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
rJleFREKDr
ICLR.cc/2020/Conference
2020
Learning to Control Latent Representations for Few-Shot Learning of Named Entities
["Omar U. Florez", "Erik Mueller"]
Humans excel in continuously learning with small data without forgetting how to solve old problems. However, neural networks require large datasets to compute latent representations across different tasks while minimizing a loss function. For example, a natural language understanding (NLU) system will often deal with ...
["Memory management", "neuroscience", "reinforcement learning", "learning with small data"]
ABSTRACTHumans excel in continuously learning with small data without forgetting how tosolve old problems. However, neural networks require large datasets to computelatent representations across different tasks while minimizing a loss function. Forexample, a natural language understanding (NLU) system will often deal w...
r1gvaG_X9B
Official Blind Review #1
1: Reject
The authors build on work regarding few-shot learning with memory augmented networks, specifically [Kaiser, et al., ICLR17] where the goal is to learn a memory address mapping such that generalization is achieved by finding the nearest neighbor memory address when predicting the label. For correct predictions, the memo...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning to Control Latent Representations for Few-Shot Learning of Named Entities ### Paper Abstract Humans excel in continuously learning with small data without forge...
gpsfGAOUs58
MIDL.io/2023/Short_Paper_Track
2023
Caption generation from histopathology whole-slide images using pre-trained transformers
["Bryan Cardenas Guevara", "Niccol\u00f2 Marini", "Stefano Marchesin", "Witali Aswolinskiy", "Robert-Jan Schlimbach", "Damian Podareanu", "Francesco Ciompi"]
The recent advent of foundation models and large language models has enabled scientists to leverage large-scale knowledge of pretrained (vision) transformers and efficiently tailor it to downstream tasks. This technology can potentially automate multiple aspects of cancer diagnosis in digital pathology, from whole-slid...
["Whole slide images", "histopathology", "multi-modal training", "caption generation", "transformers"]
Medical Imaging with Deep Learning – Under Review 2023 Short Paper – MIDL 2023 submissionCaption generation from histopathology whole-slide imagesusing pre-trained transformersBryan Cardenas Guevara1bryan.cardenasguevara@surf.nl1SURF, Amsterdam, The NetherlandsNiccol` o Marini2niccolo.marini@hevs.ch2University of Appli...
RYfXyzdG_7Y
great work
8: Top 50% of accepted papers, clear accept
This is an excelent application of LLM and ML in the context of histopathology WSI. The authors evaluate their method on a large dataset. The results are very promising and would generate very interesting discussions. Would be great to learn if this can me extended to medical imaging domains.
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 Caption generation from histopathology whole-slide images using pre-trained transformers ### Paper Abstract The recent advent of foundation models and large language mod...
HyxJ1xBYDH
ICLR.cc/2020/Conference
2020
Learning-Augmented Data Stream Algorithms
["Tanqiu Jiang", "Yi Li", "Honghao Lin", "Yisong Ruan", "David P. Woodruff"]
The data stream model is a fundamental model for processing massive data sets with limited memory and fast processing time. Recently Hsu et al. (2019) incorporated machine learning techniques into the data stream model in order to learn relevant patterns in the input data. Such techniques were encapsulated by training ...
["streaming algorithms", "heavy hitters", "F_p moment", "distinct elements", "cascaded norms"]
ABSTRACTThe data stream model is a fundamental model for processing massive data setswith limited memory and fast processing time. Recently Hsu et al. (2019) incor-porated machine learning techniques into the data stream model in order to learnrelevant patterns in the input data. Such techniques led to the training of ...
BJloMuTnFr
Official Blind Review #3
8: Accept
The paper talks about calculating various statistics over data streams. This is a very popular topic and is very relevant in big data analysis. A lot of work has been done in this general area and on the problems that are discussed in the paper. The new idea in the paper is better streaming algorithms under the assumpt...
<|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-Augmented Data Stream Algorithms ### Paper Abstract The data stream model is a fundamental model for processing massive data sets with limited memory and fast p...
3pcecaCEK-
AKBC.ws/2020/Conference
2020
Ranking vs. Classifying: Measuring Knowledge Base Completion Quality
["Marina Speranskaya", "Martin Schmitt", "Benjamin Roth"]
Knowledge base completion (KBC) methods aim at inferring missing facts from the information present in a knowledge base (KB). Such a method thus needs to estimate the likelihood of candidate facts and ultimately to distinguish between true facts and false ones to avoid compromising the KB with untrue information. In th...
["knowledge base completion", "knowledge graph embedding", "classification", "ranking"]
Automated Knowledge Base Construction (2020) Conference paperRanking vs. Classifying:Measuring Knowledge Base Completion QualityMarina Speranskaya speranskaya@cis.lmu.deMartin Schmitt martin@cis.lmu.deBenjamin Roth beroth@cis.lmu.deCenter for Information and Language Processing, LMU Munich, GermanyAbstractKnowledge bas...
x2rOzgZGDS4
The authors convincingly argue that ranking-based metrics (like MRR) for entity-relation completion are misleading for many downstream tasks.
7: Good paper, accept
The main contributions of the paper are a new train/test dataset that provides negative cases and queries to fulfill. Based on this dataset the authors show how to adapt a traditional IR methodology with F1, which is compared to current measurement techniques, especially MMR. Finally, the authors provide a variant of 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 Ranking vs. Classifying: Measuring Knowledge Base Completion Quality ### Paper Abstract Knowledge base completion (KBC) methods aim at inferring missing facts from the i...
r1eIiCNYwS
ICLR.cc/2020/Conference
2020
Transformer-XH: Multi-Evidence Reasoning with eXtra Hop Attention
["Chen Zhao", "Chenyan Xiong", "Corby Rosset", "Xia Song", "Paul Bennett", "Saurabh Tiwary"]
Transformers have achieved new heights modeling natural language as a sequence of text tokens. However, in many real world scenarios, textual data inherently exhibits structures beyond a linear sequence such as trees and graphs; many tasks require reasoning with evidence scattered across multiple pieces of texts. This ...
["Transformer-XH", "multi-hop QA", "fact verification", "extra hop attention", "structured modeling"]
ABSTRACTTransformers have achieved new heights modeling natural language as a sequenceof text tokens. However, in many real world scenarios, textual data inherently ex-hibits structures beyond a linear sequence such as trees and graphs; many tasksrequire reasoning with evidence scattered across multiple pieces of texts...
B1ghg2ACYr
Official Blind Review #1
6: Weak Accept
The paper is proposing an extension of the Transformer matching and diffusion mechanism to multi-document settings. To do so, the authors introduce a special representation for gathering document level information which is then used for propagation among documents latent representations. The extension seems quite simpl...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Transformer-XH: Multi-Evidence Reasoning with eXtra Hop Attention ### Paper Abstract Transformers have achieved new heights modeling natural language as a sequence of te...
HyWG0H5ge
ICLR.cc/2017/conference
2017
Neural Taylor Approximations: Convergence and Exploration in Rectifier Networks
["David Balduzzi", "Brian McWilliams", "Tony Butler-Yeoman"]
Modern convolutional networks, incorporating rectifiers and max-pooling, are neither smooth nor convex. Standard guarantees therefore do not apply. Nevertheless, methods from convex optimization such as gradient descent and Adam are widely used as building blocks for deep learning algorithms. This paper provides the fi...
["Deep learning", "Optimization", "Theory", "Supervised Learning"]
ABSTRACTModern convolutional networks, incorporating rectifiers and max-pooling, are nei-ther smooth nor convex. Standard guarantees therefore do not apply. Nevertheless,methods from convex optimization such as gradient descent and Adam are widelyused as building blocks for deep learning algorithms. This paper provides...
ByUX0ABNl
Good Novelty. But the main theorem of the paper is problematic.
3: Clear rejection
This paper develops a theoretical guarantee for the convergence of the training error. The result is quite general that covers the training of a wide range of neural network models. The key idea of the paper is approximate the training loss by its linear approximation. Since its linearity in the variables (thus convex)...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Neural Taylor Approximations: Convergence and Exploration in Rectifier Networks ### Paper Abstract Modern convolutional networks, incorporating rectifiers and max-poolin...
Syxwsp4KDB
ICLR.cc/2020/Conference
2020
TED: A Pretrained Unsupervised Summarization Model with Theme Modeling and Denoising
["Ziyi Yang", "Chenguang Zhu", "Michael Zeng", "Xuedong Huang", "Eric Darve"]
Text summarization aims to extract essential information from a piece of text and transform it into a concise version. Existing unsupervised abstractive summarization models use recurrent neural networks framework and ignore abundant unlabeled corpora resources. In order to address these issues, we propose TED, a trans...
["text summarization", "unsupervised learning", "natural language processing"]
ABSTRACTText summarization aims to extract essential information from a piece of text andtransform it into a concise version. Existing unsupervised abstractive summa-rization models use recurrent neural networks framework and ignore abundantunlabeled corpora resources. In order to address these issues, we propose TED, ...
rJxaA5iecS
Official Blind Review #1
6: Weak Accept
The authors propose to improve abstractive summarization models by using pretrained embeddings, theme modeling and denoising. They propose a very interesting idea: to leverage the lead bias in news article to build supervized summarization task from 21.4 M of articles. Details are given how to produce this supervized ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title TED: A Pretrained Unsupervised Summarization Model with Theme Modeling and Denoising ### Paper Abstract Text summarization aims to extract essential information from a p...
00I1UhSk55i
AAAI.org/2022/Workshop/ADAM
2022
Deep Reinforcement Learning for Engineering Design through Topology Optimization of Elementally Discretized Design Domains
["Nathan Brown", "Anthony Garland", "Georges Fadel", "Gang Li"]
Machine learning (ML) can extract patterns in design-relevant data to detect trends or make predictions that may not be inherently visible to a human designer. However, most ML-based engineering design tools rely on supervised learning, which requires the prefabrication of design domain data that may be challenging to ...
["Reinforcement Learning", "Machine Learning", "Topology Optimization", "Engineering Design"]
Deep Reinforcement Learning for Engineering Design through Topology Optimization of Elementally Discretized Design Domains Nathan Brown 1, Anthony Garland 2, Georges Fadel 1, Gang Li 1 Clemson University 1, Sandia National Laboratory 2 nkbrown@clemson.edu 1, agarlan@sandia.gov 2 Abstract Machine learning (ML) can extra...
ApVJeKgGn_j
Well written paper on RL for TO
9: Top 15% of accepted papers, strong accept
This paper presents a RL approach for solving TO problems in 2D. A sensible framing of the state space, action space is presented. By formulating the transition probability in terms of convolutional operators, the authors are able to smartly utilize a lower resolution FE solver to learn the transition probability. This...
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 Reinforcement Learning for Engineering Design through Topology Optimization of Elementally Discretized Design Domains ### Paper Abstract Machine learning (ML) can e...
QIRlze3I6hX
ICLR.cc/2021/Conference
2021
Learning Cross-Domain Correspondence for Control with Dynamics Cycle-Consistency
["Qiang Zhang", "Tete Xiao", "Alexei A Efros", "Lerrel Pinto", "Xiaolong Wang"]
At the heart of many robotics problems is the challenge of learning correspondences across domains. For instance, imitation learning requires obtaining correspondence between humans and robots; sim-to-real requires correspondence between physics simulators and real hardware; transfer learning requires correspondences b...
["self-supervised learning", "robotics"]
ABSTRACTAt the heart of many robotics problems is the challenge of learning correspon-dences across domains. For instance, imitation learning requires obtaining cor-respondence between humans and robots; sim-to-real requires correspondencebetween physics simulators and the real world; transfer learning requires corre-s...
Kn4TbMgKF86
A somewhat small yet solid contribution; paper clarity could improve
7: Good paper, accept
This paper presents a technique that leverages cycle-consistency to align data across domains given only samples of one-step trajectories (x_t, a_t, x_{t+1}) from each. The cycle-consistency is used over the space of actions across two domains: e.g., given data from X [in the form (x_t, a_t, x_{t+1})] and data from Y [...
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 Cross-Domain Correspondence for Control with Dynamics Cycle-Consistency ### Paper Abstract At the heart of many robotics problems is the challenge of learning c...
B1l3M64KwB
ICLR.cc/2020/Conference
2020
How many weights are enough : can tensor factorization learn efficient policies ?
["Pierre H. Richemond", "Arinbjorn Kolbeinsson", "Yike Guo"]
Deep reinforcement learning requires a heavy price in terms of sample efficiency and overparameterization in the neural networks used for function approximation. In this work, we employ tensor factorization in order to learn more compact representations for reinforcement learning policies. We show empirically that in t...
["reinforcement learning", "Q-learning", "tensor factorization", "low-rank approximation", "data efficiency", "second-order optimization", "scattering"]
ABSTRACTDeep reinforcement learning requires a heavy price in terms of sample efficiencyand overparameterization in the neural networks used for function approximation.In this work, we employ tensor factorization in order to learn more compact rep-resentations for reinforcement learning policies. We show empirically th...
ryxByGUaKS
Official Blind Review #2
3: Weak Reject
The paper aims at parsimonious reinforcement learning by employing 3 different techniques: using tensor regression layers (Kossaifi et al., 2017b), wavelet scattering (Mallat 2011) and using K-FAC (Kingma & Ba, 2014) as the optimization method. Learning models with fewer weights is important not only in reinforcement l...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title How many weights are enough : can tensor factorization learn efficient policies ? ### Paper Abstract Deep reinforcement learning requires a heavy price in terms of sampl...
p7OewL0RRIH
ICLR.cc/2021/Conference
2021
Sself: Robust Federated Learning against Stragglers and Adversaries
["Jungwuk Park", "Dong-Jun Han", "Minseok Choi", "Jaekyun Moon"]
While federated learning allows efficient model training with local data at edge devices, two major issues that need to be resolved are: slow devices known as stragglers and malicious attacks launched by adversaries. While the presence of both stragglers and adversaries raises serious concerns for the deployment of pra...
["Federated Learning"]
ABSTRACTWhile federated learning allows efficient model training with local data at edgedevices, two major issues that need to be resolved are: slow devices known asstragglers and malicious attacks launched by adversaries. While the presenceof both stragglers and adversaries raises serious concerns for the deployment o...
Cukvmjczi5
Official Blind Review #4
4: Ok but not good enough - rejection
This paper considers federated learning with straggling and adversarial devices. To tackle stragglers, the paper proposes semi-synchronous averaging wherein models with the same staleness are first averaged together, and then a weighted average of the results with different stateless is computed. To mitigate adversarie...
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 Sself: Robust Federated Learning against Stragglers and Adversaries ### Paper Abstract While federated learning allows efficient model training with local data at edge d...
YXHoPO33rk
logconference.io/LOG/2022/Conference
2022
Learning to Reconstruct Missing Data from Spatiotemporal Graphs with Sparse Observations
["Ivan Marisca", "Andrea Cini", "Cesare Alippi"]
Modeling multivariate time series as temporal signals over a (possibly dynamic) graph is an effective representational framework that allows for developing models for time series analysis. Spatiotemporal graphs are often highly sparse, with time series characterized by multiple, concurrent, and even long sequences of m...
["missing data", "time series imputation", "spatiotemporal graph neural networks"]
Learning to Reconstruct Missing Data from SpatiotemporalGraphs with Sparse ObservationsIvan Marisca∗The Swiss AI Lab IDSIAUniversità della Svizzera italianaivan.marisca@usi.chAndrea Cini∗The Swiss AI Lab IDSIAUniversità della Svizzera italianaandrea.cini@usi.chCesare AlippiThe Swiss AI Lab IDSIAUniversità della Sviz...
qAi5P466U5C
———————————————————— Summary of the work This short paper deals with the important problem of reconstructing multivariate time-series on graphs from so-termed sparse observations (meaning a large fraction of samples are missing). The proposed data imputation approach is based on a novel attention mechanism that takes...
<|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 Reconstruct Missing Data from Spatiotemporal Graphs with Sparse Observations ### Paper Abstract Modeling multivariate time series as temporal signals over a ...
B1QgVti6Z
ICLR.cc/2018/Conference
2018
Empirical Risk Landscape Analysis for Understanding Deep Neural Networks
["Pan Zhou", "Jiashi Feng"]
This work aims to provide comprehensive landscape analysis of empirical risk in deep neural networks (DNNs), including the convergence behavior of its gradient, its stationary points and the empirical risk itself to their corresponding population counterparts, which reveals how various network parameters determine the...
["Deep Learning Analysis", "Deep Learning Theory", "Empirical Risk", "Landscape Analysis", "Nonconvex Optimization"]
ABSTRACTThis work aims to provide comprehensive landscape analysis of empirical riskin deep neural networks (DNNs), including the convergence behavior of its gra-dient, its stationary points and the empirical risk itself to their correspondingpopulation counterparts, which reveals how various network parameters deter-m...
