance threshold
## Summary The goal of this paper is to generalize quantile generative modelling using optimal transport to the multivariate case. Quantile regression aims at reproducing the quantile function as an approximation of the data probability distribution. Previous approaches do not perform well in the multivariate case, an...
2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Conditional Generative Quantile Networks via Optimal Transport ### Paper Abstract Quantile regression has a natural extension to generative modelling by leveraging a str...
_GmNcCEAea3
eswc-conferences.org/ESWC/2021/Conference/Resources_Track
2021
ParaQA: A Question Answering Dataset with Paraphrase Responses for Single-Turn Conversation
["Endri Kacupaj", "Barshana Banerjee", "Kuldeep Singh", "Jens Lehmann"]
This paper presents ParaQA, a question answering (QA) dataset with multiple paraphrased responses for single-turn conversation over knowledge graphs (KG). The dataset was created using a semi-automated framework for generating diverse paraphrasing of the answers using techniques such as back-translation. The existing d...
["Question Answering", "Paraphrase Responses", "Single-Turn Conversation", "Knowledge Graph", "Dataset"]
ParaQA: A Question Answering Dataset withParaphrase Responses for Single-Turn ConversationEndri Kacupaj1, Barshana Banerjee1, Kuldeep Singh2, and Jens Lehmann1;31University of Bonn, Bonn, Germanyfkacupaj,jens.lehmann g@cs.uni-bonn.de s6babane@uni-bonn.de2Zerotha Research and Cerence GmbH, Germanykuldeep.singh1@cerence....
ml4Kdcmez4X
A valuable resource for KBQA improvements
2: Accept
This paper presents a new dataset for KBQA (knowledge base question answering). The unique characteristic of this dataset, ParaQA, is that each question corresponds to multiple (2 to 8) paraphrased answers, whereas all existing datasets have only one answer for each question. This one-to-many relationship allows a KBQA...
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 ParaQA: A Question Answering Dataset with Paraphrase Responses for Single-Turn Conversation ### Paper Abstract This paper presents ParaQA, a question answering (QA) data...
rkgyS0VFvr
ICLR.cc/2020/Conference
2020
DBA: Distributed Backdoor Attacks against Federated Learning
["Chulin Xie", "Keli Huang", "Pin-Yu Chen", "Bo Li"]
Backdoor attacks aim to manipulate a subset of training data by injecting adversarial triggers such that machine learning models trained on the tampered dataset will make arbitrarily (targeted) incorrect prediction on the testset with the same trigger embedded. While federated learning (FL) is capable of aggregating in...
["distributed backdoor attack", "federated learning"]
ABSTRACTBackdoor attacks aim to manipulate a subset of training data by injecting adversarialtriggers such that machine learning models trained on the tampered dataset willmake arbitrarily (targeted) incorrect prediction on the testset with the same triggerembedded. While federated learning (FL) is capable of aggregati...
SJe4yM7lcr
Official Blind Review #3
6: Weak Accept
This paper proposes a distributed backdoor attack strategy, framed differently from the previous two main approches (1) the centralised backdoor approach and (2) the (less discussed in the paper) distributed fault tolerance approach (often named "Byzantine"). The authors show through experiments how their attack is mo...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title DBA: Distributed Backdoor Attacks against Federated Learning ### Paper Abstract Backdoor attacks aim to manipulate a subset of training data by injecting adversarial tri...
K5YasWXZT3O
ICLR.cc/2021/Conference
2021
Tilted Empirical Risk Minimization
["Tian Li", "Ahmad Beirami", "Maziar Sanjabi", "Virginia Smith"]
Empirical risk minimization (ERM) is typically designed to perform well on the average loss, which can result in estimators that are sensitive to outliers, generalize poorly, or treat subgroups unfairly. While many methods aim to address these problems individually, in this work, we explore them through a unified frame...
["exponential tilting", "models of learning and generalization", "label noise robustness", "fairness"]
ABSTRACTEmpirical risk minimization (ERM) is typically designed to perform well on theaverage loss, which can result in estimators that are sensitive to outliers, generalizepoorly, or treat subgroups unfairly. While many methods aim to address theseproblems individually, in this work, we explore them through a unified ...
XV4Nntx_9Ac
The paper explore relevant applications of exponential smoothing in several ML problems. It is well written and should be interesting for ML practioners.
6: Marginally above acceptance threshold
The paper provide nice discussions of a simple modification of ERM paradigm. Mainly, it consists in applying an exponential smoothing to the loss function. The authors provide several interpretations and connexions with (robustness/fairness/quantile regression etc...) literature and show how the TERM can be adapted to ...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Tilted Empirical Risk Minimization ### Paper Abstract Empirical risk minimization (ERM) is typically designed to perform well on the average loss, which can result in es...
0vO-u0sucRF
ICLR.cc/2021/Conference
2021
Information Theoretic Meta Learning with Gaussian Processes
["Michalis Titsias", "Sotirios Nikoloutsopoulos", "Alexandre Galashov"]
We formulate meta learning using information theoretic concepts such as mutual information and the information bottleneck. The idea is to learn a stochastic representation or encoding of the task description, given by a training or support set, that is highly informative about predicting the validation set. By making u...
["Meta Learning", "Information Bottleneck", "Gaussian Processes", "Few-shot learning", "Variational Inference"]
ABSTRACTWe formulate meta learning using information theoretic concepts such as mutualinformation and the information bottleneck. The idea is to learn a stochastic rep-resentation or encoding of the task description, given by a training or support set,that is highly informative about predicting the validation set. By m...
Pqf44PumjU1
Interesting but not sure of advantages of proposed method vs. earlier methods
4: Ok but not good enough - rejection
The paper presents a method for Bayesian meta-learning. This method combines a NN feature extractor with a Gaussian Process on top. The GP kernel is linear. The information bottleneck is used to motivate a choice of approximate posterior. Using MAML to adapt the NN feature extractor weights improves performance of 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 Information Theoretic Meta Learning with Gaussian Processes ### Paper Abstract We formulate meta learning using information theoretic concepts such as mutual information...
SUyxNGzUsH
ICLR.cc/2021/Conference
2021
VilNMN: A Neural Module Network approach to Video-Grounded Language Tasks
["Hung Le", "Nancy F. Chen", "Steven Hoi"]
Neural module networks (NMN) have achieved success in image-grounded tasks such as question answering (QA) on synthetic images. However, very limited work on NMN has been studied in the video-grounded language tasks. These tasks extend the complexity of traditional visual tasks with the additional visual temporal varia...
["neural modular networks", "video-grounded dialogues", "dialogue understanding", "video understanding", "video QA", "video-grounded language tasks"]
ABSTRACTNeural module networks (NMN) have achieved success in image-grounded taskssuch as Visual Question Answering (VQA) on synthetic images. However, verylimited work on NMN has been studied in the video-grounded language tasks.These tasks extend the complexity of traditional visual tasks with the additional vi-sual ...
gIovmw9vbz
I am glad to see the progress in applying NMN in video-grounded language tasks but the paper could be improved on its current form.
5: Marginally below acceptance threshold
Summary: The paper studies the application of neural module network to video-grounded language tasks. They propose a method dubbed Visio-Linguistic Neural Module Network (VilNMN) to retrieve spatio-temporal information in a video through a linguistic-based parsed program. In particular, VilNMN first extracts entity re...
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 VilNMN: A Neural Module Network approach to Video-Grounded Language Tasks ### Paper Abstract Neural module networks (NMN) have achieved success in image-grounded tasks s...
