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
Bke4KsA5FX
ICLR.cc/2019/Conference
2019
Generative Code Modeling with Graphs
["Marc Brockschmidt", "Miltiadis Allamanis", "Alexander L. Gaunt", "Oleksandr Polozov"]
Generative models forsource code are an interesting structured prediction problem, requiring to reason about both hard syntactic and semantic constraints as well as about natural, likely programs. We present a novel model for this problem that uses a graph to represent the intermediate state of the generated output. Ou...
["Generative Model", "Source Code", "Graph Learning"]
ABSTRACTGenerative models for source code are an interesting structured prediction problem,requiring to reason about both hard syntactic and semantic constraints as well asabout natural, likely programs. We present a novel model for this problem thatuses a graph to represent the intermediate state of the generated outp...
Hyl3dbvq2X
Interesting task and dataset.
7: Good paper, accept
The paper introduces a 'code generation as hole completion' task and associated dataset, ExprGen. The authors proposed a novel extension of AST code generation which uses what they call Neural Attribute Grammars. They show the proposed method does well on this task, compared to ablations of their model (which are simil...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Generative Code Modeling with Graphs ### Paper Abstract Generative models forsource code are an interesting structured prediction problem, requiring to reason about both...
7AQUzh5ntX_
ICLR.cc/2021/Conference
2021
Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in First-person Simulated 3D Environments
["Wilka Torrico Carvalho", "Anthony Liang", "Kimin Lee", "Sungryull Sohn", "Honglak Lee", "Richard Lewis", "Satinder Singh"]
First-person object-interaction tasks in high-fidelity, 3D, simulated environments such as the AI2Thor virtual home-environment pose significant sample-efficiency challenges for reinforcement learning (RL) agents learning from sparse task rewards. To alleviate these challenges, prior work has provided extensive supervi...
["object-centric", "representation learning", "reinforcement learning", "sparse reward"]
ABSTRACTFirst-person object-interaction tasks in high-fidelity, 3D, simulated environmentssuch as the AI2Thor virtual home-environment pose significant sample-efficiencychallenges for reinforcement learning (RL) agents learning from sparse task re-wards. To alleviate these challenges, prior work has provided extensive ...
2iZzSJVchVb
Official Blind Review #3
4: Ok but not good enough - rejection
- Summary - This paper presents an architecture for learning object interaction tasks in AI2 Thor with sparse rewards. The architecture utilizes attention to select actions with an object centric representation. To handle reward sparsity, the authors propose an auxiliary objective that predicts the visual state o...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in First-person Simulated 3D Environments ### Paper Abstract First-person object-interaction tasks in h...
-ODN6SbiUU
ICLR.cc/2021/Conference
2021
In Defense of Pseudo-Labeling: An Uncertainty-Aware Pseudo-label Selection Framework for Semi-Supervised Learning
["Mamshad Nayeem Rizve", "Kevin Duarte", "Yogesh S Rawat", "Mubarak Shah"]
The recent research in semi-supervised learning (SSL) is mostly dominated by consistency regularization based methods which achieve strong performance. However, they heavily rely on domain-specific data augmentations, which are not easy to generate for all data modalities. Pseudo-labeling (PL) is a general SSL approach...
["Semi-Supervised Learning", "Pseudo-Labeling", "Uncertainty", "Calibration", "Deep Learning"]
ABSTRACTThe recent research in semi-supervised learning (SSL) is mostly dominated byconsistency regularization based methods which achieve strong performance. How-ever, they heavily rely on domain-specific data augmentations, which are not easyto generate for all data modalities. Pseudo-labeling (PL) is a general SSL a...
vh8hpbu5l-r
The paper proposes a new pseudo-label selection method for pseudo-labeling based semi-supervised learning methods. The selection process labels positive and negative samples based on confidence and filter samples based on uncertainty. Experimental results show that the proposed method outperforms previous semi-supervis...
5: Marginally below acceptance threshold
Strength: The paper notices the problem in pseudo-labeling methods of semi-supervised learning, which is the erroneous prediction of pseudo-labeling. Pseudo-labeling is an easy-to-implement method for semi-supervised learning does not require constraints needed by consistency regularization methods, so improving pseud...
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 In Defense of Pseudo-Labeling: An Uncertainty-Aware Pseudo-label Selection Framework for Semi-Supervised Learning ### Paper Abstract The recent research in semi-supervis...
rklPITVKvS
ICLR.cc/2020/Conference
2020
BRIDGING ADVERSARIAL SAMPLES AND ADVERSARIAL NETWORKS
["Faqiang Liu", "Mingkun Xu", "Guoqi Li", "Jing Pei", "Luping Shi"]
Generative adversarial networks have achieved remarkable performance on various tasks but suffer from sensitivity to hyper-parameters, training instability, and mode collapse. We find that this is partly due to gradient given by non-robust discriminator containing non-informative adversarial noise, which can hinder gen...
["ADVERSARIAL SAMPLES", "ADVERSARIAL NETWORKS"]
ABSTRACTGenerative adversarial networks have achieved remarkable performance on vari-ous tasks but suffer from sensitivity to hyper-parameters, training instability, andmode collapse. We find that this is partly due to gradient given by non-robust dis-criminator containing uninformative adversarial noise, which can hin...
ByeFqVKpdr
Official Blind Review #3
3: Weak Reject
The paper proposes a new GAN training that additionally feeds adversarial examples as the real samples to the discriminator D. A key motivation here is to regularize the target real distribution simulated by D to be robust to adversarial perturbations. Experimental results show that the proposed GAN training generally ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title BRIDGING ADVERSARIAL SAMPLES AND ADVERSARIAL NETWORKS ### Paper Abstract Generative adversarial networks have achieved remarkable performance on various tasks but suffer...
ioy7ZbTuoUo
Interspeech.org/2023/Workshop/SSW
2023
Improving robustness of spontaneous speech synthesis with linguistic speech regularization and pseudo-filled-pause insertion
["Yuta Matsunaga", "Takaaki Saeki", "Shinnosuke Takamichi", "Hiroshi Saruwatari"]
We present a training method with linguistic speech regularization that improves the robustness of spontaneous speech synthesis methods with filled pause (FP) insertion. Spontaneous speech synthesis is aimed at producing speech with human-like disfluencies, such as FPs. Because modeling the complex data distribution of...
["spontaneous speech synthesis", "filled pause", "spontaneity", "linguistic speech regularization"]
Improving robustness of spontaneous speech synthesiswith linguistic speech regularization and pseudo-filled-pause insertionYuta Matsunaga, Takaaki Saeki, Shinnosuke Takamichi, and Hiroshi SaruwatariGraduate School of Information Science and Technology, The University of Tokyo, Japan.matsunaga-yuta339@g.ecc.u-tokyo.ac.j...
X5h4pybmn_
Inspired by work in feature-matching regularization, this paper proposes a very similar regularization approach to alleviate the instability observed when synthesizing filled pauses (learned from a modest amount of training data).
5: Marginally below acceptance threshold
Inspired by work in feature-matching regularization, this paper proposes a very similar regularization approach to alleviate the instability observed when synthesizing filled pauses (learned from a modest amount of training data). The key idea is to match the learned internal embeddings between two systems: one that ha...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Improving robustness of spontaneous speech synthesis with linguistic speech regularization and pseudo-filled-pause insertion ### Paper Abstract We present a training met...
Bk6qQGWRb
ICLR.cc/2018/Conference
2018
Efficient Exploration through Bayesian Deep Q-Networks
["Kamyar Azizzadenesheli", "Emma Brunskill", "Animashree Anandkumar"]
We propose Bayesian Deep Q-Network (BDQN), a practical Thompson sampling based Reinforcement Learning (RL) Algorithm. Thompson sampling allows for targeted exploration in high dimensions through posterior sampling but is usually computationally expensive. We address this limitation by introducing uncertainty only at ...
["Deep RL", "Thompson Sampling", "Posterior update"]
ABSTRACTWe propose Bayesian Deep Q-Network ( BDQN ), a practical Thompson samplingbased Reinforcement Learning (RL) Algorithm. Thompson sampling allows fortargeted exploration in high dimensions through posterior sampling, but is usuallycomputationally expensive. We address this limitation by introducing uncertaintyonl...
SkMlZJ9gG
A nice algorithm. Decent preliminary results, but some more validation would be welcome.
6: Marginally above acceptance threshold
The authors propose a new algorithm for exploration in Deep RL. They apply Bayesian linear regression, given the last layer of a DQN network as features, to estimate the Q function for each action. Posterior weights are sampled to select actions during execution (Thompson Sampling style). I generally liked the paper an...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Efficient Exploration through Bayesian Deep Q-Networks ### Paper Abstract We propose Bayesian Deep Q-Network (BDQN), a practical Thompson sampling based Reinforcemen...
qG_x6oTIki
NeurIPS.cc/2022/Workshop/Offline_RL
2022
Let Offline RL Flow: Training Conservative Agents in the Latent Space of Normalizing Flows
["Dmitry Akimov", "Vladislav Kurenkov", "Alexander Nikulin", "Denis Tarasov", "Sergey Kolesnikov"]
Offline reinforcement learning aims to train a policy on a pre-recorded and fixed dataset without any additional environment interactions. There are two major challenges in this setting: (1) extrapolation error caused by approximating the value of state-action pairs not well-covered by the training data and (2) distrib...
["Offline Reinforcement Learning", "Normalizing Flows"]
Let Offline RL Flow: Training Conservative Agentsin the Latent Space of Normalizing FlowsDmitry AkimovTinkoffd.akimov@tinkoff.aiVladislav KurenkovTinkoffv.kurenkov@tinkoff.aiAlexander NikulinTinkoffa.p.nikulin@tinkoff.aiDenis TarasovTinkoffden.tarasov@tinkoff.aiSergey KolesnikovTinkoffs.s.kolesnikov@tinkoff.aiAbstractO...
hvazs5X5MId
A paper about learning policies in the latent action space, to ensure conservatism
6: Marginally above acceptance threshold
This paper is similar to PLAS (Zhou et al. 2020), which learns a policy in the latent action space through the use of a generative model. Specifically, it uses normalizing flows to learn the latent action space using offline dataset. In policy optimization time, the action is extracted from the latent space, along 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 Let Offline RL Flow: Training Conservative Agents in the Latent Space of Normalizing Flows ### Paper Abstract Offline reinforcement learning aims to train a policy on a ...
SkxW23NtPH
ICLR.cc/2020/Conference
2020
GDP: Generalized Device Placement for Dataflow Graphs
["Yanqi Zhou", "Sudip Roy", "Amirali Abdolrashidi", "Daniel Wong", "Peter C. Ma", "Qiumin Xu", "Ming Zhong", "Hanxiao Liu", "Anna Goldie", "Azalia Mirhoseini", "James Laudon"]
Runtime and scalability of large neural networks can be significantly affected by the placement of operations in their dataflow graphs on suitable devices. With increasingly complex neural network architectures and heterogeneous device characteristics, finding a reasonable placement is extremely challenging even for do...
["device placement", "reinforcement learning", "graph neural networks", "transformer"]
ABSTRACTRuntime and scalability of large neural networks can be significantly affected bythe placement of operations in their dataflow graphs on suitable devices. With in-creasingly complex neural network architectures and heterogeneous device char-acteristics, finding a reasonable placement is extremely challenging ev...
Skx27H9AtS
Official Blind Review #2
6: Weak Accept
In this paper the authors propose an end-to-end policy for graph placement and partitioning of computational graphs produced "under-the-hood" by platforms like Tensorflow. As the sizes of the neural networks increase, using distributed deep learning is becoming more and more necessary. Primitives like the one suggested...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title GDP: Generalized Device Placement for Dataflow Graphs ### Paper Abstract Runtime and scalability of large neural networks can be significantly affected by the placement ...
HJl0jiRqtX
ICLR.cc/2019/Conference
2019
EDDI: Efficient Dynamic Discovery of High-Value Information with Partial VAE
["Chao Ma", "Sebastian Tschiatschek", "Konstantina Palla", "Jose Miguel Hernandez Lobato", "Sebastian Nowozin", "Cheng Zhang"]
Making decisions requires information relevant to the task at hand. Many real-life decision-making situations allow acquiring further relevant information at a specific cost. For example, in assessing the health status of a patient we may decide to take additional measurements such as diagnostic tests or imaging scans ...
["active variable selection", "missing data", "amortized inference"]
ABSTRACTMaking decisions requires information relevant to the task at hand. Many real-lifedecision making situations allow acquiring further relevant information at a specificcost. For example, in assessing the health status of a patient we may decide to takeadditional measurements such as diagnostic tests or imaging s...
B1lI6Wnah7
EDDI: EFFICIENT DYNAMIC DISCOVERY OF HIGH-VALUE INFORMATION WITH PARTIAL VAE
5: Marginally below acceptance threshold
The authors present an information discovery approach based on (partial) variational autoencoders and an information theoretic acquisition function that seeks to maximize the expected information gain over a set of unobserved variables. Results are presented on image inpainting, UCI datasets and health data, namely ICU...
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 EDDI: Efficient Dynamic Discovery of High-Value Information with Partial VAE ### Paper Abstract Making decisions requires information relevant to the task at hand. Many ...
BJxpbREKvB
ICLR.cc/2020/Conference
2020
Revisiting Fine-tuning for Few-shot Learning
["Akihiro Nakamura", "Tatsuya Harada"]
Few-shot learning is the process of learning novel classes using only a few examples and it remains a challenging task in machine learning. Many sophisticated few-shot learning algorithms have been proposed based on the notion that networks can easily overfit to novel examples if they are simply fine-tuned using only a...
["few-shot learning", "fine-tuning"]
ABSTRACTFew-shot learning is the process of learning novel classes using only a few exam-ples and it remains a challenging task in machine learning. Many sophisticatedfew-shot learning algorithms have been proposed based on the notion that net-works can easily overfit to novel examples if they are simply fine-tuned usi...
B1xjB-Yy5S
Official Blind Review #2
1: Reject
This paper aims to tackle few-shot learning by revisiting the simple finetuning baseline, which is supposed to suffer from overfitting when there are only few-shot training samples. They empirically show that in the few-shot learning scenario, finetuning prefers a low learning rate and adaptive gradient optimizers. The...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Revisiting Fine-tuning for Few-shot Learning ### Paper Abstract Few-shot learning is the process of learning novel classes using only a few examples and it remains a cha...
qClL9hRDSMZ
ICLR.cc/2021/Conference
2021
Implicit bias of gradient descent for mean squared error regression with wide neural networks
["Hui Jin", "Guido Montufar"]
We investigate gradient descent training of wide neural networks and the corresponding implicit bias in function space. For 1D regression, we show that the solution of training a width-$n$ shallow ReLU network is within $n^{- 1/2}$ of the function which fits the training data and whose difference from initialization ha...
["Implicit bias", "overparametrized neural network", "cubic spline interpolation", "spatially adaptive smoothing spline", "effective capacity"]
ABSTRACTWe investigate gradient descent training of wide neural networks and the correspond-ing implicit bias in function space. For 1D regression, we show that the solutionof training a width- nshallow ReLU network is within n1=2of the function whichfits the training data and whose difference from initialization has s...
NlYk1F1VMC7
Somewhat incremental contribution? And the paper organization can be clearly improved.
5: Marginally below acceptance threshold
**Summary**: In this article, the authors characterized the implicit bias of gradient descent-based learning in the setting of wide single-hidden-layer neural networks with ReLU activation. More precisely, it was shown in Theorem 1 that the trained network output is "close" to that of the zero-training-error solution t...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Implicit bias of gradient descent for mean squared error regression with wide neural networks ### Paper Abstract We investigate gradient descent training of wide neural ...
PrzjugOsDeE
ICLR.cc/2021/Conference
2021
CcGAN: Continuous Conditional Generative Adversarial Networks for Image Generation
["Xin Ding", "Yongwei Wang", "Zuheng Xu", "William J Welch", "Z. Jane Wang"]
This work proposes the continuous conditional generative adversarial network (CcGAN), the first generative model for image generation conditional on continuous, scalar conditions (termed regression labels). Existing conditional GANs (cGANs) are mainly designed for categorical conditions (e.g., class labels); conditioni...
["Conditional generative adversarial networks", "image generation", "continuous and scalar conditions"]
ABSTRACTThis work proposes the continuous conditional generative adversarial network(CcGAN), the first generative model for image generation conditional on contin-uous ,scalar conditions (termed regression labels ). Existing conditional GANs(cGANs) are mainly designed for categorical conditions (e.g., class labels); co...
ez8yPte0LMz
A novel continuous conditional GAN
7: Good paper, accept
*** Summary: This paper proposes a novel continuous conditional GAN which takes continuous scalars (named regression labels) as conditions. There are two problems for continuous conditional GAN: (P1) cGANs with discrete labels are trained to minimize the empirical loss, but this fails for continuous conditions, becaus...