H1Wo7pKgM
This theoretical paper lacks rigor and references.
3: Clear rejection
This paper studies empirical risk in deep neural networks. Results are provided in Section 4 for linear networks and in Section 5 for nonlinear networks. Results for deep linear neural networks are puzzling. Whatever the number of layers, a deep linear NN is simply a matrix multiplication and minimizing the MSE is simp...
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 Empirical Risk Landscape Analysis for Understanding Deep Neural Networks ### Paper Abstract This work aims to provide comprehensive landscape analysis of empirical risk...
5bd-6wyFWg5
graphicsinterface.org/Graphics_Interface/2023/Conference
2023
Contextual Ambient Occlusion
["Andrey Titov", "Marta Kersten-Oertel", "Simon Drouin"]
In this paper, we present a new volumetric ambient occlusion algorithm called Contextual Ambient Occlusion (CAO) that supports real-time clipping. The algorithm produces ambient occlusion images of exactly the same quality as Local Ambient Occlusion (LAO) while enabling real-time modification to the shape used to clip ...
["Ambient occlusion", "volume rendering", "global illumination", "medical imaging", "shadowing", "transfer function", "shading"]
ABSTRACTIn this paper, we present a new volumetric ambient occlusion algo-rithm called Contextual Ambient Occlusion (CAO) that supportsreal-time clipping. The algorithm produces ambient occlusion im-ages of exactly the same quality as Local Ambient Occlusion (LAO)while enabling real-time modification to the shape used ...
yHm1m1r7uRK
A simple and effective approach for accelerating ambient occlusion computation
7: Good paper, accept
This paper presents a novel technique to compute local ambient occlusion that supports real-time geometry clipping. The main idea is to leverage the fact that LAO value will only change if nearby geometry gets clipped. Thus, one only requires local updates to the LAO value, instead of a global recomputation. This obser...
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 Contextual Ambient Occlusion ### Paper Abstract In this paper, we present a new volumetric ambient occlusion algorithm called Contextual Ambient Occlusion (CAO) that sup...
Dh29CAlnMW
ICLR.cc/2021/Conference
2021
Parsed Categoric Encodings with Automunge
["Nicholas Teague"]
The Automunge open source python library platform for tabular data pre-processing automates feature engineering data transformations of numerical encoding and missing data infill to received tidy data on bases fit to properties of columns in a designated train set for consistent and efficient application to subsequent ...
["tabular", "feature engineering"]
ABSTRACTThe Automunge open source python library platform for tabular data pre-processing automates feature engineering data transformations of numerical en-coding and missing data infill to received tidy data on bases fit to properties ofcolumns in a designated train set for consistent and efficient application to sub...
gDK6rfa_pKj
The authors present a python package to process data in the form of strings.
6: Marginally above acceptance threshold
This package aims at automating some repetitive tasks that data analysts (especially within the NLP domain) deal with. I confirm that tasks like feature selection and string encoding are beneficial for applied researchers with textual data. I can see a wide interest in the community. On the negative side, the paper lac...
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 Parsed Categoric Encodings with Automunge ### Paper Abstract The Automunge open source python library platform for tabular data pre-processing automates feature engineer...
HyAbMKwxe
ICLR.cc/2017/conference
2017
Tighter bounds lead to improved classifiers
["Nicolas Le Roux"]
The standard approach to supervised classification involves the minimization of a log-loss as an upper bound to the classification error. While this is a tight bound early on in the optimization, it overemphasizes the influence of incorrectly classified examples far from the decision boundary. Updating the upper bound ...
["upper bound", "optimization", "classifier", "tighter bounds", "classifiers tighter bounds", "classifiers", "standard", "classification", "minimization", "classification error"]
ABSTRACTThe standard approach to supervised classification involves the minimization of alog-loss as an upper bound to the classification error. While this is a tight boundearly on in the optimization, it overemphasizes the influence of incorrectly clas-sified examples far from the decision boundary. Updating the upper...
SkmefGIEg
review
8: Top 50% of accepted papers, clear accept
The paper analyses the misclassification error of discriminators and highlights the fact that while uniform probability prior of the classes makes sense early in the optimization, the distribution deviates from this prior significantly as the parameters move away from the initial values. Consequently, the optimized up...
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 Tighter bounds lead to improved classifiers ### Paper Abstract The standard approach to supervised classification involves the minimization of a log-loss as an upper bou...
HklvMJSYPB
ICLR.cc/2020/Conference
2020
Adaptive Adversarial Imitation Learning
["Yiren Lu", "Jonathan Tompson", "Sergey Levine"]
We present the ADaptive Adversarial Imitation Learning (ADAIL) algorithm for learning adaptive policies that can be transferred between environments of varying dynamics, by imitating a small number of demonstrations collected from a single source domain. This problem is important in robotic learning because in real wor...
["Imitation Learning", "Reinforcement Learning"]
ABSTRACTWe present the ADaptive Adversarial Imitation Learning (ADAIL) algorithm forlearning adaptive policies that can be transferred between environments of vary-ing dynamics, by imitating a small number of demonstrations collected from asingle source domain. This is an important problem in robotic learning becausein...
BkgSu9lhYS
Official Blind Review #2
6: Weak Accept
The paper describes an approach that combines domain adversarial neural network with generative adversarial imitation learning. In the setup, each environment is dependent on some latent context variable (e.g. physics parameters) through which latent variable dependent policy and latent variable independent discriminat...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Adaptive Adversarial Imitation Learning ### Paper Abstract We present the ADaptive Adversarial Imitation Learning (ADAIL) algorithm for learning adaptive policies that c...
SylDPJrYvS
ICLR.cc/2020/Conference
2020
EnsembleNet: A novel architecture for Incremental Learning
["Suri Bhasker Sri Harsha", "Y Kalidas"]
Deep neural networks are used in many state-of-the-art systems for machine perception. Once a network is trained to do a specific task, it cannot be easily trained to do new tasks as it leads to catastrophic forgetting of the previously learned tasks. We propose here a novel architecture called EnsembleNet that accomm...
["novel architecture", "model", "literature", "ensemblenet", "network", "incremental learning ensemblenet", "many", "systems", "machine perception"]
ABSTRACTDeep neural networks are used in many state-of-the-art systems for machine per-ception. Once a network is trained to do a specific task, it cannot be easily trainedto do new tasks as it leads to catastrophic forgetting of the previously learnedtasks.We propose here a novel architecture called EnsembleNet that a...
B1xBNKe0Kr
Official Blind Review #2
1: Reject
The authors propose an ensemble model with deep networks. The authors suggest its relevance for incremental learning. The proposed model is evaluated appropriately. There are many issues with the work. The presentation and description needs a lot of improvement. What exactly do the authors mean by rehearsal? It is co...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title EnsembleNet: A novel architecture for Incremental Learning ### Paper Abstract Deep neural networks are used in many state-of-the-art systems for machine perception. Onc...
S1xBioR5KX
ICLR.cc/2019/Conference
2019
Parameter efficient training of deep convolutional neural networks by dynamic sparse reparameterization
["Hesham Mostafa", "Xin Wang"]
Modern deep neural networks are highly overparameterized, and often of huge sizes. A number of post-training model compression techniques, such as distillation, pruning and quantization, can reduce the size of network parameters by a substantial fraction with little loss in performance. However, training a small netw...
["sparse", "reparameterization", "overparameterization", "convolutional neural network", "training", "compression", "pruning"]
ABSTRACTModern deep neural networks are highly overparameterized, and often of hugesizes. A number of post-training model compression techniques, such as distil-lation, pruning and quantization, can reduce the size of network parameters bya substantial fraction with little loss in performance. However, training a small...
Hkl9XKWc37
Efficient dynamic reparameterization but the claims need to be revisited
6: Marginally above acceptance threshold
The paper provides a dynamic sparse reparameterization method allowing small networks to be trained at a comparable accuracy as pruned network with (initially) large parameter spaces. Improper initialization along with a fewer number of parameters requires a large parameter model, to begin with (Frankle and Carbin, 201...
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 Parameter efficient training of deep convolutional neural networks by dynamic sparse reparameterization ### Paper Abstract Modern deep neural networks are highly overpar...
jYVY_piet7m
ICLR.cc/2021/Conference
2021
Hybrid-Regressive Neural Machine Translation
["Qiang Wang", "Heng Yu", "Shaohui Kuang", "Weihua Luo"]
Although the non-autoregressive translation model based on iterative refinement has achieved comparable performance to the autoregressive counterparts with faster decoding, we empirically found that such aggressive iterations make the acceleration rely heavily on small batch size (e.g., 1) and computing device (e.g., G...
["neural machine translation", "non-autoregressive translation"]
ABSTRACTAlthough the non-autoregressive translation model based on iterative refinementhas achieved comparable performance to the autoregressive counterparts withfaster decoding, we empirically found that such aggressive iterations make theacceleration rely heavily on small batch size (e.g., 1) and computing device (e....
PtNkyC-ql4y
the studied problem is novel, but the comparison is not fully convincing.
5: Marginally below acceptance threshold
Non-autoregressive decoder (NAT) greatly improves translation efficiency, but often relies on iterative refinement (IR) to retain the translation performance. Unfortunately, this strategy makes the decoding efficiency of IR-based NAT much more sensitive to batch size and computing device: with larger batch size and CPU...
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 Hybrid-Regressive Neural Machine Translation ### Paper Abstract Although the non-autoregressive translation model based on iterative refinement has achieved comparable p...
qYLp6nNU7m-
NeurIPS.cc/2022/Workshop/SVRHM
2022
The emergence of visual simulation in task-optimized recurrent neural networks
["Alekh Karkada Ashok", "Lakshmi Narasimhan Govindarajan", "Drew Linsley", "David Sheinberg", "Thomas Serre"]
Primates display remarkable prowess in making rapid visual inferences even when sensory inputs are impoverished. One hypothesis about how they accomplish this is through a process called visual simulation, in which they imagine future states of their environment using a constructed mental model. Though a growing body o...
["visual simulation", "deep learning", "recurrent neural networks"]
The emergence of visual simulation in task-optimizedrecurrent neural networksAnonymous Author(s)AffiliationAddressemailAbstractPrimates display remarkable prowess in making rapid visual inferences even when 1sensory inputs are impoverished. One hypothesis about how they accomplish 2this is through a process called visu...
ZaVoEYftPQ_
Great study! (results have to be better contextualized per existing literature)
8: Top 50% of accepted papers, clear accept
In this paper, the authors use the Planko task to show that simulation-like representations arise in deep neural networks trained to solve the task only when the task variability is high. This work shows that visual simulation may be a task-specific optimal strategy and provides a computational framework to understand ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title The emergence of visual simulation in task-optimized recurrent neural networks ### Paper Abstract Primates display remarkable prowess in making rapid visual inferences e...
H1x-x309tm
ICLR.cc/2019/Conference
2019
On the Convergence of A Class of Adam-Type Algorithms for Non-Convex Optimization
["Xiangyi Chen", "Sijia Liu", "Ruoyu Sun", "Mingyi Hong"]
This paper studies a class of adaptive gradient based momentum algorithms that update the search directions and learning rates simultaneously using past gradients. This class, which we refer to as the ''``Adam-type'', includes the popular algorithms such as Adam, AMSGrad, AdaGrad. Despite their popularity in training ...
["nonconvex optimization", "Adam", "convergence analysis"]
ABSTRACTThis paper studies a class of adaptive gradient based momentum algorithms thatupdate the search directions and learning rates simultaneously using past gradients.This class, which we refer to as the “Adam-type”, includes the popular algorithmssuch as Adam (Kingma & Ba, 2014) , AMSGrad (Reddi et al., 2018) , Ada...
SkgumbaYn7
This paper investigates the convergence condition of Adam-type optimizers in the unconstrained non-convex optimization problems.
6: Marginally above acceptance threshold
The main theory points out two scenarios causing Adam-type optimizers to diverge, which extends Reddi et al's results. The theorem in this paper applies to all Adam-type algorithms, which combine momentum with adaptive learning rates and thus are more general as compared to the recent papers, such as Zhou et al's. Th...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title On the Convergence of A Class of Adam-Type Algorithms for Non-Convex Optimization ### Paper Abstract This paper studies a class of adaptive gradient based momentum algo...
hkMoYYEkBoI
ICLR.cc/2021/Conference
2021
Computational Separation Between Convolutional and Fully-Connected Networks
["eran malach", "Shai Shalev-Shwartz"]
Convolutional neural networks (CNN) exhibit unmatched performance in a multitude of computer vision tasks. However, the advantage of using convolutional networks over fully-connected networks is not understood from a theoretical perspective. In this work, we show how convolutional networks can leverage locality in the ...
["Neural Networks", "Deep Learning", "Convolutional Networks", "Fully-Connected Networks", "Gradient Descent"]
ABSTRACTConvolutional neural networks (CNN) exhibit unmatched performance in a mul-titude of computer vision tasks. However, the advantage of using convolutionalnetworks over fully-connected networks is not understood from a theoretical per-spective. In this work, we show how convolutional networks can leverage localit...
S6LaXnha1rN
Review for separation CNN and FCNN
8: Top 50% of accepted papers, clear accept
It is well-known that neural networks (NN) perform very well in various areas and in particular if one looks at computer vision convolutional neural networks perform very well. Although convolutional neural networks (CNN) are limited in their architecture (since they only allow nearest-neighbour connections) compared 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 Computational Separation Between Convolutional and Fully-Connected Networks ### Paper Abstract Convolutional neural networks (CNN) exhibit unmatched performance in a mul...
rJvjL71DM
ICLR.cc/2018/Workshop
2018
Visual Analogies between Atari Games for Studying Transfer Learning in RL
["Doron Sobol", "Lior Wolf", "Yaniv Taigman"]
In this work, we ask the following question: can visual analogies, learned in an unsupervised way, be used in order to transfer knowledge between pairs of games and even play one game using an agent trained for another game. We attempt to answer this research question by creating visual analogies between a pair of game...
["games", "source game", "visual analogies", "atari games", "transfer learning", "pairs", "game", "target game", "mappings", "rl visual analogies"]
ABSTRACTIn this work, we ask the following question: can visual analogies, learned in anunsupervised way, be used in order to transfer knowledge between pairs of gamesand even play one game using an agent trained for another game. We attemptto answer this research question by creating visual analogies between a pair of...
rybT_w4FG
review
5: Marginally below acceptance threshold
The authors propose a brief study on the effectiveness of certain kinds of transfer-learning in RL tasks, employing visual-analogies as the basis for transferring information. The paper is reasonably straightforward to follow -- although there are a couple of places where the exposition is short of detail and difficul...
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 Visual Analogies between Atari Games for Studying Transfer Learning in RL ### Paper Abstract In this work, we ask the following question: can visual analogies, learned i...
w2mYg3d0eot
ICLR.cc/2021/Conference
2021
Fast convergence of stochastic subgradient method under interpolation
["Huang Fang", "Zhenan Fan", "Michael Friedlander"]
This paper studies the behaviour of the stochastic subgradient descent (SSGD) method applied to over-parameterized nonsmooth optimization problems that satisfy an interpolation condition. By leveraging the composite structure of the empirical risk minimization problems, we prove that SSGD converges, respectively, with ...
["Optimization", "stochastic subgradient method", "interpolation", "convergence analysis"]
ABSTRACTThis paper studies the behaviour of the stochastic subgradient descent (SSGD)method applied to over-parameterized nonsmooth optimization problems thatsatisfy an interpolation condition. By leveraging the composite structure of theempirical risk minimization problems, we prove that SSGD converges, respectively,w...
ahiJx6axFvo
A borderline paper, closer to acceptance
7: Good paper, accept
Summary: The paper considers stochastic finite-sum minimization with a special structure of composition of a smooth univariate loss with a non-smooth Lipschitz function. A generalized gradient bound is proved for this setting as well semi-smoothness. Using the former bound, sublinear 1/k rate is proved for stochastic s...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Fast convergence of stochastic subgradient method under interpolation ### Paper Abstract This paper studies the behaviour of the stochastic subgradient descent (SSGD) me...
1qdNTwXpgE
graphicsinterface.org/Graphics_Interface/2020/Conference
2020
WAAT: a Workstation AR Authoring Tool for Industry 4.0
["Anonymous"]
The use of AR in an industrial context could help for the training of new operators. To be able to use an AR guidance system, we need a tool to quickly create a 3D representation of the assembly line and of its AR annotations. This tool should be very easy to use by an operator who is not an AR or VR specialist: typica...