CF-ZIuSMXRz
ICLR.cc/2021/Conference
2021
Spatio-Temporal Graph Scattering Transform
["Chao Pan", "Siheng Chen", "Antonio Ortega"]
Although spatio-temporal graph neural networks have achieved great empirical success in handling multiple correlated time series, they may be impractical in some real-world scenarios due to a lack of sufficient high-quality training data. Furthermore, spatio-temporal graph neural networks lack theoretical interpretatio...
["scattering transform", "spatio-temporal graph", "graph neural networks", "skeleton-based action recognition"]
ABSTRACTAlthough spatio-temporal graph neural networks have achieved great empiricalsuccess in handling multiple correlated time series, they may be impractical insome real-world scenarios due to a lack of sufficient high-quality training data.Furthermore, spatio-temporal graph neural networks lack theoretical interpre...
JyDOKd6n_PA
Official review
6: Marginally above acceptance threshold
I will add it tomorrow
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 Spatio-Temporal Graph Scattering Transform ### Paper Abstract Although spatio-temporal graph neural networks have achieved great empirical success in handling multiple c...
CxGPf2BPVA
ICLR.cc/2021/Conference
2021
Regularization Shortcomings for Continual Learning
["Timothee LESORT", "Andrei Stoian"]
In most machine learning algorithms, training data is assumed to be independent and identically distributed (iid). When it is not the case, the performances of the algorithms are challenged, leading to the famous phenomenon of \textit{catastrophic forgetting}. Algorithms dealing with it are gathered in the \textit{Co...
["Continual Learning", "Regularization"]
ABSTRACTIn most machine learning algorithms, training data is assumed to be indepen-dent and identically distributed (iid). When it is not the case, the algorithmsperformances are challenged, leading to the famous phenomenon of catastrophicforgetting . Algorithms dealing with it are gathered in the “Continual Learning”...
Pjn-Nq1S91V
Interesting conclusion but insufficient experimental analysis
5: Marginally below acceptance threshold
The paper presents an interesting conclusion that kinds of regularizations cannot learn to discriminate classes from different learning tasks. Both theoretical and experimental analysis is provided. Two regularization methods are tested. Pro: 1. The conclusion is interesting and it reveals that regularizations fail wh...
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 Regularization Shortcomings for Continual Learning ### Paper Abstract In most machine learning algorithms, training data is assumed to be independent and identically dis...
r1xywsC9tQ
ICLR.cc/2019/Conference
2019
Mapping the hyponymy relation of wordnet onto vector Spaces
["Jean-Philippe Bernardy", "Aleksandre Maskharashvili"]
In this paper, we investigate mapping the hyponymy relation of wordnet to feature vectors. We aim to model lexical knowledge in such a way that it can be used as input in generic machine-learning models, such as phrase entailment predictors. We propose two models. The first one leverages an existing mapping o...
["fasttext", "hyponymy", "wordnet"]
ABSTRACTIn this paper, we investigate mapping the hyponymy relation of W ORDNETtofeature vectors. We aim to model lexical knowledge in such a way that it canbe used as input in generic machine-learning models, such as phrase entailmentpredictors. We propose two models. The first one leverages an existing mappingof word...
ryxzHUBahQ
An interesting problem, but unclear description and methodological issues
3: Clear rejection
* Summary of the paper This paper studies how hyponymy between words can be mapped to feature representations. To this end, it lists out properties of such mappings and studies two methods from the perspective of how they address these properties. * Review The goal of this paper: namely, formalizing the hypernymy rel...
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 Mapping the hyponymy relation of wordnet onto vector Spaces ### Paper Abstract In this paper, we investigate mapping the hyponymy relation of wordnet to feature vector...
gMTaia--AB2
ACM.org/ICMI/2022/Workshop/GENEA
2022
The IVI Lab entry to the GENEA Challenge 2022 -- A Tacotron2 Based Method for Co-Speech Gesture Generation With Locality-Constraint Attention Mechanism
["Che-Jui Chang", "Sen Zhang", "Mubbasir Kapadia"]
This paper describes the IVI Lab entry to the GENEA Challenge 2022. We formulate the gesture generation problem as a sequence-to-sequence conversion task with text, audio, and speaker identity as inputs and the body motion as the output. We use the Tacotron2 architecture as our backbone with the locality-constraint att...
["co-speech gesture generation", "sequence-to-sequence modeling", "locality constraint attention"]
ABSTRACTThis paper describes the IVI Lab entry to the GENEA Challenge2022. We formulate the gesture generation problem as a sequence-to-sequence conversion task with text, audio, and speaker identityas inputs and the body motion as the output. We use the Tacotron2architecture as our backbone with the locality-constrain...
iYDzLm47aOS
An interesting approach with constrained attention and combined input features
7: Good paper, accept
Paper strengths: - Overall, the paper is well written and easy to follow. - The locality constraint attention is simple but efficient and well adapted to the task of gesture generation. - The combined multiple audio features and additional silence/laughter labels help the model capture audio-gesture alignment. - The pr...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title The IVI Lab entry to the GENEA Challenge 2022 -- A Tacotron2 Based Method for Co-Speech Gesture Generation With Locality-Constraint Attention Mechanism ### Paper Abstrac...
CR1XOQ0UTh-
ICLR.cc/2021/Conference
2021
Contrastive Learning with Hard Negative Samples
["Joshua David Robinson", "Ching-Yao Chuang", "Suvrit Sra", "Stefanie Jegelka"]
We consider the question: how can you sample good negative examples for contrastive learning? We argue that, as with metric learning, learning contrastive representations benefits from hard negative samples (i.e., points that are difficult to distinguish from an anchor point). The key challenge toward using hard negati...
["contrastive learning", "unsupervised representation learning", "hard negative sampling"]
ABSTRACTHow can you sample good negative examples for contrastive learning? We arguethat, as with metric learning, contrastive learning of representations benefits fromhard negative samples (i.e., points that are difficult to distinguish from an anchorpoint). The key challenge toward using hard negatives is that contra...
hMs2EDs8oj
Official Blind Review #1
7: Good paper, accept
The paper proposes a novel noise contrastive estimation (NCE) objective that incorporates hard-negative samples without similarity supervision, e.g., assuming unsupervised learning. To this end, it modifies the denominator of the original NCE by (a) re-weighting the negative samples based on the euclidean distance from...
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 Learning with Hard Negative Samples ### Paper Abstract We consider the question: how can you sample good negative examples for contrastive learning? We argue...
ByGuynAct7
ICLR.cc/2019/Conference
2019
The Deep Weight Prior
["Andrei Atanov", "Arsenii Ashukha", "Kirill Struminsky", "Dmitriy Vetrov", "Max Welling"]
Bayesian inference is known to provide a general framework for incorporating prior knowledge or specific properties into machine learning models via carefully choosing a prior distribution. In this work, we propose a new type of prior distributions for convolutional neural networks, deep weight prior (DWP), that exploi...
["deep learning", "variational inference", "prior distributions"]
ABSTRACTBayesian inference is known to provide a general framework for incorporatingprior knowledge or specific properties into machine learning models via carefullychoosing a prior distribution. In this work, we propose a new type of prior distri-butions for convolutional neural networks, deep weight prior (dwp) , tha...
rylsLCoBnX
Deep weight prior
7: Good paper, accept
This paper considers learning informative priors for convolutional neural network models based on fits to data sets from similar problem domains. For trained networks on related datasets the authors use autoencoders to obtain an expressive prior on the filter weights, with independence assumed between different layers...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title The Deep Weight Prior ### Paper Abstract Bayesian inference is known to provide a general framework for incorporating prior knowledge or specific properties into machine...