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 CcGAN: Continuous Conditional Generative Adversarial Networks for Image Generation ### Paper Abstract This work proposes the continuous conditional generative adversaria...
rJgsskrFwH
ICLR.cc/2020/Conference
2020
Scaling Autoregressive Video Models
["Dirk Weissenborn", "Oscar T\u00e4ckstr\u00f6m", "Jakob Uszkoreit"]
Due to the statistical complexity of video, the high degree of inherent stochasticity, and the sheer amount of data, generating natural video remains a challenging task. State-of-the-art video generation models attempt to address these issues by combining sometimes complex, often video-specific neural network architect...
["autoregressive models", "video prediction", "generative models", "video generation"]
ABSTRACTDue to the statistical complexity of video, the high degree of inherent stochasticity,and the sheer amount of data, generating natural video remains a challenging task.State-of-the-art video generation models often attempt to address these issues bycombining sometimes complex, usually video-specific neural netw...
ryeoIONqKB
Official Blind Review #1
8: Accept
Summary This papers presents a pixel-autoregressive model for video generation, in the spirit of VPN (Kalchbrenner’16). The proposed method uses video transformers and is made computationally efficient by extending block-local attention (Parmar’18, Chen’18) and sub-scaling (Menick’19) to 3D volumes. The block-local a...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Scaling Autoregressive Video Models ### Paper Abstract Due to the statistical complexity of video, the high degree of inherent stochasticity, and the sheer amount of dat...
S1gUCFx4dN
ICLR.cc/2019/Workshop/drlStructPred
2019
LEARNING NEUROSYMBOLIC GENERATIVE MODELS VIA PROGRAM SYNTHESIS
["Halley Young", "Osbert Bastani", "Mayur Naik"]
Significant strides have been made toward designing better generative models in recent years. Despite this progress, however, state-of-the-art approaches are still largely unable to capture complex global structure in data. For example, images of buildings typically contain spatial patterns such as windows repeating at...
["structure", "deep learning", "generative models", "structured prediction"]
ABSTRACTSignificant strides have been made toward designing better generative modelsin recent years. Despite this progress, however, state-of-the-art approaches arestill largely unable to capture complex global structure in data. For example,images of buildings typically contain spatial patterns such as windows repeati...
BJg_lNmBKN
Interesting technique to incorporate programs representing global structures in generative models
3: Marginally above acceptance threshold
This paper presents a technique to incorporate programs (representing global structure in images) into the generative model thereby capturing more global structure in image generation and image completion tasks. For image generation tasks, it first samples a latent vector z that is used to then sample a program (s,c). ...
2: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title LEARNING NEUROSYMBOLIC GENERATIVE MODELS VIA PROGRAM SYNTHESIS ### Paper Abstract Significant strides have been made toward designing better generative models in recent ...
S1dGIXVUz
ICLR.cc/2018/Workshop
2018
Attention-Based Guided Structured Sparsity of Deep Neural Networks
["Amirsina Torfi", "Rouzbeh Asghari Shirvani"]
Network pruning is aimed at imposing sparsity in a neural network architecture by increasing the portion of zero-valued weights for reducing its size energy-efficiency consideration and increasing evaluation speed. In most of the conducted research efforts, the sparsity is enforced for network pruning without any atten...
["Deep Netwroks", "Sparsity", "Attention", "Convolutional Networks"]
ABSTRACTNetwork pruning is aimed at imposing sparsity in a neural network architectureby increasing the portion of zero-valued weights for reducing its size regardingenergy-efficiency consideration and increasing evaluation speed. In most of theconducted research efforts, the sparsity is enforced for network pruning wi...
SkD0cK-YM
Over complex regularization term, insufficient evaluation.
4: Ok but not good enough - rejection
This paper propose to prune the network weights by using a group sparsity and a variance encouraging term to regularize the weights. The text of paper is not very easy to read, for example, the name "attention mechanism" is not very well explained. I understand the idea mainly through the equation. Pro: This paper ha...
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 Attention-Based Guided Structured Sparsity of Deep Neural Networks ### Paper Abstract Network pruning is aimed at imposing sparsity in a neural network architecture by i...
ZHqpIeyJ5q
MIDL.io/2020/Conference
2020
A Keypoint-based Morphological Signature for Large-scale Neuroimage Analysis
["Laurent Chauvin", "Matthew Toews"]
We present an image keypoint-based morphological signature that can be used to efficiently assess the pair-wise whole-brain similarity for large MRI datasets. Similarity is assessed via Jaccard-like measure of set overlap based on the proportion of keypoints shared by an image pair, which may be evaluated in O(N log N)...
["morphological signature", "neuroimage analysis", "similarity", "image", "large mri datasets", "measure", "set overlap", "proportion", "keypoints", "image pair"]
Proceedings of Machine Learning Research { Under Review:1{4, 2020 Short Paper { MIDL 2020 submissionKeypoint-based morphological signature for large-scaleneuroimage analysisEditors: Under Review for MIDL 2020AbstractWe present an image keypoint-based morphological signature that can be used to ecientlyassess the pair-w...
RqQ76OXx9l
Interesting work but where is the deep learning?
2: Weak reject
The authors present a method to assess the similarity between pairs of images for large datasets that relies on the scale invariant feature transform. Both the method and the results described are interesting and of high quality. However, this work does not seem to be in line with the topic of the MIDL conference as no...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title A Keypoint-based Morphological Signature for Large-scale Neuroimage Analysis ### Paper Abstract We present an image keypoint-based morphological signature that can be us...
PKubaeJkw3
ICLR.cc/2021/Conference
2021
Rethinking Architecture Selection in Differentiable NAS
["Ruochen Wang", "Minhao Cheng", "Xiangning Chen", "Xiaocheng Tang", "Cho-Jui Hsieh"]
Differentiable Neural Architecture Search is one of the most popular Neural Architecture Search (NAS) methods for its search efficiency and simplicity, accomplished by jointly optimizing the model weight and architecture parameters in a weight-sharing supernet via gradient-based algorithms. At the end of the search pha...
["architecture selection", "supernet", "architecture parameters", "much", "optimization", "operation", "darts", "differentiable nas"]
ABSTRACTDifferentiable Neural Architecture Search is one of the most popular Neural Ar-chitecture Search (NAS) methods for its search efficiency and simplicity, accom-plished by jointly optimizing the model weight and architecture parameters in aweight-sharing supernet via gradient-based algorithms. At the end of the s...
rKa26ZHA1g2
Interesting analysis, however the proposed method not that exiting
7: Good paper, accept
-- Short Summary -- This paper proposes a new policy for selecting the optimal architecture in neural architecture search (NAS), in particular for methods that involve a one-shot model and that deploy gradient-based methods for the search. The proposed algorithm sequentially prunes and fine-tunes the one-shot (aka sup...
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 Rethinking Architecture Selection in Differentiable NAS ### Paper Abstract Differentiable Neural Architecture Search is one of the most popular Neural Architecture Searc...
XnsV9ZhsOVc
IEEE.org/2022/Workshop/altVIS
2022
Other models for data visualisations
["Anonymous"]
Other models of visualising aim at a (re)formulation of contemporary expectations and narratives concerning data and their visualisations as a very specific model of thinking data visualisation. It is precisely how and with what intention we work on and discuss visualisations that defines the conceptual space we open t...
["data", "visualisations", "critique", "models"]
Other models of visualising aim at a (re)formulation of contemporary expectations and narratives concerning data and their visualisations as a very specific model of thinking data visualisation. It is precisely how and with what intention we work on and discuss visualisations that defines the conceptual space we open t...
jQM9ZwYtcn2
As your meta review, I want to start off by saying your paper provoked a lot of discussion on and offline. As a program committee we have decided to extend a conditional acceptance to the paper. There are several points addressed in two of the reviews below that I will summarize more broadly here. We ask that you revis...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Other models for data visualisations ### Paper Abstract Other models of visualising aim at a (re)formulation of contemporary expectations and narratives concerning data ...
FefFLN5FvIM
robot-learning.org/CoRL/2023/Conference
2023
TactileVAD: Geometric Aliasing-Aware Dynamics for High-Resolution Tactile Control
["Miquel Oller", "Dmitry Berenson", "Nima Fazeli"]
Touch-based control is a promising approach to dexterous manipulation. However, existing tactile control methods often overlook tactile geometric aliasing which can compromise control performance and reliability. This type of aliasing occurs when different contact locations yield similar tactile signatures. To address ...
["Manipulation", "tactile control", "high-resolution tactile sensors"]
TactileVAD: Geometric Aliasing-Aware Dynamics forHigh-Resolution Tactile ControlMiquel Oller Dmitry Berenson Nima FazeliDepartment of Robotics, University of MichiganAnn Arbor, MI 48109, United States{oller, dmitryb, nfz }@umich.eduhttps://www.mmintlab.com/tactile-vadAbstract: Touch-based control is a promising approac...
xj6QXCDPedj
Strengths: - The problem is well-motivated – geometric aliasing is an important problem in tactile dexterous manipulation, and the proposed method is a promising approach to addressing this aliasing problem. - The experimental settings are very well designed: the simulated toy example is excellent for illustrating the ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title TactileVAD: Geometric Aliasing-Aware Dynamics for High-Resolution Tactile Control ### Paper Abstract Touch-based control is a promising approach to dexterous manipulatio...
HklE01BYDB
ICLR.cc/2020/Conference
2020
Improving Sample Efficiency in Model-Free Reinforcement Learning from Images
["Denis Yarats", "Amy Zhang", "Ilya Kostrikov", "Brandon Amos", "Joelle Pineau", "Rob Fergus"]
Training an agent to solve control tasks directly from high-dimensional images with model-free reinforcement learning (RL) has proven difficult. The agent needs to learn a latent representation together with a control policy to perform the task. Fitting a high-capacity encoder using a scarce reward signal is not only ...
["reinforcement learning", "model-free", "off-policy", "image-based reinforcement learning", "continuous control"]
ABSTRACTTraining an agent to solve control tasks directly from high-dimensional imageswith model-free reinforcement learning (RL) has proven difficult. The agent needsto learn a latent representation together with a control policy to perform the task.Fitting a high-capacity encoder using a scarce reward signal is not o...
HkeFDVFtur
Official Blind Review #2
6: Weak Accept
This work presents a simple method for model-free RL from image observations. The key component of the method is the addition of an autoencoder that is trained jointly with the policy and value function, in contrast to previous methods which separate feature learning from policy learning. Another important modification...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Improving Sample Efficiency in Model-Free Reinforcement Learning from Images ### Paper Abstract Training an agent to solve control tasks directly from high-dimensional i...
lAZiGgGpbLx
CUHK.edu.hk/2021/Course/IERG5350
2021
Reinforcement Learning for Predict+Optimize
["Xinyi HU", "Yuansen Cheng"]
Predict+Optimize (P+O) is a machine learning framework for optimization problems with unknown parameters. This paper presents a framework to tackle P+O problems using neural networks and reinforcement learning. We focus on the traveling salesman problem and train a recurrent neural network that, given a directed graph,...
["reinforcement learning", "combinatorial optimisation", "task-based learning"]
Reinforcement Learning for Predict+OptimizeXinyi HUDepartment of Computer Science and EngineeringThe Chinese University of Hong KongShatin, Hong Kongxyhu@cse.cuhk.edu.hkYuansen CHENGDepartment of Information EngineeringThe Chinese University of Hong KongShatin, Hong Kongcy019@ie.cuhk.edu.hkAbstractPredict+Optimize (P+O...
HEXpp7ivXjT
A good thought of tacking Predict+Optimize problem
8: Top 50% of accepted papers, clear accept
Summary: In this paper, the authors propose to use reinforcement learning with pointer network for tacklling the traveling salesman problem. Moreover, their proposed pointer network architecture can deal with different TSP problem with varying nodes of the input number. The result demonstrate that their RL achieves t...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Reinforcement Learning for Predict+Optimize ### Paper Abstract Predict+Optimize (P+O) is a machine learning framework for optimization problems with unknown parameters. ...
r16u6i_Xz
ICLR.cc/2018/Conference
2018
Interactive Boosting of Neural Networks for Small-sample Image Classification
["Xiaoxu Li", "Dongliang Chang", "Zheng-Hua Tan", "Zhanyu Ma", "Jun Guo", "Jie Cao"]
Neural networks have recently shown excellent performance on numerous classi- fication tasks. These networks often have a large number of parameters and thus require much data to train. When the number of training data points is small, however, a network with high flexibility will quickly overfit the training data, res...
["ensemble learning", "neural network", "small-sample", "overfitting", "variance"]
ABSTRACTNeural networks have recently shown excellent performance on numerous classi-fication tasks. These networks often have a large number of parameters and thusrequire much data to train. When the number of training data points is small,however, a network with high flexibility will quickly overfit the training data...
S1a08aYQf
The proposed method is quite a lot like mixtures of experts, and this needs to be addressed
4: Ok but not good enough - rejection
The manuscript aims to improve the performance of neural networks on small datasets, by using an ensemble approach to reduce the variance. This is a worthy goal, and the overall approach of using an ensemble method makes sense, based on the variance-reducing properties of bagging. Postive results are reported on two ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Interactive Boosting of Neural Networks for Small-sample Image Classification ### Paper Abstract Neural networks have recently shown excellent performance on numerous cl...
q4HuPoEQK_o
ICLR.cc/2021/Conference
2021
Hybrid Quantum-Classical Stochastic Networks with Boltzmann Layers
["Jonathan H Warrell", "Prashant Emani", "Mark Gerstein"]
Quantum Machine Learning (QML) has the potential to significantly advance the state-of-the-art in artificial intelligence, due to recent developments in quantum computing hardware and algorithm design. Particularly, an avenue opened up by these advances is the possibility of enhancing classical models through developi...
["Quantum Machine Learning", "Hierarchical Models", "Variational Inference", "Model Interpretation", "Computational Biology"]
ABSTRACTQuantum Machine Learning (QML) has the potential to significantly advance thestate-of-the-art in artificial intelligence, due to recent developments in quantumcomputing hardware and algorithm design. Particularly, an avenue opened up bythese advances is the possibility of enhancing classical models through deve...
PKbdR7_0ulX
Review of Hybrid Quantum-Classical Stochastic Networks with Boltzmann Layers
4: Ok but not good enough - rejection
**Summary** This paper extends previous work on Quantum Boltzmann machines (QBM) and Quantum VAEs to models with multiple stochastic layers where each layer is either a (classical) layer with a Gaussian output distribution or a Quantum Boltzman machine. The QBM differs from the classical Boltzmann machine by the presen...
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 Hybrid Quantum-Classical Stochastic Networks with Boltzmann Layers ### Paper Abstract Quantum Machine Learning (QML) has the potential to significantly advance the state...
7JSTDTZtn7-
ICLR.cc/2021/Conference
2021
Byzantine-Robust Learning on Heterogeneous Datasets via Resampling
["Lie He", "Sai Praneeth Karimireddy", "Martin Jaggi"]
In Byzantine-robust distributed optimization, a central server wants to train a machine learning model over data distributed across multiple workers. However, a fraction of these workers may deviate from the prescribed algorithm and send arbitrary messages to the server. While this problem has received significant atte...
["Byzantine robustness", "distributed training", "heterogeneous dataset"]
ABSTRACTIn Byzantine robust distributed optimization, a central server wants to train amachine learning model over data distributed across multiple workers. However,a fraction of these workers may deviate from the prescribed algorithm and sendarbitrary messages to the server. While this problem has received significant...
5H8WQjxSMc
A good attempt towards robust federated learning
6: Marginally above acceptance threshold
Paper summary: The paper studies Byzantine robustness in the context of distributed learning from heterogeneous datasets. This problem has been widely studied previously, but under the additional assumption that the data of the good workers is i.i.d.. The authors give examples of situations and poisoning attacks 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 Byzantine-Robust Learning on Heterogeneous Datasets via Resampling ### Paper Abstract In Byzantine-robust distributed optimization, a central server wants to train a mac...
-bdp_8Itjwp
ICLR.cc/2021/Conference
2021
Self-supervised Learning from a Multi-view Perspective
["Yao-Hung Hubert Tsai", "Yue Wu", "Ruslan Salakhutdinov", "Louis-Philippe Morency"]
As a subset of unsupervised representation learning, self-supervised representation learning adopts self-defined signals as supervision and uses the learned representation for downstream tasks, such as object detection and image captioning. Many proposed approaches for self-supervised learning follow naturally a multi-...