["VR", "AR", "XR", "AR Authoring", "3D Authoring", "Industry 4.0"]
ABSTRACTThe use of AR in an industrial context could help for the trainingof new operators. To be able to use an AR guidance system, weneed a tool to quickly create a 3D representation of the assemblyline and of its AR annotations. This tool should be very easy touse by an operator who is not an AR or VR specialist: ty...
sHvUwcHl4
Lacking clear research contribution and framing.
2: Strong rejection
This paper presents a system that allows users to place and position 3D objects in augmented reality. The system is motivated by a need to train factory workers at an increased rate, and demonstrated through a system which helps users align a graphics card into a computer. This paper, while it addresses an important a...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title WAAT: a Workstation AR Authoring Tool for Industry 4.0 ### Paper Abstract The use of AR in an industrial context could help for the training of new operators. To be able...
HylzTiC5Km
ICLR.cc/2019/Conference
2019
GENERATING HIGH FIDELITY IMAGES WITH SUBSCALE PIXEL NETWORKS AND MULTIDIMENSIONAL UPSCALING
["Jacob Menick", "Nal Kalchbrenner"]
The unconditional generation of high fidelity images is a longstanding benchmark for testing the performance of image decoders. Autoregressive image models have been able to generate small images unconditionally, but the extension of these methods to large images where fidelity can be more readily assessed has remained...
["high fidelity images", "size", "subscale pixel networks", "multidimensional upscaling", "unconditional generation", "able", "spn", "image", "spns", "multidimensional"]
ABSTRACTThe unconditional generation of high fidelity images is a longstanding benchmarkfor testing the performance of image decoders. Autoregressive image models havebeen able to generate small images unconditionally, but the extension of thesemethods to large images where fidelity can be more readily assessed has rem...
HkxN569T2X
Sound incremental advance
7: Good paper, accept
Authors propose a decoder arquitecture model named Subscale Pixel Network. It is meant to generate overall images as image slice sequences with memory and computation economy by using a Multidimensional Upscaling method. The paper is fairly well written and structured, and it seems technically sound. Experiments are co...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title GENERATING HIGH FIDELITY IMAGES WITH SUBSCALE PIXEL NETWORKS AND MULTIDIMENSIONAL UPSCALING ### Paper Abstract The unconditional generation of high fidelity images is a ...
bFb3V8ALx4W
MIDL.io/2023/Short_Paper_Track
2023
Visualizing chest X-ray dataset biases using GANs
["Hao Liang", "Kangqi Ni", "Guha Balakrishnan"]
Recent work demonstrates that images from various chest X-ray datasets contain visual features that are strongly correlated with protected demographic attributes like race and gender. This finding raises issues of fairness, since some of these factors may be used by downstream algorithms for clinical predictions. In th...
["Chest X-rays", "fairness", "bias", "explainability", "generative adversarial networks (GANs)"]
Medical Imaging with Deep LearningVisualizing chest X-ray dataset biases using GANsHao Liang hl106@rice.eduKangqi Ni kn22@rice.eduGuha Balakrishnan guha@rice.eduDepartment of Electrical and Computer Engineering, Rice University, USAAbstractRecent work demonstrates that images from various chest X-ray datasets contain v...
Fm-LXbCsd6I
Visualizing chest X-ray dataset biases using GANs
5: Marginally below acceptance threshold
This paper proposed a framework, using generative adversarial networks (GANs), to visualize what features are most different between X-rays belonging to two demographic subgroups. It's interesting to see such studies, which explore the potential biases in deep learning models. However, there are other explanable AI met...
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 Visualizing chest X-ray dataset biases using GANs ### Paper Abstract Recent work demonstrates that images from various chest X-ray datasets contain visual features that ...
rJEjjoR9K7
ICLR.cc/2019/Conference
2019
Learning Robust Representations by Projecting Superficial Statistics Out
["Haohan Wang", "Zexue He", "Zachary C. Lipton", "Eric P. Xing"]
Despite impressive performance as evaluated on i.i.d. holdout data, deep neural networks depend heavily on superficial statistics of the training data and are liable to break under distribution shift. For example, subtle changes to the background or texture of an image can break a seemingly powerful classifier. Buildin...
["domain generalization", "robustness"]
ABSTRACTDespite impressive performance as evaluated on i.i.d. holdout data, deep neuralnetworks depend heavily on superficial statistics of the training data and are liableto break under distribution shift. For example, subtle changes to the background ortexture of an image can break a seemingly powerful classifier. Bu...
HJee7cfwh7
The proposal is well structured and written. The quality of the paper is excellent in terms of novelty and originality.
9: Top 15% of accepted papers, strong accept
The paper proposed a novel differentiable neural GLCM network which captures the high reference textural information and discard the lower-frequency semantic information so as to solve the domain generalisation challenge. The author also proposed an approach “HEX” to discard the superficial representations. Two synthet...
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 Robust Representations by Projecting Superficial Statistics Out ### Paper Abstract Despite impressive performance as evaluated on i.i.d. holdout data, deep neur...
Hkla1eHFvS
ICLR.cc/2020/Conference
2020
Efficient Exploration via State Marginal Matching
["Lisa Lee", "Benjain Eysenbach", "Emilio Parisotto", "Erix Xing", "Sergey Levine", "Ruslan Salakhutdinov"]
Reinforcement learning agents need to explore their unknown environments to solve the tasks given to them. The Bayes optimal solution to exploration is intractable for complex environments, and while several exploration methods have been proposed as approximations, it remains unclear what underlying objective is being ...
["reinforcement learning", "exploration", "distribution matching", "robotics"]
ABSTRACTReinforcement learning agents need to explore their unknown environments tosolve the tasks given to them. The Bayes optimal solution to exploration is in-tractable for complex environments, and while several exploration methods havebeen proposed as approximations, it remains unclear what underlying objectiveis ...
SJgrEg0sKr
Official Blind Review #2
3: Weak Reject
Update: I thank the authors for their response and I think the added baselines and more in depth discussion of prior work have improved the paper. However, given the limited novelty of the technical contribution, I believe the experimental section should be further extended (i.e. add a wider variety of domains and make...
<|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 via State Marginal Matching ### Paper Abstract Reinforcement learning agents need to explore their unknown environments to solve the tasks given to...
awnQ2qTLSwn
ICLR.cc/2021/Conference
2021
Learning to Share in Multi-Agent Reinforcement Learning
["Yuxuan Yi", "Ge Li", "Yaowei Wang", "Zongqing Lu"]
In this paper, we study the problem of networked multi-agent reinforcement learning (MARL), where a number of agents are deployed as a partially connected network. Networked MARL requires all agents make decision in a decentralized manner to optimize a global objective with restricted communication between neighbors ov...
["agents", "marl", "global objective", "neighbors", "share", "reinforcement learning", "network", "ltos", "share reward", "policy"]
ABSTRACTIn this paper, we study the problem of networked multi-agent reinforcement learn-ing (MARL), where a number of agents are deployed as a partially connected net-work. Networked MARL requires all agents make decision in a decentralizedmanner to optimize a global objective with restricted communication betweenneig...
b9O_b_yBHBo
Needs some improvement
4: Ok but not good enough - rejection
The paper addresses multi-agent RL problems by presenting a decentralized approach where the agents learn to share their reward with their neighbors. In this method, a high-level policy determines a weight vector for weighting the reward of neighboring agents, and then each agent learns their own independent policy. Th...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Learning to Share in Multi-Agent Reinforcement Learning ### Paper Abstract In this paper, we study the problem of networked multi-agent reinforcement learning (MARL), wh...
M88oFvqp_9
ICLR.cc/2021/Conference
2021
Generating Furry Cars: Disentangling Object Shape and Appearance across Multiple Domains
["Utkarsh Ojha", "Krishna Kumar Singh", "Yong Jae Lee"]
We consider the novel task of learning disentangled representations of object shape and appearance across multiple domains (e.g., dogs and cars). The goal is to learn a generative model that learns an intermediate distribution, which borrows a subset of properties from each domain, enabling the generation of images th...
["multi-domain disentanglement", "generative adversarial networks", "appearance transfer"]
ABSTRACTWe consider the novel task of learning disentangled representations of objectshape and appearance across multiple domains (e.g., dogs and cars). The goalis to learn a generative model that learns an intermediate distribution, which bor-rows a subset of properties from each domain, enabling the generation of ima...
Fdm0c6C7kjY
Interesting direction, limited contribution
5: Marginally below acceptance threshold
The submission describes a method to disentangle shape and appearance of images across two domains such that new images can be generated that have appearance and shape from either of these domains while still being visually convincing. Starting from an established method (FineGAN) to disentangle shape, appearance, back...
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 Generating Furry Cars: Disentangling Object Shape and Appearance across Multiple Domains ### Paper Abstract We consider the novel task of learning disentangled represent...
B1em9h4KDS
ICLR.cc/2020/Conference
2020
Generative Imputation and Stochastic Prediction
["Mohammad Kachuee", "Kimmo K\u00e4rkk\u00e4inen", "Orpaz Goldstein", "Sajad Darabi", "Majid Sarrafzadeh"]
In many machine learning applications, we are faced with incomplete datasets. In the literature, missing data imputation techniques have been mostly concerned with filling missing values. However, the existence of missing values is synonymous with uncertainties not only over the distribution of missing values but also ...
["values", "imputations", "distribution", "simple", "generative imputation", "stochastic prediction", "many machine", "applications", "incomplete datasets"]
ABSTRACTIn many machine learning applications, we are faced with incomplete datasets. Inthe literature, missing data imputation techniques have been mostly concerned withfilling missing values. However, the existence of missing values is synonymouswith uncertainties not only over the distribution of missing values but ...
HyeAuZ3DtS
Official Blind Review #2
6: Weak Accept
The uncertainty of having a missing value is investigated on the prediction by not assigning a single imputed value but N different values generated via an imputer network (based on GAIN). Unlike old-fashion multiple imputation techniques, one predictor is trained on different samples and it induces the uncertainty. Th...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Generative Imputation and Stochastic Prediction ### Paper Abstract In many machine learning applications, we are faced with incomplete datasets. In the literature, missi...
9_J4DrgC_db
ICLR.cc/2021/Conference
2021
Deep Coherent Exploration For Continuous Control
["Yijie Zhang", "Herke van Hoof"]
In policy search methods for reinforcement learning (RL), exploration is often performed by injecting noise either in action space at each step independently or in parameter space over each full trajectory. In prior work, it has been shown that with linear policies, a more balanced trade-off between these two explorati...
["reinforcement learning", "exploration", "latent variable models"]
ABSTRACTIn policy search methods for reinforcement learning (RL), exploration is oftenperformed by injecting noise either in action space at each step independently orin parameter space over each full trajectory. In prior work, it has been shownthat with linear policies, a more balanced trade-off between these two expl...
nscZAmpeQGQ
Complex approach with little improvement
4: Ok but not good enough - rejection
Summary: This paper focuses on undirected exploration strategies in reinforcement learning. Following the prior work, this paper proposes an exploration method unifying the step-based and trajectory-based exploration. The authors propose to perturb only the last(linear) layer of the policy for exploration, instead of p...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Deep Coherent Exploration For Continuous Control ### Paper Abstract In policy search methods for reinforcement learning (RL), exploration is often performed by injecting...
BJgkbyHKDS
ICLR.cc/2020/Conference
2020
Invertible generative models for inverse problems: mitigating representation error and dataset bias
["Muhammad Asim", "Ali Ahmed", "Paul Hand"]
Trained generative models have shown remarkable performance as priors for inverse problems in imaging. For example, Generative Adversarial Network priors permit recovery of test images from 5-10x fewer measurements than sparsity priors. Unfortunately, these models may be unable to represent any particular image becau...
["Invertible generative models", "inverse problems", "generative prior", "Glow", "compressed sensing", "denoising", "inpainting."]
ABSTRACTTrained generative models have shown remarkable performance as priors forinverse problems in imaging. For example, Generative Adversarial Network priorspermit recovery of test images from 5-10x fewer measurements than sparsity priors.Unfortunately, these models may be unable to represent any particular imagebec...
Bye3icUIqB
Official Blind Review #2
3: Weak Reject
Authors extend the invertible generative model of Kingma et. al. to image inverse problems. Specifically, they use Generator trained within the Glow framework as an image prior for de-noising, inpainting and compressed sensing tasks. During training, a heuristic adjustment to the objective is made allowing optimization...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Invertible generative models for inverse problems: mitigating representation error and dataset bias ### Paper Abstract Trained generative models have shown remarkable p...
H1xSNiRcF7
ICLR.cc/2019/Conference
2019
Smoothing the Geometry of Probabilistic Box Embeddings
["Xiang Li", "Luke Vilnis", "Dongxu Zhang", "Michael Boratko", "Andrew McCallum"]
There is growing interest in geometrically-inspired embeddings for learning hierarchies, partial orders, and lattice structures, with natural applications to transitive relational data such as entailment graphs. Recent work has extended these ideas beyond deterministic hierarchies to probabilistically calibrated models...
["embeddings", "order embeddings", "knowledge graph embedding", "relational learning"]
ABSTRACTThere is growing interest in geometrically-inspired embeddings for learning hier-archies, partial orders, and lattice structures, with natural applications to transitiverelational data such as entailment graphs. Recent work has extended these ideasbeyond deterministic hierarchies to probabilistically calibrated...
H1xPEJOVsm
review
7: Good paper, accept
Post-rebuttal revision: All my concerns were adressed by the authors. This is a great paper and should be accepted. ------ The paper presents smoothing probabilistic box embeddings with softplus functions, which make the optimization landscape continuous, while also presenting the theoretical background of the propos...
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 Smoothing the Geometry of Probabilistic Box Embeddings ### Paper Abstract There is growing interest in geometrically-inspired embeddings for learning hierarchies, partia...
4SZ9Ft--pDl
ICLR.cc/2021/Conference
2021
Prior-guided Bayesian Optimization
["Artur Souza", "Luigi Nardi", "Leonardo Oliveira", "Kunle Olukotun", "Marius Lindauer", "Frank Hutter"]
While Bayesian Optimization (BO) is a very popular method for optimizing expensive black-box functions, it fails to leverage the experience of domain experts. This causes BO to waste function evaluations on bad design choices (e.g., machine learning hyperparameters) that the expert already knows to work poorly. To add...
["Bayesian Optimization", "Automated Machine Learning"]
ABSTRACTWhile Bayesian Optimization (BO) is a very popular method for optimizing ex-pensive black-box functions, it fails to leverage the experience of domain experts.This causes BO to waste function evaluations on bad design choices (e.g., machinelearning hyperparameters) that the expert already knows to work poorly. ...
p3u6RCPjNMn
Strong idea with extensive evaluation and important applications
8: Top 50% of accepted papers, clear accept
The paper presents a novel method to incorporate experts' knowledge into BO. This is done through introducing Prior-guided Bayesian Optimization (PrBO). Different experiments where conducted to compare PrBO vs different baselines and to show the effect of the user provided priors in the cases where it is well-specifie...
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 Prior-guided Bayesian Optimization ### Paper Abstract While Bayesian Optimization (BO) is a very popular method for optimizing expensive black-box functions, it fails to...
TPFhIknddKX
ICLR.cc/2021/Conference
2021
Re-examining Routing Networks for Multi-task Learning
["Limeng Cui", "Aaron Jaech"]
We re-examine Routing Networks, an approach to multi-task learning that uses reinforcement learning to decide parameter sharing with the goal of maximizing knowledge transfer between related tasks while avoiding task interference. These benefits come with the cost of solving a more difficult optimization problem. We ar...
["Routing Networks", "Multi-task Learning", "Reinforcement Learning"]
ABSTRACTWe re-examine Routing Networks, an approach to multi-task learning that usesreinforcement learning to decide parameter sharing with the goal of maximizingknowledge transfer between related tasks while avoiding task interference. Thesebenefits come with the cost of solving a more difficult optimization problem. ...
jPMh-ai_7QN
the paper is difficult to read and the novelty is not clear
3: Clear rejection
This paper studies the effectiveness of routing networks for multi-task learning. The authors argue that if the key assumptions are not satisfied, a fixed random selection routing strategy is the best. In general, the paper is not easy to follow. All the details are presented in a rough description by text. The novelty...
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 Re-examining Routing Networks for Multi-task Learning ### Paper Abstract We re-examine Routing Networks, an approach to multi-task learning that uses reinforcement learn...
HJZiRkZC-
ICLR.cc/2018/Conference
2018
Byte-Level Recursive Convolutional Auto-Encoder for Text
["Xiang Zhang", "Yann LeCun"]
This article proposes to auto-encode text at byte-level using convolutional networks with a recursive architecture. The motivation is to explore whether it is possible to have scalable and homogeneous text generation at byte-level in a non-sequential fashion through the simple task of auto-encoding. We show that non-se...
