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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ryevtyHtPr | ICLR.cc/2020/Conference | 2020 | Do Deep Neural Networks for Segmentation Understand Insideness? | ["Kimberly M Villalobos", "Vilim Stih", "Amineh Ahmadinejad", "Jamell Dozier", "Andrew Francl", "Frederico Azevedo", "Tomotake Sasaki", "Xavier Boix"] | Image segmentation aims at grouping pixels that belong to the same object or region. At the heart of image segmentation lies the problem of determining whether a pixel is inside or outside a region, which we denote as the "insideness" problem. Many Deep Neural Networks (DNNs) variants excel in segmentation benchmarks, ... | ["Image Segmentation", "Deep Networks for Spatial Relationships", "Visual Routines", "Recurrent Neural Networks"] | ABSTRACTImage segmentation aims at grouping pixels that belong to the same object or re-gion. At the heart of image segmentation lies the problem of determining whethera pixel is inside or outside a region, which we denote as the “insideness” problem.Many Deep Neural Networks (DNNs) variants excel in segmentation bench... | HyeXzhdjYS | Official Blind Review #2 | 6: Weak Accept | This submission introduces a new concept, termed insideness, to study semantic segmentation in deep learning era. The authors raise many interesting questions, such as (1) Does deep neural networks (DNN) understand insideness? (2) What representations do DNNs use to address the long-range relationships of insideness? (... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Do Deep Neural Networks for Segmentation Understand Insideness?
### Paper Abstract
Image segmentation aims at grouping pixels that belong to the same object or region. A... | |
ShGeRZBcp19 | humanrobotinteraction.org/HRI/2022/Workshop/VAM-HRI | 2022 | Head Pose as a Proxy for Gaze in Virtual Reality | ["Padraig Higgins", "Ryan Barron", "Cynthia Matuszek"] | In this work, we posit that a user's head pose can serve as a proxy for gaze in a VR object selection task. We describe a study in which participants were asked to describe a series of objects in a known order, providing approximate labels for the focus of attention. The participants' head pose was then evaluated as a ... | ["sim2real", "gaze tracking", "data generation", "virtual reality"] | Head Pose as a Proxy for Gaze in Virtual RealityPadraig HigginsCSEEUniv. of MD, Baltimore Countyphiggin1@umbc.eduRyan BarronCSEEUniv. of MD, Baltimore Countyryanb4@umbc.eduCynthia MatuszekCSEEUniv. of MD, Baltimore Countycmat@umbc.eduAbstract —In this work, we posit that a user’s head pose canserve as a proxy for gaze ... | BlejnkeVxq | Great Data Collection Idea / May Benefit from Some Robotics Framing | 8: Top 50% of accepted papers, clear accept | Summary:
This work discusses a data collection study (n=9) for measuring head pose as a proxy for eye gaze in a robot natural language grounding task. The study collected many types of data (well described within the paper), labelled times participants stared at/described an object, and measured cosine similarity of th... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Head Pose as a Proxy for Gaze in Virtual Reality
### Paper Abstract
In this work, we posit that a user's head pose can serve as a proxy for gaze in a VR object selection... |
DX-XHq9_Pa | NoDaLiDa/2023/Conference | 2023 | Machine Translation for Low-resource Finno-Ugric Languages | ["Lisa Yankovskaya", "Maali Tars", "Andre T\u00e4ttar", "Mark Fishel"] | This paper focuses on neural machine translation (NMT) for low-resource Finno-Ugric languages. Our contributions are three-fold: (1) we extend existing and collect new parallel and monolingual corpora for 20 languages, (2) we expand the 200-language translation benchmark FLORES-200 with manual translations into nine ne... | ["Finno-ugric", "multilingual", "machine translation"] | Machine Translation for Low-resource Finno-Ugric LanguagesLisa Yankovskaya Maali Tars Andre T ̈attar Mark FishelInstitute of Computer ScienceUniversity of Tartu, Estonia{lisa.yankovskaya,maali.tars,andre.tattar,mark.fishel }@ut.eeAbstractThis paper focuses on neural machinetranslation (NMT) for low-resourceFinno-Ugric ... | gzHNzwG_y0 | The paper describes existing and newly created (within this work) datasets for low-resource machine translation of Finno-Ugric languages and back-translation experiments, however it misses to answer one of the questions it raises (how much data is needed for low-resource back-translation). | 6: Marginally above acceptance threshold | The paper discusses parallel data collection for Finno-Ugric languages and experiments with back-translation.
The paper has the following contributions: 1) a new parallel dataset covering 20 languages, 2) translations of 250 sentences of the FLORES 200 dataset into nine languages, 3) experiments showing what translati... | 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 Translation for Low-resource Finno-Ugric Languages
### Paper Abstract
This paper focuses on neural machine translation (NMT) for low-resource Finno-Ugric languag... |
B1fA3oActQ | ICLR.cc/2019/Conference | 2019 | GraphSeq2Seq: Graph-Sequence-to-Sequence for Neural Machine Translation | ["Guoshuai Zhao", "Jun Li", "Lu Wang", "Xueming Qian", "Yun Fu"] | Sequence-to-Sequence (Seq2Seq) neural models have become popular for text generation problems, e.g. neural machine translation (NMT) (Bahdanau et al.,2014; Britz et al., 2017), text summarization (Nallapati et al., 2017; Wang &Ling, 2016), and image captioning (Venugopalan et al., 2015; Liu et al... | ["Neural Machine Translation", "Natural Language Generation", "Graph Embedding", "LSTM"] | ABSTRACTSequence-to-Sequence (Seq2Seq) neural models have become popular for textgeneration problems, e.g. neural machine translation (NMT) (Bahdanau et al.,2014; Britz et al., 2017), text summarization (Nallapati et al., 2017; Wang &Ling, 2016), and image captioning (Venugopalan et al., 2015; Liu et al., 2017).Though ... | r1egbGCth7 | Review of "GraphSeq2Seq: Graph-Sequence-to-Sequence for Neural Machine Translation" | 6: Marginally above acceptance threshold | This paper proposes a method for combining the Graph2Seq and Seq2Seq models into a unified model that captures the benefits of both. The paper thoroughly describes in series of experiments that demonstrate that the authors' proposed method outperforms several of the other NMT methods on a few translation tasks.
I lik... | 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
GraphSeq2Seq: Graph-Sequence-to-Sequence for Neural Machine Translation
### Paper Abstract
Sequence-to-Sequence (Seq2Seq) neural models have become popular for t... |
ouVzq2kkUiv | MICCAI.org/2021/Workshop/COMPAY | 2021 | VTBIS: Vision Transformer for Biomedical Image Segmentation | ["Anonymous"] | In this paper, we propose a novel network named Vision
Transformer for Biomedical Image Segmentation (VTBIS).
Our network splits the input feature maps into three parts
with 1 × 1, 3 × 3 and 5 × 5 convolutions in both encoder
and decoder. Concat operator is used to merge the features
before being fed to three consecuti... | ["transformer", "brain tumour segmentation"] | 00000100200300400500600700800901001101201301401501601701801902002102202302402502602702802903003103203303403503603703803904004104204304404504604704804905005105205305405505605705805906006106206306406506606706806907007107207307407507607707807908008108208308408508608708808909009109209309409509609709809910010110210310410510... | q0zYx5BQRD | Biomedical imaging but not digital pathology paper; out of scope | 3: Clear rejection | This is a fair paper but I do not two main issues for consideration in the COMPAY workshop. First, this paper is not about digital pathology and as such not in scope. Second, the method section is very short and the methods can certainly not be reproduced from the textual description. It is not entirely clear what is n... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
VTBIS: Vision Transformer for Biomedical Image Segmentation
### Paper Abstract
In this paper, we propose a novel network named Vision Transformer for Biomedical Image Se... |
N01OapqPhJl | Interspeech.org/2023/Workshop/SSW | 2023 | Situating Speech Synthesis: Investigating Contextual Factors in the Evaluation of Conversational TTS | ["Harm Lameris", "Ambika Kirkland", "Joakim Gustafson", "Eva Szekely"] | Speech synthesis evaluation methods have lagged behind the development of TTS systems, with single sentence read-speech MOS naturalness evaluation on crowdsourcing platforms being the industry standard. For TTS to successfully be applied in social contexts, evaluation methods need to be socially embedded in the situati... | ["speech synthesis", "text to speech", "evaluation", "social", "context"] | Situating Speech Synthesis: Investigating Contextual Factors in the Evaluationof Conversational TTSHarm Lameris, Ambika Kirkland, Joakim Gustafson, ́Eva Sz ́ekelyDivision of Speech, Music and Hearing, KTH Royal Institute of Technology, Stockholm, Sweden{lameris, kirkland, jkgu, szekely }@kth.seAbstractSpeech synthesis ... | 8LcabWI6u0z | SSW23 review for submission 29 | 7: Good paper, accept | The paper presents an experiment to assess the merit of including contextual factors in the evaluation of conversational TTS. The paper tries whether (a) a textual description of the context or (b) a preceding vocoded natural utterance as context or (c) both are useful to evaluate the suitability of TTS voices in addit... | 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
Situating Speech Synthesis: Investigating Contextual Factors in the Evaluation of Conversational TTS
### Paper Abstract
Speech synthesis evaluation methods have lagged b... |
Byl8BnRcYm | ICLR.cc/2019/Conference | 2019 | Capsule Graph Neural Network | ["Zhang Xinyi", "Lihui Chen"] | The high-quality node embeddings learned from the Graph Neural Networks (GNNs) have been applied to a wide range of node-based applications and some of them have achieved state-of-the-art (SOTA) performance. However, when applying node embeddings learned from GNNs to generate graph embeddings, the scalar node represent... | ["CapsNet", "Graph embedding", "GNN"] | ABSTRACTThe high-quality node embeddings learned from the Graph Neural Networks(GNNs) have been applied to a wide range of node-based applications and someof them have achieved state-of-the-art (SOTA) performance. However, when ap-plying node embeddings learned from GNNs to generate graph embeddings, thescalar node rep... | S1gMPXrcn7 | Capsule networks for graphs without convincing motivation and experimental evaluation | 6: Marginally above acceptance threshold | The authors provide an architecture that applies recent advances in the field of capsule networks in the graph neural network domain. First, hierarchical node level capsules are extracted using GCN layers. Second, after weighting each capsule by the output of a proposed attention module, graph level capsules are comput... | 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
Capsule Graph Neural Network
### Paper Abstract
The high-quality node embeddings learned from the Graph Neural Networks (GNNs) have been applied to a wide range of node-... |
csHPh2kuVP | NeurIPS.cc/2021/Workshop/DLDE | 2021 | Actor-Critic Algorithm for High-dimensional PDEs | ["Xiaohan Zhang"] | We develop a machine learning model to effectively solve high-dimensional nonlinear parabolic partial differential equations (PDE). We use Feynman-Kac formula to reformulate PDE into the equivalent stochastic control problem governed by a Backward Stochastic Differential Equation (BSDE) system. Our model is designed to... | ["Partial Differential Equations", "Reinforcement learning", "Actor-critic"] | Actor-Critic Algorithm for High-dimensional PDEsAnonymous Author(s)AffiliationAddressemailAbstractWe develop a deep neural network model to solve high-dimensional nonlinear 1parabolic partial differential equations (PDE). Our model extends and advances 2the DBSDE model in the following 3 aspects: 1) the trainable param... | x7JGm9GE-Ev | Interesting idea, but unclear research gap | Summary
The paper proposes a neural network-based model to learn solutions to parabolic backward stochastic differential equations. The proposed work claims to decrease the number of trainable parameters, hyperparameters, and convergence rate. While the general approach seems promising, the paper does not seem to artic... | 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
Actor-Critic Algorithm for High-dimensional PDEs
### Paper Abstract
We develop a machine learning model to effectively solve high-dimensional nonlinear parabolic partial... | |
dK8vOIBENa3 | logconference.io/LOG/2022/Conference | 2022 | Transductive Linear Probing: A Novel Framework for Few-Shot Node Classification | ["Zhen Tan", "Song Wang", "Kaize Ding", "Jundong Li", "huan liu"] | Few-shot node classification is tasked to provide accurate predictions for nodes from novel classes with only few representative labeled nodes. This problem has drawn tremendous attention for its projection to prevailing real-world applications, such as product categorization for newly added commodity categories on an ... | ["node classification", "few-shot learning", "graph neural networks", "graph contrastive learning"] | Transductive Linear Probing: A Novel Frameworkfor Few-Shot Node ClassificationZhen Tan∗Arizona State Universityztan36@asu.eduSong Wang∗University of Virginiasw3wv@virginia.eduKaize Ding∗Arizona State Universitykding9@asu.eduJundong LiUniversity of Virginiajundong@virginia.eduHuan LiuArizona State Universityhuanliu@asu.... | fUYGZ9rsIfD | **Summary** -
The authors propose a new solution to the problem of few-shot node classification (FSNC). They deviate from the recent SOTA literature on using meta-learning based approaches which have been shown to work well across multiple modalities. They highlight two particular caveats of meta-learning based approa... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Transductive Linear Probing: A Novel Framework for Few-Shot Node Classification
### Paper Abstract
Few-shot node classification is tasked to provide accurate predictions... | |||
S1fcnoR9K7 | ICLR.cc/2019/Conference | 2019 | Learning with Random Learning Rates. | ["L\u00e9onard Blier", "Pierre Wolinski", "Yann Ollivier"] | Hyperparameter tuning is a bothersome step in the training of deep learning mod- els. One of the most sensitive hyperparameters is the learning rate of the gradient descent. We present the All Learning Rates At Once (Alrao) optimization method for neural networks: each unit or feature in the network gets its own learni... | ["step size", "stochastic gradient descent", "hyperparameter tuning"] | ABSTRACTHyperparameter tuning is a bothersome step in the training of deep learning mod-els. One of the most sensitive hyperparameters is the learning rate of the gradientdescent. We present the All Learning Rates At Once (Alrao) optimization methodfor neural networks: each unit or feature in the network gets its own l... | HyeBGxI52Q | not convinced it's worth it | 4: Ok but not good enough - rejection | POST REBUTTAL: I think the paper is decent, there are some significant downsides to the method but it could constitute a first step towards a more mature learning-rate-free method. However, in its current state the paper is left with some gaping holes in its experiment section. The authors tried to add experiments on I... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning with Random Learning Rates.