["Self-supervised Learning", "Unsupervised Learning", "Multi-view Representation Learning"]
ABSTRACTAs a subset of unsupervised representation learning, self-supervised representationlearning adopts self-defined signals as supervision and uses the learned representa-tion for downstream tasks, such as object detection and image captioning. Manyproposed approaches for self-supervised learning follow naturally a...
z3cRCEyQc8e
An information-theoretical point to better understand SSL from a multi-view perspective, but the experimental part needs to be enhanced
6: Marginally above acceptance threshold
This paper explores both theoretical and empirical perspectives on self-supervised learning. The authors attempt to prove that self-supervised learning could extract task-relevant information while discard task-irrelevant information. They also provide a composite objective that includes contrastive and predictive loss...
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 Self-supervised Learning from a Multi-view Perspective ### Paper Abstract As a subset of unsupervised representation learning, self-supervised representation learning ad...
r17RD2oxe
ICLR.cc/2017/conference
2017
Deep Neural Networks and the Tree of Life
["Yan Wang", "Kun He", "John E. Hopcroft", "Yu Sun"]
In Evolutionary Biology, species close in the tree of evolution are identified by similar visual features. In computer vision, deep neural networks perform image classification by learning to identify similar visual features. This leads to an interesting question: is it possible to leverage the advantage of deep networ...
["Deep learning", "Computer vision", "Applications"]
ABSTRACTIn Evolutionary Biology, species close in the tree of evolution are identified bysimilar visual features. In computer vision, deep neural networks perform imageclassification by learning to identify similar visual features. This leads to an in-teresting question: is it possible to leverage the advantage of deep...
rJi_oCBVx
Nice application of using deep features but lack technical novelty
4: Ok but not good enough - rejection
This paper introduces a hierarchical clustering method using learned CNN features to build 'the tree of life'. The assumption is that the feature similarity indicates the distance in the tree. The authors tried three different ways to construct the tree: 1) approximation central point 2) minimum spanning tree and 3) mu...
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 Neural Networks and the Tree of Life ### Paper Abstract In Evolutionary Biology, species close in the tree of evolution are identified by similar visual features. I...
iktA2PtTRsK
ICLR.cc/2021/Conference
2021
Watching the World Go By: Representation Learning from Unlabeled Videos
["Daniel Gordon", "Kiana Ehsani", "Dieter Fox", "Ali Farhadi"]
Recent unsupervised representation learning techniques show remarkable success on many single image tasks by using instance discrimination: learning to differentiate between two augmented versions of the same image and a large batch of unrelated images. Prior work uses artificial data augmentation techniques such as ...
["Representation Learning", "Unsupervised Learning", "Video Analytics"]
ABSTRACTRecent unsupervised representation learning techniques show remarkable successon many single image tasks by using instance discrimination: learning to differ-entiate between two augmented versions of the same image and a large batch ofunrelated images. Prior work uses artificial data augmentation techniques suc...
XWYGO_DNtx-
Review
4: Ok but not good enough - rejection
This paper proposes an idea to use NCE for videos where positive/negative training pairs are created by temporally sampling different frames in the video. The idea is fairly straightforward, there is not anything particularly novel about the approach, other than how the training pairs are created. I think it would hav...
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 Watching the World Go By: Representation Learning from Unlabeled Videos ### Paper Abstract Recent unsupervised representation learning techniques show remarkable success...
xjXg0bnoDmS
ICLR.cc/2021/Conference
2021
Entropic gradient descent algorithms and wide flat minima
["Fabrizio Pittorino", "Carlo Lucibello", "Christoph Feinauer", "Gabriele Perugini", "Carlo Baldassi", "Elizaveta Demyanenko", "Riccardo Zecchina"]
The properties of flat minima in the empirical risk landscape of neural networks have been debated for some time. Increasing evidence suggests they possess better generalization capabilities with respect to sharp ones. In this work we first discuss the relationship between alternative measures of flatness: The local en...
["flat minima", "entropic algorithms", "statistical physics", "belief-propagation"]
Published as a conference paper at ICLR 2021Entropic gradient descent algorithmsand wide flat minimaFabrizio Pittorino1;2, Carlo Lucibello1, Christoph Feinauer1,Gabriele Perugini1, Carlo Baldassi1, Elizaveta Demyanenko1,Riccardo Zecchina11AI Lab, Institute for Data Science and Analytics, Bocconi University,20136 Milano...
zazvip4H6A
good paper but unclear what the contributions are
6: Marginally above acceptance threshold
This paper studies local entropy measures for characterizing flat regions in the energy landscape of deep networks. The paper discusses, at length, two previously proposed algorithms named Entropy-SGD and Replicated-SGD and demonstrates, using (i) controlled experiments where Belief Propagation (BP) can be used to esti...
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 Entropic gradient descent algorithms and wide flat minima ### Paper Abstract The properties of flat minima in the empirical risk landscape of neural networks have been d...
H1gZsJBYwH
ICLR.cc/2020/Conference
2020
Hybrid Weight Representation: A Quantization Method Represented with Ternary and Sparse-Large Weights
["Jinbae Park", "Sung-Ho Bae"]
Previous ternarizations such as the trained ternary quantization (TTQ), which quantized weights to three values (e.g., {−Wn, 0,+Wp}), achieved the small model size and efficient inference process. However, the extreme limit on the number of quantization steps causes some degradation in accuracy. To solve this problem...
["quantized neural networks", "centralized quantization", "hybrid weight representation", "weighted ridge", "ternary weight"]
ABSTRACTPrevious ternarizations such as the trained ternary quantization (TTQ), whichquantized weights to three values (e.g., fWn, 0,+Wpg), achieved the smallmodel size and efficient inference process. However, the extreme limit on thenumber of quantization steps causes some degradation in accuracy. To solve thisproble...
Byxk-x1AFr
Official Blind Review #1
3: Weak Reject
This paper works on weight quantization in deep networks. The authors propose to utilize the extra state in 2-bit ternary representation to encode large weight values. The authors also propose to use a weighted ridge regularizer which contains a "part of L1" term to make the weights with large values sparse. The idea ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Hybrid Weight Representation: A Quantization Method Represented with Ternary and Sparse-Large Weights ### Paper Abstract Previous ternarizations such as the trained tern...
SkeiPsAqK7
ICLR.cc/2019/Conference
2019
NA
["Dai Quoc Nguyen", "Tu Dinh Nguyen", "Dinh Phung"]
NA
[]
BylOINLqnm
decent novelty but motivations and approaches not convincing enough
5: Marginally below acceptance threshold
To solve the problem fo “preserving relative positions of nodes”, the paper applied the capsule network to the graph node embedding task and proposed the model named “CapsG”. The CapsG construct two capsule layers and utilize the dynamic routing between capsules to iteratively update the target node representation in 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 NA ### Paper Abstract NA ### Paper Keywords [] ### Paper Content <|im_end|> <|im_start|>assistant ### Review Title decent novelty but motivations and approaches not co...
ByeqyxBKvS
ICLR.cc/2020/Conference
2020
Quantum Semi-Supervised Kernel Learning
["Seyran Saeedi", "Aliakbar Panahi", "Tom Arodz"]
Quantum machine learning methods have the potential to facilitate learning using extremely large datasets. While the availability of data for training machine learning models is steadily increasing, oftentimes it is much easier to collect feature vectors that to obtain the corresponding labels. One of the approaches fo...
["quantum machine learning", "semi-supervised learning", "support vector machines"]
ABSTRACTQuantum machine learning methods have the potential to facilitate learning us-ing extremely large datasets. While the availability of data for training machinelearning models is steadily increasing, oftentimes it is much easier to collect fea-ture vectors that to obtain the corresponding labels. One of the appr...
BJxnkesP5H
Official Blind Review #4
6: Weak Accept
This paper developes a quantum algorithm for kernel-based support vector machine working in a semi-supervised learning setting. The motivation is to utilise the significant advantage of quantum computation to train machine learning models on large-scale datasets efficiently. This paper reviews the existing work on usin...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Quantum Semi-Supervised Kernel Learning ### Paper Abstract Quantum machine learning methods have the potential to facilitate learning using extremely large datasets. Whi...
sv2wC7Amb0s
ICLR.cc/2021/Conference
2021
Conditioning Trick for Training Stable GANs
["MOHAMMAD ESMAEILPOUR", "Raymel Alfonso Sallo", "Olivier St-Georges", "Patrick Cardinal", "Alessandro Lameiras Koerich"]
In this paper we propose a conditioning trick, called difference departure from normality, applied on the generator network in response to instability issues during GAN training. We force the generator to get closer to the departure from normality function of real samples computed in the spectral domain of Schur decomp...
["Generative Adversarial Network", "Departure From Normality", "Schur Decomposition", "Spectrogram Synthesis"]
ABSTRACTIn this paper we propose a conditioning trick, called difference departure fromnormality, applied on the generator network in response to instability issues dur-ing GAN training. We force the generator to get closer to the departure fromnormality function of real samples computed in the spectral domain of Schur...
R_gAct7ZAfm
Limited evaluation, some contradictory results for stabilized training
3: Clear rejection
Summary this paper propose a constraint in GAN training, to improve generated samples fidelity and stabilize training. The proposed conditioning is based on limiting the generator from departing normality function of real samples, which is computed in the spectral domain of Schur decomposition. It is claimed that this...
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 Conditioning Trick for Training Stable GANs ### Paper Abstract In this paper we propose a conditioning trick, called difference departure from normality, applied on the ...
vCEhC7nOb6
ICLR.cc/2021/Conference
2021
Inductive Bias of Gradient Descent for Exponentially Weight Normalized Smooth Homogeneous Neural Nets
["Depen Morwani", "Harish Guruprasad Ramaswamy"]
We analyze the inductive bias of gradient descent for weight normalized smooth homogeneous neural nets, when trained on exponential or cross-entropy loss. Our analysis focuses on exponential weight normalization (EWN), which encourages weight updates along the radial direction. This paper shows that the gradient flow p...
["Deep Learning Theory", "Weight Normalization", "Inductive Bias", "Gradient Descent"]
ABSTRACTWe analyze the inductive bias of gradient descent for weight normalized smoothhomogeneous neural nets, when trained on exponential or cross-entropy loss. Ouranalysis focuses on exponential weight normalization (EWN), which encouragesweight updates along the radial direction. This paper shows that the gradient f...
NPck5HgmCRX
Technical issues
4: Ok but not good enough - rejection
The paper studies the gradient flow dynamics over smooth homogeneous models with two types of weight normalized parameterization - standard weight normalization (SWN) and exponentiated weight normalization (EWN). Thm 1 shows the induced dynamics in the unnormalized parameter space resulting from gradient flow on respec...
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 Inductive Bias of Gradient Descent for Exponentially Weight Normalized Smooth Homogeneous Neural Nets ### Paper Abstract We analyze the inductive bias of gradient descen...
BJxmXhRcK7
ICLR.cc/2019/Conference
2019
TENSOR RING NETS ADAPTED DEEP MULTI-TASK LEARNING
["Xinqi Chen", "Ming Hou", "Guoxu Zhou", "Qibin Zhao"]
Recent deep multi-task learning (MTL) has been witnessed its success in alleviating data scarcity of some task by utilizing domain-specific knowledge from related tasks. Nonetheless, several major issues of deep MTL, including the effectiveness of sharing mechanisms, the efficiency of model complexity and the flexibili...
["deep learning", "deep multi-task learning", "tensor factorization", "tensor ring nets"]
ABSTRACTRecent deep multi-task learning (MTL) has been witnessed its success in alle-viating data scarcity of some task by utilizing domain-specific knowledge fromrelated tasks. Nonetheless, several major issues of deep MTL, including the effec-tiveness of sharing mechanisms, the efficiency of model complexity and the ...
H1er7SL92Q
Simple idea with interesting results
5: Marginally below acceptance threshold
The novelty and experiments are somewhat limited. Thus I am lowering my score. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- The authors proposed a variant of tensor ring formulation for multi-task...
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 TENSOR RING NETS ADAPTED DEEP MULTI-TASK LEARNING ### Paper Abstract Recent deep multi-task learning (MTL) has been witnessed its success in alleviating data scarcity of...
Skxw-REFwS
ICLR.cc/2020/Conference
2020
Unsupervised Progressive Learning and the STAM Architecture
["James Smith", "Constantine Dovrolis"]
We first pose the Unsupervised Progressive Learning (UPL) problem: learning salient representations from a non-stationary stream of unlabeled data in which the number of object classes increases with time. If some limited labeled data is also available, those representations can be associated with specific cl...
["continual learning", "unsupervised learning", "online learning"]
ABSTRACTWe first pose the Unsupervised Progressive Learning (UPL) problem: learningsalient representations from a non-stationary stream of unlabeled data in whichthe number of object classes increases with time. If some limited labeled data isalso available, those representations can be associated with specific classes...
BygbC5qqtr
Official Blind Review #1
3: Weak Reject
This paper proposes a method called Self-Taught Associative Memory (STAM) for Unsupervised Progressive Learning (UPL) , i.e., learning salient representation from streams of mostly unlabeled data with occasional class labels, where the number of class increases over time. The motivation of this paper is quite interesti...
<|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 Progressive Learning and the STAM Architecture ### Paper Abstract We first pose the Unsupervised Progressive Learning (UPL) problem: learning salie...
HJlRFlHFPS
ICLR.cc/2020/Conference
2020
Unsupervised Distillation of Syntactic Information from Contextualized Word Representations
["Shauli Ravfogel", "Yanai Elazar", "Jacob Goldberger", "Yoav Goldberg"]
Contextualized word representations, such as ELMo and BERT, were shown to perform well on a various of semantic and structural (syntactic) task. In this work, we tackle the task of unsupervised disentanglement between semantics and structure in neural language representations: we aim to learn a transformation of the co...
["dismantlement", "contextualized word representations", "language models", "representation learning"]
ABSTRACTContextualized word representations, such as ELMo and BERT, were shown toperform well on various semantic and structural (syntactic) task. In this work,we tackle the task of unsupervised disentanglement between semantics and struc-ture in neural language representations: we aim to learn a transformation of thec...
Byey-Ljz5H
Official Blind Review #4
1: Reject
Summary The authors proposed to disentangle syntactic information and semantic information from pre-trained contextualized word representations. They use BERT to generate groups of sentences that are structurally similar (have the same POS tag for each word) but semantically different. Then they use a metric-learning ...
<|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 Distillation of Syntactic Information from Contextualized Word Representations ### Paper Abstract Contextualized word representations, such as ELMo and BERT...
Hw2Za4N5hy0
ICLR.cc/2021/Conference
2021
Federated Learning with Decoupled Probabilistic-Weighted Gradient Aggregation
["Jian-hui Duan", "Wenzhong Li", "Sanglu Lu"]
In the federated learning paradigm, multiple mobile clients train local models independently based on datasets generated by edge devices, and the server aggregates parameters/gradients from local models to form a global model. However, existing model aggregation approaches suffer from high bias on both data distribu...
["Federated Learning", "Gradient Aggregation", "Variational Inference"]
ABSTRACTIn the federated learning paradigm, multiple mobile clients train local models in-dependently based on datasets generated by edge devices, and the server aggre-gates parameters/gradients from local models to form a global model. However,existing model aggregation approaches suffer from high bias on both data di...
5UndtHcTAac
An interesting view on parameter aggregation for federated learning which devises specific parameter aggregation rules to fit the nature of different layer types. The proposed solution and its fundamental idea are however a bit arbitrary (in many technical choices) and perhaps flawed. Clarifications and further elabor...
3: Clear rejection
PAPER SUMMARY This paper presents a new federated learning scheme for neural network that accounts simultaneously for different layer types and different (heterogeneous) local data distributions. In particular, the proposed method differentiates between parameter updates for conv, fc versus bn (batch-norm) layers. T...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Federated Learning with Decoupled Probabilistic-Weighted Gradient Aggregation ### Paper Abstract In the federated learning paradigm, multiple mobile clients train local...
B1x33sC9KQ
ICLR.cc/2019/Conference
2019
ACIQ: Analytical Clipping for Integer Quantization of neural networks
["Ron Banner", "Yury Nahshan", "Elad Hoffer", "Daniel Soudry"]
We analyze the trade-off between quantization noise and clipping distortion in low precision networks. We identify the statistics of various tensors, and derive exact expressions for the mean-square-error degradation due to clipping. By optimizing these expressions, we show marked improvements over standard quantizatio...