["text", "recursive convolutional", "possible", "representation", "results", "model", "article", "convolutional networks", "recursive architecture", "motivation"]
ABSTRACTThis article proposes to auto-encode text at byte-level using convolutional net-works with a recursive architecture. The motivation is to explore whether it ispossible to have scalable and homogeneous text generation at byte-level in a non-sequential fashion through the simple task of auto-encoding. We show tha...
BJuY5u9gf
Minor step forward, but good investigation & analysis
7: Good paper, accept
This paper presents a convolutional auto-encoder architecture for text encoding and generation. It works on the character level and contains a recursive structure which scales with the length of the input text. Building on the recent state-of-the-art in terms of architectural components, the paper shows the feasibilit...
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 Byte-Level Recursive Convolutional Auto-Encoder for Text ### Paper Abstract This article proposes to auto-encode text at byte-level using convolutional networks with a r...
rklEj2EFvB
ICLR.cc/2020/Conference
2020
Estimating Gradients for Discrete Random Variables by Sampling without Replacement
["Wouter Kool", "Herke van Hoof", "Max Welling"]
We derive an unbiased estimator for expectations over discrete random variables based on sampling without replacement, which reduces variance as it avoids duplicate samples. We show that our estimator can be derived as the Rao-Blackwellization of three different estimators. Combining our estimator with REINFORCE, we ob...
["gradient", "estimator", "discrete", "categorical", "sampling", "without replacement", "reinforce", "baseline", "variance", "gumbel", "vae", "structured prediction"]
ABSTRACTWe derive an unbiased estimator for expectations over discrete random variablesbased on sampling without replacement , which reduces variance as it avoidsduplicate samples. We show that our estimator can be derived as the Rao-Blackwellization of three different estimators. Combining our estimator with RE-INFORC...
B1l84u7oFS
Official Blind Review #3
8: Accept
Edited after rebuttal: I'm satisfied with the clarifications provided by the authors about where they see this work being applicable. I think they have correctly identified a setting that is of interest to a broad audience and where their approach is more attractive than the alternatives. I now recommend accepting this...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Estimating Gradients for Discrete Random Variables by Sampling without Replacement ### Paper Abstract We derive an unbiased estimator for expectations over discrete rand...
Sk7cHb-C-
ICLR.cc/2018/Conference
2018
Representing dynamically: An active process for describing sequential data
["Juan Sebastian Olier", "Emilia Barakova", "Matthias Rauterberg", "Carlo Regazzoni"]
We propose an unsupervised method for building dynamic representations of sequential data, particularly of observed interactions. The method simultaneously acquires representations of input data and its dynamics. It is based on a hierarchical generative model composed of two levels. In the first level, a model learns r...
["Generative Models", "Latent representations", "Predictive coding", "Recurrent networks", "Sequential data"]
ABSTRACTWe propose an unsupervised method for building dynamic representations of se-quential data, particularly of observed interactions. The method simultaneouslyacquires representations of input data and its dynamics. It is based on a hierar-chical generative model composed of two levels. In the first level, a model...
S1pvRc7bM
Good motivation and promising results on celebA, but very poor model description
4: Ok but not good enough - rejection
The paper proposes a hierarchical probabilistic model that learns both static representations and the dynamics of the data. The model iteratively updates its latent representations of the data in order to improve its generative power. The model is applicable to both static images (iterative improvements of the samples)...
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 Representing dynamically: An active process for describing sequential data ### Paper Abstract We propose an unsupervised method for building dynamic representations of s...
SkGT6sRcFX
ICLR.cc/2019/Conference
2019
Infinitely Deep Infinite-Width Networks
["Jovana Mitrovic", "Peter Wirnsberger", "Charles Blundell", "Dino Sejdinovic", "Yee Whye Teh"]
Infinite-width neural networks have been extensively used to study the theoretical properties underlying the extraordinary empirical success of standard, finite-width neural networks. Nevertheless, until now, infinite-width networks have been limited to at most two hidden layers. To address this shortcoming, we study t...
["Infinite-width networks", "initialisation", "kernel methods", "reproducing kernel Hilbert spaces", "Gaussian processes"]
ABSTRACTInfinite-width neural networks have been extensively used to study the theoreticalproperties underlying the extraordinary empirical success of standard, finite-widthneural networks. Nevertheless, until now, infinite-width networks have been lim-ited to at most two hidden layers. To address this shortcoming, we ...
rJlx_rSo3m
Systematic theoretical work with some supporting experiments
6: Marginally above acceptance threshold
Summary: I'm not very familiar with the work this is building off of, but my summary is as follows: The authors look at the problem of defining multilayer infinite width neural networks. The main challenge is that the weights (which are in some sense now a function) must be appropriately sampled to ensure that norms do...
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 Infinitely Deep Infinite-Width Networks ### Paper Abstract Infinite-width neural networks have been extensively used to study the theoretical properties underlying the e...
B4lUThMXhCF
ML_Reproducibility_Challenge/2021/Fall
2021
Reproduction study - DECAF: Generating Fair Synthetic Data Using Causally-Aware Generative Networks
["Anonymous"]
Reproducibility Summary Scope of Reproducibility We conduct a reproduction study on the paper "DECAF: Generating Fair Synthetic Data Using Causally-Aware Generative Networks". We investigate the following claims: (1) DECAF is versatile and can remove undesired bias using several definitions of fairness including Fairn...
["decaf", "fair synthetic data", "authors", "code", "results", "reproduction study", "undesired bias", "fairness", "ftu", "able"]
Reproduction study - DECAF: Generating Fair Synthetic DataUsing Causally-Aware Generative NetworksReproducibility Summary 1Scope of Reproducibility 2We conduct a reproduction study on the paper DECAF: Generating Fair Synthetic Data Using Causally-Aware 3Generative Networks by van Breugel et al. [2021]. We investigate t...
BZ4giIEVgXc
Review of Reproduction study - DECAF: Generating Fair Synthetic Data Using Causally-Aware Generative Networks
5: Marginally below acceptance threshold
Authors worked on replicating the paper by Boris et al. (2021) on generating fair synthetic data using casually aware generative models. Efforts were put in to replicate the Fairness Through Unawareness (FTU), Demographic Parity (DP) metrics. To verify the claim of generating fair synthetic data made in the original pa...
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 Reproduction study - DECAF: Generating Fair Synthetic Data Using Causally-Aware Generative Networks ### Paper Abstract Reproducibility Summary Scope of Reproducibility ...
S1lTg3RcFm
ICLR.cc/2019/Conference
2019
Perception-Aware Point-Based Value Iteration for Partially Observable Markov Decision Processes
["Mahsa Ghasemi", "Ufuk Topcu"]
Partially observable Markov decision processes (POMDPs) are a widely-used framework to model decision-making with uncertainty about the environment and under stochastic outcome. In conventional POMDP models, the observations that the agent receives originate from fixed known distribution. However, in a variety of real-...
["partially observable Markov decision processes", "active perception", "submodular optimization", "point-based value iteration", "reinforcement learning"]
ABSTRACTPartially observable Markov decision processes (POMDPs) are a widely-usedframework to model decision-making with uncertainty about the environment andunder stochastic outcome. In conventional POMDP models, the observations thatthe agent receives originate from fixed known distribution. However, in a varietyof r...
r1enwzJYn7
Decomposition of observation acquisition and action planning in POMDPs: Insufficient motivation and results
4: Ok but not good enough - rejection
This work addresses the problem of decomposing the observation acquisition from action planning in POMDPs. Unfortunately, the paper has two major weaknesses. First it is hindered by a confusing motivation, and the lack of clarity on the real purpose of the work is a problem throughout. Second the experiments are insu...
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 Perception-Aware Point-Based Value Iteration for Partially Observable Markov Decision Processes ### Paper Abstract Partially observable Markov decision processes (POMDPs...
B1lFa3EFwB
ICLR.cc/2020/Conference
2020
Stablizing Adversarial Invariance Induction by Discriminator Matching
["Yusuke Iwasawa", "Kei Akuzawa", "Yutaka Matsuo"]
Incorporating the desired invariance into representation learning is a key challenge in many situations, e.g., for domain generalization and privacy/fairness constraints. An adversarial invariance induction (AII) shows its power on this purpose, which maximizes the proxy of the conditional entropy between representatio...
["invariance induction", "adversarial training", "domain generalization"]
ABSTRACTIncorporating the desired invariance into representation learning is a key chal-lenge in many situations, e.g., for domain generalization and privacy/fairness con-straints. An adversarial invariance induction (AII) shows its power on this pur-pose, which maximizes the proxy of the conditional entropy between re...
BJxoDJtnKS
Official Blind Review #2
3: Weak Reject
*Summary* The paper proposes a new method to learn data-driven representations, being invariant to some specific nuisance factors which are detrimental for the selected (supervised) classification task. Authors build upon the existing probabilistic framework termed Adversarial Invariant Induction (AII) from (Xie et al...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Stablizing Adversarial Invariance Induction by Discriminator Matching ### Paper Abstract Incorporating the desired invariance into representation learning is a key chall...
By93_yAUG
ICLR.cc/2018/Workshop
2018
SufiSent - Universal Sentence Representations Using Suffix Encodings
["Siddhartha Brahma"]
Computing universal distributed representations of sentences is a fundamental task in natural language processing. We propose a method to learn such representations by encoding the suffixes of word sequences in a sentence and training on the Stanford Natural Language Inference (SNLI) dataset. We demonstrate the effecti...
["universal sentence representation", "LSTM", "natural language inference"]
ABSTRACTComputing universal distributed representations of sentences is a fundamental taskin natural language processing. We propose a method to learn such representa-tions by encoding the suffixes of word sequences in a sentence and training on theStanford Natural Language Inference (SNLI) dataset. We demonstrate the ...
Byl7eDbFf
Minor contribution with mixed results
6: Marginally above acceptance threshold
This paper proposes to extend the bi-directional LSTM (BiLSTM) for suffixes encoding. In sum, this paper is quite clear, although I would like to see more motivation for the proposed approach. In my opinion, the only difference of this model compared to the BiLSTM is that the author presents information encoded in each...
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 SufiSent - Universal Sentence Representations Using Suffix Encodings ### Paper Abstract Computing universal distributed representations of sentences is a fundamental tas...
0cmMMy8J5q
ICLR.cc/2021/Conference
2021
Zero-Cost Proxies for Lightweight NAS
["Mohamed S Abdelfattah", "Abhinav Mehrotra", "\u0141ukasz Dudziak", "Nicholas Donald Lane"]
Neural Architecture Search (NAS) is quickly becoming the standard methodology to design neural network models. However, NAS is typically compute-intensive because multiple models need to be evaluated before choosing the best one. To reduce the computational power and time needed, a proxy task is often used for evaluati...
["NAS", "AutoML", "proxy", "pruning", "efficient"]
ABSTRACTNeural Architecture Search (NAS) is quickly becoming the standard methodologyto design neural network models. However, NAS is typically compute-intensivebecause multiple models need to be evaluated before choosing the best one. Toreduce the computational power and time needed, a proxy task is often used foreval...
Gea4Pzz_mo2
This is a paper that barely meets the acceptance criteria.
6: Marginally above acceptance threshold
This paper provides an interesting direction in the neural architecture field. In particular, it proposes a series of zero-cost proxies. To design such proxies, the authors evaluate conventional reduced-training proxies and recent pruning algorithms. These proxies use just a single minibatch of training data to compute...
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 Zero-Cost Proxies for Lightweight NAS ### Paper Abstract Neural Architecture Search (NAS) is quickly becoming the standard methodology to design neural network models. H...
rkeJRhNYDH
ICLR.cc/2020/Conference
2020
TabFact: A Large-scale Dataset for Table-based Fact Verification
["Wenhu Chen", "Hongmin Wang", "Jianshu Chen", "Yunkai Zhang", "Hong Wang", "Shiyang Li", "Xiyou Zhou", "William Yang Wang"]
The problem of verifying whether a textual hypothesis holds based on the given evidence, also known as fact verification, plays an important role in the study of natural language understanding and semantic representation. However, existing studies are mainly restricted to dealing with unstructured evidence (e.g., natur...
["Fact Verification", "Tabular Data", "Symbolic Reasoning"]
ABSTRACTThe problem of verifying whether a textual hypothesis holds based on the givenevidence, also known as fact verification, plays an important role in the study ofnatural language understanding and semantic representation. However, existingstudies are mainly restricted to dealing with unstructured evidence (e.g., ...
H1gN3TMEcH
Official Blind Review #2
6: Weak Accept
This work proposes the problem of fact verification with semi-structured data source such as tables. Specifically, the authors created a new dataset TabFact and evaluated two baseline models with different variations. They applied two criteria and different rewards for workers to collect two subsets of different levels...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title TabFact: A Large-scale Dataset for Table-based Fact Verification ### Paper Abstract The problem of verifying whether a textual hypothesis holds based on the given eviden...
SkgRW64twr
ICLR.cc/2020/Conference
2020
Deep Multi-View Learning via Task-Optimal CCA
["Heather D. Couture", "Roland Kwitt", "J.S. Marron", "Melissa Troester", "Charles M. Perou", "Marc Niethammer"]
Canonical Correlation Analysis (CCA) is widely used for multimodal data analysis and, more recently, for discriminative tasks such as multi-view learning; however, it makes no use of class labels. Recent CCA methods have started to address this weakness but are limited in that they do not simultaneously optimize the C...
["multi-view", "components analysis", "CCA", "representation learning", "deep learning"]
ABSTRACTMulti-view learning seeks to form better models by making use of multiple featuresets representing the same samples. Exploiting feature correlations during train-ing can enable better models. The traditional method for computing correlationsbetween feature sets is Canonical Correlation Analysis (CCA), which fin...
SkgeAag0tH
Official Blind Review #3
6: Weak Accept
This paper addresses the problem of jointly performing CCA with task labeling. The problem is timely and important as it is challenging to perform CCA jointly with the task classification (see below) and hence previous work typically perform this in a pipeline - that is, first projecting the data using a pre-trained CC...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Deep Multi-View Learning via Task-Optimal CCA ### Paper Abstract Canonical Correlation Analysis (CCA) is widely used for multimodal data analysis and, more recently, for...
IfyIzP_GSpn
MICCAI.org/2022/Challenge/FLARE
2022
Abdominal Organ Segmentation via Self Training
["Hexin Dong", "Zifan Chen", "Jie Zhao", "Mingze Yuan", "Fei Yu", "Jiangdong Zhang", "Li Zhang"]
This paper proposes a semi-supervised abdominal organ segmentation approach based on self-training. Self-training(ST) helps improve the segmentation network’s decision boundary with unlabeled images and generated pseudo labels. Experiment results show that our method has significantly outperformed the non-ST baseline, ...
["Semantic Segmentation", "Self Training"]
Abdominal Organ Segmentation via Self TrainingHexin Dong1, Zifan Chen1, Jie Zhao3, Mingze Yuan1, Fei Yu1, JiangdongZhang1, and Li Zhang1,21Center for Data Science, Peking University, Beijing, China2Center for Data Science in Health and Medicine, Peking University, Beijing, China3National Engineering Laboratory for Big ...
o9xUnx8V0Pc
Abdominal Organ Segmentation via Self Training
5: Marginally below acceptance threshold
Strengths: The proposed method achieves effective semi-supervised learning via self-training with a mean DSC of 0.8568. Weaknesses: The two main aspects of the FLARE22 challenge are efficiency and how to use unlabeled data. Although the proposed method was a very effective semi-supervised learning strategy, the paper ...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Abdominal Organ Segmentation via Self Training ### Paper Abstract This paper proposes a semi-supervised abdominal organ segmentation approach based on self-training. Sel...
HJn_vKyPM
ICLR.cc/2018/Workshop
2018
FaceGANs: Stable Generative Adversarial Networks with High-Quality Images
["Jiayu Li", "Zheng Zhan", "Caiwen Ding", "Yanzhi Wang"]
Generative Adversarial Networks (GANs) have shown impressive performance in producing images highly similar to original dataset under unsupervised learning. However, the losses of discriminator and generator are highly fluctuated, which affects the quality of fake images produced by the generator. In this work, we prop...
["FaceGANs", "GANs", "High-Quality Images", "Stablization"]
ABSTRACTGenerative Adversarial Networks (GANs) have shown impressive performance inproducing images highly similar to original dataset under unsupervised learning.However, the losses of discriminator and generator are highly fluctuated, whichaffects the quality of fake images produced by the generator. In this work, we...
B1BgNxltz
assumptions without backing, lack of novelty and justification
4: Ok but not good enough - rejection
The paper starts off with: > "However, we find that the loss of discriminator and generator are not stable enough and highly fluctuated in GANs and DCGANs. The situation will affect the quality of images from a generator." There is no backing for this statement, and afaik this is simply not true. The loss of a GAN (G...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title FaceGANs: Stable Generative Adversarial Networks with High-Quality Images ### Paper Abstract Generative Adversarial Networks (GANs) have shown impressive performance in ...