SyVhg20cK7
ICLR.cc/2019/Conference
2019
Inducing Cooperation via Learning to reshape rewards in semi-cooperative multi-agent reinforcement learning
["David Earl Hostallero", "Daewoo Kim", "Kyunghwan Son", "Yung Yi"]
We propose a deep reinforcement learning algorithm for semi-cooperative multi-agent tasks, where agents are equipped with their separate reward functions, yet with willingness to cooperate. Under these semi-cooperative scenarios, popular methods of centralized training with decentralized execution for inducing cooperat...
["multi-agent reinforcement learning", "deep reinforcement learning", "multi-agent systems"]
ABSTRACTWe propose a deep reinforcement learning algorithm for semi-cooperative multi-agent tasks, where agents are equipped with their separate reward functions, yetwith willingness to cooperate. Under these semi-cooperative scenarios, popularmethods of centralized training with decentralized execution for inducing co...
BkgEMMW937
This paper addresses this challenge by introducing a reward-shaping mechanism and incorporating a second DQN technique which is responsible for evaluating other agents performance. No discussion about convergence.
5: Marginally below acceptance threshold
This work is well-written, but the quality of some sections can be improved significantly as suggested in the comments. I have a few main concerns that I explain in detailed comments. Among those, the paper argues that the algorithms converge without discussing why. Also, the amount of overestimation of the Q-values ar...
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 Inducing Cooperation via Learning to reshape rewards in semi-cooperative multi-agent reinforcement learning ### Paper Abstract We propose a deep reinforcement learning a...
Rhsu5qD36cL
ICLR.cc/2021/Conference
2021
Sequential Density Ratio Estimation for Simultaneous Optimization of Speed and Accuracy
["Akinori F Ebihara", "Taiki Miyagawa", "Kazuyuki Sakurai", "Hitoshi Imaoka"]
Classifying sequential data as early and as accurately as possible is a challenging yet critical problem, especially when a sampling cost is high. One algorithm that achieves this goal is the sequential probability ratio test (SPRT), which is known as Bayes-optimal: it can keep the expected number of data samples as sm...
["Sequential probability ratio test", "Early classification", "Density ratio estimation"]
ABSTRACTClassifying sequential data as early and as accurately as possible is a challengingyet critical problem, especially when a sampling cost is high. One algorithmthat achieves this goal is the sequential probability ratio test (SPRT), which isknown as Bayes-optimal: it can keep the expected number of data samples ...
vgWfiNrpu3N
The paper involves a certain interesting property connecting a flexible neural network function-approximator to conventional SPRT setting.
6: Marginally above acceptance threshold
The authors propose how to use the neural networks to estimate the posteriors for each label y for the log likelihood ratio (LLR) estimation. The LLR estimation is performed using the multivariate inputs from the windows of time series, and it is used for the SPRT criterion without conventional iid assumption. The pa...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Sequential Density Ratio Estimation for Simultaneous Optimization of Speed and Accuracy ### Paper Abstract Classifying sequential data as early and as accurately as poss...
AmXVtqN1DDe
graphicsinterface.org/Graphics_Interface/2020/Conference
2020
Image Abstraction through Overlapping Region Growth
["Rosa Azami", "David Mould"]
We propose a region-based abstraction of a photograph, where the image plane is covered by overlapping irregularly shaped regions that approximate the image content. We segment regions using a novel region growth algorithm intended to produce highly irregular regions that still respect image edges, different from conve...
["Non-photorealistic rendering", "Image stylization", "Segmentation", "Abstraction"]
ABSTRACTWe propose a region-based abstraction of a photograph, where theimage plane is covered by overlapping irregularly shaped regionsthat approximate the image content. We segment regions using anovel region growth algorithm intended to produce highly irregularregions that still respect image edges, different from c...
axF2JillPzt
Interesting work with insufficient comparisons
6: Marginally above acceptance threshold
The seed placement algorithm seems to make sense — is there any rationale behind it? If this is original, please give more details. Otherwise, if it is inspired by some other prior work, please reference them accordingly. Section 4 discusses quite a few applications. Most of them would be better demonstrated if compa...
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 Image Abstraction through Overlapping Region Growth ### Paper Abstract We propose a region-based abstraction of a photograph, where the image plane is covered by overlap...
shkmWLRBXH
AKBC.ws/2020/Conference
2020
TransINT: Embedding Implication Rules in Knowledge Graphs with Isomorphic Intersections of Linear Subspaces
["So Yeon Min", "Preethi Raghavan", "Peter Szolovits"]
Knowledge Graphs (KG), composed of entities and relations, provide a structured representation of knowledge. For easy access to statistical approaches on relational data, multiple methods to embed a KG into f(KG) ∈ R^d have been introduced. We propose TransINT, a novel and interpretable KG embedding method that isomorp...
["Knowledge Graph Embedding", "Isomorphism", "Rules", "Common Sense", "Implication Rules", "Knowledge Graph", "Isomorphic Embedding", "Semantics Mining", "Rule Mining", "TransH"]
Automated Knowledge Base Construction (2020) Conference paperTransINT: Embedding Implication Rules in KnowledgeGraphs with Isomorphic Intersections of Linear SubspacesSo Yeon Min symin95@alum.mit.eduMIT CSAIL,32 Vassar Street, Cambridge, MA 02142 USAPreethi Raghavan praghav@us.ibm.comIBM Research,75 Binney Street, Camb...
LDNrKRLwD0t
Good focused contribution, writing can be significantly improved
6: Marginally above acceptance threshold
This work describes a new approach to learn KG embeddings by preserving the "implication ordering" among relations in the embedding space. The paper provides a cute new interpretation of TransH and then uses it to extend TransH to learn entity and relation representations. The crucial novelty of the approach is to map ...
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 TransINT: Embedding Implication Rules in Knowledge Graphs with Isomorphic Intersections of Linear Subspaces ### Paper Abstract Knowledge Graphs (KG), composed of entitie...
wf8UBIR-Gwl
auai.org/UAI/2022/Workshop/CRL
2022
Selection Collider Bias in Large Language Models
["Emily McMilin"]
In this paper we motivate the causal mechanisms behind sample selection collider bias in Large Language Models (LLMs). We show that selection collider bias can be amplified in underspecified learning tasks, and that the magnitude of the resulting spurious correlations appear scale agnostic. While selection collider bia...
["large language models", "causal inference", "selection bias"]
Selection Collider Bias in Large Language ModelsEmily McMilin11Independent ResearcherAbstractIn this paper we motivate the causal mechanismsbehind sample selection collider bias in Large Lan-guage Models (LLMs). We show that selectioncollider bias can be amplified in underspecifiedlearning tasks, and that the magnitude...
9BQM0_-kRmw
Selection collider bias introduces spurious correlations in LLMs
6: Marginally above acceptance threshold
### Summary This work highlights the problem of selection collider bias in Large Language Models through experimental evaluation. The implicit conditioning on an access variable during the data collection process introduces spurious correlations between unconditionally independent variables that are causing it. To prob...
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 Selection Collider Bias in Large Language Models ### Paper Abstract In this paper we motivate the causal mechanisms behind sample selection collider bias in Large Langua...
H1xwNhCcYm
ICLR.cc/2019/Conference
2019
Do Deep Generative Models Know What They Don't Know?