### Paper Abstract
Hyperparameter tuning is a bothersome step in the training of deep learning mod- els. One of the most sensitive h... |
BJgqQ6NYvB | ICLR.cc/2020/Conference | 2020 | FasterSeg: Searching for Faster Real-time Semantic Segmentation | ["Wuyang Chen", "Xinyu Gong", "Xianming Liu", "Qian Zhang", "Yuan Li", "Zhangyang Wang"] | We present FasterSeg, an automatically designed semantic segmentation network with not only state-of-the-art performance but also faster speed than current methods. Utilizing neural architecture search (NAS), FasterSeg is discovered from a novel and broader search space integrating multi-resolution branches, that has b... | ["neural architecture search", "real-time", "segmentation"] | ABSTRACTWe present FasterSeg , an automatically designed semantic segmentation networkwith not only state-of-the-art performance but also faster speed than current meth-ods. Utilizing neural architecture search (NAS), FasterSeg is discovered froma novel and broader search space integrating multi-resolution branches, th... | SkevyWznYr | Official Blind Review #3 | 8: Accept | Summary:
- key problem: neural architecture search (NAS) to improve both accuracy and runtime efficiency of deep nets for semantic segmentation;
- contributions: 1) a novel NAS search space leveraging multi-resolution branches, efficient operators ("zoomed convolutions"), and parametrized expansion ratios, 2) a decompo... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
FasterSeg: Searching for Faster Real-time Semantic Segmentation
### Paper Abstract
We present FasterSeg, an automatically designed semantic segmentation network with not... | |
S1E3Ko09F7 | ICLR.cc/2019/Conference | 2019 | L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data | ["Jianbo Chen", "Le Song", "Martin J. Wainwright", "Michael I. Jordan"] | Instancewise feature scoring is a method for model interpretation, which yields, for each test instance, a vector of importance scores associated with features. Methods based on the Shapley score have been proposed as a fair way of computing feature attributions, but incur an exponential complexity in the number of fea... | ["Model Interpretation", "Feature Selection"] | ABSTRACTInstancewise feature scoring is a method for model interpretation, which yields,for each test instance, a vector of importance scores associated with features.Methods based on the Shapley score have been proposed as a fair way of com-puting feature attributions, but incur an exponential complexity in the number... | SyxgP3dn2Q | A new method for computing Shapely values | 7: Good paper, accept | This paper proposes two methods for instance-wise feature importance scoring, which is the task of ranking the importance of each feature in a particular example (in contrast to class-wise or overall feature importance). The approach uses Shapely values, which are a principled way of measuring the contribution of a fe... | 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
L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data
### Paper Abstract
Instancewise feature scoring is a method for model interpretation, which y... |
cy0jU8F60Hy | ICLR.cc/2021/Conference | 2021 | ACT: Asymptotic Conditional Transport | ["Huangjie Zheng", "Mingyuan Zhou"] | We propose conditional transport (CT) as a new divergence to measure the difference between two probability distributions. The CT divergence consists of the expected cost of a forward CT, which constructs a navigator to stochastically transport a data point of one distribution to the other distribution, and that of a b... | ["Statistical distance", "Divergence", "Optimal Transport", "Implicit Distribution", "Deep Generative Models", "GANs"] | ABSTRACTWe propose conditional transport (CT) as a new divergence to measure the dif-ference between two probability distributions. The CT divergence consists of theexpected cost of a forward CT, which constructs a navigator to stochastically trans-port a data point of one distribution to the other distribution, and th... | v6Ke_-8VcY | A new divergence between probability distributions and its potential application in GANs (replacing Wasserstein) | 6: Marginally above acceptance threshold | The paper proposes a new transport-based divergence between distributions (CT) and a variant for empirical distributions (ACT). The new divergence is claimed to be more suitable for learning deep generative models than existing divergences like KL, JS (as in the vanilla GAN) and Wasserstein (as used in WGAN and its var... | 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
ACT: Asymptotic Conditional Transport
### Paper Abstract
We propose conditional transport (CT) as a new divergence to measure the difference between two probability dist... |
ryjp1c9xg | ICLR.cc/2017/conference | 2017 | Extensions and Limitations of the Neural GPU | ["Eric Price", "Wojciech Zaremba", "Ilya Sutskever"] | The Neural GPU is a recent model that can learn algorithms such as multi-digit binary addition and binary multiplication in a way that generalizes to inputs of arbitrary length. We show that there are two simple ways of improving the performance of the Neural GPU: by carefully designing a curriculum, and by increasing... | ["neural gpu", "long numbers", "limitations", "able", "failure modes", "extensions", "neural gpu extensions", "recent model", "binary addition", "binary multiplication"] | ABSTRACTThe Neural GPU is a recent model that can learn algorithms such as multi-digitbinary addition and binary multiplication in a way that generalizes to inputs ofarbitrary length. We show that there are two simple ways of improving the perfor-mance of the Neural GPU: by carefully designing a curriculum, and by incr... | ry8aXlGEg | 4: Ok but not good enough - rejection | The authors investigate the neural GPU model introduced by Kaiser and Sutskever. In section 3 they claim its performance is due to the O(n^2) number of steps it can perform for each example. In the subsequent section they highlight the importance of curriculum training and empirically show that larger models generalize... | 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
Extensions and Limitations of the Neural GPU
### Paper Abstract
The Neural GPU is a recent model that can learn algorithms such as multi-digit binary addition and binary... | |
YXUV-SU6i8I | graphicsinterface.org/Graphics_Interface/2021/Conference | 2021 | Modeling One-Dimensional Touch Pointing with Nominal Target Width | ["Yu-Jung Ko", "HANG ZHAO", "I. Ramakrishnan", "Shumin Zhai", "Xiaojun Bi"] | Finger-Fitts law [6] is a variant of Fitts' law which accounts for the finger ambiguity in touch pointing. It involves the effective target width $W_e$ (i.e., $\sqrt{2 \pi e}\sigma$) in modeling touch pointing. We hypothesize that the nominal target width ($W$) can be used in lieu of $W_e$ in Finger-Fitts law. Such a m... | ["law", "touch", "model", "nominal target width", "original", "touch pointing", "distribution", "endpoints", "unknown"] | ABSTRACTFinger-Fitts law [6] is a variant of Fitts’ law which accounts for thefinger ambiguity in touch pointing. It involves the effective targetwidth We(i.e.,p2pes) in modeling touch pointing. We hypothesizethat the nominal target width ( W) can be used in lieu of Wein Finger-Fitts law. Such a model, called Finger-Fi... | KkRq9MkCBoX | Good, small contribution - but some questions about the information criteria | 5: Marginally below acceptance threshold | This work discusses a modification on the Finger-Fitts (FFitts) model, and a study which demonstrates that it slightly outperforms the previous model (despite an additional regression-fitted parameter). The new model seems fairly sensible, but not particularly novel - for instance, it is very close to an old model by W... | 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
Modeling One-Dimensional Touch Pointing with Nominal Target Width
### Paper Abstract
Finger-Fitts law [6] is a variant of Fitts' law which accounts for the finger ambigu... |
IlNA3w6pDG- | eswc-conferences.org/ESWC/2021/Conference/Research_Track | 2021 | Natural Language Inference over Tables: Enabling Explainable Data Exploration on Data Lakes | ["Mario Alberto Ramirez Orihuela", "Alex Bogatu", "Norman Paton", "Andr\u00e9 Freitas"] | Data lakes are repositories of data with potential for analysis. Data lakes aim to liberate data from silos, thereby enabling cross-cutting analyses that were hitherto out of reach. However, the cost of adding data to data lakes must be kept to a minimum, and thus data sets tend to be stored in their original form, typ... | ["semantic relations", "data sets", "data", "approaches", "natural language inference", "tables", "explainable data exploration", "data lakes", "schemes", "target"] | Natural Language Inference over Tables: EnablingExplainable Data Exploration on Data LakesMario Ramirez1, Alex Bogatu1, Norman W. Paton1, and Andr ́e Freitas1;21Department of Computer Science, University of Manchester, UK2Idiap Research Institute, Switzerlandmario.ramirezorihuela@postgrad.manchester.ac.ukfalex.bogatu,n... | PFgX2FVC2ED | Review: A Natural Language Inference Framework for Data Exploration on Data Lakes | 1: Weak Accept | The authors address the problem of finding relevant data sets in a data lake for a given task. The target dataset is defined as a set of columns. Natural Language Inference (NLI) is used to calculate the semantic relations, by obtaining several types of evidence and combining them. Two approaches for inferring the sema... | 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
Natural Language Inference over Tables: Enabling Explainable Data Exploration on Data Lakes
### Paper Abstract
Data lakes are repositories of data with potential for ana... |
BJTuHHvHG | ICLR.cc/2018/Workshop | 2018 | Withdraw | ["Withdraw"] | Withdraw | ["Withdraw"] | This page is intentionally blank. | HJSJtUWtf | review | 3: Clear rejection | This paper proposes a text classification method for zero shot sentence or document classification. The paper is poorly written with lots of missing details. For example, the authors talk about category tree and did not clearly explain where this come from (are the tags already organized in a hierarchical structure)? H... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Withdraw
### Paper Abstract
Withdraw
### Paper Keywords
["Withdraw"]
### Paper Content
This page is intentionally blank.<|im_end|>
<|im_start|>assistant
### Review Tit... |
H1xipsA5K7 | ICLR.cc/2019/Conference | 2019 | Learning Two-layer Neural Networks with Symmetric Inputs | ["Rong Ge", "Rohith Kuditipudi", "Zhize Li", "Xiang Wang"] | We give a new algorithm for learning a two-layer neural network under a very general class of input distributions. Assuming there is a ground-truth two-layer network
y = A \sigma(Wx) + \xi,
where A, W are weight matrices, \xi represents noise, and the number of neurons in the hidden layer is no larger than the input o... | ["Neural Network", "Optimization", "Symmetric Inputs", "Moment-of-moments"] | ABSTRACTWe give a new algorithm for learning a two-layer neural network under a generalclass of input distributions. Assuming there is a ground-truth two-layer networky=A(Wx) +;whereA;W are weight matrices, represents noise, and the number of neurons inthe hidden layer is no larger than the input or output, our algorit... | rJxkjtHThX | Requiring a symmetric distribution and ReLU activation functions seems to be too strong. | 6: Marginally above acceptance threshold | This paper studies the problem of learning the parameters of a two-layer (or one-hidden layer) ReLU network $y=A\sigma(Wx)$, under the assumption that the distribution of $x$ is symmetric. The main technique here is the "pure neuron detector", which is a high-order moment function of a vector. It can be proved that the... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Two-layer Neural Networks with Symmetric Inputs
### Paper Abstract
We give a new algorithm for learning a two-layer neural network under a very general class of... |
iOVomQW073 | ICLR.cc/2021/Conference | 2021 | Almost Tight L0-norm Certified Robustness of Top-k Predictions against Adversarial Perturbations | ["Jinyuan Jia", "Binghui Wang", "Xiaoyu Cao", "Hongbin Liu", "Neil Zhenqiang Gong"] | Top-k predictions are used in many real-world applications such as machine learning as a service, recommender systems, and web searches. $\ell_0$-norm adversarial perturbation characterizes an attack that arbitrarily modifies some features of an input such that a classifier makes an incorrect prediction for the perturb... | ["Certified robustness", "adversarial perturbation"] | ABSTRACTTop-kpredictions are used in many real-world applications such as machine learn-ing as a service, recommender systems, and web searches. `0-norm adversarialperturbation characterizes an attack that arbitrarily modifies some features of aninput such that a classifier makes an incorrect prediction for the perturb... | tOc9fxlYJyD | Review for "Almost Tight L0-norm Certified Robustness of Top-k Predictions against Adversarial Perturbations" | 5: Marginally below acceptance threshold | ## Summary
The paper aims to give a robustness guarantee for top-k prediction under $\ell_0$-norm bounded adversarial perturbation of the input. To give such a guarantee authors study the randomized classifier constructed from the base deterministic classifier.
Authors randomize the base classifier by randomly ablat... | 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
Almost Tight L0-norm Certified Robustness of Top-k Predictions against Adversarial Perturbations
### Paper Abstract
Top-k predictions are used in many real-world applica... |
Syl6tjAqKX | ICLR.cc/2019/Conference | 2019 | BEHAVIOR MODULE IN NEURAL NETWORKS | ["Andrey Sakryukin", "Yongkang Wong", "Mohan S. Kankanhalli"] | Prefrontal cortex (PFC) is a part of the brain which is responsible for behavior repertoire. Inspired by PFC functionality and connectivity, as well as human behavior formation process, we propose a novel modular architecture of neural networks with a Behavioral Module (BM) and corresponding end-to-end training strate... | ["Modular Networks", "Reinforcement Learning", "Task Separation", "Representation Learning", "Transfer Learning", "Adversarial Transfer"] | ABSTRACTPrefrontal cortex (PFC) is a part of the brain which is responsible for behaviorrepertoire. Inspired by PFC functionality and connectivity, as well as humanbehavior formation process, we propose a novel modular architecture of neuralnetworks with a Behavioral Module (BM) and corresponding end-to-end trainingstr... | BygeJNU5hX | Interesting Paper, underwhelming experiments | 4: Ok but not good enough - rejection | This paper introduces a "behavior module" (BM) which is a small network that encodes preferences over actions and runs parallel to the fully connected layers of a policy. The paper shows this architecture working in Atari games, where the same policy can be used to achieve different action preferences over a game while... | 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
BEHAVIOR MODULE IN NEURAL NETWORKS
### Paper Abstract
Prefrontal cortex (PFC) is a part of the brain which is responsible for behavior repertoire. Inspired by PFC functi... |
i3Ui1Csrqpm | ICLR.cc/2021/Conference | 2021 | Optimizing Transformers with Approximate Computing for Faster, Smaller and more Accurate NLP Models | ["Amrit Nagarajan", "Sanchari Sen", "Jacob R. Stevens", "Anand Raghunathan"] | Transformer models have garnered a lot of interest in recent years by delivering state-of-the-art performance in a range of Natural Language Processing (NLP) tasks. However, these models can have over a hundred billion parameters, presenting very high computational and memory requirements. We address this challenge thr... | ["BERT", "Transformer", "NLP", "Efficient", "Faster", "Smaller", "Accurate"] | ABSTRACTTransformer models have garnered a lot of interest in recent years by deliveringstate-of-the-art performance in a range of Natural Language Processing (NLP)tasks. However, these models can have over a hundred billion parameters, present-ing very high computational and memory requirements. We address this challe... | FfT4VQpGaPd | Techniques need more justification and presentation needs polishing | 4: Ok but not good enough - rejection | Thanks to the authors for the detailed feedback! I still have concerns about the clarity of the presentation, and some contributions of the papers are not strong enough, so I'll keep my score.
===
Summary:
This paper presents a framework to systematically perform pruning and layer approximation. The framework includ... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Optimizing Transformers with Approximate Computing for Faster, Smaller and more Accurate NLP Models
### Paper Abstract
Transformer models have garnered a lot of interest... |
4ja9sJJygb | ICLR.cc/2021/Conference | 2021 | Exploiting Weight Redundancy in CNNs: Beyond Pruning and Quantization | ["Yuan Wen", "David Gregg"] | Pruning and quantization are proven methods for improving the performance and storage efficiency of convolutional neural networks (CNNs). Pruning removes near-zero weights in tensors and masks weak connections between neurons in neighbouring layers. Quantization reduces the precision of weights by replacing them with n... | ["Sparse tensor", "tensor compaction", "beyond pruning and quantization"] | ABSTRACTPruning and quantization are proven methods for improving the performance andstorage efficiency of convolutional neural networks (CNNs). Pruning removesnear-zero weights in tensors and masks weak connections between neurons inneighbouring layers. Quantization reduces the precision of weights by replacingthem wi... | qk41jDNM-Df | limited novelty | 4: Ok but not good enough - rejection | This paper shows that there is still weight redundancy after pruning and quantization, and further proposes a method to reduce this redundancy by sharing same-value blocks. However, this paper has limited novelty and poor experimental support, making its contribution very limited to the research community.