SJx-j64FDr
ICLR.cc/2020/Conference
2020
In Search for a SAT-friendly Binarized Neural Network Architecture
["Nina Narodytska", "Hongce Zhang", "Aarti Gupta", "Toby Walsh"]
Analyzing the behavior of neural networks is one of the most pressing challenges in deep learning. Binarized Neural Networks are an important class of networks that allow equivalent representation in Boolean logic and can be analyzed formally with logic-based reasoning tools like SAT solvers. Such tools can be used t...
["verification", "Boolean satisfiability", "Binarized Neural Networks"]
ABSTRACTAnalyzing the behavior of neural networks is one of the most pressing challengesin deep learning. Binarized Neural Networks are an important class of networksthat allow equivalent representation in Boolean logic and can be analyzed formallywith logic-based reasoning tools like SAT solvers. Such tools can be use...
H1ep15J9OH
Official Blind Review #2
6: Weak Accept
In this paper, a new design of Binary Neural Network (BNN) is proposed. The purpose of the new design is to make the network easier to analyze using the existing Boolean satisfiability (SAT) solvers. An easier analysis is preferable since the SAT solver can be used to infer existing fragility in the network, e.g., adve...
<|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 Search for a SAT-friendly Binarized Neural Network Architecture ### Paper Abstract Analyzing the behavior of neural networks is one of the most pressing challenges i...
afF8RGcBBP
robot-learning.org/CoRL/2023/Conference
2023
PlayFusion: Skill Acquisition via Diffusion from Language-Annotated Play
["Lili Chen", "Shikhar Bahl", "Deepak Pathak"]
Learning from unstructured and uncurated data has become the dominant paradigm for generative approaches in language or vision. Such unstructured and unguided behavior data, commonly known as play, is also easier to collect in robotics but much more difficult to learn from due to its inherently multimodal, noisy, and s...
["Diffusion Models", "Learning from Play", "Language-Driven Robotics"]
PlayFusion: Skill Acquisition via Diffusion fromLanguage-Annotated PlayLili Chen* Shikhar Bahl* Deepak PathakCarnegie Mellon UniversityAbstract: Learning from unstructured and uncurated data has become the dominantparadigm for generative approaches in language and vision. Such unstructuredand unguided behavior data, co...
aSuVyyIL29
Strengths * The paper is very well presented and convincing. * Strong experimental section with proper ablations. * Results on multiple environments like CALVIN, Language conditioned Ravens, and Franka * Testing on long horizon tasks on the CALVIN evaluation suite make results more trustworthy with 128 total rollouts. ...
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 PlayFusion: Skill Acquisition via Diffusion from Language-Annotated Play ### Paper Abstract Learning from unstructured and uncurated data has become the dominant paradig...
ME5Uh_tyld5
NeurIPS.cc/2020/Workshop/SVRHM
2020
Disentangled Face Representations in Deep Generative Models and the Human Brain
["Paul Soulos", "Leyla Isik"]
How does the human brain recognize faces and represent their many features? Despite decades of research, we still lack a thorough understanding of the computations carried out in face-selective regions of the human brain. Deep networks provide good match to neural data, but lack interpretability. Here we use a new cla...
["disentangled", "representation learning", "face perception", "generative models", "fmri", "encoding model", "generalization"]
Disentangled Face Representations in DeepGenerative Models and the Human BrainPaul SoulosJohns Hopkins Universitypsoulos1@jhu.eduLeyla IsikJohns Hopkins Universitylisik@jhu.eduAbstractHow does the human brain recognize faces and represent their many features?Despite decades of research, we still lack a thorough underst...
gLupe_i1tkx
interesting idea with limited results
6: Marginally above acceptance threshold
This paper is investigating VAEs and disentangled VAEs as potential representation of faces in the human brain. It claims that disentangled VAEs similarly well match fmri data as VAEs but have the benefit of additional control and disentanglement. They further claim that gender and 3D rotation (yaw) is represented dise...
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 Disentangled Face Representations in Deep Generative Models and the Human Brain ### Paper Abstract How does the human brain recognize faces and represent their many feat...
kic8cng35wX
ICLR.cc/2021/Conference
2021
Weak NAS Predictor Is All You Need
["Junru Wu", "Xiyang Dai", "Dongdong Chen", "Yinpeng Chen", "Mengchen Liu", "Ye Yu", "Zhangyang Wang", "Zicheng Liu", "Mei Chen", "Lu Yuan"]
Neural Architecture Search (NAS) finds the best network architecture by exploring the architecture-to-performance manifold. It often trains and evaluates a large amount of architectures, causing tremendous computation cost. Recent predictor-based NAS approaches attempt to solve this problem with two key steps: sampling...
["performance predictor", "neural architecture search"]
ABSTRACTNeural Architecture Search (NAS) finds the best network architecture by explor-ing the architecture-to-performance manifold. It often trains and evaluates a largeamount of architectures, causing tremendous computation cost. Recent predictor-based NAS approaches attempt to solve this problem with two key steps: ...
flTv2z2U-bl
An interesting and inspirational paper
6: Marginally above acceptance threshold
The paper proposes an idea of jointly optimizing the sampling policy and predictor in NAS. With this method, we avoid to train a predictor which performs well on the whole search space and search a model with less queries. ### advantages The authors propose that training a good accuracy predictor for models in the who...
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 Weak NAS Predictor Is All You Need ### Paper Abstract Neural Architecture Search (NAS) finds the best network architecture by exploring the architecture-to-performance m...
B1eiJyrtDB
ICLR.cc/2020/Conference
2020
Improved Generalization Bound of Permutation Invariant Deep Neural Networks
["Akiyoshi Sannai", "Masaaki Imaizumi"]
We theoretically prove that a permutation invariant property of deep neural networks largely improves its generalization performance. Learning problems with data that are invariant to permutations are frequently observed in various applications, for example, point cloud data and graph neural networks. Numerous methodol...
["Deep Neural Network", "Invariance", "Symmetry", "Group", "Generalization"]
ABSTRACTWe theoretically prove that learning with permutation invariant deep neural net-works largely improves their generalization performance. Learning problems withdata that are invariant to permutations are frequently observed in various applica-tions, for example, point cloud data and graph data. Numerous methodol...
rJx0u1p19B
Official Blind Review #3
1: Reject
This paper provides generalization bounds for permutation invariant neural networks where the learning problem is invariant to the permutation of input data. Unfortunately, the technical value of the content and its novelty is very limited since the proof reduces to a very basic argument that counts invariances (whic...
<|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 Generalization Bound of Permutation Invariant Deep Neural Networks ### Paper Abstract We theoretically prove that a permutation invariant property of deep neura...
Sklia3EFPH
ICLR.cc/2020/Conference
2020
Input Alignment along Chaotic directions increases Stability in Recurrent Neural Networks
["Priyadarshini Panda", "Kaushik Roy"]
Anatomical studies demonstrate that brain reformats input information to generate reliable responses for performing computations. However, it remains unclear how neural circuits encode complex spatio-temporal patterns. We show that neural dynamics are strongly influenced by the phase alignment between the input and the...
["Reservoir Models", "Stability", "Chaos", "Input Alignment", "Mean Field Analysis"]
ABSTRACTAnatomical studies demonstrate that brain reformats input information to gener-ate reliable responses for performing computations. However, it remains unclearhow neural circuits encode complex spatio-temporal patterns. We show that neu-ral dynamics are strongly influenced by the phase alignment between the inpu...
BJx8E8ZWcH
Official Blind Review #2
1: Reject
This paper presents an analysis of recurrent networks with random weights, showing that certain input phases can lead to more stable neuron trajectories over time. To explain the phenomenon, the author’s present a mean field analysis, and experimentally show that certain input phases lead to more robust readouts. Majo...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Input Alignment along Chaotic directions increases Stability in Recurrent Neural Networks ### Paper Abstract Anatomical studies demonstrate that brain reformats input in...
Hke_f0EYPH
ICLR.cc/2020/Conference
2020
Efficient Training of Robust and Verifiable Neural Networks
["Akhilan Boopathy", "Lily Weng", "Sijia Liu", "Pin-Yu Chen", "Luca Daniel"]
Recent works have developed several methods of defending neural networks against adversarial attacks with certified guarantees. We propose that many common certified defenses can be viewed under a unified framework of regularization. This unified framework provides a technique for comparing different certified defenses...
["robust", "unified framework", "certified defenses", "efficient training", "several methods", "neural networks", "adversarial attacks", "certified guarantees"]
ABSTRACTRecent works have developed several methods of defending neural networks againstadversarial attacks with certified guarantees. We propose that many commoncertified defenses can be viewed under a unified framework of regularization. Thisunified framework provides a technique for comparing different certified def...
B1gnQjh5KH
Official Blind Review #2
1: Reject
Summary: This paper studies the problem of certified robustness to adversarial examples and proposes a new training method to obtain better certified robustness. This new method is based on using a double margin regularizer. Decision: I vot for rejecting this paper. I think the paper studies an important problem. How...
<|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 Training of Robust and Verifiable Neural Networks ### Paper Abstract Recent works have developed several methods of defending neural networks against adversari...
0XXpJ4OtjW
ICLR.cc/2021/Conference
2021
Evolving Reinforcement Learning Algorithms
["John D Co-Reyes", "Yingjie Miao", "Daiyi Peng", "Esteban Real", "Quoc V Le", "Sergey Levine", "Honglak Lee", "Aleksandra Faust"]
We propose a method for meta-learning reinforcement learning algorithms by searching over the space of computational graphs which compute the loss function for a value-based model-free RL agent to optimize. The learned algorithms are domain-agnostic and can generalize to new environments not seen during training. Our m...
["reinforcement learning", "evolutionary algorithms", "meta-learning", "genetic programming"]
ABSTRACTWe propose a method for meta-learning reinforcement learning algorithms bysearching over the space of computational graphs which compute the loss functionfor a value-based model-free RL agent to optimize. The learned algorithms aredomain-agnostic and can generalize to new environments not seen during train-ing....
1XCJWKC70of
A bit old-fashioned idea, but checks outs
9: Top 15% of accepted papers, strong accept
The paper proposes an approach to develop new Reinforcement Learning (RL) algorithms through a population-based method very reminiscent of Genetic Programming (GP). The authors "evolve" loss functions that can be used across different RL scenarios and providing good generalization. Pros: - The paper is well written, w...
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 Evolving Reinforcement Learning Algorithms ### Paper Abstract We propose a method for meta-learning reinforcement learning algorithms by searching over the space of comp...
Skl3SkSKDr
ICLR.cc/2020/Conference
2020
Generating valid Euclidean distance matrices
["Moritz Hoffmann", "Frank Noe"]
Generating point clouds, e.g., molecular structures, in arbitrary rotations, translations, and enumerations remains a challenging task. Meanwhile, neural networks utilizing symmetry invariant layers have been shown to be able to optimize their training objective in a data-efficient way. In this spirit, we present an ar...
["euclidean distance matrices", "wgan", "point clouds", "molecular structures"]
ABSTRACTGenerating point clouds, e.g., molecular structures, in arbitrary rotations, transla-tions, and enumerations remains a challenging task. Meanwhile, neural networksutilizing symmetry invariant layers have been shown to be able to optimize theirtraining objective in a data-efficient way. In this spirit, we presen...
Skx9wBpHFB
Official Blind Review #2
8: Accept
The paper is extremely interesting, solid and very well written. The idea is simple but nonetheless developed in a smart and effective fashion. The underlying theory is solid, even if some choices should have been discussed more deeply (e.g. the chosen loss function). Introduction and references are adequate, and the p...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Generating valid Euclidean distance matrices ### Paper Abstract Generating point clouds, e.g., molecular structures, in arbitrary rotations, translations, and enumeratio...
EVV259WQuFG
ICLR.cc/2021/Conference
2021
Machine Reading Comprehension with Enhanced Linguistic Verifiers
["Xianchao Wu"]
We propose two linguistic verifiers for span-extraction style machine reading comprehension to respectively tackle two challenges: how to evaluate the syntactic completeness of predicted answers and how to utilize the rich context of long documents. Our first verifier rewrites a question through replacing its interroga...
["machine reading comprehension", "BERT", "linguistic verifiers", "hierarchical attention networks"]
ABSTRACTWe propose two linguistic verifiers for span-extraction style machine readingcomprehension to respectively tackle two challenges: how to evaluate the syntacticcompleteness of predicted answers and how to utilize the rich context of longdocuments. Our first verifier rewrites a question through replacing its inte...
Jb4GJKX84a2
New modules with strong empirical results on MRC
7: Good paper, accept
In this paper, two linguistic verifiers are proposed to improve the model performance on machine reading comprehension datasets, such as SQuAD v2, NewsQA and TriviaQA. The first verifier rewrites the question by replacing its interrogatives with the predicted answer phrases. Then it computes a score between the rewritt...
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 Machine Reading Comprehension with Enhanced Linguistic Verifiers ### Paper Abstract We propose two linguistic verifiers for span-extraction style machine reading compreh...
fB2GZQajQ2b
ICLR.cc/2021/Conference
2021
Information-Theoretic Odometry Learning
["Sen Zhang", "Jing Zhang", "Dacheng Tao"]
In this paper, we propose a unified information-theoretic framework for odometry learning, a crucial component of many robotics and vision tasks such as navigation and virtual reality where 6-DOF poses are required in real time. We formulate this problem as optimizing a variational information bottleneck objective func...
["Odometry Learning", "Information Bottleneck", "Generalization Bound"]
ABSTRACTIn this paper, we propose a unified information-theoretic framework for odometrylearning, a crucial component of many robotics and vision tasks such as navigationand virtual reality where 6-DOF poses are required in real time. We formulatethis problem as optimizing a variational information bottleneck objective...
_UTTvUmpn93
Good contribution using an information bottleneck constraint to improve genealization in learning-based VO, but the exposition is a bit difficult to follow at times.
6: Marginally above acceptance threshold
Summary The paper addresses the odometry problem i.e. estimating the relative pose of a camera or device as it moves through the environment. The authors propose an information-theoretic model for visual and visual inertial odometry. The method is based on the hypothesis that features contain pose-irrelevant informati...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Information-Theoretic Odometry Learning ### Paper Abstract In this paper, we propose a unified information-theoretic framework for odometry learning, a crucial component...
P84ryxVG6tR
ICLR.cc/2021/Conference
2021
REPAINT: Knowledge Transfer in Deep Actor-Critic Reinforcement Learning
["Yunzhe Tao", "Sahika Genc", "TAO SUN", "Sunil Mallya"]
Accelerating the learning processes for complex tasks by leveraging previously learned tasks has been one of the most challenging problems in reinforcement learning, especially when the similarity between source and target tasks is low or unknown. In this work, we propose a REPresentation-And-INstance Transfer algorith...
["reinforcement learning", "transfer learning", "actor-critic RL", "representation transfer", "instance transfer", "task similarity", "MuJoCo", "DeepRacer"]
ABSTRACTAccelerating the learning processes for complex tasks by leveraging previouslylearned tasks has been one of the most challenging problems in reinforcementlearning, especially when the similarity between source and target tasks is lowor unknown. In this work, we propose a REPresentation-And-INstance Transferalgo...
lAGJrDcGV6S
A well written paper suggesting a hybrid approach for transfer learning
6: Marginally above acceptance threshold
This paper deals with transfer learning in RL. The problem broadly defined is to improve performance of an agent on a new task, given an agent (teacher) trained on a previous (different) task. There are multiple approaches to this from fine-tuning, distillation to instance transfer. A central question in this field is ...
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 REPAINT: Knowledge Transfer in Deep Actor-Critic Reinforcement Learning ### Paper Abstract Accelerating the learning processes for complex tasks by leveraging previously...
-vshFhHpKhX
logconference.io/LOG/2022/Conference
2022
Dynamic Network Reconfiguration for Entropy Maximization using Deep Reinforcement Learning
["Christoffel Doorman", "Victor-Alexandru Darvariu", "Stephen Hailes", "Mirco Musolesi"]
A key problem in network theory is how to reconfigure a graph in order to optimize a quantifiable objective. Given the ubiquity of networked systems, such work has broad practical applications in a variety of situations, ranging from drug and material design to telecommunications. The large decision space of possible r...
["network reconfiguration", "deep reinforcement learning", "graph learning", "moving target defense", "cybersecurity"]
Dynamic Network Reconfiguration for Entropy Maximizationusing Deep Reinforcement LearningChristoffel Doorman1, Victor-Alexandru Darvariu1,2, Stephen Hailes1, Mirco Musolesi1,2,31University College London2The Alan Turing Institute3University of Bologna{christoffel.doorman.20, v.darvariu, s.hailes, m.musolesi}@ucl.ac.ukA...