["Eric Nalisnick", "Akihiro Matsukawa", "Yee Whye Teh", "Dilan Gorur", "Balaji Lakshminarayanan"]
A neural network deployed in the wild may be asked to make predictions for inputs that were drawn from a different distribution than that of the training data. A plethora of work has demonstrated that it is easy to find or synthesize inputs for which a neural network is highly confident yet wrong. Generative models a...
["deep generative models", "out-of-distribution inputs", "flow-based models", "uncertainty", "density"]
ABSTRACTA neural network deployed in the wild may be asked to make predictions for inputsthat were drawn from a different distribution than that of the training data. Aplethora of work has demonstrated that it is easy to find or synthesize inputs forwhich a neural network is highly confident yet wrong. Generative model...
BJe__C5d3Q
Interesting example of density modelling shortcoming
6: Marginally above acceptance threshold
This paper displays an occurrence of density models assigning higher likelihood to out-of-distribution inputs compared to the training distribution. Specifically, density models trained on CIFAR10 have higher likelihood on SVHN than CIFAR10. This is an interesting observation because the prevailing assumption is that ...
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 Do Deep Generative Models Know What They Don't Know? ### Paper Abstract A neural network deployed in the wild may be asked to make predictions for inputs that were draw...
rkzjUoAcFX
ICLR.cc/2019/Conference
2019
Sample Efficient Adaptive Text-to-Speech
["Yutian Chen", "Yannis Assael", "Brendan Shillingford", "David Budden", "Scott Reed", "Heiga Zen", "Quan Wang", "Luis C. Cobo", "Andrew Trask", "Ben Laurie", "Caglar Gulcehre", "A\u00e4ron van den Oord", "Oriol Vinyals", "Nando de Freitas"]
We present a meta-learning approach for adaptive text-to-speech (TTS) with few data. During training, we learn a multi-speaker model using a shared conditional WaveNet core and independent learned embeddings for each speaker. The aim of training is not to produce a neural network with fixed weights, which is then deplo...
["few shot", "meta learning", "text to speech", "wavenet"]
ABSTRACTWe present a meta-learning approach for adaptive text-to-speech (TTS) with fewdata. During training, we learn a multi-speaker model using a shared conditionalWaveNet core and independent learned embeddings for each speaker. The aimof training is not to produce a neural network with fixed weights, which is thend...
rJgr4cv03X
nice paper, publish
7: Good paper, accept
This paper presents an approach to customize or adapt a text-to-speech synthesis system to a new speaker, given relatively small amount of data from that speaker. It is a very well written paper with rather strong results indicating high quality, naturalness, and similarity with real speech from a speaker can be achie...
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 Sample Efficient Adaptive Text-to-Speech ### Paper Abstract We present a meta-learning approach for adaptive text-to-speech (TTS) with few data. During training, we lear...
qhWU3OOOx7G
ICLR.cc/2022/Track/Blog
2022
Physics-Aware Difference Graph Networks for Sparsely-Observed Dynamics
["Anonymous"]
Sparsely available data points cause numerical error on finite differences which hinders us from modeling the dynamics of physical systems. The discretization error becomes even larger when the sparse data are irregularly distributed or defined on an unstructured grid, making it hard to build deep learning models to ha...
["PA-DGN", "PDE", "Graph Networks", "Sparse Data", "RGN"]
rzlnzV2gGq
A good summary of the cited paper
6: Marginally above acceptance threshold
The authors provide a clear and reasonable summary of the cited paper, which helps readers understand it more easily. However, the visualization can be improved, and reviews of recent follow-up works after the publication of the cited paper can be enhanced. Pros: 1. A clear and thorough introduction of the cited paper...
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 Physics-Aware Difference Graph Networks for Sparsely-Observed Dynamics ### Paper Abstract Sparsely available data points cause numerical error on finite differences whic...
rL11psvzvO
MPM/2022/Workshop
2022
Error Estimation for the Material Point Method
["Martin Berzins"]
A common feature of many methods in computational mechanics is that there is often a way of estimating the error in the computed solution. The situation for computational mechanics codes based upon the Material Point Method is very different in that there has been comparatively little work on computable error estimates...
["Error", "estimation", "Material", "Point", "Method"]
ABSTRACTThe challenge of choosing the best time step for the Material Point Method (MPM) is oftenaddressed by using a simple stability criterion, such as the speed of sound. While in many instancesthis works well it is important to understand how this relates to the overall errors present in themethod. This is particul...
ERPLGfkf9dH
Review
9: Top 15% of accepted papers, strong accept
Error estimation is an important topic that needs addressing.
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 Error Estimation for the Material Point Method ### Paper Abstract A common feature of many methods in computational mechanics is that there is often a way of estimating ...
FNMbe2vLvev
MICCAI.org/2022/Challenge/FLARE
2022
Revisiting nnU-Net for Iterative Pseudo Labeling and Efficient Sliding Window Inference
["Ziyan Huang", "Haoyu Wang", "Jin Ye", "Jingqi Niu", "Can Tu", "Yuncheng Yang", "Shiyi Du", "Zhongying Deng", "Lixu Gu", "Junjun He"]
nnU-Net serves as a good baseline for many medical image segmentation challenges in recent years. It works pretty well for fully-supervised segmentation tasks. However, it is less efficient for inference and cannot effectively make full use of unlabeled data, both of which are vital in real clinical scenarios. To this ...
["Segmentation", "Semi-supervised learning", "Computational Efficiency"]
Revisiting nnU-Net for Iterative PseudoLabeling and Efficient Sliding Window InferenceZiyan Huang1,2,†[0000−0002−1533−5239], Haoyu Wang1,2,†, Jin Ye2, Jingqi Niu1,2,Can Tu1,2, Yuncheng Yang1,2, Shiyi Du2,3, Zhongying Deng2,4, Lixu Gu1, andJunjun He2⋆1 Shanghai Jiao Tong University, Shanghai, China{ziyanhuang, small_dar...
fjuU3Nz_lEC
Iterative pseudo labeling with nnUNet
8: Top 50% of accepted papers, clear accept
The authors propose an iterative pseudo-labelling process based on nnU-Net to leverage the unlabeled data. Two nnU-Net models are developed: a high-performance, high-resource nnU-Net for iterative pseudo label generation a lower-performing, lower-resource nnU-Net for efficient inference The authors also change the slid...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Revisiting nnU-Net for Iterative Pseudo Labeling and Efficient Sliding Window Inference ### Paper Abstract nnU-Net serves as a good baseline for many medical image segme...
BkgzMCVtPB
ICLR.cc/2020/Conference
2020
Optimal Strategies Against Generative Attacks
["Roy Mor", "Erez Peterfreund", "Matan Gavish", "Amir Globerson"]
Generative neural models have improved dramatically recently. With this progress comes the risk that such models will be used to attack systems that rely on sensor data for authentication and anomaly detection. Many such learning systems are installed worldwide, protecting critical infrastructure or private data agains...
["optimal strategies", "attacker", "authenticator", "models", "progress", "risk", "systems", "sensor data"]
ABSTRACTGenerative neural models have improved dramatically recently. With this progresscomes the risk that such models will be used to attack systems that rely on sensordata for authentication and anomaly detection. Many such learning systems areinstalled worldwide, protecting critical infrastructure or private data a...