Pros:
1. 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
Exploiting Weight Redundancy in CNNs: Beyond Pruning and Quantization
### Paper Abstract
Pruning and quantization are proven methods for improving the performance and st... |
1VP6e2gxSjs | vivoconference.org/VIVO/2021/Conference | 2021 | Customized implementation of VIVO for working with philosophical texts | ["Georgy Litvinov"] | Institute of philosophy Russian Academy of Sciences is working on Electronic Philosophical Encyclopedia. VIVO was chosen as a research information system and platform for representing Encyclopedia articles. To present Electronic Philosophical Encyclopedia articles in VIVO we created “text structures” ontology. It prese... | ["vivo", "philosophical texts", "complex logical search"] | DZH5gxycPP0 | Interesting submission | 7: Good paper, accept | This use case is very interesting to that part of the community, who thinks a bit out of the box with VIVO. It would be interesting to learn about the ontology, and the workflows. Are different user groups (roles) involved? What skills are necessary to do this? How can the data be viewed/searched. | 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
Customized implementation of VIVO for working with philosophical texts
### Paper Abstract
Institute of philosophy Russian Academy of Sciences is working on Electronic Ph... | |
DBI4ML2mgif | MICCAI.org/2022/Challenge/FLARE | 2022 | A Noisy nnU-Net Student for Semi-supervised Abdominal Organ Segmentation | ["Gregor Koehler", "Fabian Isensee", "Klaus Maier-Hein"] | While deep learning methods have shown great potential in the context of medical image segmentation, it remains both time-consuming and expensive to collect sufficient data with expert annotations required to train large neural networks effectively. However, large amounts of unlabeled medical image data is available du... | ["Organ Segmentation", "Self-training", "Noisy student"] | A Noisy nnU-Net Student for Semi-supervisedAbdominal Organ SegmentationGregor Koehler1,2[0000−0002−5263−6786],Fabian Isensee1,2,3[0000−0002−3519−5886], andKlaus Maier-Hein1,2[0000−0002−6626−2463]1German Cancer Research Center (DKFZ)2Division of Medical Image Computing, German Cancer Research Center,Heidelberg, Germany3... | 4HYI2_3nfGv | Points are clear generally speaking, just need some minor changes | 7: Good paper, accept | The paper showed their points clearly, and writing in such manner that it stick with their list of contributions. That's being said, the structure of paper is coherence. However, I have some advice about content arrangement.
Abstract: Author articulate the challenges and the method abundantly. But, about the results, ... | 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 Noisy nnU-Net Student for Semi-supervised Abdominal Organ Segmentation
### Paper Abstract
While deep learning methods have shown great potential in the context of medi... |
HygHtpVtPH | ICLR.cc/2020/Conference | 2020 | Laplacian Denoising Autoencoder | ["Jianbo Jiao", "Linchao Bao", "Yunchao Wei", "Shengfeng He", "Honghui Shi", "Rynson Lau", "Thomas Huang"] | While deep neural networks have been shown to perform remarkably well in many machine learning tasks, labeling a large amount of supervised data is usually very costly to scale. Therefore, learning robust representations with unlabeled data is critical in relieving human effort and vital for many downstream application... | ["unsupervised", "representation learning", "Laplacian"] | ABSTRACTWhile deep neural networks have been shown to perform remarkably well in manymachine learning tasks, labeling a large amount of supervised data is usually verycostly to scale. Therefore, learning robust representations with unlabeled data iscritical in relieving human effort and vital for many downstream applic... | S1g7TmTQFr | Official Blind Review #2 | 6: Weak Accept | Denoising auto-encoder (DAE) is a representation learning framework proposes in [1], which uses noisy input to reconstruct the clean “repaired” input. In the conventional DAE, the noise is directly added to the input space. This paper introduces a novel type of noise, which corrupts the Laplacian pyramid representatio... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Laplacian Denoising Autoencoder
### Paper Abstract
While deep neural networks have been shown to perform remarkably well in many machine learning tasks, labeling a large... | |
_Pw7HplUPPP | eswc-conferences.org/ESWC/2021/Workshop/KGCW | 2021 | Collaborative-AI Knowledge Graph Generation: Taxonomization of IATE, the EU Terminology | ["Alena Vasilevich", "Michael Wetzel", "Stelios Mathioulakis", "Jochen Hummel"] | Formalized knowledge is a powerful resource for AI projects, but it is usually created at great expense. Taxonomization is linking a flat set of concepts into a hierarchical knowledge graph, and in this work, we present our approach to semi-automatic generation of such concept maps, elevating a sub-domain of IATE termi... | ["Collaborative taxonomization", "Knowledge Graph UI", "KG Construction Methods", "Machine Learning"] | Collaborative-AI Knowledge Graph Generation:Taxonomization of IATE, the EU TerminologyMichael Wetzel1, Stelios Mathioulakis2, Jochen Hummel1;2, and AlenaVasilevich1[0000000297691885]1Coreon GmbH, Berlin, Germanymichael@coreon.com alena@coreon.comhttp://www.coreon.com2ESTeam AB, Gothenburg, Swedenstelios@esteam.se joche... | VwmCLobieO | Clearly written paper showing the practicality of the presented approach, interesting future directions, but more distinction from related work desirable | 7: Good paper, accept | This paper presents a collaborative approach to create taxonomies from flat list of hierarchies
by leveraging a machine learning algorithm generating taxonomies which are refined by an expert.
The approach was compared to a manual curation by an expert and the results indicate that an expert is five times faster using ... | 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
Collaborative-AI Knowledge Graph Generation: Taxonomization of IATE, the EU Terminology
### Paper Abstract
Formalized knowledge is a powerful resource for AI projects, b... |
BxYsP-7ggf | NeurIPS.cc/2022/Workshop/Offline_RL | 2022 | Choreographer: Learning and Adapting Skills in Imagination | ["Pietro Mazzaglia", "Tim Verbelen", "Bart Dhoedt", "Alexandre Lacoste", "Sai Rajeswar"] | Unsupervised skill learning aims to learn a rich repertoire of behaviors without external supervision, providing artificial agents with the ability to control and influence the environment. However, without appropriate knowledge and exploration, skills may provide control only over a restricted area of the environment,... | ["fast adaptation", "skill learning", "world models"] | Choreographer: Learning andAdapting Skills in ImaginationPietro Mazzaglia∗1Tim Verbelen1Bart Dhoedt1Alexandre Lacoste2Sai Rajeswar21Ghent University - imec2ServiceNow ResearchAbstractUnsupervised skill learning aims to learn a rich repertoire of behaviors withoutexternal supervision, providing artificial agents with th... | FXVHcejlxzV | The method is interesting, but the baseline comparison is insufficient | 7: Good paper, accept | Summary
This paper proposed an offline skill-learning algorithm leveraging a world model. The method first pretrain a world model that learns latent space dynamics. Then they train the skill encoder and decoder which encodes/decodes between the skill space and the model’s latent space. The author also proposed a code ... | 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
Choreographer: Learning and Adapting Skills in Imagination
### Paper Abstract
Unsupervised skill learning aims to learn a rich repertoire of behaviors without external s... |
z2RNsvYZZTf | NeurIPS.cc/2022/Workshop/Federated_Learning | 2022 | Self-Supervised Vertical Federated Learning | ["Timothy Castiglia", "Shiqiang Wang", "Stacy Patterson"] | We consider a system where parties store vertically-partitioned data with a partially overlapping sample space, and a server stores labels on a subset of data samples. Supervised Vertical Federated Learning (VFL) algorithms are limited to training models using only overlapping labeled data, which can lead to poor model... | ["Machine learning", "distributed learning", "self-supervised learning", "vertically-partitioned data"] | Self-Supervised Vertical Federated LearningTimothy CastigliaDepartment of Computer ScienceRensselaer Polytechnic InstituteTroy, NY 12180castit@rpi.eduShiqiang WangIBM ResearchThomas J. Watson Research CenterYorktown Heights, NYwangshiq@us.ibm.comStacy PattersonDepartment of Computer ScienceRensselaer Polytechnic Instit... | PvYKKGjhMcH | interesting topic, limited novelty, lack of experiments | This paper proposes to leverage self-supervised learning to train representation models to improve the performance of VFL when the labeled data is scarce. The paper presents two self-supervised VFL algorithms, SS-VFL-I and SS-VFL-C. The experiment results on two image datasets demonstrate that both algorithms can achie... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Self-Supervised Vertical Federated Learning
### Paper Abstract
We consider a system where parties store vertically-partitioned data with a partially overlapping sample s... | ||
rkeDJ04Mf | ICLR.cc/2018/Conference | 2018 | HyperNetworks with statistical filtering for defending adversarial examples | ["Anonymous"] | Deep learning algorithms have been known to be vulnerable to adversarial perturbations in various tasks such as image classification. This problem was addressed by employing several defense methods for detection and rejection of particular types of attacks. However, training and manipulating networks according to parti... | ["attacks", "cnns", "statistical filtering", "adversarial attacks", "hypernetworks", "adversarial examples hypernetworks", "adversarial examples", "algorithms", "vulnerable", "adversarial perturbations"] | ABSTRACTDeep learning algorithms have been known to be vulnerable to adversarial perturba-tions in various tasks such as image classification. This problem was addressed byemploying several defense methods for detection and rejection of particular typesof attacks. However, training and manipulating networks according t... | rycbv6tQM | Unjustified application of existing technique (hypernetwork) in defending CNN against adversarial attacks. | 4: Ok but not good enough - rejection | This paper proposed to improve the robustness and reliability of deep convolution neural networks (with regard to adversarial attacks such as noise or perturbations) by using data-dependent convolution kernels. Basically the authors designed a new type of hypernetwork that takes as input the statistical properties of... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
HyperNetworks with statistical filtering for defending adversarial examples
### Paper Abstract
Deep learning algorithms have been known to be vulnerable to adversarial p... |
aGmEDl1NWJ- | ICLR.cc/2021/Conference | 2021 | Luring of transferable adversarial perturbations in the black-box paradigm | ["R\u00e9mi Bernhard", "Pierre-Alain Mo\u00ebllic", "Jean-Max Dutertre"] | The growing interest for adversarial examples, i.e. maliciously modified examples which fool a classifier, has resulted in many defenses intended to detect them, render them inoffensive or make the model more robust against them. In this paper, we pave the way towards a new approach to improve the robustness of a model... | ["Neural Networks", "Adversarial Machine Learning", "Security"] | ABSTRACTThe growing interest for adversarial examples, i.e. maliciously modified exam-ples which fool a classifier, has resulted in many defenses intended to detect them,render them inoffensive or make the model more robust against them. In thispaper, we pave the way towards a new approach to improve the robustness of ... | aZ2mbprq4Lv | A conceptually new framework for addressing the problem of adversaries in black box settings with promising results | 8: Top 50% of accepted papers, clear accept | Summary :
The paper proposes a new framework for addressing the problem of adversaries in black box settings in order to improve model robustness. Leveraging classical deception frameworks used in network security, the authors propose to fool the attacker by training what they call a `luring component’ that is augment... | 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
Luring of transferable adversarial perturbations in the black-box paradigm
### Paper Abstract
The growing interest for adversarial examples, i.e. maliciously modified ex... |
r1g5b2RcKm | ICLR.cc/2019/Conference | 2019 | MLPrune: Multi-Layer Pruning for Automated Neural Network Compression | ["Wenyuan Zeng", "Raquel Urtasun"] | Model compression can significantly reduce the computation and memory footprint of large neural networks. To achieve a good trade-off between model size and accuracy, popular compression techniques usually rely on hand-crafted heuristics and
require manually setting the compression ratio of each layer. This process is ... | ["Automated Model Compression", "Neural Network Pruning"] | ABSTRACTModel compression can significantly reduce the computation and memory footprintof large neural networks. To achieve a good trade-off between model size andaccuracy, popular compression techniques usually rely on hand-crafted heuristicsand require manually setting the compression ratio of each layer. This proces... | SJgffqb2h7 | Second order method for pruning multiple layers | 5: Marginally below acceptance threshold | This paper proposes a multi-layer pruning technique based on the Hessian. The main claims are performing better than other second order pruning methods and be principled.
Main concerns / comments are:
- Part of the novelty relays on computing the Hessian, and the algorithm goes for very large networks (parameter wis... | 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
MLPrune: Multi-Layer Pruning for Automated Neural Network Compression
### Paper Abstract
Model compression can significantly reduce the computation and memory footprint ... |
SkfrvsA9FX | ICLR.cc/2019/Conference | 2019 | Reward Constrained Policy Optimization | ["Chen Tessler", "Daniel J. Mankowitz", "Shie Mannor"] | Solving tasks in Reinforcement Learning is no easy feat. As the goal of the agent is to maximize the accumulated reward, it often learns to exploit loopholes and misspecifications in the reward signal resulting in unwanted behavior. While constraints may solve this issue, there is no closed form solution for general co... | ["reinforcement learning", "markov decision process", "constrained markov decision process", "deep learning"] | Published as a conference paper at ICLR 2019Reward Constrained Policy OptimizationChen Tessler1, Daniel J. Mankowitz2, and Shie Mannor11Technion Israel Institute of Technology, Haifa, Israel2DeepMind, London, Englandchen.tessler@campus.technion.ac.il, dmankowitz@google.com, shie@ee.technion.ac.ilAbstractSolving tasks i... | r1eEFRcAnm | Interesting paper, with some details not so clear to me | 6: Marginally above acceptance threshold | This paper proposed a general framework for policy optimization of constrained MDP. Compared with the traditional methods, such as Lagrangian multiplier methods and trust region approach, this method shows better empirical results and theoretical merits.
Major concerns:
My major concern is about the time-scales. RCP... | 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
Reward Constrained Policy Optimization
### Paper Abstract
Solving tasks in Reinforcement Learning is no easy feat. As the goal of the agent is to maximize the accumulate... |
rksfwnFxl | ICLR.cc/2017/conference | 2017 | LSTM-Based System-Call Language Modeling and Ensemble Method for Host-Based Intrusion Detection | ["Gyuwan Kim", "Hayoon Yi", "Jangho Lee", "Yunheung Paek", "Sungroh Yoon"] | In computer security, designing a robust intrusion detection system is one of the most fundamental and important problems. In this paper, we propose a system-call language-modeling approach for designing anomaly-based host intrusion detection systems. To remedy the issue of high false-alarm rates commonly arising in co... | ["language modeling", "ensemble", "intrusion detection", "computer security", "fundamental", "important problems", "issue", "high"] | ABSTRACTIn computer security, designing a robust intrusion detection system is one of themost fundamental and important problems. In this paper, we propose a system-calllanguage-modeling approach for designing anomaly-based host intrusion detec-tion systems. To remedy the issue of high false-alarm rates commonly arisin... | HJLV5d2me | neat idea, clearly written | 8: Top 50% of accepted papers, clear accept | In this paper a novel approach for anomaly detection is considered for the task of intrusion detection based on system call sequence.
The system call sequence is regarded as a language, and multiple lstm-rnn language models are trained and ensembled.
Diversity in the ensemble is achieved by choosing different hyper 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
LSTM-Based System-Call Language Modeling and Ensemble Method for Host-Based Intrusion Detection
### Paper Abstract
In computer security, designing a robust intrusion det... |
4jJI8Sqwz9V | ICLR.cc/2021/Conference | 2021 | Generalization and Stability of GANs: A theory and promise from data augmentation | ["Khoat Than", "Nghia Vu"] | The instability when training generative adversarial networks (GANs) is a notoriously difficult issue, and the generalization of GANs remains open. In this paper, we will analyze various sources of instability which not only come from the discriminator but also the generator. We then point out that the requirement ... | ["generative adversarial networks", "generalization", "stability", "data augmentation"] | ABSTRACTThe instability when training generative adversarial networks (GANs) is a no-toriously difficult issue, and the generalization of GANs remains open. In thispaper, we will analyze various sources of instability which not only come fromthe discriminator but also the generator. We then point out that the requireme... | ch4swvsiCRa | Official Review | 3: Clear rejection | This paper studies the generalization properties of generative adversarial networks (GANs). The paper's goal is to demonstrate that the generalization error of GANs correlates with the Lipschitz constant of the discriminator and generator players. To show this, the paper bounds the generalization error with a function ... | 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
Generalization and Stability of GANs: A theory and promise from data augmentation
### Paper Abstract
The instability when training generative adversarial networks (GANs... |
anKt6unvU- | icaps-conference.org/ICAPS/2020/Workshop/HSDIP | 2020 | Beating LM-cut with LM-cut: Quick Cutting and Practical Tie Breaking for the Precondition Choice Function | ["Anonymous"] | LM-cut is one of the most popular heuristics in optimal planning that computes strong admissible estimates of the perfect delete relaxation heuristic $h^+$. The heuristic iteratively computes disjunctive action landmarks for the current state, reducing their action costs until no more landmarks with remaining action co... | ["precondition choice function", "quick cutting", "practical tie", "heuristic", "action costs", "landmarks", "computation", "popular heuristics", "optimal planning", "strong admissible estimates"] | Beating LM-cut with LM-cut:Quick Cutting and Practical Tie Breaking for the Precondition Choice FunctionPascal Lauer andMaximilian FickertSaarland UniversitySaarland Informatics CampusSaarbr ̈ucken, Germanys8palaue@stud.uni-saarland.de, fickert@cs.uni-saarland.deAbstractLM-cut is one of the most popular heuristics in o... | 2AnDcA8gAMA | Initial assessment | 7: Good paper, accept | I think it is a nice paper with interesting results that provide more insight
into the LM-Cut heuristic.