PQJODBYGjdA
## Overview: The authors consider the problem of optimizing structural properties of a graph by rewiring edges. They propose a reinforcement learning algorithm combining deep Q-networks (DQN) with graph neural networks (GNNs). They evaluate their method on a cybersecurity task, which they show is equivalent to a certai...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Dynamic Network Reconfiguration for Entropy Maximization using Deep Reinforcement Learning ### Paper Abstract A key problem in network theory is how to reconfigure a gra...
fGiKxvF-eub
ICLR.cc/2021/Conference
2021
Oblivious Sketching-based Central Path Method for Solving Linear Programming Problems
["Zhao Song", "Zheng Yu"]
In this work, we propose a sketching-based central path method for solving linear programmings, whose running time matches the state of art results [Cohen, Lee, Song STOC 19; Lee, Song, Zhang COLT 19]. Our method opens up the iterations of the central path method and deploys an "iterate and sketch" approach towards the...
["optimization", "sketching", "linear programming", "central path method", "running time complexity"]
Under review as a conference paper at ICLR 2021Oblivious Sketching-based Central PathMethod for Solving Linear ProgrammingProblemsAnonymous authorsPaper under double-blind reviewAbstractIn this work, we propose a sketching-based central path method for solvinglinear programmings, whose running time matches the state of...
e-tz6KX24X8
Oblivious sketching based interior point method matching state-of-the-art LPs
4: Ok but not good enough - rejection
SUMMARY: The paper is based on the works of Lee et al and Cohen et al. Building upon these works, the paper comes up with an interior point method that matches the state-of-the-art in LPs. The paper's contribution is in a new type of sketching used inside the interior point method, that demonstrates some advantages ove...
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 Oblivious Sketching-based Central Path Method for Solving Linear Programming Problems ### Paper Abstract In this work, we propose a sketching-based central path method f...
dpuLRRQ7zC
ICLR.cc/2021/Conference
2021
Enhancing Certified Robustness of Smoothed Classifiers via Weighted Model Ensembling
["Chizhou Liu", "Yunzhen Feng", "Ranran Wang", "Bin Dong"]
Randomized smoothing has achieved state-of-the-art certified robustness against $l_2$-norm adversarial attacks. However, it is not wholly resolved on how to find the optimal base classifier for randomized smoothing. In this work, we employ a Smoothed WEighted ENsembling (SWEEN) scheme to improve the performance of rand...
["Adversairal Robustness", "Randomized Smoothing", "Ensembling"]
ABSTRACTRandomized smoothing has achieved state-of-the-art certified robustness againstl2-norm adversarial attacks. However, it is not wholly resolved on how to findthe optimal base classifier for randomized smoothing. In this work, we employ aSmoothed WEighted ENsembling (SWEEN) scheme to improve the performanceof ran...
JLw5UsQ2CSh
Initial review
6: Marginally above acceptance threshold
In this paper, the authors introduce SWEEN, an ensembling scheme for smoothed classifiers. This scheme uses pre-trained models and learns the weights used to sum the output predictions. During deployment, smoothed models rely on sampling a large number of input perturbations and ensembling exacerbates the computation b...
3: The reviewer is fairly confident that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Enhancing Certified Robustness of Smoothed Classifiers via Weighted Model Ensembling ### Paper Abstract Randomized smoothing has achieved state-of-the-art certified robu...
r1RQdCg0W
ICLR.cc/2018/Conference
2018
MACH: Embarrassingly parallel $K$-class classification in $O(d\log{K})$ memory and $O(K\log{K} + d\log{K})$ time, instead of $O(Kd)$
["Qixuan Huang", "Anshumali Shrivastava", "Yiqiu Wang"]
We present Merged-Averaged Classifiers via Hashing (MACH) for $K$-classification with large $K$. Compared to traditional one-vs-all classifiers that require $O(Kd)$ memory and inference cost, MACH only need $O(d\log{K})$ memory while only requiring $O(K\log{K} + d\log{K})$ operation for inference. MACH is the first gen...
["Extreme Classification", "Large-scale learning", "hashing", "GPU", "High Performance Computing"]
ABSTRACTWe present Merged-Averaged Classifiers via Hashing (MACH) for K-classification with large K. Compared to traditional one-vs-all classifiers thatrequireO(Kd)memory and inference cost, MACH only need O(dlogK)mem-ory while only requiring O(KlogK+dlogK)operation for inference. MACH isthe first generic K-classificat...
H1VwD15lG
MACH: Embarrassingly parallel $K$-class classification
6: Marginally above acceptance threshold
The paper presents a hashing based scheme (MACH) for reducing memory and computation time for K-way classification when K is large. The main idea is to use R hash functions to generate R different datasets/classifiers where the K classes are mapped into a small number of buckets (B). During inference the probabilities ...
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 MACH: Embarrassingly parallel $K$-class classification in $O(d\log{K})$ memory and $O(K\log{K} + d\log{K})$ time, instead of $O(Kd)$ ### Paper Abstract We present Merged...
HylpqA4FwS
ICLR.cc/2020/Conference
2020
RNNs Incrementally Evolving on an Equilibrium Manifold: A Panacea for Vanishing and Exploding Gradients?
["Anil Kag", "Ziming Zhang", "Venkatesh Saligrama"]
Recurrent neural networks (RNNs) are particularly well-suited for modeling long-term dependencies in sequential data, but are notoriously hard to train because the error backpropagated in time either vanishes or explodes at an exponential rate. While a number of works attempt to mitigate this effect through gated recur...
["novel recurrent neural architectures", "learning representations of outputs or states"]
ABSTRACTRecurrent neural networks (RNNs) are particularly well-suited for modeling long-term dependencies in sequential data, but are notoriously hard to train because theerror backpropagated in time either vanishes or explodes at an exponential rate.While a number of works attempt to mitigate this effect through gated...
SJgfNIk3YB
Official Blind Review #2
6: Weak Accept
The authors present a novel work to address the problem of signal propagation in the recurrent neural networks. The idea is to build a attractor system for the signal transition from state h_{k-1} to h_k. If the attractor system converges to a equilibrium, then the hidden to hidden gradient is an identity matrix. This ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title RNNs Incrementally Evolving on an Equilibrium Manifold: A Panacea for Vanishing and Exploding Gradients? ### Paper Abstract Recurrent neural networks (RNNs) are particul...
Syg-YfWCW
ICLR.cc/2018/Conference
2018
Go for a Walk and Arrive at the Answer: Reasoning Over Paths in Knowledge Bases using Reinforcement Learning
["Rajarshi Das", "Shehzaad Dhuliawala", "Manzil Zaheer", "Luke Vilnis", "Ishan Durugkar", "Akshay Krishnamurthy", "Alex Smola", "Andrew McCallum"]
Knowledge bases (KB), both automatically and manually constructed, are often incomplete --- many valid facts can be inferred from the KB by synthesizing existing information. A popular approach to KB completion is to infer new relations by combinatory reasoning over the information found along other paths connecting a ...
["Knowledge Graphs", "Reinforcement Learning", "Query Answering"]
ABSTRACTKnowledge bases (KB), both automatically and manually constructed, are oftenincomplete — many valid facts can be inferred from the KB by synthesizingexisting information. A popular approach to KB completion is to infer new relationsby combinatory reasoning over the information found along other paths connecting...
SymbMC2xf
Simple RL approach for reasoning on Knowledge bases, good performance on variety of datasets but need to be slightly more thorough in their comparisons with prior work
7: Good paper, accept
The paper present a RL based approach to walk on a knowledge graph to answer queries. The idea is novel, the paper is clear in its exposition, and the authors provide a number of experimental comparisons with prior work on a variety of datasets . Pros: 1. The approach is simple (no pre-training, no reward shaping, ju...
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 Go for a Walk and Arrive at the Answer: Reasoning Over Paths in Knowledge Bases using Reinforcement Learning ### Paper Abstract Knowledge bases (KB), both automatically ...
rJg_1L5gg
ICLR.cc/2017/conference
2017
Incremental Sequence Learning
["Edwin D. de Jong"]
Deep learning research over the past years has shown that by increasing the scope or difficulty of the learning problem over time, increasingly complex learning problems can be addressed. We study incremental learning in the context of sequence learning, using generative RNNs in the form of multi-layer recurrent Mixtur...
["Deep learning", "Supervised Learning"]
ABSTRACTDeep learning research over the past years has shown that by increasing the scopeor difficulty of the learning problem over time, increasingly complex learningproblems can be addressed. We study incremental learning in the context ofsequence learning, using generative RNNs in the form of multi-layer recurrentMi...
S1CHraWNg
5: Marginally below acceptance threshold
This paper presents a thorough analysis of different methods to do curriculum learning. The major issue I have with it is that the dataset used seems very specific and does not necessarily justified, as mentioned by AnonReviewer3. It would have been great to see experiments on more standard tasks. Also, I really can'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 Incremental Sequence Learning ### Paper Abstract Deep learning research over the past years has shown that by increasing the scope or difficulty of the learning problem ...
6YEQUn0QICG
ICLR.cc/2021/Conference
2021
FedBN: Federated Learning on Non-IID Features via Local Batch Normalization
["Xiaoxiao Li", "Meirui JIANG", "Xiaofei Zhang", "Michael Kamp", "Qi Dou"]
The emerging paradigm of federated learning (FL) strives to enable collaborative training of deep models on the network edge without centrally aggregating raw data and hence improving data privacy. In most cases, the assumption of independent and identically distributed samples across local clients does not hold for fe...
["Federated Learning", "Non-IID", "Batch Normalization"]
ABSTRACTThe emerging paradigm of federated learning (FL) strives to enable collabora-tive training of deep models on the network edge without centrally aggregatingraw data and hence improving data privacy. In most cases, the assumption of in-dependent and identically distributed samples across local clients does not ho...
ugDU0fozHAW
Approaching the important problem of batch normalisation in federated learning
7: Good paper, accept
This work proposes an extremely simple approach to a well-know problem within Federated Learning: batch normalisation. Indeed, FL usually imply an averaging of different model parameters trained on different distributed devices. But what happen with the running statistics that some training methods have ? Such as batch...
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 FedBN: Federated Learning on Non-IID Features via Local Batch Normalization ### Paper Abstract The emerging paradigm of federated learning (FL) strives to enable collabo...
rJg6ssC5Y7
ICLR.cc/2019/Conference
2019
DeepOBS: A Deep Learning Optimizer Benchmark Suite
["Frank Schneider", "Lukas Balles", "Philipp Hennig"]
Because the choice and tuning of the optimizer affects the speed, and ultimately the performance of deep learning, there is significant past and recent research in this area. Yet, perhaps surprisingly, there is no generally agreed-upon protocol for the quantitative and reproducible evaluation of optimization strategies...
["deep learning", "optimization"]
ABSTRACTThere is significant past and ongoing research on optimization methods for deeplearning. Yet, perhaps surprisingly, there is no generally agreed-upon protocolfor the quantitative and reproducible evaluation of such optimizers. We suggestroutines and benchmarks for stochastic optimization, with special focus on ...
r1eC9mIc2X
Good initiative at its beginning stage
6: Marginally above acceptance threshold
As the paper claims there is no common accept system for benchmarking deep learning optimizer. It is also hard to repeat others' results. The paper describes a benchmarking framework for deep learning optimizer. It proposes three performance indicators, and includes 20 test problems and a core set of benchmarks. Pro:...
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 DeepOBS: A Deep Learning Optimizer Benchmark Suite ### Paper Abstract Because the choice and tuning of the optimizer affects the speed, and ultimately the performance of...
E-PcUeaDbzv
graphicsinterface.org/Graphics_Interface/2022/Conference
2022
Automatic Asymmetric Weight Distribution Detection and Correction Utilizing Electrical Muscle Stimulation
["Ravi Kiran Kattoju", "Eugene M Taranta", "Ryan Ghamandi", "Joseph LaViola"]
Postural control is a constant re-establishment process for the maintenance of balance and stability. Asymmetric weight distribution (AWD), characterized by uneven leg loading, leads to increased instability, injury, and progressive deterioration of posture and gait. Postural self-correction is automatically affected b...
["Wearable computing", "Preventive Healthcare", "Posture correction", "Asymmetric weight distribution", "Electrical Muscle Stimulation"]
ABSTRACTPostural control is a constant re-establishment process for the main-tenance of balance and stability. Asymmetric weight distribution(AWD), characterized by uneven leg loading, leads to increasedinstability, injury, and progressive deterioration of posture and gait.Postural self-correction is automatically affe...
tRPIptLYHhc
Good contribution overall, pairwise comparison not presented or not done (with it paper would be "clear accept"), some limitations not discussed
7: Good paper, accept
The manuscript presents an electrical muscle stimulation system that automatically stimulates lower limb muscles in order to automatically restore symmetrical weight distribution (a center of pressure that is centralized, measure using a wii balance board). The device is presented as is a user study that compares it to...
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 Automatic Asymmetric Weight Distribution Detection and Correction Utilizing Electrical Muscle Stimulation ### Paper Abstract Postural control is a constant re-establishm...
SJg4Y3VFPS
ICLR.cc/2020/Conference
2020
Group-Connected Multilayer Perceptron Networks
["Mohammad Kachuee", "Sajad Darabi", "Shayan Fazeli", "Majid Sarrafzadeh"]
Despite the success of deep learning in domains such as image, voice, and graphs, there has been little progress in deep representation learning for domains without a known structure between features. For instance, a tabular dataset of different demographic and clinical factors where the feature interactions are not gi...
["gmlp", "domains", "operations", "multilayer perceptron networks", "deep representation", "expressive feature combinations", "groups", "success", "deep learning", "image"]
ABSTRACTDespite the success of deep learning in domains such as image, voice, and graphs,there has been little progress in deep representation learning for domains withouta known structure between features. For instance, a tabular dataset of differentdemographic and clinical factors where the feature interactions are n...
BJgm18Gaqr
Official Blind Review #3
3: Weak Reject
This paper addresses the problem of learning expressive feature combinations in order to improve learning for domains where there is no known structure between features. These settings would normally lead to the use of fully-connected MLP networks, which unfortunately have problems with efficient training and generaliz...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Group-Connected Multilayer Perceptron Networks ### Paper Abstract Despite the success of deep learning in domains such as image, voice, and graphs, there has been little...
ryl1r1BYDS
ICLR.cc/2020/Conference
2020
Multiagent Reinforcement Learning in Games with an Iterated Dominance Solution
["Yoram Bachrach", "Tor Lattimore", "Marta Garnelo", "Julien Perolat", "David Balduzzi", "Thomas Anthony", "Satinder Singh", "Thore Graepel"]
Multiagent reinforcement learning (MARL) attempts to optimize policies of intelligent agents interacting in the same environment. However, it may fail to converge to a Nash equilibrium in some games. We study independent MARL under the more demanding solution concept of iterated elimination of strictly dominated strat...
["multiagent", "reinforcement learning", "iterated dominance", "mechanism design", "Nash equilibrium"]
ABSTRACTMultiagent reinforcement learning (MARL) attempts to optimize policies of intelli-gent agents interacting in the same environment. However, it may fail to convergeto a Nash equilibrium in some games. We study independent MARL under themore demanding solution concept of iterated elimination of strictly dominated...
H1luZksD9r
Official Blind Review #4
6: Weak Accept
This paper studies reinforcement learning algorithms in a specific subset of multi-agent environments that are 'dominance solvable'. This means that, given an initial set of strategies in the game, if we iteratively remove 'dominated strategies' (those whose utility is strictly less than another strategy independent of...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Multiagent Reinforcement Learning in Games with an Iterated Dominance Solution ### Paper Abstract Multiagent reinforcement learning (MARL) attempts to optimize policies ...
rygfnn4twS
ICLR.cc/2020/Conference
2020
AutoQ: Automated Kernel-Wise Neural Network Quantization
["Qian Lou", "Feng Guo", "Minje Kim", "Lantao Liu", "Lei Jiang."]
Network quantization is one of the most hardware friendly techniques to enable the deployment of convolutional neural networks (CNNs) on low-power mobile devices. Recent network quantization techniques quantize each weight kernel in a convolutional layer independently for higher inference accuracy, since the weight ker...
["AutoML", "Kernel-Wise Neural Networks Quantization", "Hierarchical Deep Reinforcement Learning"]
ABSTRACTNetwork quantization is one of the most hardware friendly techniques to enablethe deployment of convolutional neural networks (CNNs) on low-power mobiledevices. Recent network quantization techniques quantize each weight kernel in aconvolutional layer independently for higher inference accuracy, since the weigh...