["quantization", "reduced precision", "training", "inference", "activation"]
ABSTRACTUnlike traditional approaches that focus on the quantization at the network level,in this work we propose to minimize the quantization effect at the tensor level.We analyze the trade-off between quantization noise and clipping distortion inlow precision networks. We identify the statistics of various tensors, a...
HJegw7SEhQ
A simple but not very convincing clipping method for activation quantization in deep networks
5: Marginally below acceptance threshold
This paper empirically finds that the distribution of activations in quantized networks follow Gaussian or Laplacian distribution, and proposes to determine the optimal clipping factor by minimizing the quantization error based on the distribution assumption. The pros of the work are its simplicity, the proposed clip...
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 ACIQ: Analytical Clipping for Integer Quantization of neural networks ### Paper Abstract We analyze the trade-off between quantization noise and clipping distortion in l...
BJx1SsAcYQ
ICLR.cc/2019/Conference
2019
Discovering Low-Precision Networks Close to Full-Precision Networks for Efficient Embedded Inference
["Jeffrey L. McKinstry", "Steven K. Esser", "Rathinakumar Appuswamy", "Deepika Bablani", "John V. Arthur", "Izzet B. Yildiz", "Dharmendra S. Modha"]
To realize the promise of ubiquitous embedded deep network inference, it is essential to seek limits of energy and area efficiency. To this end, low-precision networks offer tremendous promise because both energy and area scale down quadratically with the reduction in precision. Here, for the first time, we demonstra...
["Deep Learning", "Convolutional Neural Networks", "Low-precision inference", "Network quantization"]
ABSTRACTTo realize the promise of ubiquitous embedded deep network inference, it is es-sential to seek limits of energy and area efficiency. To this end, low-precisionnetworks offer tremendous promise because both energy and area scale downquadratically with the reduction in precision. Here, for the first time, we demo...
rkxJpxdMh7
A good but narrow contribution in a crowded space
6: Marginally above acceptance threshold
This manuscript joins a crowded space of methods for low bit quantization to enable inference on more efficient hardware. In the past, these methods often were limited to 8-bit quantization, or smaller networks, or result in accuracy degradation. This paper is part of a recent crop of methods that achieve full accuracy...
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 Discovering Low-Precision Networks Close to Full-Precision Networks for Efficient Embedded Inference ### Paper Abstract To realize the promise of ubiquitous embedded dee...
eEeyRrKVfbL
ICLR.cc/2021/Conference
2021
Balancing training time vs. performance with Bayesian Early Pruning
["Mohit Rajpal", "Yehong Zhang", "Bryan Kian Hsiang Low"]
Pruning is an approach to alleviate overparameterization of deep neural networks (DNN) by zeroing out or pruning DNN elements with little to no efficacy at a given task. In contrast to related works that do pruning before or after training, this paper presents a novel method to perform early pruning of DNN elements (e....
["Efficient Training", "Multi-Output Gaussian Process", "Gaussian Process", "Bayesian", "Single-shot network pruning", "Dynamic Sparse Reparameterization", "Lottery Ticket Hypothesis"]
ABSTRACTPruning is an approach to alleviate overparameterization of deep neural network(DNN) by zeroing out or pruning DNN elements with little to no efficacy at a giventask. In contrast to related works that do pruning before or after training, this paperpresents a novel method to perform early pruning of DNN elements...
m38SFfTc7-Y
Review - Bayesian early pruning
6: Marginally above acceptance threshold
This paper introduces a new method to accelerate training by saliency-based pruning. The method predicts future saliency for neurons based on observed saliency with a multi-output Gaussian process (MOGP), then greedily prunes neurons with least saliency at fixed intervals during training. The authors provide extensive ...
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 Balancing training time vs. performance with Bayesian Early Pruning ### Paper Abstract Pruning is an approach to alleviate overparameterization of deep neural networks (...
UcWZrerHDCe
NoDaLiDa/2023/Conference
2023
NoCoLA: The Norwegian Corpus of Linguistic Acceptability
["Matias Jentoft", "David Samuel"]
While there has been a surge of large language models for Norwegian in recent years, we lack any tool to evaluate their understanding of grammaticality. We present two new Norwegian datasets for this task. NoCoLA-class is a supervised binary classification task where the goal is to discriminate between acceptable and n...
["nocola", "linguistic acceptability", "language models", "nlp", "norwegian", "evaluation", "ask"]
NoCoLA: The Norwegian Corpus of Linguistic AcceptabilityMatias Jentoft and David SamuelUniversity of Oslo, Language Technology Group{matiasj, davisamu }@ifi.uio.noAbstractWhile there has been a surge of large lan-guage models for Norwegian in recentyears, we lack any tool to evaluate their un-derstanding of grammatical...
RMrI8slSrb
Good dataset, limited context
6: Marginally above acceptance threshold
The dataset presented is useful, and it's applaudable that other languages than English establish benchmarks for recognized NLP tasks. However, the paper provides way too limited context of previous work on similar datasets and on the area of linguistic acceptability, and especially on corresponding GLUE / SuperGLUE be...
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 NoCoLA: The Norwegian Corpus of Linguistic Acceptability ### Paper Abstract While there has been a surge of large language models for Norwegian in recent years, we lack ...
UJRFjuJDsIO
ICLR.cc/2021/Conference
2021
Why Convolutional Networks Learn Oriented Bandpass Filters: Theory and Empirical Support
["Isma Hadji", "Richard Wildes"]
It has been repeatedly observed that convolutional architectures when applied to image understanding tasks learn oriented bandpass filters. A standard explanation of this result is that these filters reflect the structure of the images that they have been exposed to during training: Natural images typically are locally...
["bandpass filters", "convolutional architectures", "filters", "structure", "convolutional networks", "empirical support", "eigenfunctions", "theory", "images", "filter characteristics"]
ABSTRACTIt has been repeatedly observed that convolutional architectures when applied toimage understanding tasks learn oriented bandpass filters. A standard explanationof this result is that these filters reflect the structure of the images that they havebeen exposed to during training: Natural images typically are lo...
lznvMxmNww2
Bandlimited concern is confusing for standard convnets (purely discrete-time formalism) : discrete sub-selection and distorsions are not necessarily associated with bandlimited processes
3: Clear rejection
The continuous domain formalism is not adapted to what the authors would like to prove because convnet discrete kernels make inadequate and questionable, any inference through a continuous formalism: one can say anything we want, giving that one can choose the intrinsic underlying continuous kernel as we want. Let us ...
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 Why Convolutional Networks Learn Oriented Bandpass Filters: Theory and Empirical Support ### Paper Abstract It has been repeatedly observed that convolutional architectu...
Xh_BzLS_3p
graphicsinterface.org/Graphics_Interface/2021/Conference/Second_Cycle
2021
AmbiTeam: Providing Team Awareness Through Ambient Displays
["Sarah Morrison-Smith", "Jaime Ruiz", "Lydia Chilton"]
Due to the COVID-19 pandemic, research is increasingly conducted remotely without the benefit of informal interactions that help maintain awareness of each collaborator's work progress. We developed AmbiTeam, an ambient display that shows activity related to the files of a team project, to help collaborations preserve ...
["Collaboration", "remote work", "awareness", "ambient display"]
ABSTRACTDue to the COVID-19 pandemic, research is increasingly conductedremotely without the benefit of informal interactions that help main-tain awareness of each collaborator’s work progress. We developedAmbiTeam, an ambient display that shows activity related to thefiles of a team project, to help collaborations pre...
8t3uBcFeykq
Review of "AmbiTeam: Providing Team Awareness Through Ambient Displays"
5: Marginally below acceptance threshold
The paper presents the AmbiTeam, an ambient display that shows activity related to the files of a team project, to help collaborations preserve a sense of the team’s involvement while working remotely. A 10-participant user study was conducted to determine whether AmbiTeam facilitates team awareness. The study results ...
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 AmbiTeam: Providing Team Awareness Through Ambient Displays ### Paper Abstract Due to the COVID-19 pandemic, research is increasingly conducted remotely without the bene...
HJxN0CNFPB
ICLR.cc/2020/Conference
2020
Ladder Polynomial Neural Networks
["Li-Ping Liu", "Ruiyuan Gu", "Xiaozhe Hu"]
The underlying functions of polynomial neural networks are polynomial functions. These networks are shown to have nice theoretical properties by previous analysis, but they are actually hard to train when their polynomial orders are high. In this work, we devise a new type of activations and then create the Ladder Poly...
["polynomial neural networks"]
ABSTRACTPolynomial neural networks are essentially polynomial functions. Thesenetworks are shown to have nice theoretical properties by previous analysis,but they are hard to train when their polynomial orders are high. In this work,we devise a new type of activations and then create the Ladder PolynomialNeural Network...
B1lRcpQ79H
Official Blind Review #1
6: Weak Accept
In this paper, the authors a new type of polynomial neural networks LPNN that can have an arbitrary polynomial order. The network has a feedforward structure which provides a good control of its polynomial order. Empirical study shows that deep LPNN models achieve good performances in regression and classification task...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Ladder Polynomial Neural Networks ### Paper Abstract The underlying functions of polynomial neural networks are polynomial functions. These networks are shown to have ni...
rJe-LiA5YX
ICLR.cc/2019/Conference
2019
Exponentially Decaying Flows for Optimization in Deep Learning
["Mitsuharu Takeori", "Kenta Nakamura"]
The field of deep learning has been craving for an optimization method that shows outstanding property for both optimization and generalization. We propose a method for mathematical optimization based on flows along geodesics, that is, the shortest paths between two points, with respect to the Riemannian metric induce...
["optimization", "deep learning"]
ABSTRACTThe field of deep learning has been craving for an optimization method that showsoutstanding property for both optimization and generalization. We propose amethod for mathematical optimization based on flows along geodesics, that is,the shortest paths between two points, with respect to the Riemannian metric in...
SJl2fxm937
Missing critical references; lacking novelty
3: Clear rejection
The first abstract reads, "The field of deep learning has been craving for an optimization method that shows outstanding property for both optimization and generalization", what does it mean to say that the field is "craving"? Yes, a rigorously (mathematically proven or empirically tested extensively) better algorithm ...
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 Exponentially Decaying Flows for Optimization in Deep Learning ### Paper Abstract The field of deep learning has been craving for an optimization method that shows outst...
rke3TJrtPS
ICLR.cc/2020/Conference
2020
Projection-Based Constrained Policy Optimization
["Tsung-Yen Yang", "Justinian Rosca", "Karthik Narasimhan", "Peter J. Ramadge"]
We consider the problem of learning control policies that optimize a reward function while satisfying constraints due to considerations of safety, fairness, or other costs. We propose a new algorithm - Projection-Based Constrained Policy Optimization (PCPO), an iterative method for optimizing policies in a two-step pro...
["Reinforcement learning with constraints", "Safe reinforcement learning"]
ABSTRACTWe consider the problem of learning control policies that optimize a reward func-tion while satisfying constraints due to considerations of safety, fairness, or othercosts. We propose a new algorithm, Projection-Based Constrained Policy Opti-mization (PCPO). This is an iterative method for optimizing policies i...
BkgdO3S0FS
Official Blind Review #3
6: Weak Accept
The paper proposes a technique to handle a certain type of constraints involved in Markov Decision Processes (MDP). The problem is well-motivated, and according to the authors, there is not much relevant work. The authors compare with the competing methods that they think are most appropriate. The numerical experiments...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Projection-Based Constrained Policy Optimization ### Paper Abstract We consider the problem of learning control policies that optimize a reward function while satisfying...
80FMcTSZ6J0
ICLR.cc/2021/Conference
2021
Noise against noise: stochastic label noise helps combat inherent label noise
["Pengfei Chen", "Guangyong Chen", "Junjie Ye", "jingwei zhao", "Pheng-Ann Heng"]
The noise in stochastic gradient descent (SGD) provides a crucial implicit regularization effect, previously studied in optimization by analyzing the dynamics of parameter updates. In this paper, we are interested in learning with noisy labels, where we have a collection of samples with potential mislabeling. We show t...
["Noisy Labels", "Robust Learning", "SGD noise", "Regularization"]
ABSTRACTThe noise in stochastic gradient descent (SGD) provides a crucial implicit regu-larization effect, previously studied in optimization by analyzing the dynamics ofparameter updates. In this paper, we are interested in learning with noisy labels,where we have a collection of samples with potential mislabeling. We...
7rU7TtgTBAw
Very interest idea, but the empirical studies are not sufficient
6: Marginally above acceptance threshold
This paper studies the noisy stochastic gradient descent algorithm in noisy label learning. Concretely, the authors added Gaussian noise on the labels rather than the gradient itself. By comparing different noisy SGD algorithms, the authors demonstrate that the proposed SLN algorithm not only help the model to escape f...
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 Noise against noise: stochastic label noise helps combat inherent label noise ### Paper Abstract The noise in stochastic gradient descent (SGD) provides a crucial implic...
6KeoIx2G0o
NeurIPS.cc/2022/Workshop/SVRHM
2022
Image-computable Bayesian model for 3D motion estimation with natural stimuli explains human biases
["Daniel Herrera-Esposito", "Johannes Burge"]
Estimating the motion of objects in depth is important for behavior, and is strongly supported by binocular visual cues. To understand how the brain should estimate motion in depth, we develop image-computable ideal observer models from naturalistic binocular video clips of two 3D motion tasks. The observers spatio-tem...
["3D vision", "Bayesian models", "Supervised learning", "Motion", "Canonical computations"]
Image-computable Bayesian model for 3D motionestimation with natural stimuli explains human biasesDaniel Herrera-EspositoDepartment of PsychologyUniversity of PennsylvaniaPhiladelphia, Pennsylvania 19104dherresp@sas.upenn.eduJohannes BurgeDepartment of PsychologyUniversity of PennsylvaniaPhiladelphia, Pennsylvania 1910...
XKK9CHSV960
Image-computable Bayesian model for 3D motion estimation with natural stimuli explains human biases
6: Marginally above acceptance threshold
This work presents a Bayesian approach to model 3D motion estimation. More specifically the paper proposes two tasks to estimate motion having speed and direction as latent variables for each task respectively. The authors train these models by learning filters to reproduce human response taken from a reference. Pro...
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 Image-computable Bayesian model for 3D motion estimation with natural stimuli explains human biases ### Paper Abstract Estimating the motion of objects in depth is impor...
rkxDon4Yvr
ICLR.cc/2020/Conference
2020
Discriminator Based Corpus Generation for General Code Synthesis
["Alexander Wild", "Barry Porter"]
Current work on neural code synthesis consists of increasingly sophisticated architectures being trained on highly simplified domain-specific languages, using uniform sampling across program space of those languages for training. By comparison, program space for a C-like language is vast, and extremely sparsely populat...
["Code Synthesis", "Neural Code Synthesis"]
ABSTRACTCurrent work on neural code synthesis consists of increasingly sophisticated archi-tectures being trained on highly simplified domain-specific languages, using uni-form sampling across program space of those languages for training. By compari-son, program space for a C-like language is vast, and extremely spars...
B1lGk2hsFB
Official Blind Review #2
1: Reject
This paper presents a technique based on genetic programming to generate a suitable training corpus of programs and I/O examples using a trained discriminator network. Given a set of human useful programs and the corresponding I/O examples, the main idea of the approach is to use genetic programming to iteratively cons...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Discriminator Based Corpus Generation for General Code Synthesis ### Paper Abstract Current work on neural code synthesis consists of increasingly sophisticated architec...
Bke13pVKPS
ICLR.cc/2020/Conference
2020
Improved Training Speed, Accuracy, and Data Utilization via Loss Function Optimization
["Santiago Gonzalez", "Risto Miikkulainen"]
As the complexity of neural network models has grown, it has become increasingly important to optimize their design automatically through metalearning. Methods for discovering hyperparameters, topologies, and learning rate schedules have lead to significant increases in performance. This paper shows that loss functions...
["metalearning", "evolutionary computation", "loss functions", "optimization", "genetic programming"]
ABSTRACTAs the complexity of neural network models has grown, it has become increas-ingly important to optimize their design automatically through metalearning.Methods for discovering hyperparameters, topologies, and learning rate scheduleshave lead to significant increases in performance. This paper shows that loss fu...
HJenQ5wnYS
Official Blind Review #1
3: Weak Reject
This paper proposes a very interesting idea of loss function optimization. At first sight, loss function is the goal of optimization and can not be optimized directly. However, the true goal of optimization is the final accuracy (for classification). So lots of loss functions can be designed and combined to form a larg...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Improved Training Speed, Accuracy, and Data Utilization via Loss Function Optimization ### Paper Abstract As the complexity of neural network models has grown, it has be...