However, I think the following (rather minor) issues should be addressed:
1. Bonet and Helmert (2010) is correctly cited as proposing to use maximum of
multiple passes of LM-Cut heuristics with different pcfs. But t... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Beating LM-cut with LM-cut: Quick Cutting and Practical Tie Breaking for the Precondition Choice Function
### Paper Abstract
LM-cut is one of the most popular heuristics... |
FP9kKyNWwwE | ICLR.cc/2021/Conference | 2021 | Zero-shot Transfer Learning for Gray-box Hyper-parameter Optimization | ["Hadi Samer Jomaa", "Lars Schmidt-Thieme", "Josif Grabocka"] | Zero-shot hyper-parameter optimization refers to the process of selecting hyper- parameter configurations that are expected to perform well for a given dataset upfront, without access to any observations of the losses of the target response. Existing zero-shot approaches are posed as initialization strategies for Bayes... | ["Hyper-parameter Optimization", "Transfer Learning", "Meta-learning"] | ABSTRACTZero-shot hyper-parameter optimization refers to the process of selecting hyper-parameter configurations that are expected to perform well for a given datasetupfront, without access to any observations of the losses of the target response.Existing zero-shot approaches are posed as initialization strategies for ... | A5kgZGbJUiU | This paper proposed a regularized regression using meta-feature embeddings of certain datasets as regularization for standard HBO. | 6: Marginally above acceptance threshold | In this papar, the authors formulated a new objective function for HBO, which included an additional regularization term based on the dataset similarity. The authors used the distance between the meta-features of selected datasets to measure this dataset similarity and assumpted that similar datasets should have simila... | 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
Zero-shot Transfer Learning for Gray-box Hyper-parameter Optimization
### Paper Abstract
Zero-shot hyper-parameter optimization refers to the process of selecting hyper-... |
3MrDT4bTycn | graphicsinterface.org/Graphics_Interface/2022/Conference | 2022 | VR Stylus and Controller Combination Was Preferred Over the Mouse or Hand Tracking for Object Manipulation and Marking Tasks in Virtual Reality | ["Anonymous"] | For medical surgery planning, virtual reality (VR) provides a new kind of user experience, where 3D images of the operation area can be utilized. Using VR, it is possible to view the 3D models in a more realistic 3D environment, which would reduce the perception problems and increase spatial understanding. In this expe... | ["3D visualization", "virtual reality", "hand tracking", "controller interaction", "mouse interaction", "3D imaging"] | ABSTRACTFor medical surgery planning, virtual reality (VR) provides a newkind of user experience, where 3D images of the operation area canbe utilized. Using VR, it is possible to view the 3D models in amore realistic 3D environment, which would reduce the perceptionproblems and increase spatial understanding. In this ... | Icv6zBsXYNg | Seems like a clear accept after trivial issues are fixed. | 9: Top 15% of accepted papers, strong accept | The paper is presents a study in which 3 different 3D object manipulation techniques are presented: mouse, tracker+stylus, and hands.
I am not aware of any studies in which a similar technique (3D controller and stylus) were used, and thus the study has an important practical value, as it can generalize to other com... | 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
VR Stylus and Controller Combination Was Preferred Over the Mouse or Hand Tracking for Object Manipulation and Marking Tasks in Virtual Reality
### Paper Abstract
For me... |
Bye5SiAqKX | ICLR.cc/2019/Conference | 2019 | Preconditioner on Matrix Lie Group for SGD | ["Xi-Lin Li"] | We study two types of preconditioners and preconditioned stochastic gradient descent (SGD) methods in a unified framework. We call the first one the Newton type due to its close relationship to the Newton method, and the second one the Fisher type as its preconditioner is closely related to the inverse of Fisher inform... | ["preconditioner", "stochastic gradient descent", "Newton method", "Fisher information", "natural gradient", "Lie group"] | ABSTRACTWe study two types of preconditioners and preconditioned stochastic gradient de-scent (SGD) methods in a unified framework. We call the first one the Newton typedue to its close relationship to the Newton method, and the second one the Fishertype as its preconditioner is closely related to the inverse of Fisher... | S1e9NsdXsm | Solid work | 7: Good paper, accept | The authors suggest and analyse two types of preconditioners for optimization, a Newton type and a Fisher type preconditioner.
The paper is well written, the analysis is clear and the significance is arguably given. The authors run their optimizers on a synthetic benchmark data set and on imagenet.
The originality is ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Preconditioner on Matrix Lie Group for SGD
### Paper Abstract
We study two types of preconditioners and preconditioned stochastic gradient descent (SGD) methods in a uni... |
MyHwDabUHZm | ICLR.cc/2021/Conference | 2021 | Beyond Categorical Label Representations for Image Classification | ["Boyuan Chen", "Yu Li", "Sunand Raghupathi", "Hod Lipson"] | We find that the way we choose to represent data labels can have a profound effect on the quality of trained models. For example, training an image classifier to regress audio labels rather than traditional categorical probabilities produces a more reliable classification. This result is surprising, considering that au... | ["Label Representation", "Image Classification", "Representation Learning"] | ABSTRACTWe find that the way we choose to represent data labels can have a profound ef-fect on the quality of trained models. For example, training an image classifierto regress audio labels rather than traditional categorical probabilities produces amore reliable classification. This result is surprising, considering ... | 7bh7nxvdEm | TTS-generated audio can be used to train classification model instead of one-hot class label. | 7: Good paper, accept | The paper proposed to use other high-dimensional representation instead of direct one-hot class labels in image classification problem. They used spectrogram of the pronunciations (TTS-based generated speech) of class labels as a high-dimensional representation. Then, they performed regression using the spectrogram as ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Beyond Categorical Label Representations for Image Classification
### Paper Abstract
We find that the way we choose to represent data labels can have a profound effect o... |
Z4R1vxLbRLO | ICLR.cc/2021/Conference | 2021 | Extreme Memorization via Scale of Initialization | ["Harsh Mehta", "Ashok Cutkosky", "Behnam Neyshabur"] | We construct an experimental setup in which changing the scale of initialization strongly impacts the implicit regularization induced by SGD, interpolating from good generalization performance to completely memorizing the training set while making little progress on the test set. Moreover, we find that the extent and m... | ["Scale of initialization", "Memorization", "Overfitting", "Generalization", "Generalization Measure", "Understanding Deep Learning"] | ABSTRACTWe construct an experimental setup in which changing the scale of initializationstrongly impacts the implicit regularization induced by SGD, interpolating fromgood generalization performance to completely memorizing the training set whilemaking little progress on the test set. Moreover, we find that the extent ... | EHVIZ3OAzcn | Paper is interesting, but I am not sure about the experiment setup | 9: Top 15% of accepted papers, strong accept | ## Overview
The paper studies how the generalization of the neural network trained with SGD is affected by the scale of the random initialization. Specifically, that when the scale of the initialization is big, the network overfits to the training set with bad performance on the test set.
The paper also provides a h... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Extreme Memorization via Scale of Initialization
### Paper Abstract
We construct an experimental setup in which changing the scale of initialization strongly impacts the... |
ziGdL0BwPL | NeurIPS.cc/2020/Workshop/DL-IG | 2020 | Implicit Regularization via Neural Feature Alignment | ["Aristide Baratin", "Thomas George", "C\u00e9sar Laurent", "R Devon Hjelm", "Guillaume Lajoie", "Pascal Vincent", "Simon Lacoste-Julien"] | We approach the problem of implicit regularization in deep learning from a geometrical viewpoint. We highlight a regularization effect induced by a dynamical alignment of the neural tangent features introduced by Jacot et al (2018), along a small number of task-relevant directions. This can be interpreted as a combined... | ["deep learning", "learning dynamics", "implicit regularization"] | Implicit Regularization via Feature AlignmentOne important property of deep neural networks is their ability to generalize well on real data. 1Surprisingly,thisiseventruewithveryhigh-capacitynetworks withoutexplicitregularization [42, 261,29]. Thisseemsatoddswiththeusualunderstandingofthebias-variancetrade-off[ 24,41,1... | IhDk0MFz55 | Review of "Implicit Regularization via Neural Feature Alignment" | 9: Top 15% of accepted papers, strong accept | The geometric point of view to explain generalizability of deep networks is indeed an interesting avenue to explore in this research area. The experimental evidences appear to support the authors claim that the neural tangent features are concentrated on a small number of eigenvectors of the tangent kernel basis which ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Implicit Regularization via Neural Feature Alignment
### Paper Abstract
We approach the problem of implicit regularization in deep learning from a geometrical viewpoint.... |
rkgO66VKDS | ICLR.cc/2020/Conference | 2020 | LEARNED STEP SIZE QUANTIZATION | ["Steven K. Esser", "Jeffrey L. McKinstry", "Deepika Bablani", "Rathinakumar Appuswamy", "Dharmendra S. Modha"] | Deep networks run with low precision operations at inference time offer power and space advantages over high precision alternatives, but need to overcome the challenge of maintaining high accuracy as precision decreases. Here, we present a method for training such networks, Learned Step Size Quantization, that achieves... | ["deep learning", "low precision", "classification", "quantization"] | ABSTRACTDeep networks run with low precision operations at inference time offer powerand space advantages over high precision alternatives, but need to overcome thechallenge of maintaining high accuracy as precision decreases. Here, we present amethod for training such networks, Learned Step Size Quantization, that ach... | rJx7-yzYqB | Official Blind Review #2 | 6: Weak Accept | The motivation of the paper is to be able to train low precision networks to a high-accuracy. Quantization is a useful tool in model compression, and doing it well for very low-precision models (2-3 bit precision specifically), is challenging.
The main contribution of the paper comes from:
a) Step Size Gradient: They ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
LEARNED STEP SIZE QUANTIZATION
### Paper Abstract
Deep networks run with low precision operations at inference time offer power and space advantages over high precision ... | |
4Xo8nv5DNS | JSYS/2021/Mar_Papers | 2021 | SoK: A Generalized Multi-Leader State Machine Replication Tutorial | ["Michael Whittaker", "Neil Giridharan", "Adriana Szekeres", "Joseph Hellerstein", "Ion Stoica"] | MultiPaxos and Raft are the two most popular and widely deployed state machine replication protocols. There is a more sophisticated family of generalized multi-leader state machine replication protocols like EPaxos, Caesar, and Atlas that have better performance, but they are extremely complicated and hard to understan... | ["consensus", "state machine replication", "generalized", "multi-leader", "Paxos", "Raft", "EPaxos", "Caesar", "Atlas"] | SOK: A G ENERALIZED MULTI -LEADER STATE MACHINE REPLICATION TUTORIALMichael WhittakerUC BerkeleyNeil GiridharanUC BerkeleyAdriana SzekeresUniversity of WashingtonJoseph M. HellersteinUC BerkeleyIon StoicaUC BerkeleyAbstractMultiPaxos and Raft are the two most popular and widelydeployed state machine replication protoco... | _m60k8yI8_A | Review for paper titled, "SoK: A Generalized Multi-Leader State Machine Replication Tutorial" | ## Summary:
This paper aims to explain the space of multi-leader consensus protocols. The key advantage of adopting multi-leader protocols is that they facilitate out-of-order execution of commutating. In specific, if two commands do not conflict, then these protocols allow replicas to execute them in their desired ord... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
SoK: A Generalized Multi-Leader State Machine Replication Tutorial
### Paper Abstract
MultiPaxos and Raft are the two most popular and widely deployed state machine repl... | ||
SJxbps09K7 | ICLR.cc/2019/Conference | 2019 | Empirical observations on the instability of aligning word vector spaces with GANs | ["Mareike Hartmann", "Yova Kementchedjhieva", "Anders S\u00f8gaard"] | Unsupervised bilingual dictionary induction (UBDI) is useful for unsupervised machine translation and for cross-lingual transfer of models into low-resource languages. One approach to UBDI is to align word vector spaces in different languages using Generative adversarial networks (GANs) with linear generators, achievin... | ["natural language processing", "bilingual dictionary induction", "unsupervised learning", "generative adversarial networks"] | ABSTRACTUnsupervised bilingual dictionary induction (UBDI) is useful for unsupervisedmachine translation and for cross-lingual transfer of models into low-resource lan-guages. One approach to UBDI is to align word vector spaces in different languagesusing Generative Adversarial Networks (GANs) with linear generators, a... | HyghBqh83Q | This paper provides interesting insight into the usefulness of GANs to perform alignment of embedding spaces across languages. The paper is well motivated and introduces the problem well. While existing experiments are sufficient, there is scope to add more experiments. | 6: Marginally above acceptance threshold | This paper is clear and well written, except for the occasional typos, eg. missing a bracket in the very first equation.