H1ebnpXOnH
Official Blind Review #4
6: Weak Accept
Summary This paper proposes a network quantization method. Different from previous methods focusing on network-level or layer-lever quantization, this work pays attention to kernel-level quantization. Specifically, they use a hierarchical reinforcement learning framework to search in the search space related with quant...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title AutoQ: Automated Kernel-Wise Neural Network Quantization ### Paper Abstract Network quantization is one of the most hardware friendly techniques to enable the deploymen...
Z_L9j0HW3QM
MICCAI.org/2021/Workshop/COMPAY
2021
Attention-based Multiple Instance Learning with Mixed Supervision on the Camelyon16 Dataset
["Paul Tourniaire", "Marius Ilie", "Paul Hofman", "Nicholas Ayache", "Herv\u00e9 Delingette"]
Since the standardization of Whole Slide Images (WSIs) digitization, the use of deep learning methods for the analysis of histological images has shown much potential. However, the sheer size of WSIs is a real challenge, as they are often up to 100,000 pixels wide and high at the highest resolution, and therefore canno...
["Attention Mechanism", "Mixed Supervision", "Histopathology"]
Proceedings of Machine Learning Research 156, 2021 MICCAI Computational Pathology (COMPAY) WorkshopAttention-based Multiple Instance Learning with MixedSupervision on the Camelyon16 DatasetPaul Tourniaire paul.tourniaire@inria.frUniversite Cote d'Azur, Inria, Epione Team, Sophia Antipolis, FranceMarius Ilie ilie.m@chu-...
UUnd8QIWyJ1
The paper introduces a mixed supervision approach to attention-based MIL, that improves the diagnosis of cancerous WSI, as well as the localization of tumor regions within the WSI.
8: Top 50% of accepted papers, clear accept
**Strengths** The paper contributions are clearly listed and are significant. The paper is coherent and well written and includes a comprehensive literature study. The results convincingly show that their method improves classification performance and reduces the number of false positive tumor tiles. **Weaknesses** ...
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 Attention-based Multiple Instance Learning with Mixed Supervision on the Camelyon16 Dataset ### Paper Abstract Since the standardization of Whole Slide Images (WSIs) dig...
BJe02gRiY4
MIDL.io/2019/Conference/Abstract
2019
Transfer learning from synthetic data reduces need for labels to segment brain vasculature and neural pathways in 3D
["Johannes C. Paetzold", "Oliver Schoppe", "Rami Al-Maskari", "Giles Tetteh", "Velizar Efremov", "Mihail I. Todorov", "Ruiyao Cai", "Hongcheng Mai", "Zhouyi Rong", "Ali Ertuerk", "Bjoern H. Menze"]
Novel microscopic techniques yield high-resolution volumetric scans of complex anatomical structures such as the blood vasculature or the nervous system. Here, we show how transfer learning and synthetic data generation can be used to train deep neural networks to segment these structures successfully in the absence of...
["Deep learning", "transfer learning", "synthetic data", "vasculature", "neural pathways"]
Proceedings of Machine Learning Research { Under Review:1{5, 2019 Extended Abstract { MIDL 2019 submissionTransfer learning from synthetic data reduces need for labelsto segment brain vasculature and neural pathways in 3DJohannes C. Paetzold1;y, Oliver Schoppe1, Rami Al-Maskari1, Giles Tetteh1,Velizar Efremov1, Mihail ...
BylrDpRSo4
Interesting application of transfer learning to novel dataset
3: Accept
The authors present an application of transfer learning for vessel segmentation to tackle the issue of labour-intensive dense data labeling. They demonstrate the utility of pre-training on synthetic data as well as on different datasets - namely human brain vessels from MRI, and mouse brain vessels and mouse periphera...
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 Transfer learning from synthetic data reduces need for labels to segment brain vasculature and neural pathways in 3D ### Paper Abstract Novel microscopic techniques yiel...
YbDGyviJkrL
ICLR.cc/2021/Conference
2021
Transformers for Modeling Physical Systems
["Nicholas Geneva", "Nicholas Zabaras"]
Transformers are widely used in neural language processing due to their ability to model longer-term dependencies in text. Although these models achieve state-of-the-art performance for many language related tasks, their applicability outside of the neural language processing field has been minimal. In this work, we pr...
["Scientific Machine Learning", "Deep Learning", "Physics", "Surrogate Modeling", "Koopman", "Transformers", "Attention"]
ABSTRACTTransformers are widely used in neural language processing due to their ability tomodel longer-term dependencies in text. Although these models achieve state-of-the-art performance for many language related tasks, their applicability outside ofthe neural language processing field has been minimal. In this work,...
mYUIMKI32uA
A new application of transformer models (physical systems) with domain-specific embeddings
6: Marginally above acceptance threshold
**Summary**: The paper proposes applying transformer models to modeling physical systems. The state at each time step is embedded into a continuous vector using a pretrained encoder-decoder model based on Koopman’s theory. The experiments are performed on three physical systems and generally show that (1) a transformer...
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 Transformers for Modeling Physical Systems ### Paper Abstract Transformers are widely used in neural language processing due to their ability to model longer-term depend...
GPuvhWrEdUn
ICLR.cc/2021/Conference
2021
MixCon: Adjusting the Separability of Data Representations for Harder Data Recovery
["Xiaoxiao Li", "Yangsibo Huang", "Binghui Peng", "Zhao Song", "Kai Li"]
To address the issue that deep neural networks (DNNs) are vulnerable to model inversion attacks, we design an objective function to adjust the separability of the hidden data representations as a way to control the trade-off between data utility and vulnerability to inversion attacks. Our method is motivated by the the...
["Data Recovery", "Data Separability", "Distributed Deep Learning"]
ABSTRACTTo address the issue that deep neural networks (DNNs) are vulnerable to modelinversion attacks, we design an objective function, which adjusts the separabilityof the hidden data representations, as a way to control the trade-off between datautility and vulnerability to inversion attacks. Our method is motivated...
ttLRPm_iWhM
ICLR 2020 Conference Paper995 AnonReviewer
5: Marginally below acceptance threshold
Post response update Thanks for the response. However, my major concern is still that the technical contribution of this paper is limited. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- This...
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 MixCon: Adjusting the Separability of Data Representations for Harder Data Recovery ### Paper Abstract To address the issue that deep neural networks (DNNs) are vulnerab...
ZzwDy_wiWv
ICLR.cc/2021/Conference
2021
Knowledge distillation via softmax regression representation learning
["Jing Yang", "Brais Martinez", "Adrian Bulat", "Georgios Tzimiropoulos"]
This paper addresses the problem of model compression via knowledge distillation. We advocate for a method that optimizes the output feature of the penultimate layer of the student network and hence is directly related to representation learning. Previous distillation methods which typically impose direct feature match...
["teacher", "knowledge distillation", "student", "softmax regression representation", "representation learning", "classifier", "simple", "problem", "model compression", "output feature"]
ABSTRACTThis paper addresses the problem of model compression via knowledge distilla-tion. We advocate for a method that optimizes the output feature of the penulti-mate layer of the student network and hence is directly related to representationlearning. To this end, we firstly propose a direct feature matching approa...
dmw6KydW9KI
A simple yet effective form of loss for general knowledge distillation
7: Good paper, accept
This paper proposed a novel method for knowledge distillation. The idea is to utilize the teacher’s pre-trained classifier to train the student’s penultimate layer feature by adopting two losses: (a) the Feature Matching loss LFM and (b) the Softmax Regression loss LSR. The latter is designed to take into account the c...
5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Knowledge distillation via softmax regression representation learning ### Paper Abstract This paper addresses the problem of model compression via knowledge distillation...
Bklg1grtDr
ICLR.cc/2020/Conference
2020
Neural Design of Contests and All-Pay Auctions using Multi-Agent Simulation
["Thomas Anthony", "Ian Gemp", "Janos Kramar", "Tom Eccles", "Andrea Tacchetti", "Yoram Bachrach"]
We propose a multi-agent learning approach for designing crowdsourcing contests and all-pay auctions. Prizes in contests incentivise contestants to expend effort on their entries, with different prize allocations resulting in different incentives and bidding behaviors. In contrast to auctions designed manually by econo...
["Auctions", "Mechanism Design", "Multi-Agent", "Fictitious Play"]
ABSTRACTWe propose a multi-agent learning approach for designing crowdsourcing contestsand all-pay auctions. Prizes in contests incentivise contestants to expend effort ontheir entries, with different prize allocations resulting in different incentives andbidding behaviors. In contrast to auctions designed manually by ...
r1g0zMbRKS
Official Blind Review #2
3: Weak Reject
1. Summary The authors employ a multi-agent learning approach for learning how to set payoffs optimally for crowdsourcing contests and auctions. Optimality means e.g. incentive alignment (the principal problem) between the principal (e.g. the organizer) and participants (e.g. bidders), assuming e.g. that participants ...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Neural Design of Contests and All-Pay Auctions using Multi-Agent Simulation ### Paper Abstract We propose a multi-agent learning approach for designing crowdsourcing con...
JzG0n48hRf
ICLR.cc/2021/Conference
2021
Uncertainty for deep image classifiers on out of distribution data.
["Tiago Salvador", "Alexander Iannantuono", "Adam M Oberman"]
In addition to achieving high accuracy, in many applications, it is important to estimate the probability that a model prediction is correct. Predictive uncertainty is particularly important on out of distribution (OOD) data where accuracy degrades. However, models are typically overconfident, and model calibration ...
["uncertainty", "confidence", "out of distribution", "outlier exposure", "classification"]
ABSTRACTIn addition to achieving high accuracy, in many applications, it is important toestimate the probability that a model prediction is correct. Predictive uncertaintyis particularly important on out-of-distribution (OOD) data where accuracy de-grades. However, models are typically overconfident, and model calibrat...
Kppxu78l1q
Post-hoc calibration looks promising, but many open questions remain about applications and generalization
4: Ok but not good enough - rejection
In this work, the authors propose a post-hoc calibration method for potentially OOD data that relies on estimation of the "degree" of corruption for new test data. The rely on the benchmark provided in Ovadia et al. as a basis for their analysis. Ovadia et al. assessed common measurements of uncertainty such as Brier s...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Uncertainty for deep image classifiers on out of distribution data. ### Paper Abstract In addition to achieving high accuracy, in many applications, it is important to...
mCtadqIxOJ
ICLR.cc/2021/Conference
2021
Representing Partial Programs with Blended Abstract Semantics
["Maxwell Nye", "Yewen Pu", "Matthew Bowers", "Jacob Andreas", "Joshua B. Tenenbaum", "Armando Solar-Lezama"]
Synthesizing programs from examples requires searching over a vast, combinatorial space of possible programs. In this search process, a key challenge is representing the behavior of a partially written program before it can be executed, to judge if it is on the right track and predict where to search next. We introduce...
["program synthesis", "representation learning", "abstract interpretation", "modular neural networks"]
ABSTRACT SEMANTICSMaxwell NyeYewen Pu Matthew BowersJacob Andreas Joshua B. Tenenbaum Armando Solar-LezamaMassachusetts Institute of TechnologyABSTRACTSynthesizing programs from examples requires searching over a vast, combinato-rial space of possible programs. In this search process, a key challenge is repre-senting t...
5DyqmgA5jIg
Nice technique but incomplete semantics, and unclear scope and generalizability
6: Marginally above acceptance threshold
This paper proposes an embedding mechanism for partial programs for search space exploration in example-driven synthesis. It executes a sub-expression concretely whenever possible and applies neural module networks on vector representations otherwise. The embeddings of partial programs and goal states are used for dete...
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 Representing Partial Programs with Blended Abstract Semantics ### Paper Abstract Synthesizing programs from examples requires searching over a vast, combinatorial space ...
r5LS3fmh0t
ML_Reproducibility_Challenge/2021/Fall
2021
Thompson Sampling for Bandits with Clustered Arms
["Andra\u017e De Luisa"]
Scope of Reproducibility This report covers our reproduction of the paper `Thompson Sampling for Bandits with Clustered Arms' by Carlsson et al. (IJCAI 2021). The authors propose a new set of algorithms for the stochastic multi-armed bandit problem (and its contextual variant with linear expected rewards) in settings w...
["Multi-armed bandits", "Thompson sampling"]
Thompson Sampling for Bandits with Clustered ArmsReproducibility SummaryScope of ReproducibilityThis report covers our reproduction of the paper ‘Thompson Sampling for Bandits with Clustered Arms’ by Carlssonet al. (IJCAI 2021) [ 1]. The authors propose a new set of algorithms for the stochastic multi-armed bandit prob...
rBrlLzGrse9
great report
7: Good paper, accept
Reproducibility summary: The major findings are summarized in the summary. Scope of reproducibility: The scope is clearly stated, but it would be great to include more important information other than the background of the original paper. Code: The authors implemented everything from scratch. Communication with orig...
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 Thompson Sampling for Bandits with Clustered Arms ### Paper Abstract Scope of Reproducibility This report covers our reproduction of the paper `Thompson Sampling for Ban...
H1gN6kSFwS
ICLR.cc/2020/Conference
2020
Learning Neural Causal Models from Unknown Interventions
["Nan Rosemary Ke", "Olexa Bilaniuk", "Anirudh Goyal", "Stephan Bauer", "Hugol Larochelle", "Chris Pal", "Yoshua Bengio"]
Meta-learning over a set of distributions can be interpreted as learning different types of parameters corresponding to short-term vs long-term aspects of the mechanisms underlying the generation of data. These are respectively captured by quickly-changing \textit{parameters} and slowly-changing \textit{meta-parameters...
["deep learning", "graphical models", "meta learning"]
ABSTRACTMeta-learning over a set of distributions can be interpreted as learning differenttypes of parameters corresponding to short-term vs long-term aspects of the mech-anisms underlying the generation of data. These are respectively captured byquickly-changing parameters and slowly-changing meta-parameters . We pres...
HJxO6r5HYr
Official Blind Review #2
8: Accept
The paper develops a learning-based causal inference method that performs multiple tasks jointly: i) scalable discovery of the underlying Structured Causal Model (SCM) by modeling both structural assignments and the SCM as a continuously parameterized chain of distributions, ii) identification of the intervened var...
<|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 Neural Causal Models from Unknown Interventions ### Paper Abstract Meta-learning over a set of distributions can be interpreted as learning different types of p...
S1xim8UFuV
ICLR.cc/2019/Workshop/DeepGenStruct
2019
Point Cloud GAN
["Chun-Liang Li", "Manzil Zaheer", "Yang Zhang", "Barnab\u00e1s P\u00f3czos", "Ruslan Salakhutdinov"]
Generative Adversarial Networks (GAN) can achieve promising performance on learning complex data distributions on different types of data. In this paper, we first show that a straightforward extension of an existing GAN algorithm is not applicable to point clouds, because the constraint required for discriminators is u...
["Point Cloud", "GAN"]
ABSTRACTGenerative Adversarial Networks (GAN) can achieve promising performance onlearning complex data distributions on different types of data. In this paper, wefirst show that a straightforward extension of an existing GAN algorithm is notapplicable to point clouds, because the constraint required for discriminators...
rkl9XBCfcN
Interesting generative model for set data
4: Top 50% of accepted papers, clear accept
This paper proposes a new generative model for unordered data, with a particular application to point clouds. This model includes an inference method and and a novel objective function based on sandwiching the Wasserstein distance between an upper and lower bound. The paper includes a clear description of the problem ...
1: The reviewer's evaluation is an educated guess
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Point Cloud GAN ### Paper Abstract Generative Adversarial Networks (GAN) can achieve promising performance on learning complex data distributions on different types of d...
EzGiUyWYayS
NeurIPS.cc/2022/Workshop/GMML
2022
SecNet: Semantic Eye Completion in Implicit Field
["Yida Wang", "Yiru Shen", "David Joseph Tan", "Federico Tombari", "Sachin S. Talathi"]
If we take a depth image of an eye, noise artifacts and holes significantly affect the depth values on the eye due to the specularity of the sclera. This paper aims at solving this problem through semantic shape completion. We propose an end-to-end approach to train a neural network, called \emph{SecNet} (semantic eye...
["Eye completion", "Implicit Field", "Semantic Completion"]
Proceedings of Machine Learning Research 1:–, 2022 NeurIPS 2022 Gaze Meets ML WorkshopSecNet: Semantic Eye Completion in Implicit FieldYida Wang∗yida@fb.comYiru Shen†shenyirustar@gmail.comDavid Joseph Tan djtan@google.comFederico Tombari tombari@in.tum.deSachin Talathi stalathi@fb.comEditor: Editor’s nameAbstractIf we ...
8_iaA82GNWR
A novel take on Gaze estimation
9: Top 15% of accepted papers, strong accept
The authors present a novel approach to reconstructing the complete the eye region from a noisy partial eye scan. They describe the short comings in existing RGB based approaches and point out that 3D semantic surface modeling can lead to a more practical use-case, i.e. gaze estimation. The authors describe their metho...