BJge4FdRFH
Official Blind Review #2
8: Accept
This paper proposes a new threat model for generative impersonation attacks: The attacker has access to several leaked images of a person; the authenticator knows several registration images per person and decides a person's identify by comparing some newly-sampled images from that person with corresponding registratio...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Optimal Strategies Against Generative Attacks ### Paper Abstract Generative neural models have improved dramatically recently. With this progress comes the risk that suc...
S1xLuRVFvr
ICLR.cc/2020/Conference
2020
Visual Explanation for Deep Metric Learning
["Sijie Zhu", "Taojiannan Yang", "Chen Chen"]
This work explores the visual explanation for deep metric learning and its applications. As an important problem for learning representation, metric learning has attracted much attention recently, while the interpretation of such model is not as well studied as classification. To this end, we propose an intuitive idea ...
["Metric Learning", "Visual Explanation"]
ABSTRACTThis work explores the visual explanation for deep metric learning and its appli-cations. As an important problem for learning representation, metric learning hasattracted much attention recently, while the interpretation of such model is not aswell studied as classification. To this end, we propose an intuitiv...
B1gxS8ARFS
Official Blind Review #2
8: Accept
This paper proposes a visualization method for deep metric learning, which derived by analyzing the inner product of two globally averaged activations. The proposed method can generate an overall activation map which highlights the regions contributing most to the similarity. Also, it can generate a partial activation ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Visual Explanation for Deep Metric Learning ### Paper Abstract This work explores the visual explanation for deep metric learning and its applications. As an important p...
Ms9zjhVB5R
ICLR.cc/2021/Conference
2021
SOAR: Second-Order Adversarial Regularization
["Avery Ma", "Fartash Faghri", "Nicolas Papernot", "Amir-massoud Farahmand"]
Adversarial training is a common approach to improving the robustness of deep neural networks against adversarial examples. In this work, we propose a novel regularization approach as an alternative. To derive the regularizer, we formulate the adversarial robustness problem under the robust optimization framework and a...
["Adversarial Robustness"]
ABSTRACTAdversarial training is a common approach to improving the robustness of deepneural networks against adversarial examples. In this work, we propose a novelregularization approach as an alternative. To derive the regularizer, we formulatethe adversarial robustness problem under the robust optimization framework ...
2qkrEUk9CA7
Some Concerns Needed to be Addressed
4: Ok but not good enough - rejection
This paper introduced a regularization scheme based on the second-order Taylor expansion of the loss objective to improve the robustness of the trained models. Theoretically, it used the simple linear model as an example to demonstrate 1) with a proper regularization term, the training algorithm can encode robustness ...
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 SOAR: Second-Order Adversarial Regularization ### Paper Abstract Adversarial training is a common approach to improving the robustness of deep neural networks against ad...
m9yojWL64gD
AIS-J.org/2021/Journal
2021
The study of Building Information Model server focusing on IFC to RDB conversion
["Zhenglu ZHU", "Kazuya Shide"]
This study proposes that the Building Information Model should aim to become a data service provider in order to expand the possibilities of utilizing BIM data. In order to construct a BIM server, this study devised a design method for creating a database of IFC, an international standard specification. We have develop...
["Building Information Model", "buildingSMART International", "IFC Server", "Relational Database", "SQL", "Auto-mapping"]
1 a The study of Building Information Model server focusing on IFC to RDB conversion Zhenglu ZHU1, Kazuya SHIDE2 1 Ph.D. Candidate, Graduate School of Eng, Shibaura Institute of Technology Inkx, Inc 2 Prof, School of Architecture, Shibaura Institute of Technology * na21102@shibaura- it.ac.jp Abstract This study defines...
QKojykGoOaB
2: Conditional acceptance
- Section 1 Background covers the necessity of BIM data service by touching on the latest 3D city model use cases. The author also states that the difficulties to realize the BIM data service as wide range of data, and complexities of discipline BIM data handling process and specifications. It can be valid statement, a...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title The study of Building Information Model server focusing on IFC to RDB conversion ### Paper Abstract This study proposes that the Building Information Model should aim to...
SkxG-CVFDH
ICLR.cc/2020/Conference
2020
GraphMix: Regularized Training of Graph Neural Networks for Semi-Supervised Learning
["Vikas Verma", "Meng Qu", "Alex Lamb", "Yoshua Bengio", "Juho Kannala", "Jian Tang"]
We present GraphMix, a regularization technique for Graph Neural Network based semi-supervised object classification, leveraging the recent advances in the regularization of classical deep neural networks. Specifically, we propose a unified approach in which we train a fully-connected network jointly with the graph neu...
["Regularization", "Graph Neural Networks", "Mixup", "Manifold Mixup", "Semi-supervised Object Classification over graph Data"]
ABSTRACTWe present GraphMix , a regularized training scheme for Graph Neural Networkbased semi-supervised object classification, leveraging the recent advances inthe regularization of classical deep neural networks. Specifically, we propose aunified approach in which we train a fully-connected network jointly with theg...
B1g6Dm86FH
Official Blind Review #3
3: Weak Reject
In this paper, the authors developed GNN by integrating an interpolation based regularization. In particular, the proposed GraphMix model has two components, one GNN and one MLP. For the MLP, the manifold mixup loss is used for training. The two components share parameters so that the GCN training can be regularized by...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title GraphMix: Regularized Training of Graph Neural Networks for Semi-Supervised Learning ### Paper Abstract We present GraphMix, a regularization technique for Graph Neural ...
Ek7qrYhJMbn
ICLR.cc/2021/Conference
2021
Central Server Free Federated Learning over Single-sided Trust Social Networks
["Chaoyang He", "Conghui Tan", "Hanlin Tang", "Shuang Qiu", "Ji Liu"]
Federated learning has become increasingly important for modern machine learning, especially for data privacy-sensitive scenarios. Existing federated learning mostly adopts the central server-based architecture or centralized architecture. However, in many social network scenarios, centralized federated learning is not...
["central", "free", "federated learning", "users", "central server", "trust social networks", "learning", "important", "modern machine learning"]
ABSTRACTFederated learning has become increasingly important for modern machine learn-ing, especially for data privacy-sensitive scenarios. Existing federated learningmostly adopts the central server-based architecture or centralized architecture.However, in many social network scenarios, centralized federated learning...
txJeSrhiMo3
The studied problem of decentralized federated learning in the context of social network seems to be interesting, but I think the novelty is low.
4: Ok but not good enough - rejection
**Paper Summary:** This paper proposes a decentralized federated learning algorithm called Online Push-Sum (OPS) for peer-to-peer learning in the context of social networks with the property of single-sided trust . **Questions for the authors** 1. Sec 1, it is mentioned that “Only models rather than local gradients a...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Central Server Free Federated Learning over Single-sided Trust Social Networks ### Paper Abstract Federated learning has become increasingly important for modern machine...
SkgscaNYPS
ICLR.cc/2020/Conference
2020
The asymptotic spectrum of the Hessian of DNN throughout training
["Arthur Jacot", "Franck Gabriel", "Clement Hongler"]
The dynamics of DNNs during gradient descent is described by the so-called Neural Tangent Kernel (NTK). In this article, we show that the NTK allows one to gain precise insight into the Hessian of the cost of DNNs: we obtain a full characterization of the asymptotics of the spectrum of the Hessian, at initialization an...
["theory of deep learning", "loss surface", "training", "fisher information matrix"]
Published as a conference paper at ICLR 2020The asymptotic spectrum of the Hessian ofDNN throughout trainingArthur Jacot, Franck Gabriel & Cl ement HonglerChair of Statistical Field TheoryEcole Polytechnique F ed erale de Lausannefarthur.jacot,franck.grabriel,clement.hongler g@epfl.chAbstractThe dynamics of DNNs during...