The problem of learning transformations to align vector spaces across languages is extremely interesting and relevant. While the paper does a thorough job in explaining the metrics used and in detail... | 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 observations on the instability of aligning word vector spaces with GANs
### Paper Abstract
Unsupervised bilingual dictionary induction (UBDI) is useful for un... |
3_2Zf8Rr1N | thecvf.com/ECCV/2020/Workshop/BIC | 2020 | Bionic Tracking: Using Eye Tracking to Track Biological Cells in Virtual Reality | ["Ulrik G\u00fcnther", "Kyle I.S. Harrington", "Raimund Dachselt", "Ivo Sbalzarini"] | We present Bionic Tracking, a novel method for solving biological cell tracking problems with eye tracking in virtual reality using commodity hardware. Using gaze data, and especially smooth pursuit eye movements, we are able to track cells in time series of 3D volumetric datasets. The problem of tracking cells is ubiq... | ["Cell tracking", "virtual reality", "eye tracking"] | 000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044000001002003004005006007008009010011012013014015016017018019020021022023024025026027028029030031032033034035036037038039040041042043044ECCV#23ECCV#23Bionic Tracking: Using Eye Tracking ... | baEU5z6Hw9r | A new approach for annotating 3D+T cell tracking data | 6: Marginally above acceptance threshold | Cell tracking is an important but challenging step for many biological research. By combining VR and eye tracking, the authors developed a new method for tracking cells in 3D and time. With the help of the two devices, users can generate cell trajectories by simply looking at a cell in a 3D movie. In the manuscript, de... | 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
Bionic Tracking: Using Eye Tracking to Track Biological Cells in Virtual Reality
### Paper Abstract
We present Bionic Tracking, a novel method for solving biological cel... |
HkewNJStDr | ICLR.cc/2020/Conference | 2020 | Efficient High-Dimensional Data Representation Learning via Semi-Stochastic Block Coordinate Descent Methods | ["Bingkun Wei", "Yangyang Li", "Fanhua Shang", "Yuanyuan Liu", "Hongying Liu", "Shengmei Shen"] | With the increase of data volume and data dimension, sparse representation learning attracts more and more attention. For high-dimensional data, randomized block coordinate descent methods perform well because they do not need to calculate the gradient along the whole dimension. Existing hard thresholding algorithms ev... | ["Sparse learning", "Hard thresholding", "High-dimensional regression"] | ABSTRACTWith the increase of data volume and data dimension, sparse representation learn-ing attracts more and more attention. For high-dimensional data, randomizedblock coordinate descent methods perform well because they do not need to cal-culate the gradient along the whole dimension. Existing hard thresholding algo... | Hyg-tqxRtr | Official Blind Review #2 | 3: Weak Reject | This paper proposes a semi-stochastic block coordinate descent hard thresholding pursuit (SBCD-HTP) algorithm for solving l0 sparsity-constrained minimization with decomposable convex objective. The key idea is to introduce BCD into SVRG type of algorithm to speed up the convergence as well as the hard thresholding ope... | <|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 High-Dimensional Data Representation Learning via Semi-Stochastic Block Coordinate Descent Methods
### Paper Abstract
With the increase of data volume and data... | |
v8ioFR4fqpr | aclweb.org/ACL/2020/Workshop/NLP-COVID | 2020 | Enabling Low-Resource Transfer Learning across COVID-19 Corpora by Combining Event-Extraction and Co-Training | ["Alexander Spangher", "Nanyun Peng", "Jonathan May", "Emilio Ferrara"] | Social-science investigations can benefit from a direct comparison of heterogenous corpora: in this work, we compare U.S. state-level COVID-19 policy announcements with policy discussions on Twitter. To perform this task, we require classifiers with high transfer accuracy to both (1) classify policy announcements and (... | ["co-training", "cross domain learning", "natural language processing", "covid-19", "coronavirus", "social science", "event-extraction"] | Enabling Low-Resource Transfer Learning across COVID-19 Corpora byCombining Event-Extraction and Co-TrainingAlexander Spangher, Nanyun Peng, Jonathan May, Emilio FerraraInformation Sciences Institute, University of Southern California, Marina del Rey, USAfspangher, npeng, jonmay, ferrarae g@isi.eduAbstractSocial-scienc... | -T1hlFj9e2_ | Interesting work; some questions could be addressed | 5: Marginally below acceptance threshold | Summary: The paper describes a topic classification task for policy announcements and tweets. The authors experiment with logistic regression and RoBERTa classifier. Co-training based on views based on events (indicated by a subset of words in the text) is used to increase the dataset size. This results in an improveme... | 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
Enabling Low-Resource Transfer Learning across COVID-19 Corpora by Combining Event-Extraction and Co-Training
### Paper Abstract
Social-science investigations can benefi... |
B1suU-bAW | ICLR.cc/2018/Conference | 2018 | Learning Covariate-Specific Embeddings with Tensor Decompositions | ["Kevin Tian", "Teng Zhang", "James Zou"] | Word embedding is a useful approach to capture co-occurrence structures in a large corpus of text. In addition to the text data itself, we often have additional covariates associated with individual documents in the corpus---e.g. the demographic of the author, time and venue of publication, etc.---and we would like the... | ["Word embedding", "tensor decomposition"] | ABSTRACTWord embedding is a useful approach to capture co-occurrence structures in alarge corpus of text. In addition to the text data itself, we often have additionalcovariates associated with individual documents in the corpus—e.g. the demo-graphic of the author, time and venue of publication, etc.—and we would like ... | HkqIGZGlG | Good idea, not well evaluated against other methods | 5: Marginally below acceptance threshold | This paper produces word embedding tensors where the third order gives covariate information, via venue or author. The model is simple: tensor factorization, where the covariate can be viewed as warping the cosine distance to favor that covariate's more commonly cooccuring vocabulary (e.g. trump on hillary and crooked)... | 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 Covariate-Specific Embeddings with Tensor Decompositions
### Paper Abstract
Word embedding is a useful approach to capture co-occurrence structures in a large c... |
Hynn8SHOx | ICLR.cc/2017/workshop | 2017 | DL-gleaning: An approach for Improving inference speed and accuracy | ["HyunYong Lee and Byung-Tak Lee"] | Improving inference speed and accuracy is one of the main objectives of current deep learning-related research. In this paper, we introduce our approach using middle output layer for this purpose. From the feasibility study using Inception-v4, we found that our approach has potential to reduce the average inference tim... | ["Deep learning", "Supervised Learning"] | ABSTRACTImproving inference speed and accuracy is one of the main objectives of currentdeep learning-related research. In this paper, we introduce our approach usingmiddle output layer for this purpose. From the feasibility study using Inception-v4, we found that our approach has potential to reduce the average inferen... | B1OJyGKcx | Not promising | 3: Clear rejection | This paper proposes to add output layers from middle layers to speed up inference. The paper showed that in the oracle condition speedup is possible. However, because of errors in making decisions on whether to pick the middle layer result, it's very unlikely that the proposed approach would work. There is no free lunc... | 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
DL-gleaning: An approach for Improving inference speed and accuracy
### Paper Abstract
Improving inference speed and accuracy is one of the main objectives of current de... |
Byht0GbRZ | ICLR.cc/2018/Conference | 2018 | STRUCTURED ALIGNMENT NETWORKS | ["Yang Liu", "Matt Gardner"] | Many tasks in natural language processing involve comparing two sentences to compute some notion of relevance, entailment, or similarity. Typically this comparison is done either at the word level or at the sentence level, with no attempt to leverage the inherent structure of the sentence. When sentence structure is u... | ["structured attention", "sentence matching"] | ABSTRACTMany tasks in natural language processing involve comparing two sentences tocompute some notion of relevance, entailment, or similarity. Typically this com-parison is done either at the word level or at the sentence level, with no attempt toleverage the inherent structure of the sentence. When sentence structur... | Sybe_7qlG | Structured alignment networks | 5: Marginally below acceptance threshold | This paper proposes a model of "structured alignments" between sentences as a means of comparing two sentences by matching their latent structures. Overall, this paper seems a straightforward application of the model first proposed by Kim et al. 2017 with latent tree attention.
In section 3.1, the formula for p(c|x) l... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
STRUCTURED ALIGNMENT NETWORKS
### Paper Abstract
Many tasks in natural language processing involve comparing two sentences to compute some notion of relevance, entailme... |
Sye2c3NYDB | ICLR.cc/2020/Conference | 2020 | Boosting Ticket: Towards Practical Pruning for Adversarial Training with Lottery Ticket Hypothesis | ["Bai Li", "Shiqi Wang", "Yunhan Jia", "Yantao Lu", "Zhenyu Zhong", "Lawrence Carin", "Suman Jana"] | Recent research has proposed the lottery ticket hypothesis, suggesting that for a deep neural network, there exist trainable sub-networks performing equally or better than the original model with commensurate training steps. While this discovery is insightful, finding proper sub-networks requires iterative training and... | ["neural networks", "adversarial training", "prune"] | ABSTRACTRecent research has proposed the lottery ticket hypothesis, suggesting that for adeep neural network, there exist trainable sub-networks performing equally or bet-ter than the original model with commensurate training steps. While this discoveryis insightful, finding proper sub-networks requires iterative train... | Hkew_FETFr | Official Blind Review #2 | 1: Reject | The paper presents some follow-up work on the lottery ticket hypothesis (LTH). The iterative magnitude-based pruning (IMP) used in the original LTH paper by Frankle and Carbin (FC19) failed to find winning tickets on large-scale vision networks without modifying the training procedure. This paper describes a change in ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Boosting Ticket: Towards Practical Pruning for Adversarial Training with Lottery Ticket Hypothesis
### Paper Abstract
Recent research has proposed the lottery ticket hyp... | |
BklZ9A4YDr | ICLR.cc/2020/Conference | 2020 | Learning to Generate 3D Training Data through Hybrid Gradient | ["Dawei Yang", "Jia Deng"] | Synthetic images rendered by graphics engines are a promising source for training deep networks. However, it is challenging to ensure that they can help train a network to perform well on real images, because a graphics-based generation pipeline requires numerous design decisions such as the selection of 3D shapes and ... | ["training data", "hybrid gradient", "generation", "graphics engines", "promising source", "deep networks", "network", "real images", "generation pipeline"] | ABSTRACTSynthetic images rendered by graphics engines are a promising source for train-ing deep networks. However, it is challenging to ensure that they can help traina network to perform well on real images, because a graphics-based generationpipeline requires numerous design decisions such as the selection of 3D shap... | H1lMs4tqYH | Official Blind Review #3 | 8: Accept | The paper presents a method for optimizing synthetically-generated scenes for transferability to real scenes. The synthetic scene is parameterized by a vector of parameters (beta), and the Authors propose a method to optimize the evaluation loss with respect to beta by gradient descent. The key contribution is a method... | <|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 Generate 3D Training Data through Hybrid Gradient
### Paper Abstract
Synthetic images rendered by graphics engines are a promising source for training deep n... | |
CVZMcRg_bd | ICLR.cc/2021/Conference | 2021 | Differentiable Programming for Piecewise Polynomial Functions | ["Minsu Cho", "Ameya Joshi", "Xian Yeow Lee", "Aditya Balu", "Adarsh Krishnamurthy", "Baskar Ganapathysubramanian", "Soumik Sarkar", "Chinmay Hegde"] | The paradigm of differentiable programming has considerably enhanced the scope of machine learning via the judicious use of gradient-based optimization. However, standard differentiable programming methods (such as autodiff) typically require that the models be differentiable, limiting their applicability. We introduce... | ["Differentiable Programming", "piecewise polynomial regression", "generative models", "segmentation"] | ABSTRACTThe paradigm of differentiable programming has significantly enhanced the scopeof machine learning via the judicious use of gradient-based optimization. How-ever, standard differentiable programming methods (such as autodiff) typicallyrequire that the machine learning models be differentiable, limiting their ap... | pPUQFl2r4q | A hard to read article, with too light experiments | 4: Ok but not good enough - rejection | Summary
The authors consider the approximation of a 1-d signal x and derive the Jacobian of the function that maps x to a piecewise polynomial approximation. This jacobian has a convenient block sparse structure which allows for fast computations. Then, the authors use the polynomial approximation as a layer in a mach... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Differentiable Programming for Piecewise Polynomial Functions
### Paper Abstract
The paradigm of differentiable programming has considerably enhanced the scope of machin... |
ryHlUtqge | ICLR.cc/2017/conference | 2017 | Generalizing Skills with Semi-Supervised Reinforcement Learning | ["Chelsea Finn", "Tianhe Yu", "Justin Fu", "Pieter Abbeel", "Sergey Levine"] | Deep reinforcement learning (RL) can acquire complex behaviors from low-level inputs, such as images. However, real-world applications of such methods require generalizing to the vast variability of the real world. Deep networks are known to achieve remarkable generalization when provided with massive amounts of labele... | ["Reinforcement Learning"] | ABSTRACTDeep reinforcement learning (RL) can acquire complex behaviors from low-levelinputs, such as images. However, real-world applications of such methods requiregeneralizing to the vast variability of the real world. Deep networks are knownto achieve remarkable generalization when provided with massive amounts of l... | rkrSDoZ4e | Review | 8: Top 50% of accepted papers, clear accept | The paper proposes to study the problem of semi-supervised RL where one has to distinguish between labelled MDPs that provide rewards, and unlabelled MDPs that are not associated with any reward signal. The underlying is very simple since it aims at simultaneously learning a policy based on the REINFORCE+entropy regula... | 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
Generalizing Skills with Semi-Supervised Reinforcement Learning
### Paper Abstract
Deep reinforcement learning (RL) can acquire complex behaviors from low-level inputs, ... |
ol_xwLR2uWD | ICLR.cc/2021/Conference | 2021 | Reviving Autoencoder Pretraining | ["You Xie", "Nils Thuerey"] | The pressing need for pretraining algorithms has been diminished by numerous advances in terms of regularization, architectures, and optimizers. Despite this trend, we re-visit the classic idea of unsupervised autoencoder pretraining and propose a modified variant that relies on a full reverse pass trained in conjuncti... | ["unsupervised pretraining", "greedy layer-wise pretraining", "transfer learning", "orthogonality"] | ABSTRACTThe pressing need for pretraining algorithms has been diminished by numerousadvances in terms of regularization, architectures, and optimizers. Despite thistrend, we re-visit the classic idea of unsupervised autoencoder pretraining andpropose a modified variant that relies on a full reverse pass trained in conj... | SNUhhiKXQXf | An interesting and well described approach for unsupervised pretraining of neural networks | 9: Top 15% of accepted papers, strong accept | Quality
The paper is well written and mathematically precise. Statements are supported through helpful illustrations and the appendix provides the reader with sufficient information about the experimental setup and evaluations.
Clarity
The approach is clearly motivated and is described in full mathematical detail. 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
Reviving Autoencoder Pretraining
### Paper Abstract
The pressing need for pretraining algorithms has been diminished by numerous advances in terms of regularization, arc... |
BJgpDyHKwH | ICLR.cc/2020/Conference | 2020 | Compressive Hyperspherical Energy Minimization | ["Rongmei Lin", "Weiyang Liu", "Zhen Liu", "Chen Feng", "Zhiding Yu", "James M. Rehg", "Li Xiong", "Le Song"] | Minimum hyperspherical energy (MHE) has demonstrated its potential in regularizing neural networks and improving the generalization. MHE was inspired by the Thomson problem in physics where the distribution of multiple propelling electrons on a unit sphere can be modeled via minimizing some potential energy. Despite it... | ["comhe", "mhe", "potential", "neural networks", "dimensionality", "projection mapping", "generalization", "thomson problem", "physics"] | ABSTRACTMinimum hyperspherical energy (MHE) has demonstrated its potential in regu-larizing neural networks and improving the generalization. MHE was inspiredby the Thomson problem in physics where the distribution of multiple propellingelectrons on a unit sphere can be modeled via minimizing some potential energy.Desp... | ryxVwAyjcS | Official Blind Review #2 | 3: Weak Reject | The paper builds on the hyperspherical energy minimization line of research and proposes that instead of optimizing the distance between actual filters (or network parameters) it optimizes the distance between a projection of the filters (or network parameters). The idea is quite straightforward. Nevertheless, I do not... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Compressive Hyperspherical Energy Minimization
### Paper Abstract
Minimum hyperspherical energy (MHE) has demonstrated its potential in regularizing neural networks and ... | |
ryl5khRcKm | ICLR.cc/2019/Conference | 2019 | Human-level Protein Localization with Convolutional Neural Networks | ["Elisabeth Rumetshofer", "Markus Hofmarcher", "Clemens R\u00f6hrl", "Sepp Hochreiter", "G\u00fcnter Klambauer"] | Localizing a specific protein in a human cell is essential for understanding cellular functions and biological processes of underlying diseases. A promising, low-cost,and time-efficient biotechnology for localizing proteins is high-throughput fluorescence microscopy imaging (HTI). This imaging technique stains the prot... | ["Convolutional Neural Networks", "High-resolution images", "Multiple-Instance Learning", "Microscopy Imaging", "Protein Localization"] | ABSTRACTLocalizing a specific protein in a human cell is essential for understanding cellularfunctions and biological processes of underlying diseases. A promising, low-cost,and time-efficient biotechnology for localizing proteins is high-throughput fluo-rescence microscopy imaging (HTI). This imaging technique stains ... | SJlJ53Vqnm | This is an application oriented paper with little technical contribution | 4: Ok but not good enough - rejection | This paper designed a GapNet-PL architecture and applied GapNet-PL, DenseNet, Multi-scale CNN etc. to the protein image (multi-labels) classification dataset.