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 SecNet: Semantic Eye Completion in Implicit Field ### Paper Abstract If we take a depth image of an eye, noise artifacts and holes significantly affect the depth values ...
rJgQkT4twH
ICLR.cc/2020/Conference
2020
Analysis of Video Feature Learning in Two-Stream CNNs on the Example of Zebrafish Swim Bout Classification
["Bennet Breier", "Arno Onken"]
Semmelhack et al. (2014) have achieved high classification accuracy in distinguishing swim bouts of zebrafish using a Support Vector Machine (SVM). Convolutional Neural Networks (CNNs) have reached superior performance in various image recognition tasks over SVMs, but these powerful networks remain a black box. Reachin...
["convolutional neural networks", "neural network transparency", "AI explainability", "deep Taylor decomposition", "supervised classification", "zebrafish", "transparency", "behavioral research", "optical flow"]
ABSTRACTSemmelhack et al. (2014) have achieved high classification accuracy in distin-guishing swim bouts of zebrafish using a Support Vector Machine (SVM). Con-volutional Neural Networks (CNNs) have reached superior performance in variousimage recognition tasks over SVMs, but their learnt features are not immediatelyv...
SkxFdq2Y9r
Official Blind Review #4
6: Weak Accept
The paper presents a case study of training a video classifier using convolutional networks, and on how the learned features related to previous, hand-designed ones. The particular domain considered is of importance for biologists/medical doctors/neuroscientists: zebra fish swim bout classification. In order to identi...
<|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 Video Feature Learning in Two-Stream CNNs on the Example of Zebrafish Swim Bout Classification ### Paper Abstract Semmelhack et al. (2014) have achieved high...
IkYEJ5Cps5H
ICLR.cc/2021/Conference
2021
Succinct Network Channel and Spatial Pruning via Discrete Variable QCQP
["Yeonwoo Jeong", "Deokjae Lee", "Gaon An", "Changyong Son", "Hyun Oh Song"]
Reducing the heavy computational cost of large convolutional neural networks is crucial when deploying the networks to resource-constrained environments. In this context, recent works propose channel pruning via greedy channel selection to achieve practical acceleration and memory footprint reduction. We first show...
["Network Pruning", "Channel pruning", "Spatial pruning", "Network Compression", "MIQCQP", "Specified target resource constraint"]
ABSTRACTReducing the heavy computational cost of large convolutional neural networks iscrucial when deploying the networks to resource-constrained environments. Inthis context, recent works propose channel pruning via greedy channel selection toachieve practical acceleration and memory footprint reduction. We first sho...
g2cLyBv_lV7
Good submission focusing on a valuable topic
7: Good paper, accept
The authors proposed a pruning method that aims to reduce the parameters and heavy computational cost of large convolutional neural networks (CNNs). According to the comparison experiments performed on several widely-used network fashions, the proposed strategy could help to efficiently reduce the numbers of parameters...
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 Succinct Network Channel and Spatial Pruning via Discrete Variable QCQP ### Paper Abstract Reducing the heavy computational cost of large convolutional neural networ...
BkgosiRcKm
ICLR.cc/2019/Conference
2019
Deep Recurrent Gaussian Process with Variational Sparse Spectrum Approximation
["Roman F\u00f6ll", "Bernard Haasdonk", "Markus Hanselmann", "Holger Ulmer"]
Modeling sequential data has become more and more important in practice. Some applications are autonomous driving, virtual sensors and weather forecasting. To model such systems, so called recurrent models are frequently used. In this paper we introduce several new Deep Recurrent Gaussian Process (DRGP) models based on...
["Deep Gaussian Process Model", "Recurrent Model", "State-Space Model", "Nonlinear system identification", "Dynamical modeling"]
ABSTRACTModeling sequential data has become more and more important in practice. Someapplications are autonomous driving, virtual sensors and weather forecasting. Tomodel such systems, so called recurrent models are frequently used. In this pa-per we introduce several new Deep Recurrent Gaussian Process (DRGP) modelsba...
rJgI340tn7
Combination of known ideas, hard to read
5: Marginally below acceptance threshold
This paper proposes deep recurrent GP models based on the existing DRGP framework, two works on sparse spectrum approximation as well as that of inducing points. In these models, uncertainty is propagated by marginalizing out the hidden inputs at every layer. The authors have combined a series of known ideas in the pr...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Deep Recurrent Gaussian Process with Variational Sparse Spectrum Approximation ### Paper Abstract Modeling sequential data has become more and more important in practice...
LXMSvPmsm0g
ICLR.cc/2021/Conference
2021
Long Live the Lottery: The Existence of Winning Tickets in Lifelong Learning
["Tianlong Chen", "Zhenyu Zhang", "Sijia Liu", "Shiyu Chang", "Zhangyang Wang"]
The lottery ticket hypothesis states that a highly sparsified sub-network can be trained in isolation, given the appropriate weight initialization. This paper extends that hypothesis from one-shot task learning, and demonstrates for the first time that such extremely compact and independently trainable sub-networks can...
["lottery tickets", "winning tickets", "lifelong learning"]
ABSTRACTThe lottery ticket hypothesis states that a highly sparsified sub-network can betrained in isolation, given the appropriate weight initialization. This paper ex-tends that hypothesis from one-shot task learning, and demonstrates for the firsttime that such extremely compact and independently trainable sub-netwo...
evHg6vPdZ6q
Good contributions on the lottery ticket hypothesis to life-long learning with a few open questions
8: Top 50% of accepted papers, clear accept
[EDIT AFTER DISCUSSIONS] I thank the authors for their answers to my comments. Having read the various threads, I confirm my score and see interesting work in this paper. [/EDIT] ########################################################################## Summary: this paper extends the lottery ticket hypothesis to life...
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 Long Live the Lottery: The Existence of Winning Tickets in Lifelong Learning ### Paper Abstract The lottery ticket hypothesis states that a highly sparsified sub-network...
HyiAuyb0b
ICLR.cc/2018/Conference
2018
TD or not TD: Analyzing the Role of Temporal Differencing in Deep Reinforcement Learning
["Artemij Amiranashvili", "Alexey Dosovitskiy", "Vladlen Koltun", "Thomas Brox"]
Our understanding of reinforcement learning (RL) has been shaped by theoretical and empirical results that were obtained decades ago using tabular representations and linear function approximators. These results suggest that RL methods that use temporal differencing (TD) are superior to direct Monte Carlo estimation (M...
["deep learning", "reinforcement learning", "temporal difference"]
ABSTRACTOur understanding of reinforcement learning (RL) has been shaped by theoreticaland empirical results that were obtained decades ago using tabular representationsand linear function approximators. These results suggest that RL methods that usetemporal differencing (TD) are superior to direct Monte Carlo estimati...
Sku18wYgz
A thoughtful and well executed investigation into deep-RL techniques and challenges for robust approaches, including the relative benefits of TD versus MC.
7: Good paper, accept
The authors present a testing framework for deep RL methods in which difficulty can be controlled along a number of dimensions, including: reward delay, reward sparsity, episode length with terminating rewards, binary vs real rewards and perceptual complexity. The authors then experiment with a variety of TD and MC bas...
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 TD or not TD: Analyzing the Role of Temporal Differencing in Deep Reinforcement Learning ### Paper Abstract Our understanding of reinforcement learning (RL) has been sha...
HyfyN30qt7
ICLR.cc/2019/Conference
2019
NICE: noise injection and clamping estimation for neural network quantization
["Chaim Baskin", "Natan Liss", "Yoav Chai", "Evgenii Zheltonozhskii", "Eli Schwartz", "Raja Girayes", "Avi Mendelson", "Alexander M.Bronstein"]
Convolutional Neural Networks (CNN) are very popular in many fields including computer vision, speech recognition, natural language processing, to name a few. Though deep learning leads to groundbreaking performance in these domains, the networks used are very demanding computationally and are far from real-time even o...
["Efficient inference", "Hardware-efficient model architectures", "Quantization"]
Under review as a conference paper at ICLR 2019N I C E : N O I S E I N J E C T I O N A N D C L A M P I N G E S T I M A -T I O N F O R N E U R A L N E T W O R K Q U A N T I Z AT I O NAnonymous authorsPaper under double-blind reviewAB S T R A C TConvolutional Neural Networks (CNN) are very popular in many fields includin...
B1xOIQXcnm
Low novelty and weak experiments.
4: Ok but not good enough - rejection
[Summary] Neural network quantization is can enable many practical applications for deep learning, therefore it is an important research problem. The paper claims two contributions: 1. Injecting noise during training to make it more robust to quantization errors. 2. Clamping the parameter values in a layer as well as 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 NICE: noise injection and clamping estimation for neural network quantization ### Paper Abstract Convolutional Neural Networks (CNN) are very popular in many fields incl...
rk8wKk-R-
ICLR.cc/2018/Conference
2018
Convolutional Sequence Modeling Revisited
["Shaojie Bai", "J. Zico Kolter", "Vladlen Koltun"]
This paper revisits the problem of sequence modeling using convolutional architectures. Although both convolutional and recurrent architectures have a long history in sequence prediction, the current "default" mindset in much of the deep learning community is that generic sequence modeling is best handled using recur...
["Temporal Convolutional Network", "Sequence Modeling", "Deep Learning"]
ABSTRACTThis paper revisits the problem of sequence modeling using convolutional archi-tectures. Although both convolutional and recurrent architectures have a longhistory in sequence prediction, the current “default” mindset in much of the deeplearning community is that generic sequence modeling is best handled using ...
HkTNLM5gM
Convolutional networks are good for solving sequence tasks
4: Ok but not good enough - rejection
This paper argues that convolutional networks should be the default approach for sequence modeling. The paper is nicely done and rather easy to understand. Nevertheless, I find it difficult to assess its significance. In order to support the original hypothesis, I think that a much larger and more diverse set of exper...
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 Convolutional Sequence Modeling Revisited ### Paper Abstract This paper revisits the problem of sequence modeling using convolutional architectures. Although both conv...
COq9CMgo_Xd
eswc-conferences.org/ESWC/2021/Conference/Industry_Track
2021
A Virtual Knowledge Graph for Enabling Defect Traceability and Customer Service Analytics
["Nico Wilhelm", "Diego Collarana", "Jens Lehmann"]
In this paper, we showcase the implementation of a semantic information model and a virtual knowledge graph at ZF Friedrichshafen AG company, with two main goals in mind: 1) integration of heterogeneous data sources following a pay-as-you-go approach; and the 2) combination core domain concepts from ZF’s production lin...
["Virtual Knowledge Graph", "Semantic Data Model", "Data Integration"]
A Virtual Knowledge Graph for Enabling DefectTraceability and Customer Service AnalyticsNico Wilhelm1(B), Diego Collarana2, Jens Lehmann21ZF Friedrichshafen AGnico.wilhelm@zf.com2Fraunhofer IAIS and University of Bonn, Germany{diego.collarana.vargas|jens.lehmann}@iais.fraunhofer.deAbstract. In this paper, we showcase t...
RHtGJj4GSL
Experience using knowledge graphs for data integration in industry
9: Top 15% of accepted papers, strong accept
This paper presents the implementation of a semantic model and a virtual graph evaluated for two use cases at ZF Friedrichshafen AG company. This solution aimed at integrating heterogeneous data sources and combination of domain concepts with internal metadata-data. The paper is very well written and clear. The author...
4: The reviewer is confident but not absolutely certain that the evaluation is correct
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title A Virtual Knowledge Graph for Enabling Defect Traceability and Customer Service Analytics ### Paper Abstract In this paper, we showcase the implementation of a semantic ...
HJxt76nmcV
icaps-conference.org/ICAPS/2019/Workshop/XAIP
2019
Explaining the Space of Plans through Plan-Property Dependencies
["Anonymous"]
A key problem in explainable AI planning is to elucidate decision rationales. User questions in this context are often contrastive, taking the form “Why do A rather than B?”. Answering such a question requires a statement about the space of possible plans. We propose to do so through plan-property dependencies, where p...
["dependencies", "space", "plans", "question", "key problem", "explainable ai planning", "decision rationales", "user questions", "context", "contrastive"]
Explaining the Space of Plans through Plan-Property DependenciesRebecca Eifler,1Michael Cashmore,2J ̈org Hoffmann,1Daniele Magazzeni,2Marcel Steinmetz,11Saarland University, Saarland Informatics Campus, Saarbr ̈ucken, Germany,2King’s College London, Department of Informatics, London, UK,flastnameg@cs.uni-saarland.de, f...
r1lQVyIAoE
Provides a nice conceptual framework for contrastive explanation
4: Top 50% of accepted papers, clear accept
This paper describes a general approach to contrastive explanation using plan-property dependencies. The key concept in this paper is plan-property depenency order, in which dependencies between properties over a set of plans are related. This provides a nice conceptual framework for asking contrastive questions over d...
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 Explaining the Space of Plans through Plan-Property Dependencies ### Paper Abstract A key problem in explainable AI planning is to elucidate decision rationales. User qu...
Ogga20D2HO-
ICLR.cc/2021/Conference
2021
FedMix: Approximation of Mixup under Mean Augmented Federated Learning
["Tehrim Yoon", "Sumin Shin", "Sung Ju Hwang", "Eunho Yang"]
Federated learning (FL) allows edge devices to collectively learn a model without directly sharing data within each device, thus preserving privacy and eliminating the need to store data globally. While there are promising results under the assumption of independent and identically distributed (iid) local data, current...
["federated learning", "mixup"]
ABSTRACTFederated learning (FL) allows edge devices to collectively learn a model with-out directly sharing data within each device, thus preserving privacy and elim-inating the need to store data globally. While there are promising results underthe assumption of independent and identically distributed (iid) local data...
9qWbfatB0VE
A novel method, but the experimental justification can be improved
6: Marginally above acceptance threshold
########################################################################## Summary:   The paper proposed MAFL, a novel approach to conduct Mixup under the federated learning setting whiling preserving data privacy. The proposed FedMix scheme is inspired by Taylor’s expansion of the global Mixup formulation. The effect...
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 FedMix: Approximation of Mixup under Mean Augmented Federated Learning ### Paper Abstract Federated learning (FL) allows edge devices to collectively learn a model witho...
mdiNjHHdoQ7
ecmlpkdd.org/ECMLPKDD/2021/Workshop/TeachML
2021
Developing Open Source Educational Resources for Machine Learning and Data Science
["Anonymous"]
Education should not be a privilege but a common good. It should be openly accessible to everyone, with as few barriers as possible; even more so for key technologies such as Machine Learning (ML) and Data Science (DS). Open Educational Resources (OER) are a crucial factor for greater educational equity. In this paper,...
["OER", "OSER", "Open Teaching", "Machine Learning", "Data Science", "MOOC", "Blended Learning"]
000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044045046047048049050051052053054Developing Open Source Educational Resourcesfor Machine Learning and Data ScienceAnonymous Authors1AbstractEducation should not be a privilege but a common...
8SLsx79m2w6
Problem well described but no new insights
4: Ok but not good enough - rejection
The paper "Developing Open Source Educational Resources for Machine Learning and Data Science" is well structured and easy to read. The authors describe a general approach to make OER materials qualitatively and quantitatively more usable for the ML and DS community. Thereby, the argumentation is followed that, beside...
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 Developing Open Source Educational Resources for Machine Learning and Data Science ### Paper Abstract Education should not be a privilege but a common good. It should be...
HyeG9lHYwH
ICLR.cc/2020/Conference
2020
Compression without Quantization
["Gergely Flamich", "Marton Havasi", "Jos\u00e9 Miguel Hern\u00e1ndez-Lobato"]
Standard compression algorithms work by mapping an image to discrete code using an encoder from which the original image can be reconstructed through a decoder. This process, due to the quantization step, is inherently non-differentiable so these algorithms must rely on approximate methods to train the encoder and dec...
["Image Compression", "Bits-back efficient", "Quantization"]
ABSTRACTStandard compression algorithms work by mapping an image to discrete code us-ing an encoder from which the original image can be reconstructed through a de-coder. This process, due to the quantization step, is inherently non-differentiableso these algorithms must rely on approximate methods to train the encoder...
SkxaSBcaFH
Official Blind Review #3
3: Weak Reject
Summary: This paper aims to circumvent the quantization step and associated gradient approximations for compression algorithms that make use of entropy coding for compression. Entropy coding requires a probability mass function over discrete symbols. As an alternative approach the authors adapt the MIRACLE algorithm by...
<|im_start|>system You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|> <|im_start|>user ### Paper Title Compression without Quantization ### Paper Abstract Standard compression algorithms work by mapping an image to discrete code using an encoder from which the original i...