SkgmG-iQ9B
Official Blind Review #1
3: Weak Reject
This paper uses the Neural Tangent Kernel (NTK) to presents an asymptotic analysis of the evolution of Hessian of the loss (w.r.t. model parameters) throughout training. The authors leverage the Neural Tangent Kernel to analyze the evolution of the Hessian of the loss w.r.t the model parameters. Specifically 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 The asymptotic spectrum of the Hessian of DNN throughout training ### Paper Abstract The dynamics of DNNs during gradient descent is described by the so-called Neural Ta...
OZgVHzdKicb
ICLR.cc/2021/Conference
2021
Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples
["Kevin Li", "Abhishek Gupta", "Vitchyr H. Pong", "Ashwin Reddy", "Aurick Zhou", "Justin Yu", "Sergey Levine"]
Exploration in reinforcement learning is, in general, a challenging problem. In this work, we study a more tractable class of reinforcement learning problems defined by data that provides examples of successful outcome states. In this case, the reward function can be obtained automatically by training a classifier to c...
["Reinforcement Learning", "Goal Reaching", "Bayesian Classification", "Reward Inference"]
ABSTRACTExploration in reinforcement learning is, in general, a challenging problem. Inthis work, we study a more tractable class of reinforcement learning problemsdefined by data that provides examples of successful outcome states. In this case,the reward function can be obtained automatically by training a classifier...
0mkkGx_usM
An interesting paper, but some concerns regarding motivation and experiments
5: Marginally below acceptance threshold
Summary ------- This paper addresses a reinforcement learning problem where the reward function is learned through a classifier that decides whether states are successful or not based on previous examples (i.e. RL after inverse RL). The authors show that this requires uncertainty-aware predictions, which are difficult...
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 with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples ### Paper Abstract Exploration in reinforcement learning is, in general,...
BJerQWcp6Q
AKBC.ws/2019/Conference
2019
NormCo: Deep Disease Normalization for Biomedical Knowledge Base Construction
["Dustin Wright", "Yannis Katsis", "Raghav Mehta", "Chun-Nan Hsu"]
Biomedical knowledge bases are crucial in modern data-driven biomedical sciences, but auto-mated biomedical knowledge base construction remains challenging. In this paper, we consider the problem of disease entity normalization, an essential task in constructing a biomedical knowledge base. We present NormCo, a deep c...
["Entity Normalization", "Biomedical Knowledge Base Construction"]
Automated Knowledge Base Construction (2019) Conference paperNormCo: Deep Disease Normalization forBiomedical Knowledge Base ConstructionDustin Wright DBW 003@ ENG.UCSD .EDUYannis Katsis YANNIS .KATSIS @IBM.COMRaghav Mehta R3MEHTA @ENG.UCSD .EDUChun-Nan Hsu CHUNNAN @UCSD .EDUAbstractBiomedical knowledge bases are cruci...
Skg8RfIzfN
Very interesting paper that describes a positive contribution to the state of the art in BioNLP
9: Top 15% of accepted papers, strong accept
This paper proposes a deep-learning-based method to solve the known BioNLP task of disease normalization on the NCBI disease benchmark (where disease named entities are normalized/disambiguated against the MeSH and OMIM disease controlled vocabularies and taxonomies). The best known methods (DNorm, TaggerOne) are based...
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 NormCo: Deep Disease Normalization for Biomedical Knowledge Base Construction ### Paper Abstract Biomedical knowledge bases are crucial in modern data-driven biomedical ...
Hyls7h05FQ
ICLR.cc/2019/Conference
2019
A Differentiable Self-disambiguated Sense Embedding Model via Scaled Gumbel Softmax
["Fenfei Guo", "Mohit Iyyer", "Leah Findlater", "Jordan Boyd-Graber"]
We present a differentiable multi-prototype word representation model that disentangles senses of polysemous words and produces meaningful sense-specific embeddings without external resources. It jointly learns how to disambiguate senses given local context and how to represent senses using hard attention. Unlike previ...
["unsupervised representation learning", "sense embedding", "word sense disambiguation", "human evaluation"]
ABSTRACTWe present a differentiable multi-prototype word representation model that dis-entangles senses of polysemous words and produces meaningful sense-specificembeddings without external resources. It jointly learns how to disambiguatesenses given local context and how to represent senses using hard attention. Unlik...
BkgDZsDThQ
Interesting paper, promising results
6: Marginally above acceptance threshold
* Summary This paper extends the skipgram model using one vector per sense of a word. Based on this, the paper proposes two models for training sense embeddings: One where the word senses are marginalized out with attention over the senses, and the second where only the sense with highest value of attention contribu...
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 Differentiable Self-disambiguated Sense Embedding Model via Scaled Gumbel Softmax ### Paper Abstract We present a differentiable multi-prototype word representation mo...
SJl47yBYPS
ICLR.cc/2020/Conference
2020
Towards Simplicity in Deep Reinforcement Learning: Streamlined Off-Policy Learning
["Che Wang", "Yanqiu Wu", "Quan Vuong", "Keith Ross"]
The field of Deep Reinforcement Learning (DRL) has recently seen a surge in the popularity of maximum entropy reinforcement learning algorithms. Their popularity stems from the intuitive interpretation of the maximum entropy objective and their superior sample efficiency on standard benchmarks. In this paper, we seek ...
["Deep Reinforcement Learning", "Sample Efficiency", "Off-Policy Algorithms"]
ABSTRACTThe field of Deep Reinforcement Learning (DRL) has recently seen a surge in thepopularity of maximum entropy reinforcement learning algorithms. Their pop-ularity stems from the intuitive interpretation of the maximum entropy objectiveand their superior sample efficiency on standard benchmarks. In this paper, we...
HyeO-W-0YB
Official Blind Review #1
3: Weak Reject
The main contribution of this paper is a normalization scheme to avoid saturating the squashing function typically used to constrain actions within a bounded range in continuous control problems. It is argued that algorithms like DDPG and TD3 suffer from such saturation, which prevents proper exploration during trainin...
<|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 Simplicity in Deep Reinforcement Learning: Streamlined Off-Policy Learning ### Paper Abstract The field of Deep Reinforcement Learning (DRL) has recently seen a ...
B1gNKxrYPB
ICLR.cc/2020/Conference
2020
Attributed Graph Learning with 2-D Graph Convolution
["Qimai Li", "Xiaotong Zhang", "Han Liu", "Xiao-Ming Wu"]
Graph convolutional neural networks have demonstrated promising performance in attributed graph learning, thanks to the use of graph convolution that effectively combines graph structures and node features for learning node representations. However, one intrinsic limitation of the commonly adopted 1-D graph convolution...
["2-D Graph Convolution", "Attributed Graph", "Representation learning"]
ABSTRACTGraph convolutional neural networks have demonstrated promising performancein attributed graph learning, thanks to the use of graph convolution that effective-ly combines graph structures and node features for learning node representations.However, one intrinsic limitation of the commonly adopted 1-D graph conv...