Pros:
1. The proposed method has a good performance on the given task. Compared to the claimed baselines (Liimatainen et al. and human experts), the proposed a... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Human-level Protein Localization with Convolutional Neural Networks
### Paper Abstract
Localizing a specific protein in a human cell is essential for understanding cellu... |
Cue2ZEBf12 | ICLR.cc/2021/Conference | 2021 | Towards Adversarial Robustness of Bayesian Neural Network through Hierarchical Variational Inference | ["Byung-Kwan Lee", "Youngjoon Yu", "Yong Man Ro"] | Recent works have applied Bayesian Neural Network (BNN) to adversarial training, and shown the improvement of adversarial robustness via the BNN's strength of stochastic gradient defense. However, we have found that in general, the BNN loses its stochasticity after its training with the BNN's posterior. As a result, th... | ["bnn", "bayesian neural network", "adversarial robustness", "stochasticity", "adversarial training", "improvement", "kl divergence", "attack", "towards adversarial robustness"] | ABSTRACTRecent works have applied Bayesian Neural Network (BNN) to adversarial train-ing, and shown the improvement of adversarial robustness via the BNN’s strengthof stochastic gradient defense. However, we have found that in general, the BNNloses its stochasticity after its training with the BNN’s posterior. As a res... | vp6VvPN7d8 | Improving Adversarial-BNN with hierarchical variational inference | 6: Marginally above acceptance threshold | Updates:
The author addressed my concerns about the experiments. Though the improvement is marginal and I still have some concerns, I’m ok to accept the paper. I’ll change my score to 6.
========================
Summary:
The paper studied the adversarial Bayesian Neural Network and found that the stochasticity of it v... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Adversarial Robustness of Bayesian Neural Network through Hierarchical Variational Inference
### Paper Abstract
Recent works have applied Bayesian Neural Network... |
KxUlUb26-P3 | ICLR.cc/2021/Conference | 2021 | PABI: A Unified PAC-Bayesian Informativeness Measure for Incidental Supervision Signals | ["Hangfeng He", "Mingyuan Zhang", "Qiang Ning", "Dan Roth"] | Real-world applications often require making use of {\em a range of incidental supervision signals}. However, we currently lack a principled way to measure the benefit an incidental training dataset can bring, and the common practice of using indirect, weaker signals is through exhaustive experiments with various model... | ["informativeness measure", "incidental supervision", "natural language processing"] | ABSTRACTReal-world applications often require making use of a range of incidental supervi-sion signals . However, we currently lack a principled way to measure the benefitan incidental training dataset can bring, and the common practice of using indirect,weak signals is through exhaustive experiments with various model... | Gn3_Bi66Lan | A nice unified framework to measure informativeness for various incidental supervision signals, but with the lack of qualitative evaluation of the correlation studies. | 5: Marginally below acceptance threshold | ##########################################################################
Summary:
This paper proposes a unified PAC-Bayesian-based informativeness measure (PABI) to quantify the value of incidental signals. PABI can measure various types of incidental signals such as partial labels, noisy labels, constraints, auxili... | 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
PABI: A Unified PAC-Bayesian Informativeness Measure for Incidental Supervision Signals
### Paper Abstract
Real-world applications often require making use of {\em a ran... |
BJx-ZeSKDB | ICLR.cc/2020/Conference | 2020 | Compositional Embeddings: Joint Perception and Comparison of Class Label Sets | ["Zeqian Li", "Jacob Whitehill"] | We explore the idea of compositional set embeddings that can be used to infer not
just a single class, but the set of classes associated with the input data (e.g., image,
video, audio signal). This can be useful, for example, in multi-object detection in
images, or multi-speaker diarization (one-shot learning) in audio... | ["Embedding", "One-shot Learning", "Compositional Representation"] | ABSTRACTWe explore the idea of compositional set embeddings that can be used to infer notjust a single class, but the setof classes associated with the input data (e.g., image,video, audio signal). This can be useful, for example, in multi-object detection inimages, or multi-speaker diarization (one-shot learning) in a... | B1lxwbMTtS | Official Blind Review #1 | 6: Weak Accept | Summary:
=======
This paper proposes compositional embeddings i.e. embeddings that can be used to infer multiple classes from the data. In particular, the paper deals with two types of composite functions for embeddings, one that computes union of the different classes represented by each embedding vector, and the othe... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Compositional Embeddings: Joint Perception and Comparison of Class Label Sets
### Paper Abstract
We explore the idea of compositional set embeddings that can be used to ... | |
BJRxfZbAW | ICLR.cc/2018/Conference | 2018 | The Context-Aware Learner | ["Conor Durkan", "Amos Storkey", "Harrison Edwards"] | One important aspect of generalization in machine learning involves reasoning about previously seen data in new settings. Such reasoning requires learning disentangled representations of data which are interpretable in isolation, but can also be combined in a new, unseen scenario. To this end, we introduce the context-... | ["learner", "data", "model", "representations", "important aspect", "generalization", "machine", "involves", "new settings", "disentangled representations"] | ABSTRACTOne important aspect of generalization in machine learning involves reasoningabout previously seen data in new settings. Such reasoning requires learningdisentangled representations of data which are interpretable in isolation, but canalso be combined in a new, unseen scenario. To this end, we introduce the con... | BkzesZcxG | The authors propose an interesting disentangling version of the neural statistician with richer discrete structures. | 6: Marginally above acceptance threshold | The authors propose an extension to the Neural Statistician which can model contexts with multiple partially overlapping features. This model can explain datasets by taking into account covariate structure needed to explain away factors of variation and it can also share this structure partially between datasets.
A pa... | 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
The Context-Aware Learner
### Paper Abstract
One important aspect of generalization in machine learning involves reasoning about previously seen data in new settings. Su... |
r1kjEuHpZ | ICLR.cc/2018/Conference | 2018 | Learning Less-Overlapping Representations | ["Hongbao Zhang", "Pengtao Xie", "Eric Xing"] | In representation learning (RL), how to make the learned representations easy to interpret and less overfitted to training data are two important but challenging issues. To address these problems, we study a new type of regularization approach that encourages the supports of weight vectors in RL models to have small ov... | ["Less-overlapness", "regularization", "near-orthogonality", "sparsity"] | ABSTRACTIn representation learning (RL), how to make the learned representations easy tointerpret and less overfitted to training data are two important but challenging is-sues. To address these problems, we study a new type of regularization approachthat encourages the supports of weight vectors in RL models to have s... | ByL47G5lM | logdet for diversity is not novel | 3: Clear rejection | The paper proposed a new regularization approach that simultaneously encourages the weight vectors (W) to be sparse and orthogonal to each other. The argument is that the sparsity helps to eliminate the irrelevant feature vectors by making the corresponding weights zero. Nearly orthogonal sparse vectors will have zeros... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Less-Overlapping Representations
### Paper Abstract
In representation learning (RL), how to make the learned representations easy to interpret and less overfitt... |
BJgrxbqp67 | AKBC.ws/2019/Conference | 2019 | Improving Relation Extraction by Pre-trained Language Representations | ["Christoph Alt", "Marc H\u00fcbner", "Leonhard Hennig"] | Current state-of-the-art relation extraction methods typically rely on a set of lexical, syntactic, and semantic features, explicitly computed in a pre-processing step. Training feature extraction models requires additional annotated language resources, which severely restricts the applicability and portability of rela... | ["relation extraction", "deep language representations", "transformer", "transfer learning", "unsupervised pre-training"] | Automated Knowledge Base Construction (2019) Conference paperImproving Relation Extraction by Pre-trained LanguageRepresentationsChristoph Altchristoph.alt@dfki.deMarc H ubnermarc.huebner@dfki.deLeonhard Hennig leonhard.hennig@dfki.deGerman Research Center for Articial Intelligence (DFKI)Speech and Language Technology ... | S1ercHpmMV | Incremental but solid contribution | 7: Good paper, accept | This paper presents a transformer-based relation extraction model that leverages pre-training on unlabeled text with a language modeling objective.
The proposed approach is essentially an application of the OpenAI GPT to relation extraction. Although this work is rather incremental, the experiments and analysis are 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
Improving Relation Extraction by Pre-trained Language Representations
### Paper Abstract
Current state-of-the-art relation extraction methods typically rely on a set of ... |
mmmFSYM3k5 | Interspeech.org/2023/Workshop/SSW | 2023 | On the Use of Self-Supervised Speech Representations in Spontaneous Speech Synthesis | ["Siyang Wang", "Gustav Eje Henter", "Joakim Gustafson", "Eva Szekely"] | Self-supervised learning (SSL) speech representations learned
from large amounts of diverse, mixed-quality speech data
without transcriptions are gaining ground in many speech-
technology applications. Prior work has shown that SSL is
an effective intermediate representation in two-stage text-to-
speech (TTS) for both ... | ["spontaneous speech synthesis", "text-to-speech", "self-supervised learning", "mean-opinion-score prediction"] | On the Use of Self-Supervised Speech Representationsin Spontaneous Speech SynthesisSiyang Wang*, Gustav Eje Henter, Joakim Gustafson, ́Eva Sz ́ekelyDivision of Speech, Music and Hearing, KTH Royal Institute of Technology, Stockholm, Sweden∗siyangw@kth.seAbstractSelf-supervised learning (SSL) speech representations lear... | DDII0YA0d1 | This paper mainly investigates the use of self-supervised speech representations in spontaneous speech synthesis by experiments. | 6: Marginally above acceptance threshold | This paper presents a study on the use of self-supervised speech representations in spontaneous speech synthesis. The models used in this paper are existing ones. The main contribution of this paper is to build synthesis models using different SSLs and different layers and to investigate their performance by objective... | 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
On the Use of Self-Supervised Speech Representations in Spontaneous Speech Synthesis
### Paper Abstract
Self-supervised learning (SSL) speech representations learned fro... |
H92-E4kFwbR | ICLR.cc/2021/Conference | 2021 | Composite Adversarial Training for Multiple Adversarial Perturbations and Beyond | ["Xinyang Zhang", "Zheng Zhang", "Ting Wang"] | One intriguing property of deep neural networks (DNNs) is their vulnerability to adversarial perturbations. Despite the plethora of work on defending against individual perturbation models, improving DNN robustness against the combinations of multiple perturbations is still fairly under-studied. In this paper, we propo... | ["adversarial examples", "deep learning", "robustness"] | ABSTRACTOne intriguing property of deep neural networks (DNNs) is their vulnerability toadversarial perturbations. Despite the plethora of work on defending against indi-vidual perturbation models, improving DNN robustness against the combinationsof multiple perturbations is still fairly under-studied. In this paper, w... | wA3w_pMvSTs | [Official Review] | 5: Marginally below acceptance threshold | #### Summary ####
This paper tackles the problem of adversarial training for the image classification task. It proposed a novel adversarial training method called composite adversarial training (CAT) against combined attacks constructed by multiple perturbations. First, CAT is based on the composite adversarial attacks... | 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
Composite Adversarial Training for Multiple Adversarial Perturbations and Beyond
### Paper Abstract
One intriguing property of deep neural networks (DNNs) is their vulne... |
SygXPaEYvH | ICLR.cc/2020/Conference | 2020 | VL-BERT: Pre-training of Generic Visual-Linguistic Representations | ["Weijie Su", "Xizhou Zhu", "Yue Cao", "Bin Li", "Lewei Lu", "Furu Wei", "Jifeng Dai"] | We introduce a new pre-trainable generic representation for visual-linguistic tasks, called Visual-Linguistic BERT (VL-BERT for short). VL-BERT adopts the simple yet powerful Transformer model as the backbone, and extends it to take both visual and linguistic embedded features as input. In it, each element of the input... | ["Visual-Linguistic", "Generic Representation", "Pre-training"] | ABSTRACTWe introduce a new pre-trainable generic representation for visual-linguistic tasks,called Visual-Linguistic BERT (VL-BERT for short). VL-BERT adopts the sim-ple yet powerful Transformer model as the backbone, and extends it to take bothvisual and linguistic embedded features as input. In it, each element of th... | S1ludY4GqS | Official Blind Review #2 | 6: Weak Accept | This paper proposed a pre-trainable generic representation for visual-linguistic tasks call VL-BERT. VL-BERT extend BERT by changing the input from subsequent sentence to image regions and modify the caption words has the additional visual feature embedding. The authors pre-train the VL-BERT on the conceptual caption d... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
VL-BERT: Pre-training of Generic Visual-Linguistic Representations
### Paper Abstract
We introduce a new pre-trainable generic representation for visual-linguistic tasks... | |
QBGYYu3l3dG | logconference.io/LOG/2022/Conference | 2022 | Reasoning-Modulated Representations | ["Petar Veli\u010dkovi\u0107", "Matko Bo\u0161njak", "Thomas Kipf", "Alexander Lerchner", "Raia Hadsell", "Razvan Pascanu", "Charles Blundell"] | Neural networks leverage robust internal representations in order to generalise. Learning them is difficult, and often requires a large training set that covers the data distribution densely. We study a common setting where our task is not purely opaque. Indeed, very often we may have access to information about the un... | ["representation learning", "algorithmic reasoning", "graph neural networks", "relational learning"] | Reasoning-Modulated RepresentationsPetar Veli ˇckovi ́cDeepMindpetarv@deepmind.comMatko BošnjakDeepMindmatko@deepmind.comThomas KipfGoogle Research, Brain TeamAlexander LerchnerDeepMindRaia HadsellDeepMindRazvan PascanuDeepMindCharles BlundellDeepMindAbstractNeural networks leverage robust internal representations in ... | aJuWZQGOras | In the real world, bouncing balls’ movements in specific spaces must obey the laws of physics, which motivates this work. This work injects knowledge from a massive abstract dataset of (x ̄, y ̄) pairs, directly through the pre-trained parameters of P. In addition, they build a neural network pipeline to reduce the dep... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Reasoning-Modulated Representations
### Paper Abstract
Neural networks leverage robust internal representations in order to generalise. Learning them is difficult, and o... | |||
jiCgp4qEvFS | eswc-conferences.org/ESWC/2021/Workshop/SeDiT | 2021 | Steps towards a Dislocation Ontology for Crystalline Materials | ["Ahmad Zainul Ihsan", "Danilo Dess\u00ec", "Mehwish Alam", "Harald Sack", "Stefan Sandfeld"] | The field of Materials Science is concerned with, e.g., properties and performance of materials. An important class of materials are crystalline materials that usually contain "dislocation" -- a line-like defect type. Dislocation decisively determine many important materials properties. Over the past decades, significa... | ["Dislocation Ontology", "Ontology Design"] | Steps towards a Dislocation Ontology forCrystalline MaterialsAhmad Zainul Ihsan1, Danilo Dess 2;3, Mehwish Alam2;3, Harald Sack2;3, andStefan Sandfeld1;41Forschungszentrum J ulich, Institute of Advanced Simulations { Materials DataScience and Informatics (IAS-9), Germanyfa.ihsan, s.sandfeld g@fz-juelich.de2FIZ Karlsruh... | lRuFOfs-Biy | Interesting - more detail requested on the domain ontology | 6: Marginally above acceptance threshold | This article proposes 'steps towards' a dislocation ontology for crystalline materials. It does so by proposing an ontology that is meant to be linked to the EMMO ontology. It includes particularly a number of classes that allow to describe the relevant geometry for the representations of lattices (slipPlane etc.). Whi... | 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
Steps towards a Dislocation Ontology for Crystalline Materials
### Paper Abstract
The field of Materials Science is concerned with, e.g., properties and performance of m... |
S1XFzYyPM | ICLR.cc/2018/Workshop | 2018 | Learning and Memorization | ["Satrajit Chatterjee"] | In the machine learning research community, it is generally believed that
there is a tension between memorization and generalization. In this work we