B1xFxh0cKX
ICLR.cc/2019/Conference
2019
Guided Evolutionary Strategies: Escaping the curse of dimensionality in random search
["Niru Maheswaranathan", "Luke Metz", "George Tucker", "Dami Choi", "Jascha Sohl-Dickstein"]
Many applications in machine learning require optimizing a function whose true gradient is unknown, but where surrogate gradient information (directions that may be correlated with, but not necessarily identical to, the true gradient) is available instead. This arises when an approximate gradient is easier to compute t...
["evolutionary strategies", "optimization", "gradient estimators", "biased gradients"]
ABSTRACTMany applications in machine learning require optimizing a function whose truegradient is unknown, but where surrogate gradient information (directions thatmay be correlated with, but not necessarily identical to, the true gradient) is avail-able instead. This arises when an approximate gradient is easier to co...
rJeWFl_Y37
Interesting idea, but not really new
6: Marginally above acceptance threshold
The idea of this paper is to accelerate the OpenAI type evolution strategy by introducing 1. non-isotropic distribution, where the covariance matrix is of form I + UU^t; and 2. external information such as a surrogate gradient to determine U. The experiments show promising results. I think it is the right direction to ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Guided Evolutionary Strategies: Escaping the curse of dimensionality in random search ### Paper Abstract Many applications in machine learning require optimizing a funct...
ByJIWUnpW
ICLR.cc/2018/Conference
2018
Automatically Inferring Data Quality for Spatiotemporal Forecasting
["Sungyong Seo", "Arash Mohegh", "George Ban-Weiss", "Yan Liu"]
Spatiotemporal forecasting has become an increasingly important prediction task in machine learning and statistics due to its vast applications, such as climate modeling, traffic prediction, video caching predictions, and so on. While numerous studies have been conducted, most existing works assume that the data from d...
["spatiotemporal data", "graph convolutional network", "data quality"]
ABSTRACTSpatiotemporal forecasting has become an increasingly important prediction taskin machine learning and statistics due to its vast applications, such as climatemodeling, traffic prediction, video caching predictions, and so on. While numer-ous studies have been conducted, most existing works assume that the data...
rJDUzhtxf
Nice idea and application, but needs significant work before publication.
6: Marginally above acceptance threshold
The paper is an application of neural nets to data quality assessment. The authors introduce a new definition of data quality that relies on the notion of local variation defined in (Zhou and Schölkopf, 2004), and they extend it to multiple heterogenous data sources. The data quality function is learned using a GCN 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 Automatically Inferring Data Quality for Spatiotemporal Forecasting ### Paper Abstract Spatiotemporal forecasting has become an increasingly important prediction task in...
ERb8dfghQKX
graphicsinterface.org/Graphics_Interface/2021/Conference/Second_Cycle
2021
Library of Apollo: A Virtual Library Experience in your Web Browser
["Arda Ege Unlu", "Robert Xiao"]
Research libraries that house a large number of books organize, classify and lay out their collections by using increasingly rigorous classification systems. These classification systems make relations between books physically explicit within the dense library shelves, and allow for fast, easy and high quality serendip...
["human-computer interaction", "virtual reality", "libraries"]
ABSTRACTResearch libraries that house a large number of books organize,classify and lay out their collections by using increasingly rigorousclassification systems. These classification systems make relationsbetween books physically explicit within the dense library shelves,and allow for fast, easy and high quality sere...
pURMg7KfbC
Interesting system targeting a well-articulated problem, but limitations in the literature review weakens the scholarly contribution.
6: Marginally above acceptance threshold
This paper presents a 3D virtual library aimed at recreating the capabilities physical libraries offer in supporting browsing and serendipitous discovery. The need to support browsing and serendipity have been well argued in the library and information studies, and the paper does a good job of summarizing and articulat...
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 Library of Apollo: A Virtual Library Experience in your Web Browser ### Paper Abstract Research libraries that house a large number of books organize, classify and lay o...
r1g7xT4Kwr
ICLR.cc/2020/Conference
2020
PLEX: PLanner and EXecutor for Embodied Learning in Navigation
["Gil Avraham", "Yan Zuo", "Tom Drummond"]
We present a method for policy learning to navigate indoor environments. We adopt a hierarchical policy approach, where two agents are trained to work in cohesion with one another to perform a complex navigation task. A Planner agent operates at a higher level and proposes sub-goals for an Executor agent. The Executor ...
["Hierarchical Reinforcement Learning", "Embodied Learning", "Navigation"]
ABSTRACTWe present a method for policy learning to navigate indoor environments. Weadopt a hierarchical policy approach, where two agents are trained to work in co-hesion with one another to perform a complex navigation task. A Planner agentoperates at a higher level and proposes sub-goals for an Executor agent. TheExe...
BJxmdd2AFH
Official Blind Review #3
3: Weak Reject
This paper proposes a hierarchical policy consisting of a planner and an executor policy for solving navigation tasks. A planner policy consumes the desired target goal and a ELI context from the executor to propose subgoals for the executor. The executor consumes these subgoals to output actions that the agent should ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title PLEX: PLanner and EXecutor for Embodied Learning in Navigation ### Paper Abstract We present a method for policy learning to navigate indoor environments. We adopt a hie...
BuO4I7FgUbq
aclweb.org/ACL/2022/Workshop/CMCL
2022
Codenames as a Game of Co-occurrence Counting
["Anonymous"]
Codenames is a popular board game, in which knowledge and cooperation between players play an important role. The task of a player playing as a spymaster is to find words (clues) that a teammate finds related to as many of some given words as possible, but not to other specified words. This is a hard challenge even wit...
["co-occurrences", "associations", "word relatedness", "Codenames", "games", "human-machine cooperation"]
Codenames as a Game of Co-occurrence CountingRéka Cserháti andIstván Kolláth andAndrás Kicsi andGábor BerendInstitute of InformaticsUniversity of Szeged{cserhatir,kollathistvan}@gmail.com{akicsi,berendg}@inf.u-szeged.huAbstractCodenames is a popular board game, in whichknowledge and cooperation between playerspla...
BF-xYQQYizc
Solid work but some questions about conclusions and prior literature
6: Marginally above acceptance threshold
This paper contributes to a small but growing literature investigating the codenames board game as a way to evaluate models of human semantic memory. The authors separate sources of semantic relatedness and decision rules and evaluate both of these in their ability to guide human players in an online game. The direct...
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 Codenames as a Game of Co-occurrence Counting ### Paper Abstract Codenames is a popular board game, in which knowledge and cooperation between players play an important ...
FUtMxDTJ_h
ICLR.cc/2021/Conference
2021
Symmetry Control Neural Networks
["Marc Syvaeri", "Sven Krippendorf"]
This paper continues the quest for designing the optimal physics bias for neural networks predicting the dynamics of systems when the underlying dynamics shall be inferred from the data directly. The description of physical systems is greatly simplified when the underlying symmetries of the system are taken into accoun...
["Inductive (symmetry) Bias", "Predictive Models", "Hamiltonian Dynamics", "Physics"]
Under review as a conference paper at ICLR 2021Symmetry Control Neural NetworksAnonymous authorsPaper under double-blind reviewAbstractThis paper continues the quest for designing the optimal physics bias forneural networks predicting the dynamics of systems when the underlyingdynamics shall be inferred from the data d...
mfQO2UF7f0Q
an exploration of learning symmetries in the HNN paradigm
5: Marginally below acceptance threshold
This submission explores the question of identifying conserved quantities in Hamiltonian dynamics for physical systems by attempting to learn canonical transformations. The approach closely resembles previous work on "Hamiltonian neural networks" but the loss is augmented with a term enforcing the invariance of the dyn...
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 Symmetry Control Neural Networks ### Paper Abstract This paper continues the quest for designing the optimal physics bias for neural networks predicting the dynamics of ...
HkxJHlrFvr
ICLR.cc/2020/Conference
2020
Angular Visual Hardness
["Beidi Chen", "Weiyang Liu", "Animesh Garg", "Zhiding Yu", "Anshumali Shrivastava", "Jan Kautz", "Anima Anandkumar"]
The mechanisms behind human visual systems and convolutional neural networks (CNNs) are vastly different. Hence, it is expected that they have different notions of ambiguity or hardness. In this paper, we make a surprising discovery: there exists a (nearly) universal score function for CNNs whose correlation with huma...
["angular similarity", "self-training", "hard samples mining"]
ABSTRACTAlthough convolutional neural networks (CNNs) are inspired by the mechanismsbehind human visual systems, they diverge on many measures such as ambiguityor hardness. In this paper, we make a surprising discovery: there exists a (nearly)universal score function for CNNs whose correlation is statistically signific...
BJeifxHT5r
Official Blind Review #3
8: Accept
Main Contribution: This paper is trying to bridge the gap between CNN and the human visual system by proposing a metric (angular visual distance) and validate that this metric is correlated to the human visual hardness and this metric has a stronger relation compared to the softmax score which has been viewed as a ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Angular Visual Hardness ### Paper Abstract The mechanisms behind human visual systems and convolutional neural networks (CNNs) are vastly different. Hence, it is expect...
6Y_I9qc6JBD
eswc-conferences.org/ESWC/2021/Conference/Resources_Track
2021
A Knowledge Organization System for the United Nations Sustainable Development Goals
["Amit Joshi", "Luis Gerardo Gonzalez Morales", "Szymon Klarman", "Armando Stellato", "Aaron Helton", "Christopher Sean Lovell", "Artur Haczek"]
This paper presents a formal knowledge organization system (KOS) to represent the United Nations Sustainable Development Goals (SDGs). The SDGs are a set of objectives adopted by all United Nations member states in 2015 to achieve a better and sustainable future. The developed KOS consists of an ontology that models th...
["United Nations", "Sustainable Development Goals", "SDGs", "Ontology", "Linked Data", "Knowledge Organization Systems", "Metadata"]
A Knowledge Organization System for theUnited Nations Sustainable Development GoalsAmit Joshi1, Luis Gonzalez Morales1, Szymon Klarman2, ArmandoStellato3, Aaron Helton4, Sean Lovell1, and Artur Haczek21United Nations, Department of Economic and Social Aairs, New York, NY, USAfjoshi6, gonzalezmorales, lovells g@un.org2E...
r4AXqVW9xAp
Good work
2: Accept
The paper describes the developments of the ontology, data and applications about UN Sustainable Development Goals. The work is carefully and professionally done, with the open schemas, data and applications available for everyone's use. The paper is well written, and the descriptions are clear. The developments are ...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title A Knowledge Organization System for the United Nations Sustainable Development Goals ### Paper Abstract This paper presents a formal knowledge organization system (KOS) ...
FVhZIBWqykk
ICLR.cc/2021/Conference
2021
Resurrecting Submodularity for Neural Text Generation
["SIMENG HAN", "Xiang Lin", "Shafiq Joty"]
Submodularity is a desirable property for a variety of objectives in content selection where the current neural encoder-decoder framework is inadequate. We define a class of novel attention mechanisms with submodular functions and in turn, prove the submodularity of the effective neural coverage. The resulting attentio...
["submodularity", "text generation", "attention"]
ABSTRACTSubmodularity is desirable for a variety of objectives in content selection wherethe current neural encoder-decoder framework is inadequate. However, it has sofar not been explored in the neural encoder-decoder system for text generation. Inthis work, we define diminishing attentions with submodular functions a...
t7SYQb-AALz
A clean well-motivated approach to attention
6: Marginally above acceptance threshold
This paper describes a novel attention method that makes use of submodular functions to effectively give diminishing attention on an input item over time. Positives: it's a clean, motivated approach, with a good empirical study showing gains across several tasks. Negatives: it doesn't feel like a very large step over...
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 Resurrecting Submodularity for Neural Text Generation ### Paper Abstract Submodularity is a desirable property for a variety of objectives in content selection where the...
z9k8BWL-_2u
ICLR.cc/2021/Conference
2021
Statistical inference for individual fairness
["Subha Maity", "Songkai Xue", "Mikhail Yurochkin", "Yuekai Sun"]
As we rely on machine learning (ML) models to make more consequential decisions, the issue of ML models perpetuating unwanted social biases has come to the fore of the public's and the research community's attention. In this paper, we focus on the problem of detecting violations of individual fairness in ML models. We ...
["ml models", "individual fairness", "problem", "adversarial loss", "tools", "statistical inference", "machine learning", "models"]
ABSTRACTAs we rely on machine learning (ML) models to make more consequential decisions,the issue of ML models perpetuating or even exacerbating undesirable historicalbiases ( e.g.gender and racial biases) has come to the fore of the public’s attention.In this paper, we focus on the problem of detecting violations of i...
twqwvfTQeF6
Weak Accept
6: Marginally above acceptance threshold
The paper introduces a framework to statistically test whether a given model is individually fair or not. In particular, given a model, a distance metric over individuals, and a data point z, the authors propose an algorithm that finds a new data point z' such that z' is similar to z but their corresponding losses 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 Statistical inference for individual fairness ### Paper Abstract As we rely on machine learning (ML) models to make more consequential decisions, the issue of ML models ...
jE6SlVTOFPV
NeurIPS.cc/2020/Workshop/SVRHM
2020
Iterative VAE as a predictive brain model for out-of-distribution generalization
["Victor Boutin", "Aimen Zerroug", "Minju Jung", "Thomas Serre"]
Our ability to generalize beyond training data to novel, out-of-distribution, image degradations is a hallmark of primate vision. The predictive brain, exemplified by predictive coding networks (PCNs), has become a prominent neuroscience theory of neural computation. Motivated by the recent successes of variational aut...
["VAE", "Iterative VAE", "Predictive Coding", "Out-of-distribution generalization", "Generative models", "Visual perception"]
Iterative V AE as a predictive brain model forout-of-distribution generalizationVictor Boutin1,2, Aimen Zerroug1,2, Minju Jung2, Thomas Serre1,21Artificial and Natural Intelligence Toulouse Institute, Université de Toulouse, France2Carney Institute for Brain Science, Dpt. of Cognitive Linguistic & Psychological Scienc...
xiuEUjXkHYg
Interesting links to human studies of categorization
6: Marginally above acceptance threshold
The paper is very interesting in the sense that it takes the idea of semi amortization in context of variational autoencoders (VAEs) and studies how it might suggest computational accounts of human beings in recognizing out of distribution samples, specifically in a tradeoff of inference time and accuracy on the harder...
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 Iterative VAE as a predictive brain model for out-of-distribution generalization ### Paper Abstract Our ability to generalize beyond training data to novel, out-of-distr...
hnJSgY7p33a
ICLR.cc/2021/Conference
2021
Graph-Based Neural Network Models with Multiple Self-Supervised Auxiliary Tasks
["Franco Manessi", "Alessandro Rozza"]
Self-supervised learning is currently gaining a lot of attention, as it allows neural networks to learn robust representations from large quantities of unlabeled data. Additionally, multi-task learning can further improve representation learning by training networks simultaneously on related tasks, leading to significa...
["Graph Neural Networks", "Self-supervised learning", "Multi-task learning", "Graph Convolutional Networks", "Semi-supervised learning"]
ABSTRACTSelf-supervised learning is currently gaining a lot of attention, as it allows neuralnetworks to learn robust representations from large quantities of unlabeled data.Additionally, multi-task learning can further improve representation learning bytraining networks simultaneously on related tasks, leading to sign...
7zCH5afQDx
GNN + SSL in an interesting and trending topic, but the paper misses related works and the experiments are insufficient
4: Ok but not good enough - rejection
Summary: This paper aims to study how to enhance GNNs using multi-task learning and self-supervised learning. Specifically, the paper proposes a framework to improve graph-based semi-supervised node classification by jointly training on three self-supervised tasks, i.e., auto-encoding, corrupted embeddings reconstruc...
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 Graph-Based Neural Network Models with Multiple Self-Supervised Auxiliary Tasks ### Paper Abstract Self-supervised learning is currently gaining a lot of attention, as i...
HkzRQhR9YX
ICLR.cc/2019/Conference
2019
Tree-Structured Recurrent Switching Linear Dynamical Systems for Multi-Scale Modeling
["Josue Nassar", "Scott Linderman", "Monica Bugallo", "Il Memming Park"]
Many real-world systems studied are governed by complex, nonlinear dynamics. By modeling these dynamics, we can gain insight into how these systems work, make predictions about how they will behave, and develop strategies for controlling them. While there are many methods for modeling nonlinear dynamical systems, exis...