H1lM6AJEqB
Official Blind Review #3
6: Weak Accept
The paper proposes a new 2-D graph convolution method to aggregate information using both the node relation graph and the attribute graph generated using, e.g., PMI and KNN. The motivation does make sense that using 1-D convolution along the node dimension might not enough for learning representation for those low-degr...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Attributed Graph Learning with 2-D Graph Convolution ### Paper Abstract Graph convolutional neural networks have demonstrated promising performance in attributed graph l...
rkexLAH0FE
MIDL.io/2019/Conference/Abstract
2019
GANs 'N Lungs: improving pneumonia prediction
["Tatiana Malygina", "Elena Ericheva", "Ivan Drokin"]
We propose a novel method to improve deep learning model performance on highly-imbalanced tasks. The proposed method is based on CycleGAN to achieve balanced dataset. We show that data augmentation with GAN helps to improve accuracy of pneumonia binary classification task even if the generative network was trained on t...
["X-ray medical image classification", "data augmentation", "CycleGAN"]
Medical Imaging with Deep Learning 2019 MIDL 2019 { Extended Abstract TrackGANs 'N Lungs: improving pneumonia predictionTatiana Malygina1;2tanya.malygina@botkin.ai1Intellogic Limited Liability Company (Intellogic LLC), oce 1/334/63, building 1, 42 Bolshoiblvd., territory of Skolkovo Innovation Center, 121205, Moscow, R...
BkeOLXqNjV
Interesting premise and results, broader application and human inspection would add significant value
3: Accept
The authors use cycleGANs to generate fake pneumonia images in the CheXpert dataset and show that the fake images help improve classification performance on the validation set. Although the experiments have a very preliminary nature (e.g. does this only work for pneumonia and also for the other classes, how realistic d...
3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title GANs 'N Lungs: improving pneumonia prediction ### Paper Abstract We propose a novel method to improve deep learning model performance on highly-imbalanced tasks. The pro...
jM76BCb6F9m
ICLR.cc/2021/Conference
2021
LEAF: A Learnable Frontend for Audio Classification
["Neil Zeghidour", "Olivier Teboul", "F\u00e9lix de Chaumont Quitry", "Marco Tagliasacchi"]
Mel-filterbanks are fixed, engineered audio features which emulate human perception and have been used through the history of audio understanding up to today. However, their undeniable qualities are counterbalanced by the fundamental limitations of handmade representations. In this work we show that we can train a sing...
["audio understanding", "frontend", "learnable", "mel-filterbanks", "time-frequency representations", "sound classification"]
ABSTRACTMel-filterbanks are fixed, engineered audio features which emulate human percep-tion and have been used through the history of audio understanding up to today.However, their undeniable qualities are counterbalanced by the fundamental lim-itations of handmade representations. In this work we show that we can tra...
NYlNefvv5X
A simple but interesting frontend for audio classification tasks
7: Good paper, accept
The paper proposes a learnable frontend for classification tasks on audio signals. The proposed learnable audio frontend (LEAF) is a generalization of a mel filterbank, used commonly in machine audition. LEAF consists of a Gabor filterbank, magnitude-squared nonlinearity, Gaussian lowpass filter and previously-proposed...
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 LEAF: A Learnable Frontend for Audio Classification ### Paper Abstract Mel-filterbanks are fixed, engineered audio features which emulate human perception and have been ...
VyENEGiEYAQ
ICLR.cc/2021/Conference
2021
Cluster-Former: Clustering-based Sparse Transformer for Question Answering
["Shuohang Wang", "Luowei Zhou", "Zhe Gan", "Yen-Chun Chen", "Yuwei Fang", "Siqi Sun", "Yu Cheng", "Jingjing Liu"]
Transformer has become ubiquitous in the deep learning field. One of the key ingredients that destined its success is the self-attention mechanism, which allows fully-connected contextual encoding over input tokens. However, despite its effectiveness in modeling short sequences, self-attention suffers when handling in...
["Transformer", "Question Answering"]
ABSTRACTTransformer has become ubiquitous in the deep learning field. One of the keyingredients that destined its success is the self-attention mechanism, which allowsfully-connected contextual encoding over input tokens. However, despite its effec-tiveness in modeling short sequences, self-attention suffers when handl...
cc1VP6NTky
Good results, but not a new work
2: Strong rejection
The paper proposes ClusterFormer to address the problem of quadratic compute requirements of the attention mechanism in a Transformer model. To this end this paper proposes to combine local attention to promote local consistency and proposes KMeans clustering to gather global information for every token. The paper esta...
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 Cluster-Former: Clustering-based Sparse Transformer for Question Answering ### Paper Abstract Transformer has become ubiquitous in the deep learning field. One of the ke...
F-mvpFpn_0q
ICLR.cc/2021/Conference
2021
Rapid Task-Solving in Novel Environments
["Samuel Ritter", "Ryan Faulkner", "Laurent Sartran", "Adam Santoro", "Matthew Botvinick", "David Raposo"]
We propose the challenge of rapid task-solving in novel environments (RTS), wherein an agent must solve a series of tasks as rapidly as possible in an unfamiliar environment. An effective RTS agent must balance between exploring the unfamiliar environment and solving its current task, all while building a model of the ...
["deep reinforcement learning", "meta learning", "deep learning", "exploration", "planning"]
ABSTRACTWe propose the challenge of rapid task-solving in novel environments (RTS),wherein an agent must solve a series of tasks as rapidly as possible in an unfa-miliar environment. An effective RTS agent must balance between exploring theunfamiliar environment and solving its current task, all while building a model ...
jv82-mHFV8-
Clear hypothesis and demonstration of the effectiveness of transformers as an episodic memory
7: Good paper, accept
The authors propose a non-parametric memory based on the transformer/self-attention architecture to learn over tasks that require planning from previously experienced tasks. They describe this style of learning as a form of meta reinforcement learning where individual episodes are a collection of tasks in the same envi...
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 Rapid Task-Solving in Novel Environments ### Paper Abstract We propose the challenge of rapid task-solving in novel environments (RTS), wherein an agent must solve a ser...
agyFqcmgl6y
ICLR.cc/2021/Conference
2021
Disentangled Generative Causal Representation Learning
["Xinwei Shen", "Furui Liu", "Hanze Dong", "Qing LIAN", "Zhitang Chen", "Tong Zhang"]
This paper proposes a Disentangled gEnerative cAusal Representation (DEAR) learning method. Unlike existing disentanglement methods that enforce independence of the latent variables, we consider the general case where the underlying factors of interests can be causally correlated. We show that previous methods with ind...
["disentanglement", "causality", "representation learning", "generative model"]
ABSTRACTThispaperproposesaDisentangledgEnerativecAusalRepresentation(DEAR)learningmethod.Unlikeexistingdisentanglementmethodsthatenforceindepen-denceofthelatentvariables,weconsiderthegeneralcasewheretheunderlyingfactorsofinterestscanbecausallycorrelated.Weshowthatpreviousmethodswithindependentpriorsfailtodisentanglecau...
UdQ41APz-HT
Disentangled Generative Causal Representation Learning
6: Marginally above acceptance threshold
The paper propose to learning causal disentangled representation which conforms human's cognition. The learned representation can be applied in causal controllable generation and benefit downstream tasks. The uses of this method is interesting. The paper has some advantages: It only requires part of sample is supervise...
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 Disentangled Generative Causal Representation Learning ### Paper Abstract This paper proposes a Disentangled gEnerative cAusal Representation (DEAR) learning method. Unl...
rmd-D7h_2zP
ICLR.cc/2021/Conference
2021
Domain-slot Relationship Modeling using a Pre-trained Language Encoder for Multi-Domain Dialogue State Tracking
["Jinwon An", "Misuk Kim", "Sungzoon Cho", "Junseong Bang"]
Dialogue state tracking for multi-domain dialogues is challenging because the model should be able to track dialogue states across multiple domains and slots. Past studies had its limitations in that they did not factor in the relationship among different domain-slot pairs. Although recent approaches did support relati...