examine to what extent this tension exists, by exploring if it is
possible to generalize through memorization alone. Although direct memorization
with a lookup table obvi... | ["deep learning", "network architecture", "memorization", "generalization error"] | ABSTRACT )Satrajit ChatterjeeTwo Sigmafsatrajit.chatterjee g@twosigma.comABSTRACTIn the machine learning research community, it is generally believed that there isa tension between memorization and generalization. In this work we examine towhat extent this tension exists, by exploring if it is possible to generalize th... | rJGaKvMFG | Interesting model, merits discussion | 8: Top 50% of accepted papers, clear accept | The paper proposes an interesting way of building a hierarchy of features using memorization alone. Each feature is a lookup table that maps k-bit binary strings to {0, 1} corresponding to the two classes in a binary classification problem. The entry for a particular k-bit string is 0 or 1 depending on the majority cla... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning and Memorization
### Paper Abstract
In the machine learning research community, it is generally believed that there is a tension between memorization and genera... |
WznmQa42ZAx | ICLR.cc/2021/Conference | 2021 | Interpreting Graph Neural Networks for NLP With Differentiable Edge Masking | ["Michael Sejr Schlichtkrull", "Nicola De Cao", "Ivan Titov"] | Graph neural networks (GNNs) have become a popular approach to integrating structural inductive biases into NLP models. However, there has been little work on interpreting them, and specifically on understanding which parts of the graphs (e.g. syntactic trees or co-reference structures) contribute to a prediction. In t... | ["Graph neural networks", "interpretability", "sparse stochastic gates", "semantic role labeling", "question answering"] | ABSTRACTGraph neural networks (GNNs) have become a popular approach to integratingstructural inductive biases into NLP models. However, there has been little workon interpreting them, and specifically on understanding which parts of the graphs(e.g. syntactic trees or co-reference structures) contribute to a prediction.... | inK6Tvu11R1 | I am waiting for the author's response. | 6: Marginally above acceptance threshold | Summary
This paper presents a new interpreting algorithm for understanding graph neural network (GNN) models for natural language processing (NLP) tasks. The main idea is to remove redundant edges for each layer of GNN after training the model. To find the redundant edges, the authors suggest minimizing the difference ... | 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
Interpreting Graph Neural Networks for NLP With Differentiable Edge Masking
### Paper Abstract
Graph neural networks (GNNs) have become a popular approach to integrating... |
24-DxeAe2af | ICLR.cc/2021/Conference | 2021 | Accurate and fast detection of copy number variations from short-read whole-genome sequencing with deep convolutional neural network | ["Jiajin Li", "Stephen Hwang", "Luke Zhang", "Jae Hoon Sul"] | A copy number variant (CNV) is a type of genetic mutation where a stretch of DNA is lost or duplicated once or multiple times. CNVs play important roles in the development of diseases and complex traits. CNV detection with short-read DNA sequencing technology is challenging because CNVs significantly vary in size and a... | ["copy number variation", "deep learning", "convolutional neural network", "computational biology", "DNA sequencing"] | ABSTRACTA copy number variant (CNV) is a type of genetic mutation where a stretch ofDNA is lost or duplicated once or multiple times. CNVs play important roles in thedevelopment of diseases and complex traits. CNV detection with short-read DNAsequencing technology is challenging because CNVs significantly vary in size ... | bV13DskJMfC | A very short, trival paper, recommend rejection | 2: Strong rejection | The authors in the paper describe a deep learning approach to detect copy number variants (CNVs) from DNA sequencing data, CNV-Net. It described the approach by transforming the pileups into images and pass them through a CNN. This strategy has been proposed four or five years back to do SNPs and indels calling (DeepV... | 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
Accurate and fast detection of copy number variations from short-read whole-genome sequencing with deep convolutional neural network
### Paper Abstract
A copy number var... |
SkzeJ3A9F7 | ICLR.cc/2019/Conference | 2019 | Beyond Games: Bringing Exploration to Robots in Real-world | ["Deepak Pathak", "Dhiraj Gandhi", "Abhinav Gupta"] | Exploration has been a long standing problem in both model-based and model-free learning methods for sensorimotor control. While there has been major advances over the years, most of these successes have been demonstrated in either video games or simulation environments. This is primarily because the rewards (even the ... | ["Exploration", "curiosity", "manipulation"] | ABSTRACTExploration has been a long standing problem in both model-based and model-freelearning methods for sensorimotor control. While there has been major advancesover the years, most of these successes have been demonstrated in either videogames or simulation environments. This is primarily because the rewards (even... | rklLqv7q2X | Review | 5: Marginally below acceptance threshold | This paper presents an interesting way to reformulate intrinsic curiosity as a differentiable function. The authors compare the differentiable function against using prediction error via REINFORCE and DQN, showing that their intrinsic curiosity method results in more interactions with unseen objects than the other two ... | 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
Beyond Games: Bringing Exploration to Robots in Real-world
### Paper Abstract
Exploration has been a long standing problem in both model-based and model-free learning me... |
Abuzft2FfH | NeurIPS.cc/2022/Workshop/Offline_RL | 2022 | On- and Offline Multi-agent Reinforcement Learning for Disease Mitigation using Human Mobility Data | ["Sofia Hurtado", "Radu Marculescu"] | The COVID-19 pandemic generates new real-world data-driven problems such as predicting case surges, managing resource depletion, or modeling geo-spatial infection spreading. Though reinforcement learning (RL) has been previously proposed to optimize regional lock-downs, the availability of mobility tracking data with o... | ["Offline Reinforcement Learning", "Infectious Diseases", "Multi-agent Reinforcement Learning", "Human Mobility"] | On- and Offline Multi-agent Reinforcement Learningfor Disease Mitigation using Human Mobility DataSofia HurtadoUniversity of Texas at AustinAustin TX, 78712slhurtad@utexas.eduRadu MarculescuUniversity of Texas at AustinAustin TX, 78712radum@utexas.eduAbstractThe COVID-19 pandemic generates new real-world data-driven pr... | QgqiuzdJJR | Missing the mark for me | 4: Ok but not good enough - rejection | Inspired by Covid-19, authors simulate multiple agents with known disease states and incentives to discourage risky behaviour. Simulations indicate reductions in predicted disease rates relative to a naive population.
Overall, when I review a paper, I look for either strong theoretical or applied results. This 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
On- and Offline Multi-agent Reinforcement Learning for Disease Mitigation using Human Mobility Data
### Paper Abstract
The COVID-19 pandemic generates new real-world da... |
HyxQ3gSKvr | ICLR.cc/2020/Conference | 2020 | Variational Information Bottleneck for Unsupervised Clustering: Deep Gaussian Mixture Embedding | ["Yigit Ugur", "George Arvanitakis", "Abdellatif Zaidi"] | In this paper, we develop an unsupervised generative clustering framework that combines variational information bottleneck and the Gaussian Mixture Model. Specifically, in our approach we use the variational information bottleneck method and model the latent space as a mixture of Gaussians. We derive a bound on the cos... | ["clustering", "Variational Information Bottleneck", "Gaussian Mixture Model"] | ABSTRACTIn this paper, we develop an unsupervised generative clustering framework thatcombines variational information bottleneck and the Gaussian Mixture Model.Specifically, in our approach we use the variational information bottleneck methodand model the latent space as a mixture of Gaussians. We derive a bound on th... | rkgw2B4Jqr | Official Blind Review #1 | 3: Weak Reject | This paper purposes to cluster data in an unsupervised manner that estimates the distribution with GMM in latent space instead of original data space. Also, to better describe the distribution of latent code, the authors involve VIB to constraint the latent code.
The whole pipeline is clear and makes sense. And the ex... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Variational Information Bottleneck for Unsupervised Clustering: Deep Gaussian Mixture Embedding
### Paper Abstract
In this paper, we develop an unsupervised generative c... | |
3JI45wPuReY | ICLR.cc/2021/Conference | 2021 | Neural Network Surgery: Combining Training with Topology Optimization | ["Elisabeth Schiessler", "Roland Aydin", "Kevin Linka", "Christian Cyron"] | With ever increasing computational capacities, neural networks become more and more proficient at solving complex tasks. However, picking a sufficiently good network topology usually relies on expert human knowledge. Neural architecture search aims to reduce the extent of expertise that is needed. Modern architecture s... | ["Neural Architecture Search", "Genetic Algorithm", "SVD"] | ABSTRACTWith ever increasing computational capacities, neural networks become more andmore proficient at solving complex tasks. However, picking a sufficiently goodnetwork topology usually relies on expert human knowledge. Neural architecturesearch aims to reduce the extent of expertise that is needed. Modern architect... | DgjgMZvsBke | novelty & experiment | 4: Ok but not good enough - rejection | this paper presents the Surgeon, a ANN/evolutionary algorithm hybrid optimization de- signed for neural architecture search. On SVHN and CIFAR-10, the network generated by the Surgeon is able to outperform the baseline in case of suboptimal topologies, or reach comparable accuracies while pruning the underlying network... | 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 Network Surgery: Combining Training with Topology Optimization
### Paper Abstract
With ever increasing computational capacities, neural networks become more and m... |
OpUJ46CNv43 | ICLR.cc/2021/Conference | 2021 | MOFA: Modular Factorial Design for Hyperparameter Optimization | ["Bo Xiong", "Yimin Huang", "Steffen Staab", "Zhenguo Li"] | Automated hyperparameter optimization (HPO) has shown great power in many machine learning applications. While existing methods suffer from model selection, parallelism, or sample efficiency, this paper presents a new HPO method, MOdular FActorial Design (MOFA), to address these issues simultaneously. The major idea is... | ["Automated Hyperparameter Optimization", "Factorial Analysis", "Model-Free", "Sample Efficiency", "Orthogonal Latin Hypercubes"] | MOFA: Modular Factorial Design forHyperparameter OptimizationBo Xiong1,Yimin Huang2,Hanrong Ye2,Steffen Staab1,3, and Zhenguo Li21University of Stuttgart, Germany2Huawei Noah’s Ark Lab, China3University of Southampton, United KingdomAbstractThis paper presents a novel and lightweight hyperparameter optimization (HPO)me... | d9XUgUEgSo | Hyperparameter optimization via modular factorial designs. | 4: Ok but not good enough - rejection | MOFA: Modular Factorial Design for Hyperparameter Optimization
**(Summary)**
This paper investigates the use of modular factorial designs (MOFA) for hyperparameter optimization (HPO) and more generally black-box optimization. Their method does not use a surrogate to model the output of the unknown black-box function... | 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
MOFA: Modular Factorial Design for Hyperparameter Optimization
### Paper Abstract
Automated hyperparameter optimization (HPO) has shown great power in many machine learn... |
x0zleacdnee | NeurIPS.cc/2022/Workshop/SVRHM | 2022 | Adapting Brain-Like Neural Networks for Modeling Cortical Visual Prostheses | ["Jacob Granley", "Alexander Riedel", "Michael Beyeler"] | Cortical prostheses are devices implanted in the visual cortex that attempt to restore lost vision by electrically stimulating neurons. Currently, the vision provided by these devices is limited, and accurately predicting the visual percepts resulting from stimulation is an open challenge. We propose to address this ch... | ["visual prostheses", "cortical prostheses", "phosphene model", "brain-like neural network", "deep neural network", "cortical responses", "bionic vision", "neuroAI"] | Adapting Brain-Like Neural Networks for ModelingCortical Visual ProsthesesJacob GranleyDepartment of Computer ScienceUniversity of California Santa Barbarajgranley@ucsb.eduAlexander RiedelErnst-Abbe-University, Jena, Germanyalexander.riedel@eah-jena.deMichael BeyelerDepartment of Computer ScienceDepartment of Psycholog... | JnU5_lxR3T9 | Interesting visual prostheses application of neural network model | 6: Marginally above acceptance threshold | One of the challenges in cortical visual prostheses is predicting the visual percepts from the stimulation. In this work, the authors adopt a “brain-like’ neural network model to simulate the percept of the stimulation and argued that the phosphenes created by the neural network model is qualitatively similar to the ph... | 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
Adapting Brain-Like Neural Networks for Modeling Cortical Visual Prostheses
### Paper Abstract
Cortical prostheses are devices implanted in the visual cortex that attemp... |
0qbEq5UBfGD | ICLR.cc/2021/Conference | 2021 | Latent Space Semi-Supervised Time Series Data Clustering | ["Andrew Hill", "Katerina Kechris", "Russell Bowler", "Farnoush Kashani"] | Time series data is abundantly available in the real world, but there is a distinct lack of large, labeled datasets available for many types of learning tasks. Semi-supervised models, which can leverage small amounts of expert-labeled data along with a larger unlabeled dataset, have been shown to improve performance ov... | ["Semi-supervised clustering", "clustering", "deep learning", "autoencoder"] | ABSTRACTTime series data is abundantly available in the real world, but there is a distinctlack of large, labeled datasets available for many types of learning tasks. Semi-supervised models, which can leverage small amounts of expert-labeled data alongwith a larger unlabeled dataset, have been shown to improve performa... | PMEYgqXAIEK | Benchmarking different clustering losses in semi-supervised time series data clustering | 5: Marginally below acceptance threshold | This paper benchmarked three different existing losses, DB/Propotype/Sihouette, on time series clustering. Although there is no technique innovation, this is an interesting topic and the results look good to me. However, there are some concerns:
1. I am not an expert in time series clustering, especially in the semi-... | 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
Latent Space Semi-Supervised Time Series Data Clustering
### Paper Abstract
Time series data is abundantly available in the real world, but there is a distinct lack of l... |
s7-yAok04gI | CUHK.edu.hk/2021/Course/IERG5350 | 2021 | Reinforcement Learning for Contact-Rich Tasks: Robotic Peg Insertion Strategies | ["Jianbang LIU", "Ang ZHANG"] | Vision and touch are especially important when doing contact-rich manipulation
tasks in unstructured environments. It is non-trivial to manually design a robot con-
troller that combines these modalities which have very different characteristics. In
this project, to connect vision and touch, we first equip robots with ... | ["Reinforcement Learning", "robotic control", "deep learning"] | Reinforcement Learning for Contact-Rich Tasks:Robotic Peg Insertion StrategiesAng ZhangDepartment of Electronic EngineeringThe Chinese University of Hong KongShatin, Hong Kong1155118934@link.cuhk.edu.hkJianbang LiuDepartment of Electronic EngineeringThe Chinese University of Hong KongShatin, Hong Kong1155071948@link.cu... | PEcrOTIiJpA | Nice work but could be better | 7: Good paper, accept | General:
Significance: This paper uses proximal policy optimization (PPO) to learn robotic peg insertion strategies, uses PyBullet library to construct a simulation environment, and compares the experimental results between the proposed pre-train model and the proposed model without pre-train.
Novelty: Using reinforc... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Reinforcement Learning for Contact-Rich Tasks: Robotic Peg Insertion Strategies
### Paper Abstract
Vision and touch are especially important when doing contact-rich mani... |
Id4b5SY1Y8 | robot-learning.org/CoRL/2023/Conference | 2023 | PairwiseNet: Pairwise Collision Distance Learning for High-dof Robot Systems | ["Jihwan Kim", "Frank C. Park"] | Motion planning for robot manipulation systems operating in complex environments remains a challenging problem. It requires the evaluation of both the collision distance and its derivative. Owing to its computational complexity, recent studies have attempted to utilize data-driven approaches to learn the collision dist... | ["Robot Collision", "Collision Distance", "Machine Learning"] | PairwiseNet: Pairwise Collision Distance Learningfor High-dof Robot SystemsJihwan KimSeoul National Universityjihwankim@robotics.snu.ac.krFrank Chongwoo ParkSeoul National Universityfcp@snu.ac.krAbstract: Motion planning for robot manipulation systems operating in complexenvironments remains a challenging problem. It r... | laYIVpmGdW | The major strengths of these work are:
- Computation of collision distance as a pairwise problem for multi-body system is intuitive. This is how even deterministic method of computation of collision distance is performed.