["machine learning", "bayesian statistics", "dynamical systems"]
ABSTRACTMany real-world systems studied are governed by complex, nonlinear dynamics. Bymodeling these dynamics, we can gain insight into how these systems work, makepredictions about how they will behave, and develop strategies for controlling them.While there are many methods for modeling nonlinear dynamical systems, ...
ryxsWdw927
Interesting extension to switching linear dynamical systems; scalability concerns and issues with experiments.
6: Marginally above acceptance threshold
The authors develop a tree structured extension to the recently proposed recurrent switching linear dynamical systems. Like switching linear dynamical systems (sLDS) the proposed models capture non-linear dynamics by switching between a collection of linear regimes. However, unlike SLDS, the transition between the regi...
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 Tree-Structured Recurrent Switching Linear Dynamical Systems for Multi-Scale Modeling ### Paper Abstract Many real-world systems studied are governed by complex, nonline...
kHromd7SNA
ICLR.cc/2021/Conference
2021
Motion Representations for Articulated Animation
["Aliaksandr Siarohin", "Oliver J. Woodford", "Jian Ren", "Menglei Chai", "Sergey Tulyakov"]
We propose novel motion representations for animating articulated objects consisting of distinct parts. In a completely unsupervised manner, our method identifies meaningful object parts, tracks them in a driving video, and infers their motions by considering their principal axes. In contrast to the previous keypoint-b...
["deep learning", "imaga animation", "generative modeling"]
ABSTRACTWe propose novel motion representations for animating articulated objects consist-ing of distinct parts. In a completely unsupervised manner, our method identifiesmeaningful object parts, tracks them in a driving video, and infers their motions byconsidering their principal axes. In contrast to the previous key...
dFtHLBI09rT
Interesting topic, unconvincing results, incremental.
4: Ok but not good enough - rejection
The proposed paper introduces a novel method for learning motion representations for articulated characters shown in videos. The main objective of the introduced approach is to identify object parts in an unsupervised manner and to track these parts based on a driving video. The goal of the method is to apply the motio...
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 Motion Representations for Articulated Animation ### Paper Abstract We propose novel motion representations for animating articulated objects consisting of distinct part...
r1f0YiCctm
ICLR.cc/2019/Conference
2019
Minimal Random Code Learning: Getting Bits Back from Compressed Model Parameters
["Marton Havasi", "Robert Peharz", "Jos\u00e9 Miguel Hern\u00e1ndez-Lobato"]
While deep neural networks are a highly successful model class, their large memory footprint puts considerable strain on energy consumption, communication bandwidth, and storage requirements. Consequently, model size reduction has become an utmost goal in deep learning. A typical approach is to train a set of determini...
["compression", "neural networks", "bits-back argument", "Bayesian", "Shannon", "information theory"]
ABSTRACTWhile deep neural networks are a highly successful model class, their large mem-ory footprint puts considerable strain on energy consumption, communicationbandwidth, and storage requirements. Consequently, model size reduction hasbecome an utmost goal in deep learning. A typical approach is to train a setof det...
S1lV-njc2m
The proposed approach has interesting formulation and good performance tradeoff while the main theorems are based on existing works.
7: Good paper, accept
This paper considers the compression of the model parameters in deep neural networks. The authors propose minimal random code learning (MIRACLE), which uses a random sample of weights and the variational framework interpreted by the bits-back argument. The authors introduce two theorems characterizing the properties 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 Minimal Random Code Learning: Getting Bits Back from Compressed Model Parameters ### Paper Abstract While deep neural networks are a highly successful model class, their...
Byx9p2EtDH
ICLR.cc/2020/Conference
2020
MULTIPOLAR: Multi-Source Policy Aggregation for Transfer Reinforcement Learning between Diverse Environmental Dynamics
["Mohammadamin Barekatain", "Ryo Yonetani", "Masashi Hamaya"]
Transfer reinforcement learning (RL) aims at improving learning efficiency of an agent by exploiting knowledge from other source agents trained on relevant tasks. However, it remains challenging to transfer knowledge between different environmental dynamics without having access to the source environments. In this work...
["reinforcement learning", "transfer learning", "policy aggregation", "residual policy learning"]
ABSTRACTTransfer reinforcement learning (RL) aims at improving learning efficiency of anagent by exploiting knowledge from other source agents trained on relevant tasks.However, it remains challenging to transfer knowledge between different environ-mental dynamics without having access to the source environments. In th...
rJxxKMssdB
Official Blind Review #3
1: Reject
The paper introduces a novel multi-source policy transfer problem, where we want to utilize policies from multiple source domains with different dynamics to improve the performance of the policy on our target dynamics. The paper addresses the problem by adaptively aggregating the deterministic actions produced by sou...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title MULTIPOLAR: Multi-Source Policy Aggregation for Transfer Reinforcement Learning between Diverse Environmental Dynamics ### Paper Abstract Transfer reinforcement learning...
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...
BJFoMD9eG
Incremental yet interesting advance in geometric CNNs. But, some core technical aspects and experiments are missing.
5: Marginally below acceptance threshold
Paper Summary: This work proposes a new geometric CNN model to process spatially sparse data. Like several existing geometric CNNs, convolutions are performed on each point using nearest neighbors. Instead of using a fixed or Gaussian parametric filters, this work proposes to predict filter weights using a multi-layer ...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning Graph Convolution Filters from Data Manifold ### Paper Abstract Convolution Neural Network (CNN) has gained tremendous success in computer vision tasks with its...
a-xFK8Ymz5J
ICLR.cc/2021/Conference
2021
DiffWave: A Versatile Diffusion Model for Audio Synthesis
["Zhifeng Kong", "Wei Ping", "Jiaji Huang", "Kexin Zhao", "Bryan Catanzaro"]
In this work, we propose DiffWave, a versatile diffusion probabilistic model for conditional and unconditional waveform generation. The model is non-autoregressive, and converts the white noise signal into structured waveform through a Markov chain with a constant number of steps at synthesis. It is efficiently trained...
["diffusion probabilistic models", "audio synthesis", "speech synthesis", "generative models"]
ABSTRACTIn this work, we propose DiffWave, a versatile diffusion probabilistic model for con-ditional and unconditional waveform generation. The model is non-autoregressive,and converts the white noise signal into structured waveform through a Markovchain with a constant number of steps at synthesis. It is efficiently ...
LTDmB3vxjc0
State of the art neural vocoder that is fast, small, accurate, and works with little supervision.
9: Top 15% of accepted papers, strong accept
This paper describes a neural vocoder based on a diffusion probabilistic model. The model utilizes a fixed-length markov chain to convert between a latent uncorrelated Gaussian vector and a full-length observation. The conversion from observation to latent is fixed and amounts to adding noise at each step. The conversi...
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 DiffWave: A Versatile Diffusion Model for Audio Synthesis ### Paper Abstract In this work, we propose DiffWave, a versatile diffusion probabilistic model for conditional...
BkXMikqxx
ICLR.cc/2017/conference
2017
Cortical-Inspired Open-Bigram Representation for Handwritten Word Recognition
["Th\u00e9odore Bluche", "Christopher Kermorvant", "Claude Touzet", "Herv\u00e9 Glotin"]
Recent research in the cognitive process of reading hypothesized that we do not read words by sequentially recognizing letters, but rather by identifing open-bigrams, i.e. couple of letters that are not necessarily next to each other. In this paper, we evaluate an handwritten word recognition method based on original ...
["representation", "letters", "handwritten word recognition", "cognitive process", "words", "couple", "next", "original", "long"]
ABSTRACTRecent research in the cognitive process of reading hypothesized that we do notread words by sequentially recognizing letters, but rather by identifing open-bigrams, i.e.couple of letters that are not necessarily next to each other. In thispaper, we evaluate an handwritten word recognition method based on origi...
ByaIT9WEg
Review
7: Good paper, accept
This submission investigates the usability of cortical-inspired distant bigram representations for handwritten word recognition. Instead of generating neural network based posterior features for character (optionally in local context), sets posterior for character bigrams of different length are used to represent words...
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 Cortical-Inspired Open-Bigram Representation for Handwritten Word Recognition ### Paper Abstract Recent research in the cognitive process of reading hypothesized that we...
BJluGHcee
ICLR.cc/2017/conference
2017
Tensorial Mixture Models
["Or Sharir", "Ronen Tamari", "Nadav Cohen", "Amnon Shashua"]
We introduce a generative model, we call Tensorial Mixture Models (TMMs) based on mixtures of basic component distributions over local structures (e.g. patches in an image) where the dependencies between the local-structures are represented by a "priors tensor" holding the prior probabilities of assigning a component d...
["Deep learning", "Supervised Learning", "Unsupervised Learning"]
ABSTRACTWe introduce a generative model, we call Tensorial Mixture Models (TMMs)based on mixtures of basic component distributions over local structures (e.g.patches in an image) where the dependencies between the local-structures are rep-resented by a ”priors tensor” holding the prior probabilities of assigning a comp...
SkGQ_LwEl
Interesting approach but ...
4: Ok but not good enough - rejection
This paper proposes a generative model for mixtures of basic local structures where the dependency between local structures is a tensor. They use tensor decomposition and the result of their earlier paper on expressive power of CNNs along with hierarchical Tucker to provide an inference mechanism. However, this is cond...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Tensorial Mixture Models ### Paper Abstract We introduce a generative model, we call Tensorial Mixture Models (TMMs) based on mixtures of basic component distributions o...
UqamDYtuh9
logconference.io/LOG/2022/Conference
2022
DIGRAC: Digraph Clustering Based on Flow Imbalance
["Yixuan He", "Gesine Reinert", "Mihai Cucuringu"]
Node clustering is a powerful tool in the analysis of networks. We introduce a graph neural network framework, named DIGRAC, to obtain node embeddings for directed networks in a self-supervised manner, including a novel probabilistic imbalance loss, which can be used for network clustering. Here, we propose \textit{dir...
["flow imbalance", "directed networks", "graph neural networks", "clustering", "directed stochastic block models"]
DIGRAC: Digraph Clustering Based on Flow ImbalanceYixuan HeDepartment of StatisticsUniversity of Oxfordyixuan.he@stats.ox.ac.ukGesine ReinertDepartment of StatisticsUniversity of Oxford &The Alan Turing Institute, London, UKreinert@stats.ox.ac.ukMihai CucuringuDepartment of Statistics and Mathematical InstituteUniversi...
KaQi2KNUuqU
Summary: The paper presents DIGRAC, an algorithm for clustering directed graphs. The algorithm has several advantages such as not requiring node labels and its ability to compute node embeddings. To accomplish this, DIGRAC introduces a directed flow imbalance measure and a graph neural network (GNN) architecture to...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title DIGRAC: Digraph Clustering Based on Flow Imbalance ### Paper Abstract Node clustering is a powerful tool in the analysis of networks. We introduce a graph neural netwo...
lIQ45G9P_zh
NeurIPS.cc/2021/Workshop/SVRHM
2021
Learning to perceive objects by prediction
["Tushar Arora", "Li Erran Li", "Ming Bo Cai"]
The representation of objects is the building block of higher-level concepts. Infants develop the notion of objects without supervision. The prediction error of future sensory input is likely the major teaching signal for infants. Inspired by this, we propose a new framework to extract object-centric representation fro...
["Biologically-driven self-supervision", "predictive learning", "object-centric representation", "3D perception", "sensory grounding"]
Learning to perceive objects by predictionTushar AroraIRCN, The University of Tokyoarora.tushar1208@gmail.comLi Erran LiAlexa AI, Amazonerranli@gmail.comMing Bo CaiIRCN, The University of Tokyomingbo.cai@ircn.jpAbstractThe representation of objects is the building block of higher-level concepts. Infantsdevelop the noti...
ykB_rC71m7_
Great submission. Minor suggestions on how to place the work in context of past studies at the intersection of neuroscience and machine learning.
9: Top 15% of accepted papers, strong accept
This is a strong submission, and I appreciate how the clarity which with the authors step through the paper from background, motivation, methods, and discussion. I also would like to thank the authors for carefully explain each component of the OPPLE network via a schematic figure and via text in the main paper as well...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning to perceive objects by prediction ### Paper Abstract The representation of objects is the building block of higher-level concepts. Infants develop the notion of...
ryeLXFe494
MIDL.io/2019/Conference/Abstract
2019
Do Lateral Views Help Automated Chest X-ray Predictions?
["Hadrien Bertrand", "Mohammad Hashir", "Joseph Paul Cohen"]
Most convolutional neural networks in chest radiology use only the frontal posteroanterior (PA) view to make a prediction. However the lateral view is known to help the diagnosis of certain diseases and conditions. The recently released PadChest dataset contains paired PA and lateral views, allowing us to study for whi...
["Chest X-ray", "classification", "multilabel", "multi-view"]
Proceedings of Machine Learning Research { Under Review:1{4, 2019 Extended Abstract { MIDL 2019 submissionDo Lateral Views Help Automated Chest X-ray Predictions?Hadrien Bertrand1hadrien.bertrand@mila.quebecMohammad Hashir1;2mohammad.hashir.khan@umontreal.caJoseph Paul Cohen1;2joseph.paul.cohen@mila.quebec1Mila, Quebec...
HkgK20_yjE
empirical study, interesting for a narrow audience
2: Reject
This abstract presents an empirical study, testing whether the use of a less frequent lateral radiograph may help image classification of chest X-ray, where posterior-anterior (PA) images are much more frequent. The results are interesting, but limited to a narrow audience (chest X-ray, or a handful of other imaging pr...
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 Do Lateral Views Help Automated Chest X-ray Predictions? ### Paper Abstract Most convolutional neural networks in chest radiology use only the frontal posteroanterior (P...
BTzgpgtNaGq
graphicsinterface.org/Graphics_Interface/2022/Conference
2022
Comparison of a VR Stylus with a Controller, Hand Tracking and a Mouse for Object Manipulation and Medical Marking Tasks in Virtual Reality
["Anonymous"]
For medical surgery planning, virtual reality (VR) provides a new kind of user experience, where 3D images of the operation area can be utilized. Using VR, it is possible to view the 3D models in a more realistic 3D environment, which would reduce the perception problems and increase spatial understanding. In the prese...
["3D visualization", "virtual reality", "hand tracking", "controller interaction", "mouse interaction", "3D imaging"]
ABSTRACTFor medical surgery planning, virtual reality (VR) provides a newkind of user experience, where 3D images of the operation area canbe utilized. Using VR, it is possible to view the 3D models in amore realistic 3D environment, which would reduce the perceptionproblems and increase spatial understanding. In the p...
OZkmL47CaiR
Improved from previous submission, now above the bar
6: Marginally above acceptance threshold
(Note: this review considers changes made for the second GI submission cycle; I also reviewed the initial submission). The current submission addresses several of the problems seen in the original version. The first review identified the following issues: * related work inadequate: the revision adds several papers t...
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 Comparison of a VR Stylus with a Controller, Hand Tracking and a Mouse for Object Manipulation and Medical Marking Tasks in Virtual Reality ### Paper Abstract For medica...
ryx4TlHKDS
ICLR.cc/2020/Conference
2020
EXACT ANALYSIS OF CURVATURE CORRECTED LEARNING DYNAMICS IN DEEP LINEAR NETWORKS
["Dongsung Huh"]
Deep neural networks exhibit complex learning dynamics due to the highly non-convex loss landscape, which causes slow convergence and vanishing gradient problems. Second order approaches, such as natural gradient descent, mitigate such problems by neutralizing the effect of potentially ill-conditioned curvature on the ...
["dynamics", "curvature", "gradient problems", "natural gradient descent", "gradient descent", "path", "approximations", "exact analysis"]
ABSTRACTDeep neural networks exhibit complex learning dynamics due to highly non-convex loss landscape. Second order approaches, such as natural gradient descent,mitigate such problems by neutralizing the effect of potentially ill-conditionedcurvature, yet it is largely unknown how the current theory of deep learning g...
BJx3CVotqH
Official Blind Review #2
1: Reject
In this manuscript, the authors analyze the dynamics of training deep linear neural networks under a generalized family of natural gradient methods that apply curvature corrections. They first show that the learning trajectory (direction of singular mode dynamics) in natural gradient descent follows the same path as gr...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title EXACT ANALYSIS OF CURVATURE CORRECTED LEARNING DYNAMICS IN DEEP LINEAR NETWORKS ### Paper Abstract Deep neural networks exhibit complex learning dynamics due to the high...
hsFN92eQEla
ICLR.cc/2021/Conference
2021
EVALUATION OF NEURAL ARCHITECTURES TRAINED WITH SQUARE LOSS VS CROSS-ENTROPY IN CLASSIFICATION TASKS
["Like Hui", "Mikhail Belkin"]
Modern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the square loss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of stand...