["Dialogue State Tracking", "Domain-slot Relationship Modeling", "Pre-trained Language Encoder"]
ABSTRACTDialogue state tracking for multi-domain dialogues is challenging because themodel should be able to track dialogue states across multiple domains and slots.Past studies had its limitations in that they did not factor in the relationship amongdifferent domain-slot pairs. Although recent approaches did support r...
n0_EQ7h4Kkh
Interesting approach using pre-trained representations to encode domain-slot pairs for DST
7: Good paper, accept
Summary: This paper showcases how pre-training can help with Dialogue State Tracking. The authors explicitly model the relationship between domain-slot pairs. With their encoding and using strong pre-trained initializations they are able to improve the joint goal accuracy by almost 1.5 points which is impressive. Reas...
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 Domain-slot Relationship Modeling using a Pre-trained Language Encoder for Multi-Domain Dialogue State Tracking ### Paper Abstract Dialogue state tracking for multi-dom...
eNSpdJeR_J
ICLR.cc/2021/Conference
2021
Deep Learning with Data Privacy via Residual Perturbation
["Wenqi Tao", "Huaming Ling", "Zuoqiang Shi", "Bao Wang"]
Protecting data privacy in deep learning (DL) is at its urgency. Several celebrated privacy notions have been established and used for privacy-preserving DL. However, many of the existing mechanisms achieve data privacy at the cost of significant utility degradation. In this paper, we propose a stochastic differential ...
["Data Privacy", "Residual Perturbation", "Deep Learning"]
ABSTRACTProtecting data privacy in deep learning (DL) is at its urgency. Several celebratedprivacy notions have been established and used for privacy-preserving DL. However,many of the existing mechanisms achieve data privacy at the cost of significantutility degradation. In this paper, we propose a stochastic differen...
waKnmaPJa6I
Interesting ideas but results seem preliminary.
6: Marginally above acceptance threshold
### Summary The paper presents a method for training ResNets with differential privacy. Rather than the usual methods based on noisy gradient descent, the authors propose adding noise at each layer of the network during both training and testing. The authors prove differential privacy guarantees for two strategies 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 Deep Learning with Data Privacy via Residual Perturbation ### Paper Abstract Protecting data privacy in deep learning (DL) is at its urgency. Several celebrated privacy ...
jxUrueqBd5B
NeurIPS.cc/2022/Workshop/Federated_Learning
2022
Accelerating Federated Learning Through Attention on Local Model Updates
["Parsa Assadi", "Byung Hoon Ahn", "Hadi Esmaeilzadeh"]
Federated learning is used widely for privacy-preserving training. It performs well if the client datasets are both balanced and IID. However, in real-world settings, client datasets are non-IID and imbalanced. They may also experience significant distribution shifts. These non-idealities can hinder the performance of ...
["Federated Learning", "Clustering", "Communication Reduction", "Knowledge Extraction", "Attention"]
Accelerating Federated Learning Through Attention onLocal Model UpdatesParsa Assadi, Byung Hoon Ahn, Hadi EsmaeilzadehUniversity of California, San Diego{passadi, bhahn221, hadi}@ucsd.eduAbstractFederated learning is used widely for privacy-preserving training. It performs well if theclient datasets are both balanced a...
a2vO62u_D4
concern about using attention in parameter space
The author proposed an attention based method seeking to solve data distributional shift problem in federated learning which will reduce training training, performance and often lead to sub-optimal solution due to local minimum. To address this problem, the authors utilize an attention mechanism to understand pattern...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Accelerating Federated Learning Through Attention on Local Model Updates ### Paper Abstract Federated learning is used widely for privacy-preserving training. It perform...
rJXIPK1PM
ICLR.cc/2018/Workshop
2018
Easing non-convex optimization with neural networks
["David Lopez-Paz", "Levent Sagun"]
Despite being non-convex, deep neural networks are surprisingly amenable to optimization by gradient descent. In this note, we use a deep neural network with $D$ parameters to parametrize the input space of a generic $d$-dimensional non-convex optimization problem. Our experiments show that minimizing the over-parametr...
["nonconvex optimization", "deep neural networks", "overparametrized models"]
ABSTRACTDespite being non-convex, deep neural networks are surprisingly amenable tooptimization by gradient descent. In this note, we use a deep neural networkwith Dparameters to parametrize the input space of a generic d-dimensional non-convex optimization problem. Our experiments show that minimizing the over-paramet...
ByCELQxtM
Overall good but needs some improvements
7: Good paper, accept
This paper considers the problem of solving non-convex problems using deep neural network. The authors propose that the deep neural networks can ease and accelerate the optimization process because of the redundancy of parameters. This idea is interesting and the results are promising. The experiments show that if D>>d...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Easing non-convex optimization with neural networks ### Paper Abstract Despite being non-convex, deep neural networks are surprisingly amenable to optimization by gradie...
rkLyJl-0-
ICLR.cc/2018/Conference
2018
Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks
["Shankar Krishnan", "Ying Xiao", "Rif. A. Saurous"]
Progress in deep learning is slowed by the days or weeks it takes to train large models. The natural solution of using more hardware is limited by diminishing returns, and leads to inefficient use of additional resources. In this paper, we present a large batch, stochastic optimization algorithm that is both faster tha...
["Deep Learning", "Optimization"]
ABSTRACTProgress in deep learning is slowed by the days or weeks it takes to train largemodels. The natural solution of using more hardware is limited by diminishing re-turns, and leads to inefficient use of additional resources. In this paper, we presenta large batch, stochastic optimization algorithm that is both fas...
BkEmOSvef
Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks
6: Marginally above acceptance threshold
Summary: The paper proposes Neumman optimizer, which makes some adjustments to the idealized Neumman algorithm to improve performance and stability in training. The paper also provides the effectiveness of the algorithm by training ImageNet models (Inception-V3, Resnet-50, Resnet-101, and Inception-Resnet-V2). Comm...
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 Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks ### Paper Abstract Progress in deep learning is slowed by the days or weeks it takes to tr...
BJe4JJBYwS
ICLR.cc/2020/Conference
2020
CROSS-DOMAIN CASCADED DEEP TRANSLATION
["Oren Katzir", "Dani Lischinski", "Daniel Cohen-Or"]
In recent years we have witnessed tremendous progress in unpaired image-to-image translation methods, propelled by the emergence of DNNs and adversarial training strategies. However, most existing methods focus on transfer of style and appearance, rather than on shape translation. The latter task is challenging, due to...
["computer vision", "image translation", "generative adversarial networks"]
ABSTRACTIn recent years we have witnessed tremendous progress in unpaired image-to-image translation methods, propelled by the emergence of DNNs and adversarialtraining strategies. However, most existing methods focus on transfer of styleandappearance , rather than on shape translation. The latter task is challenging,d...
rJljMjEctH
Official Blind Review #1
6: Weak Accept
The paper proposes a new method for image-to-image translation. The problem of most existing methods is that they are good in translating style (e.g from photo to Van Gogh) but do not allow for significant changes in shape (e.g. from zebra to giraffe). The authors address this by performing the translation in a cascad...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title CROSS-DOMAIN CASCADED DEEP TRANSLATION ### Paper Abstract In recent years we have witnessed tremendous progress in unpaired image-to-image translation methods, propelled...