["large scale learning", "square loss vs cross-entropy", "classification", "experimental evaluation"]
ABSTRACTModern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the squareloss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of...
UtutCgLBgXe
Interesting paper that challenges the conventional wisdom of CE loss being superior to MSE loss in clasffication tasks.
8: Top 50% of accepted papers, clear accept
I think this a very good contribution to ICLR given the topic and the quality of the submission (originality, contribution to the stare of the art, experimental evidence, etc) Some of the strong points of the submission are summarized as follows: 1. The paper tackles a very interesting subject, questioning the con...
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 EVALUATION OF NEURAL ARCHITECTURES TRAINED WITH SQUARE LOSS VS CROSS-ENTROPY IN CLASSIFICATION TASKS ### Paper Abstract Modern neural architectures for classification ta...
e3bhF_p0T7c
ICLR.cc/2021/Conference
2021
Analysis of Alignment Phenomenon in Simple Teacher-student Networks with Finite Width
["Hanlin Zhu", "Chengyang Ying", "Song Zuo"]
Recent theoretical analysis suggests that ultra-wide neural networks always converge to global minima near the initialization under first order methods. However, the convergence property of neural networks with finite width could be very different. The simplest experiment with two-layer teacher-student networks shows t...
["alignment", "finite width network", "teacher student model", "angular distance function"]
ABSTRACTRecent theoretical analysis suggests that ultra-wide neural networks always con-verge to global minima near the initialization under first order methods. However,the convergence property of neural networks with finite width could be very dif-ferent. The simplest experiment with two-layer teacher-student network...
6-0mu0kqvvT
Interesting ideas for characterizing the critical points of over-parametrized student network for a teacher network with one neuron
4: Ok but not good enough - rejection
Summary of review: This paper studies the optimization landscape of over-parametrized two-layer ReLU neural networks under an (isotropic) Gaussian input distribution. The main goal is to characterize the critical points of the over-parametrized student network, and show that these critical points "align" with the neur...
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 Analysis of Alignment Phenomenon in Simple Teacher-student Networks with Finite Width ### Paper Abstract Recent theoretical analysis suggests that ultra-wide neural netw...
SkGpW3C5KX
ICLR.cc/2019/Conference
2019
Heated-Up Softmax Embedding
["Xu Zhang", "Felix Xinnan Yu", "Svebor Karaman", "Wei Zhang", "Shih-Fu Chang"]
Metric learning aims at learning a distance which is consistent with the semantic meaning of the samples. The problem is generally solved by learning an embedding, such that the samples of the same category are close (compact) while samples from different categories are far away (spread-out) in the embedding space. One...
["softmax", "samples", "metric", "embeddings", "classifier", "aims", "distance", "consistent", "semantic meaning", "problem"]
ABSTRACTMetric learning aims at learning a distance which is consistent with the semanticmeaning of the samples. The problem is generally solved by learning an embedding,such that the samples of the same category are close (compact) while samples fromdifferent categories are far away (spread-out) in the embedding space...
S1loo0Von7
Novelty
3: Clear rejection
The introduction and the title does not match. Metric learning does not require to specify the dimension; while the embedding has to specify the reduced dimension. I feel confused that the authors mix these two concepts. The objective in (1) is very close to that of t-SNE[5], where it uses the KL as the objective. The...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Heated-Up Softmax Embedding ### Paper Abstract Metric learning aims at learning a distance which is consistent with the semantic meaning of the samples. The problem is g...
XMoyS8zm6GA
ICLR.cc/2021/Conference
2021
Slice, Dice, and Optimize: Measuring the Dimension of Neural Network Class Manifolds
["Stanislav Fort", "Ekin Dogus Cubuk", "Surya Ganguli", "Samuel Stern Schoenholz"]
Deep neural network classifiers naturally partition input space into regions belonging to different classes. The geometry of these class manifolds (CMs) is widely studied and is intimately related to model performance; for example, the margin is defined via boundaries between these CMs. We present a simple technique to...
["input space", "random hyperplane", "optimization", "robustness", "dimension", "codimension", "manifold"]
ABSTRACTDeep neural network classifiers naturally partition input space into regions belong-ing to different classes. The geometry of these class manifolds (CMs) is widelystudied and is intimately related to model performance; for example, the marginis defined via boundaries between these CMs. We present a simple techn...
yETsvdDxtb-
Interesting and solid work, but the fundamental assumption seems problematic
4: Ok but not good enough - rejection
This paper proposes to understand the behavior of deep networks for classification tasks by studying the dimensionality of the "class manifolds", i.e., regions in the data space that are mapped to the same one-hot output. To measure such dimensionality, the paper proposes a method that is based on intersecting the clas...
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 Slice, Dice, and Optimize: Measuring the Dimension of Neural Network Class Manifolds ### Paper Abstract Deep neural network classifiers naturally partition input space i...
Uqu9yHvqlRf
ICLR.cc/2021/Conference
2021
What Preserves the Emergence of Language?
["Ziluo Ding", "Tiejun Huang", "Zongqing Lu"]
The emergence of language is a mystery. One dominant theory is that cooperation boosts language to emerge. However, as a means of giving out information, language seems not to be an evolutionarily stable strategy. To ensure the survival advantage of many competitors, animals are selfish in nature. From the perspective ...
["emergence of language", "reinforcement learning"]
ABSTRACTThe emergence of language is a mystery. One dominant theory is that coopera-tion boosts language to emerge. However, as a means of giving out information,language seems not to be an evolutionarily stable strategy. To ensure the survivaladvantage of many competitors, animals are selfish in nature. From the persp...
a0iw7YBEPt1
Interesting, but needs clearer motivation and detail
5: Marginally below acceptance threshold
# Overall review This paper attempts to address a question in the emergent communication literature: what preserves / maintains the stability of emerged communication protocols. The authors manipulate the prevalence of lying behavior in a community of agents playing a variant of a Nash bargaining game. The main take...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title What Preserves the Emergence of Language? ### Paper Abstract The emergence of language is a mystery. One dominant theory is that cooperation boosts language to emerge. H...
ByRWCqvT-
ICLR.cc/2018/Conference
2018
Learning to cluster in order to transfer across domains and tasks
["Yen-Chang Hsu", "Zhaoyang Lv", "Zsolt Kira"]
This paper introduces a novel method to perform transfer learning across domains and tasks, formulating it as a problem of learning to cluster. The key insight is that, in addition to features, we can transfer similarity information and this is sufficient to learn a similarity function and clustering network to perform...
["transfer learning", "similarity prediction", "clustering", "domain adaptation", "unsupervised learning", "computer vision", "deep learning", "constrained clustering"]
ABSTRACTThis paper introduces a novel method to perform transfer learning across domainsand tasks, formulating it as a problem of learning to cluster. The key insight isthat, in addition to features, we can transfer similarity information and this is suffi-cient to learn a similarity function and clustering network to ...
BJbSJYcgG
New method with a rough presentation
7: Good paper, accept
The authors propose a method for performing transfer learning and domain adaptation via a clustering approach. The primary contribution is the introduction of a Learnable Clustering Objective (LCO) that is trained on an auxiliary set of labeled data to correctly identify whether pairs of data belong to the same class. ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning to cluster in order to transfer across domains and tasks ### Paper Abstract This paper introduces a novel method to perform transfer learning across domains and...
rJxGGlSKwH
ICLR.cc/2020/Conference
2020
Sentence embedding with contrastive multi-views learning
["Antoine Simoulin"]
In this work, we propose a self-supervised method to learn sentence representations with an injection of linguistic knowledge. Multiple linguistic frameworks propose diverse sentence structures from which semantic meaning might be expressed out of compositional words operations. We aim to take advantage of this linguis...
["contrastive", "multi-views", "linguistic", "embedding"]
ABSTRACTIn this work, we propose a self-supervised method to learn sentence representa-tions with an injection of linguistic knowledge. Multiple linguistic frameworkspropose diverse sentence structures from which semantic meaning might be ex-pressed out of compositional words operations. We aim to take advantage ofthis...
BJejdUk3qr
Official Blind Review #4
3: Weak Reject
This paper describes a self-supervised sentence embedding approach that incorporates a different view from plain text where some extent of linguistic knowledge is incorporated through the application of tree LSTM. The training procedure is standard contrastive framework where the model is encouraged to distinguish betw...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Sentence embedding with contrastive multi-views learning ### Paper Abstract In this work, we propose a self-supervised method to learn sentence representations with an i...
dx85vTSjh1h
NeurIPS.cc/2022/Workshop/Federated_Learning
2022
DASH: Decentralized CASH for Federated Learning
["Md Ibrahim Ibne Alam", "Koushik Kar", "Theodoros Salonidis", "Horst Samulowitz"]
We present DASH, a decentralized framework that addresses for the first time the Combined Algorithm Selection and HyperParameter Optimization (CASH) problem in Federated Learning (FL) settings. DASH generates a set of algorithm-hyper-parameter (Alg-HP) pairs using existing centralized HPO algorithms which are then eval...
["Decentralized FL", "Decentralized CASH."]
DASH: Decentralized CASH for Federated LearningMd Ibrahim Ibne AlamDepartment of ECSERensselaer Polytechnic InstituteTroy, NY , USA - 12180alamm2@rpi.eduKoushik KarDepartment of ECSERensselaer Polytechnic InstituteTroy, NY , USA - 12180kark@rpi.eduTheodoros SalonidisIBM T.J. Watson Research CenterYorktown Heights, NY ,...
cwVPf0invT1
Review
The paper proposes a decentralized hyper-parameter optimization and algorithm selection method for the federated learning framework. Specifically, because of the high dimension and communication cost of the centralized version, the paper proposes to sent the parameters copies to every clients and aggregate them in the ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title DASH: Decentralized CASH for Federated Learning ### Paper Abstract We present DASH, a decentralized framework that addresses for the first time the Combined Algorithm Se...
0N8jUH4JMv6
ICLR.cc/2021/Conference
2021
Implicit Convex Regularizers of CNN Architectures: Convex Optimization of Two- and Three-Layer Networks in Polynomial Time
["Tolga Ergen", "Mert Pilanci"]
We study training of Convolutional Neural Networks (CNNs) with ReLU activations and introduce exact convex optimization formulations with a polynomial complexity with respect to the number of data samples, the number of neurons, and data dimension. More specifically, we develop a convex analytic framework utilizing sem...
["Convex optimization", "non-convex optimization", "group sparsity", "$\\ell_1$ norm", "convex duality", "polynomial time", "deep learning"]
ABSTRACTWe study training of Convolutional Neural Networks (CNNs) with ReLU activa-tions and introduce exact convex optimization formulations with a polynomialcomplexity with respect to the number of data samples, the number of neurons,and data dimension. More specifically, we develop a convex analytic frameworkutilizi...
Zz8q4LhbbU9
Official Blind Review #3
6: Marginally above acceptance threshold
[Summary] This paper focuses on training convolutional neural networks (CNNs) by using convex optimization techniques. By taking the dual of the nonconvex training problems, (and the dual of its dual), the main contribution of the paper is to show the strong duality between the convex problem and its original nonconvex...
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 Implicit Convex Regularizers of CNN Architectures: Convex Optimization of Two- and Three-Layer Networks in Polynomial Time ### Paper Abstract We study training of Convol...
H1gDaa4YwS
ICLR.cc/2020/Conference
2020
Learning General and Reusable Features via Racecar-Training
["You Xie", "Nils Thuerey"]
We propose a novel training approach for improving the learning of generalizing features in neural networks. We augment the network with a reverse pass which aims for reconstructing the full sequence of internal states of the network. Despite being a surprisingly simple change, we demonstrate that this forward-backward...
["transfer learning", "neural networks", "generalization", "reusable features"]
ABSTRACTWe propose a novel training approach for improving the learning of generalizingfeatures in neural networks. We augment the network with a reverse pass whichaims for reconstructing the full sequence of internal states of the network. Despitebeing a surprisingly simple change, we demonstrate that this forward-bac...
B1lfbmsKtH
Official Blind Review #2
1: Reject
This paper attempts to learn general and reusable features for transfer learning tasks. The authors propose a training paradigm called Racecar Training. The core idea of it is to operate a reverse pass for the network. The authors use mutual information to analyze the network for its improved generalizing capabilities....
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning General and Reusable Features via Racecar-Training ### Paper Abstract We propose a novel training approach for improving the learning of generalizing features i...
gHsr-v8Tz6l
ICLR.cc/2021/Conference
2021
Variational Invariant Learning for Bayesian Domain Generalization
["Zehao Xiao", "Jiayi Shen", "Xiantong Zhen", "Ling Shao", "Cees G. M. Snoek"]
Domain generalization addresses the out-of-distribution problem, which is challenging due to the domain shift and the uncertainty caused by the inaccessibility to data from the target domains. In this paper, we propose variational invariant learning, a probabilistic inference framework that jointly models domain invari...
["domain generalization", "variational invariant learning", "Bayesian inference"]
ABSTRACTDomain generalization addresses the out-of-distribution problem, which is chal-lenging due to the domain shift and the uncertainty caused by the inaccessibilityto data from the target domains. In this paper, we propose variational invariantlearning , a probabilistic inference framework that jointly models domai...
bCeFSKfymgw
Although I like the research direction of this paper, I have some concerns about the technical novelty of this method
5: Marginally below acceptance threshold
Summary: - The paper proposes a Bayesian inference framework for domain generalization (DG) that deals with both the uncertainty and domain shit. The proposed method treats the uncertainty effectively by applying the variational Bayesian inference to the last two layers of the model (feature representations and classi...
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 Variational Invariant Learning for Bayesian Domain Generalization ### Paper Abstract Domain generalization addresses the out-of-distribution problem, which is challengin...
SkeRTsAcYm
ICLR.cc/2019/Conference
2019
Phase-Aware Speech Enhancement with Deep Complex U-Net
["Hyeong-Seok Choi", "Jang-Hyun Kim", "Jaesung Huh", "Adrian Kim", "Jung-Woo Ha", "Kyogu Lee"]
Most deep learning-based models for speech enhancement have mainly focused on estimating the magnitude of spectrogram while reusing the phase from noisy speech for reconstruction. This is due to the difficulty of estimating the phase of clean speech. To improve speech enhancement performance, we tackle the phase estima...
["speech enhancement", "deep learning", "complex neural networks", "phase estimation"]
ABSTRACTMost deep learning-based models for speech enhancement have mainly focusedon estimating the magnitude of spectrogram while reusing the phase from noisyspeech for reconstruction. This is due to the difficulty of estimating the phaseof clean speech. To improve speech enhancement performance, we tackle thephase es...
S1lX3gYqhm
generally good paper on speech enhancement using complex operations
7: Good paper, accept
This paper used a complex-valued network to learn the modified complex ratio mask with a weighted SDR loss for the speech enhancement task. It can get good enhancement performance. For me, the complex-valued network is already there and weighted SDR loss is not difficult to think. The modified complex ratio mask is 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 Phase-Aware Speech Enhancement with Deep Complex U-Net ### Paper Abstract Most deep learning-based models for speech enhancement have mainly focused on estimating the ma...
icP8jy6ayy8
JSYS/2021/Nov_Papers
2021
Solution: End-to-end Scheduling of Real-time Task Pipelines on Multiprocessors
["Soham Sinha", "Richard West"]
Task pipelines are common in today’s embedded systems, as data moves from source to sink in sensing-processing-actuation task chains. A real-time task pipeline is constructed by connecting a series of periodic tasks with data buffers. In a time-critical system, end-to-end timing and data-transfer properties of a task p...
["real-time task pipelines", "constraint optimization", "end-to-end scheduling"]
END-TO-END SCHEDULING OF REAL-TIME TASK PIPELINES ONMULTIPROCESSORSSOHAM SINHADepartment of Computer ScienceBoston University, Boston, MAsoham1@bu.eduRICHARD WESTDepartment of Computer ScienceBoston University, Boston, MArichwest@bu.eduAbstractTask pipelines are common in today’s embedded systems,as data moves from sou...
eTS7gGNEH4M
Unmotivated contribution with significant problems
*** Summary *** The paper presents the formulation and solution of a mixed integer non linear programming problem to determine the best periods for pipelines of tasks that share data with one another. *** Overall impression *** The paper is certainly not ready for publication: 1) the relevance of the problem is not...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Solution: End-to-end Scheduling of Real-time Task Pipelines on Multiprocessors ### Paper Abstract Task pipelines are common in today’s embedded systems, as data moves fr...
rJlPdcY38B
NeurIPS.cc/2019/Workshop/Program_Transformations
2019
Sparsity Programming: Automated Sparsity-Aware Optimizations in Differentiable Programming
["Shashi Gowda", "Yingbo Ma", "Valentin Churavy", "Alan Edelman", "Christopher Rackauckas"]
Previous studies in numerical analysis have shown how the calculation of a Jacobian, Hessian, and their factorizations can be accelerated when their sparsity pattern is known. However, accurate Jacobian and Hessian sparsity patterns cannot be computed numerically, leaving the burden on the user to provide them. As sci...
["program transformation", "differentiable programming", "automatic differentiation", "sparsity detection"]
Sparsity Programming: Automated Sparsity-AwareOptimizations in Differentiable ProgrammingShashi GowdaMassachusetts Institute of TechnologyValentin ChuravyMassachusetts Institute of TechnologyAlan EdelmanMassachusetts Institute of TechnologyYingbo MaJulia ComputingChristopher RackauckasMassachusetts Institute of Technol...
Hye70AciDH
Straightforward use of operator overloading (non-standard interpretation) for sparsity detection
6: Marginally above acceptance threshold
The authors implement an operator overloading method in Julia which detects sparsity. Variables are tagged with their dependencies and these are propagated throughout the program. For Hessians, the framework requires a set of mathematical primitives which are marked as linear or non-linear. A few things that are not di...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Sparsity Programming: Automated Sparsity-Aware Optimizations in Differentiable Programming ### Paper Abstract Previous studies in numerical analysis have shown how the c...
M9hdyCNlWaf
ICLR.cc/2021/Conference
2021
Sparse Uncertainty Representation in Deep Learning with Inducing Weights
["Hippolyt Ritter", "Martin Kukla", "Cheng Zhang", "Yingzhen Li"]
Bayesian neural networks and deep ensembles represent two modern paradigms of uncertainty quantification in deep learning. Yet these approaches struggle to scale mainly due to memory inefficiency issues, since they require parameter storage several times higher than their deterministic counterparts. To address this, we...
["Bayesian neural networks", "uncertainty estimation", "memory efficiency"]
ABSTRACTBayesian neural networks and deep ensembles represent two modern paradigmsof uncertainty quantification in deep learning. Yet these approaches struggle toscale mainly due to memory inefficiency issues, since they require parameter stor-age several times higher than their deterministic counterparts. To address t...
n4DQA_680-M
A compact variational BNN posterior for memory efficiency
6: Marginally above acceptance threshold
Bayesian deep learning attempts to incorporate uncertainty estimation in the modern neural network models. However, common approaches such as Bayesian neural networks and Deep Ensembles incur large memory overhead because of the increased parameter sizes. Motivated by the inducing points approach in the sparse variatio...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Sparse Uncertainty Representation in Deep Learning with Inducing Weights ### Paper Abstract Bayesian neural networks and deep ensembles represent two modern paradigms of...
B1xwv1StvS
ICLR.cc/2020/Conference
2020
Few-shot Learning by Focusing on Differences
["Muhammad Rizki Maulana", "Lee Wee Sun"]
Few-shot classification may involve differentiating data that belongs to a different level of labels granularity. Compounded by the fact that the number of available labeled examples are scarce in the novel classification set, relying solely on the loss function to implicitly guide the classifier to separate data based...
["Deep learning", "few-shot learning"]
ABSTRACTFew-shot classification may involve differentiating data that belongs to a differentlevel of labels granularity. Compounded by the fact that the number of availablelabeled examples are scarce in the novel classification set, relying solely on theloss function to implicitly guide the classifier to separate data ...
S1l8TWFRFS
Official Blind Review #2
3: Weak Reject
The stated contributions of the paper are: (1) a method for performing few-shot learning and (2) an approach for building harder few-shot learning datasets from existing datasets. The authors describe a model for creating a task-aware embedding for different novel sets (for different image classification settings) usin...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Few-shot Learning by Focusing on Differences ### Paper Abstract Few-shot classification may involve differentiating data that belongs to a different level of labels gran...
ryGvcoA5YX
ICLR.cc/2019/Conference
2019
Overcoming Catastrophic Forgetting for Continual Learning via Model Adaptation
["Wenpeng Hu", "Zhou Lin", "Bing Liu", "Chongyang Tao", "Zhengwei Tao", "Jinwen Ma", "Dongyan Zhao", "Rui Yan"]
Learning multiple tasks sequentially is important for the development of AI and lifelong learning systems. However, standard neural network architectures suffer from catastrophic forgetting which makes it difficult for them to learn a sequence of tasks. Several continual learning methods have been proposed to address t...
["overcoming forgetting", "model adaptation", "continual learning"]
ABSTRACTLearning multiple tasks sequentially is important for the development of AI andlifelong learning systems. However, standard neural network architectures sufferfrom catastrophic forgetting which makes it difficult for them to learn a sequenceof tasks. Several continual learning methods have been proposed to addr...
SJxhhsOmhm
Interesting ideas and results. Needs improvement in writing, formalization and possibly more ablation studies.
7: Good paper, accept
This paper proposes a Dynamic Parameter Generator (DPG) that given a test input modifies the parameters of a classification model. They also propose to regularize the training using a Data Generator (DG) to slow down catastrophic forgetting. DG is used to constrain the training that the internal representations of data...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Overcoming Catastrophic Forgetting for Continual Learning via Model Adaptation ### Paper Abstract Learning multiple tasks sequentially is important for the development o...
PlUA_mgGaPq
aclweb.org/ACL/2020/Workshop/NLP-COVID
2020
Rapidly Deploying a Neural Search Engine for the COVID-19 Open Research Dataset: Preliminary Thoughts and Lessons Learned
["Edwin Zhang", "Nikhil Gupta", "Rodrigo Nogueira", "Kyunghyun Cho", "Jimmy Lin"]
We present the Neural Covidex, a search engine that exploits the latest neural ranking architectures to provide information access to the COVID-19 Open Research Dataset curated by the Allen Institute for AI. This web application exists as part of a suite of tools that we have developed over the past few weeks to help d...
["open research dataset", "lessons", "neural search engine", "preliminary thoughts", "neural covidex", "search engine", "information access", "allen institute"]
Rapidly Deploying a Neural Search Engine for the COVID-19 OpenResearch Dataset: Preliminary Thoughts and Lessons LearnedEdwin Zhang,1Nikhil Gupta,1Rodrigo Nogueira,1Kyunghyun Cho,2;3;4;5andJimmy Lin11David R. Cheriton School of Computer Science, University of Waterloo2Courant Institute of Mathematical Sciences, New Yor...
r405CzUub_7
Useful resources collected and introduced
6: Marginally above acceptance threshold
This work is an introduction to a suite of resources and their corresponding ideas revolving around the COVID data provided by the Allen institute. Authors in particular provide Neural Covidex which allows exploration of information retrieval and NLP techniques over the COVID dataset. The positives of this manuscript ...
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 Rapidly Deploying a Neural Search Engine for the COVID-19 Open Research Dataset: Preliminary Thoughts and Lessons Learned ### Paper Abstract We present the Neural Covide...
Hy5QRt1wz
ICLR.cc/2018/Workshop
2018
Semi-Supervised Learning With GANs: Revisiting Manifold Regularization
["Bruno Lecouat", "Chuan Sheng Foo", "Houssam Zenati", "Vijay Ramaseshan Chandrasekhar"]
GANS are powerful generative models that are able to model the manifold of natural images. We leverage this property to perform manifold regularization by approximating the Laplacian norm using a Monte Carlo approximation that is easily computed with the GAN. When incorporated into the feature-matching GAN of Salimans ...
["Generative Adversarial Networks", "Semi Supervised Learning"]
ABSTRACTGANS are powerful generative models that are able to model the manifold ofnatural images. We leverage this property to perform manifold regularization byapproximating the Laplacian norm using a Monte Carlo approximation that is eas-ily computed with the GAN. When incorporated into the feature-matching GANof Sal...
B1hFJATOf
interesting synthesis of prior methods
8: Top 50% of accepted papers, clear accept
I think this is a neat implementation of an intuitively-appealing idea -- i.e., we can regularize a classifier by encouraging its predictions to change slowly along trajectories that follow the data manifold. The general idea of optimizing monte-carlo approximations of a classifier's sensitivity to perturbations -- in...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Semi-Supervised Learning With GANs: Revisiting Manifold Regularization ### Paper Abstract GANS are powerful generative models that are able to model the manifold of natu...
rkTBjG-AZ
ICLR.cc/2018/Conference
2018
DeepArchitect: Automatically Designing and Training Deep Architectures
["Renato Negrinho", "Geoff Gordon"]
In deep learning, performance is strongly affected by the choice of architecture and hyperparameters. While there has been extensive work on automatic hyperpa- rameter optimization for simple spaces, complex spaces such as the space of deep architectures remain largely unexplored. As a result, the choice of architectur...
["architecture search", "deep learning", "hyperparameter tuning"]
ABSTRACTIn deep learning, performance is strongly affected by the choice of architectureand hyperparameters. While there has been extensive work on automatic hyperpa-rameter optimization for simple spaces, complex spaces such as the space of deeparchitectures remain largely unexplored. As a result, the choice of archit...
Sy4q4vBgf
The authors propose to automatically design and train deep architectures.
5: Marginally below acceptance threshold
This paper introduces a DeepArchitect framework to build and train deep models automatically. Specifically, the authors proposes three components, i.e., model search specification language, model search algorithm, model evaluation algorithm. The paper is written well, and the proposed framework provides us with a syste...
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 DeepArchitect: Automatically Designing and Training Deep Architectures ### Paper Abstract In deep learning, performance is strongly affected by the choice of architectur...
eyykI3UIHa
robot-learning.org/CoRL/2023/Conference
2023
NOIR: Neural Signal Operated Intelligent Robots for Everyday Activities
["Ruohan Zhang", "Sharon Lee", "Minjune Hwang", "Ayano Hiranaka", "Chen Wang", "Wensi Ai", "Jin Jie Ryan Tan", "Shreya Gupta", "Yilun Hao", "Gabrael Levine", "Ruohan Gao", "Anthony Norcia", "Li Fei-Fei", "Jiajun Wu"]
We present Neural Signal Operated Intelligent Robots (NOIR), a general-purpose, intelligent brain-robot interface system that enables humans to command robots to perform everyday activities through brain signals. Through this interface, humans communicate their intended objects of interest and actions to the robots usi...
["Brain-Robot Interface", "Human-Robot Interaction"]
NOIR: Neural Signal OperatedIntelligent Robots for Everyday ActivitiesRuohan Zhang*1,4, Sharon Lee*1, Minjune Hwang*1, Ayano Hiranaka*2,Chen Wang1, Wensi Ai1, Jin Jie Ryan Tan1, Shreya Gupta1, Yilun Hao1,Gabrael Levine1, Ruohan Gao1, Anthony Norcia3, Li Fei-Fei1,4, Jiajun Wu1,4*Equally contributed; zharu@stanford.edu1D...
zUB1TskpgS
The paper in my opinion is well written and straightforward to understand. The line of argument behind the approach is reasonable and the design decisions made by the authors are well explained and reasonable as well. The approach is evaluated in various experiments and the results seem to be convincing. Citations seem...
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 NOIR: Neural Signal Operated Intelligent Robots for Everyday Activities ### Paper Abstract We present Neural Signal Operated Intelligent Robots (NOIR), a general-purpose...
udqDDCjdjs5
ICLR.cc/2021/Conference
2021
Unified analytic forms for Convolutional Neural Networks and Wavelet Filter Banks
["Abdourrahmane M ATTO"]
This paper provides a unified analytic framework integrating expressions of several variants of convolutional neural networks and wavelet filter banks. The expressions are derived recursively, from dowstream to upstream layers, for the sequences of features returned at the nodes of a general form of network architectur...
["Convolutional neural network", "Multiserial convolution filter bank", "Analytic expressions"]
ABSTRACTThis paper provides a unified analytic framework integrating expressions of sev-eral variants of convolutional neural networks and wavelet filter banks. The ex-pressions are derived recursively, from downstream to upstream layers, for thesequences of features returned at the nodes of a general form of network a...
_nxR-1JrqKD
The idea may be promissing but the motivation is missing
2: Strong rejection
The paper proposes to unify wavelet filter banks and convolutional neural networks. They authors start from presenting a brief theory of M-Band Discrete Wavelet Packet Transform. Then, after a series of relaxations and transformations they connect the derived equations to the equations of convolutional neural networks...
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 Unified analytic forms for Convolutional Neural Networks and Wavelet Filter Banks ### Paper Abstract This paper provides a unified analytic framework integrating express...
ByePUo05K7
ICLR.cc/2019/Conference
2019
What a difference a pixel makes: An empirical examination of features used by CNNs for categorisation
["Gaurav Malhotra", "Jeffrey Bowers"]
Convolutional neural networks (CNNs) were inspired by human vision and, in some settings, achieve a performance comparable to human object recognition. This has lead to the speculation that both systems use similar mechanisms to perform recognition. In this study, we conducted a series of simulations that indicate that...
["deep learning", "shape bias", "vision", "feature selection"]
ABSTRACTConvolutional neural networks (CNNs) were inspired by human vision and, insome settings, achieve a performance comparable to human object recognition.This has lead to the speculation that both systems use similar mechanisms to per-form recognition. In this study, we conducted a series of simulations that indica...
H1ebFhuRnQ
Shape-bias or shortcut-bias and catastrophic forgetting?
4: Ok but not good enough - rejection
The paper seeks to establish via a series of well-designed experiments that CNNs trained for image classification differ in a fundamental way from human vision – they don’t encode shape-bias like human vision. Towards this goal, the authors modified the training data with ‘shortcut’ features to be functions of the cate...
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 What a difference a pixel makes: An empirical examination of features used by CNNs for categorisation ### Paper Abstract Convolutional neural networks (CNNs) were inspir...
BJlLQlrFwS
ICLR.cc/2020/Conference
2020
Minimizing Change in Classifier Likelihood to Mitigate Catastrophic Forgetting
["Ashish Gaurav", "Sachin Vernekar", "Sean Sedwards", "Jaeyoung Lee", "Vahdat Abdelzad", "Krzysztof Czarnecki"]
Continual learning is a longstanding goal of artificial intelligence, but is often counfounded by catastrophic forgetting that prevents neural networks from learning tasks sequentially. Previous methods in continual learning have demonstrated how to mitigate catastrophic forgetting, and learn new tasks while retaining ...
["catastrophic forgetting", "continual learning", "classification", "regularization"]
ABSTRACTContinual learning is a longstanding goal of artificial intelligence, but is oftenconfounded by “catastrophic forgetting” that prevents neural networks from learn-ing tasks sequentially. Previous methods in continual learning have demonstratedhow to mitigate catastrophic forgetting, and learn new tasks while re...
Bkg9_8antS
Official Blind Review #3
3: Weak Reject
This paper introduces a framework for combatting catastrophic forgetting, based upon changing the loss term to minimise changes in classifier likelihood, obtained via a Taylor series approximation. They compare different versions of their method with EWC and a method based on freezing important weights on three benchma...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Minimizing Change in Classifier Likelihood to Mitigate Catastrophic Forgetting ### Paper Abstract Continual learning is a longstanding goal of artificial intelligence, b...
HHSxCet46fq
graphicsinterface.org/Graphics_Interface/2022/Conference
2022
DisRnet: Discrete Relevance semantic segmentation network based on Discrete Data Distribution
["Anonymous"]
In the current pixel-level semantic segmentation tasks, the current popular algorithm models are generally based on CNNs and combine contextual information to achieve semantic segmentation of images. However, when these algorithms extract features, they are all affected by the mean pooling layer or the maximum pooling ...
["Discrete Data Analysis", "Residual structure", "Pool layer", "Image features", "Related Features", "Discrete features."]
ABSTRACT In the current pixel-level semantic segmentation tasks, the current popular algorithm models are generally based on CNNs and combine contextual information to achieve semantic segmentation of images. However, when these algorithms extract features, they are all affected by the mean pooling layer or the maximum...
ry3Qh_dD-Et
The paper presents a novel pixel-level image segmentation DNN. which extends ResNet with a so-called "discrete feature" layer and a correlation layer. Through experiments with some standardized data sets (Cityscapes, PASCAL VOC) they show the superiority of their approach.
3: Clear rejection
The paper modifies an existing architecture for pixel-level segmenation of images with a "Dis pool" layer and a "Rel pool" layer. While the experimental validation looks promising, the details in the paper are hard to follow. Besides a number of grammar and spelling problems, some basics remain unclear that cloud a pro...
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 DisRnet: Discrete Relevance semantic segmentation network based on Discrete Data Distribution ### Paper Abstract In the current pixel-level semantic segmentation tasks, ...