- Separation of encoder and the regressor in the system architecture is useful in reducing traini... | 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
PairwiseNet: Pairwise Collision Distance Learning for High-dof Robot Systems
### Paper Abstract
Motion planning for robot manipulation systems operating in complex envir... | ||
kEnBH98BGs5 | ICLR.cc/2021/Conference | 2021 | Estimating informativeness of samples with Smooth Unique Information | ["Hrayr Harutyunyan", "Alessandro Achille", "Giovanni Paolini", "Orchid Majumder", "Avinash Ravichandran", "Rahul Bhotika", "Stefano Soatto"] | We define a notion of information that an individual sample provides to the training of a neural network, and we specialize it to measure both how much a sample informs the final weights and how much it informs the function computed by the weights. Though related, we show that these quantities have a qualitatively dif... | ["sample information", "information theory", "stability theory", "ntk", "dataset summarization"] | ABSTRACTWe define a notion of information that an individual sample provides to the train-ing of a neural network, and we specialize it to measure both how much a sam-ple informs the final weights and how much it informs the function computedby the weights. Though related, we show that these quantities have a qualita-t... | PNTbyD7UImR | A new way to address an old issue with interesting theoretical insights but unclear practical merits | 6: Marginally above acceptance threshold | The paper proposes a way to estimating sample information in the
context of neural networks. The authors propose to simplify the definition
by using a first-order approximation of an arbitrary network's architecture
and the mean squared error as the loss function. In addition, a smoothing
matrix obtained around the ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Estimating informativeness of samples with Smooth Unique Information
### Paper Abstract
We define a notion of information that an individual sample provides to the train... |
B1EGg7ZCb | ICLR.cc/2018/Conference | 2018 | Autonomous Vehicle Fleet Coordination With Deep Reinforcement Learning | ["Cane Punma"] | Autonomous vehicles are becoming more common in city transportation. Companies will begin to find a need to teach these vehicles smart city fleet coordination. Currently, simulation based modeling along with hand coded rules dictate the decision making of these autonomous vehicles. We believe that complex intelligent... | ["Deep Reinforcement Learning", "mult-agent systems"] | ABSTRACTAutonomous vehicles are becoming more common in city transportation. Com-panies will begin to find a need to teach these vehicles smart city fleet coordina-tion. Currently, simulation based modeling along with hand coded rules dictatethe decision making of these autonomous vehicles. We believe that complex inte... | HyqzL3Ogz | Interesting problem and approach, but not ready for ICLR | 3: Clear rejection | In this paper, the authors define a simulated, multi-agent “taxi pickup” task in a GridWorld environment. In the task, there are multiple taxi agents that a model must learn to control. “Customers” randomly appear throughout the task and the taxi agents receive reward for moving to the same square as a customer. Since ... | 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
Autonomous Vehicle Fleet Coordination With Deep Reinforcement Learning
### Paper Abstract
Autonomous vehicles are becoming more common in city transportation. Companies... |
B1xZD1rtPr | ICLR.cc/2020/Conference | 2020 | The Dual Information Bottleneck | ["Zoe Piran", "Naftali Tishby"] | The Information-Bottleneck (IB) framework suggests a general characterization of optimal representations in learning, and deep learning in particular. It is based on the optimal trade off between the representation complexity and accuracy, both of which are quantified by mutual information. The problem is solved by alt... | ["optimal prediction learning", "exponential families", "critical points", "information theory"] | ABSTRACTThe Information Bottleneck (IB) framework is a general characterization of op-timal representations in large scale learning that suggests an intriguing interpre-tation of efficient representations in deep learning in particular. It is based onthe optimal trade off between the representation complexity and accur... | Byl0gW8RFH | Official Blind Review #3 | 6: Weak Accept | This paper proposes a new "dual" variant of the Information Bottleneck framework. The IB framework has been the subject of many papers in past years, with a focus on understanding the inner workings of deep learning. The framework poses machine learning as optimizing an internal representation to tradeoff between retai... | <|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 Dual Information Bottleneck
### Paper Abstract
The Information-Bottleneck (IB) framework suggests a general characterization of optimal representations in learning, ... | |
B2E4hqYfKWq | aclweb.org/ACL/2022/Workshop/FL4NLP | 2022 | Adaptive Differential Privacy for Language Model Training | ["Xinwei Wu", "li gong", "Deyi Xiong"] | Although differential privacy (DP) can protect language models from leaking privacy, its indiscriminative protection on all data points reduces its practical utility. Previous works improve DP training by discriminating privacy and non-privacy data. But these works rely on datasets with prior privacy information, whic... | ["differential privacy", "language model"] | Adaptive Differential Privacy for Language ModelingXinwei Wu1†, Li Gong2, Deyi Xiong1∗1College of Intelligence and Computing, Tianjin University, Tianjin, China2ByteDance Lark AI, Beijing, China{wuxw2021,dyxiong}@tju.edu.cn, franck.gong@bytedance.comAbstractAlthough differential privacy (DP) can protectlanguage models ... | B_blH-rtTz9 | Concerns about the privacy definition | 4: Ok but not good enough - rejection | This paper proposed adaptive differential privacy. For a DP-SGD like algorithm, when adding noise, this paper proposed to use the "frequency" to scale the noise. The "frequency" is estimated by estimating perplexity based on a trustworthy language model.
My main concern is that the privacy definition is not clear to ... | 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
Adaptive Differential Privacy for Language Model Training
### Paper Abstract
Although differential privacy (DP) can protect language models from leaking privacy, its ind... |
JthLaV0RsV | ICLR.cc/2021/Conference | 2021 | Refine and Imitate: Reducing Repetition and Inconsistency in Dialogue Generation via Reinforcement Learning and Human Demonstration | ["Weiyan Shi", "Yu Li", "Saurav Sahay", "Zhou Yu"] | Despite the recent success of large-scale language models on various downstream NLP tasks, the repetition and inconsistency problems still persist in dialogue response generation. Previous approaches have attempted to avoid repetition by penalizing the language model's undesirable behaviors in the loss function. Howeve... | ["Dialogue System", "Persuasion", "Conversation"] | ABSTRACTDespite the recent success of large-scale language models on various downstreamNLP tasks, the repetition and inconsistency problems still persist in dialogue re-sponse generation. Previous approaches have attempted to avoid repetition bypenalizing the language model’s undesirable behaviors in the loss function.... | cuAl5oCRqOz | Addresses important problem, but approach may not generalize to other dialogue tasks | 4: Ok but not good enough - rejection | Summary:
The paper addresses the problem of inconsistency and repetition with large-scale language models applied to dialogue. It uses an RL objective to refine MLE-based models and does so without any user simulator. It also proposes several reward functions to identify repetitions and inconsistencies. Experiments o... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Refine and Imitate: Reducing Repetition and Inconsistency in Dialogue Generation via Reinforcement Learning and Human Demonstration
### Paper Abstract
Despite the recent... |
KD0M-BbuCYx | MICCAI.org/2022/Challenge/FLARE | 2022 | Semi-Supervised Detection, Identification and Segmentation for Abdominal Organs | ["Mingze Sun", "Yankai Jiang", "Heng Guo"] | Abdominal organ segmentation is an important prerequisite in many medical image analysis applications. Methods based on U-Net have demonstrated their scalability and achieved great success in different organ segmentation tasks. However, the limited number of data and labels hinders the training process of these methods... | ["Semi-supervised", "Organ detection", "Organ segmentation"] | Semi-Supervised Detection, Identification andSegmentation for Abdominal OrgansMingze Sun, Yankai Jiang, and Heng GuoAlibaba DAMO Academygh205191@alibaba-inc.comAbstract. Abdominal organ segmentation is an important prerequi-site in many medical image analysis applications. Methods based onU-Net have demonstrated their ... | 4YMAxeQGxi5 | A semi-surpervised segment method based on multiple segmentation networks | 6: Marginally above acceptance threshold | Their method utilizes multiple segmentation networks to segment each organ on a bounding box after organ detection and identification. They adopt a consistency voting strategy to ensure the quality of pseudo labels in teacher networks, which avoids poor quality input data affecting the performance of the student networ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Semi-Supervised Detection, Identification and Segmentation for Abdominal Organs
### Paper Abstract
Abdominal organ segmentation is an important prerequisite in many medi... |
SJggZnRcFQ | ICLR.cc/2019/Conference | 2019 | Learning Programmatically Structured Representations with Perceptor Gradients | ["Svetlin Penkov", "Subramanian Ramamoorthy"] | We present the perceptor gradients algorithm -- a novel approach to learning symbolic representations based on the idea of decomposing an agent's policy into i) a perceptor network extracting symbols from raw observation data and ii) a task encoding program which maps the input symbols to output actions. We show that t... | ["representation learning", "structured representations", "symbols", "programs"] | ABSTRACTWe present the perceptor gradients algorithm – a novel approach to learning sym-bolic representations based on the idea of decomposing an agent’s policy into i)a perceptor network extracting symbols from raw observation data and ii) a taskencoding program which maps the input symbols to output actions. We show ... | r1xbCrZgpQ | Learning Programmatically Structured Representations with Perceptor Gradients | 6: Marginally above acceptance threshold | This paper proposes the perceptor gradients algorithm to learn symbolic representations for devising autonomous agent policies to act.
The perceptor gradients algorithm decomposes a typical policy into a perceptor network that maps observations to symbolic representations and a user-provided task encoding program which... | 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
Learning Programmatically Structured Representations with Perceptor Gradients
### Paper Abstract
We present the perceptor gradients algorithm -- a novel approach to lear... |
HJgDulS7cV | MIDL.io/2019/Conference/Abstract | 2019 | Automatic prostate and prostate zones segmentation of magnetic resonance images using convolutional neural networks | ["Nader Aldoj", "Federico Biavati", "Miriam Rutz", "Florian Michallek", "Sebastian Stober", "Marc Dewey"] | Magnetic resonance imaging (MRI) provides detailed anatomical images of the prostate (PR)
and its zones. The importance of segmenting the prostate and the prostate zones, such as the
central zone (CZ) and the peripheral zone (PZ) lies in the fact that the diagnostic guidelines
differ depending on in which zone the lesi... | ["prostate gland", "image processing", "neural networks", "magnetic resonance imaging."] | Proceedings of Machine Learning Research { Under Review:1{4, 2019 Abstract { MIDL 2019 submissionAutomatic prostate and prostate zones segmentation ofmagnetic resonance images using convolutional neural networksNader Aldoj1nader.aldoj@charite.de1Charite, Department of Radiology, Berlin, GermanyFederico Biavati1federico... | B1eZ3SHR54 | this work mainly shows that a U-Net with or without extensions might be a good choice for image segmentation | 2: Reject | This abstract combines a U-Net with a DenseNet and evaluates this setup on a prostate zone segmentation task.
The results section is hard to read. It would be better if these would be presented only in tabular form like Table 1, highlighting the best performing in bold. Furthermore, which of these results are signific... | 3: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Automatic prostate and prostate zones segmentation of magnetic resonance images using convolutional neural networks
### Paper Abstract
Magnetic resonance imaging (MRI) p... |
B1eEKi0qYQ | ICLR.cc/2019/Conference | 2019 | Interactive Parallel Exploration for Reinforcement Learning in Continuous Action Spaces | ["Whiyoung Jung", "Giseung Park", "Youngchul Sung"] | In this paper, a new interactive parallel learning scheme is proposed to enhance the performance of off-policy continuous-action reinforcement learning. In the proposed interactive parallel learning scheme, multiple identical learners with their own value-functions and policies share a common experience replay buffer,... | ["reinforcement learning", "continuous action space RL"] | ABSTRACTIn this paper, a new interactive parallel learning scheme is proposed to enhance theperformance of off-policy continuous-action reinforceme nt learning. In the pro-posed interactive parallel learning scheme, multiple iden tical learners with theirown value-functions and policies share a common experience replay... | Syg5AYAt27 | Potential for impact but not well framed in prior work and limited evaluation | 6: Marginally above acceptance threshold | Revision: The authors added many references to prior work to the paper and did some additional experiments that certainly improved the quality. However, the additional results also show that the shared experience buffer doesn't have that much influence and that for the original tasks (the humanoid results in the append... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Interactive Parallel Exploration for Reinforcement Learning in Continuous Action Spaces
### Paper Abstract
In this paper, a new interactive parallel learning scheme is p... |
BJg_roAcK7 | ICLR.cc/2019/Conference | 2019 | INVASE: Instance-wise Variable Selection using Neural Networks | ["Jinsung Yoon", "James Jordon", "Mihaela van der Schaar"] | The advent of big data brings with it data with more and more dimensions and thus a growing need to be able to efficiently select which features to use for a variety of problems. While global feature selection has been a well-studied problem for quite some time, only recently has the paradigm of instance-wise feature s... | ["Instance-wise feature selection", "interpretability", "actor-critic methodology"] | ABSTRACTThe advent of big data brings with it data with more and more dimensions and thusa growing need to be able to efficiently select which features to use for a varietyof problems. While global feature selection has been a well-studied problem forquite some time, only recently has the paradigm of instance-wise feat... | BkehMR6shm | Instance-wise feature selection | 6: Marginally above acceptance threshold | This paper proposes an instance-wise feature selection method, which chooses relevant features for each individual sample. The basic idea is to minimize the KL divergence between the distribution p(Y|X) and p(Y|X^{(s)}). The authors consider the classification problem and construct three frameworks: 1) a selector netwo... | 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
INVASE: Instance-wise Variable Selection using Neural Networks
### Paper Abstract
The advent of big data brings with it data with more and more dimensions and thus a gro... |
qU-eouoIyAy | ICLR.cc/2021/Conference | 2021 | Hyperrealistic neural decoding: Reconstruction of face stimuli from fMRI measurements via the GAN latent space | ["Thirza Dado", "Ya\u011fmur G\u00fc\u00e7l\u00fct\u00fcrk", "Luca Ambrogioni", "Gabrielle Ras", "Sander E. Bosch", "Marcel van Gerven", "Umut G\u00fc\u00e7l\u00fc"] | We introduce a new framework for hyperrealistic reconstruction of perceived naturalistic stimuli from brain recordings. To this end, we embrace the use of generative adversarial networks (GANs) at the earliest step of our neural decoding pipeline by acquiring functional magnetic resonance imaging data as subjects perce... | ["Deep learning", "Face perception", "fMRI", "Generative Adversarial Networks", "Neural decoding"] | ABSTRACTWe introduce a new framework for hyperrealistic reconstruction of perceived natu-ralistic stimuli from brain recordings. To this end, we embrace the use of generativeadversarial networks (GANs) at the earliest step of our neural decoding pipelineby acquiring functional magnetic resonance imaging data as subject... | a4bOFmQ8WM4 | The stimuli were generated with the same GAN model | 4: Ok but not good enough - rejection | The paper proposes to reconstruct images of faces from fMRI measurements, using GANs. The authors collected a new dataset, showing static faces generated by a GAN model to human subjects, and recording their brain BOLD responses with MRI. Then, they learned a model to reconstruct the stimulus based on the brain respons... | 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
Hyperrealistic neural decoding: Reconstruction of face stimuli from fMRI measurements via the GAN latent space
### Paper Abstract
We introduce a new framework for hyperr... |
DRc-o6DUGVf | ICLR.cc/2021/Conference | 2021 | cross-modal knowledge enhancement mechanism for few-shot learning | ["Haiyang Zhang", "Jiaming Duan", "liang liu"] | Few-shot learning problems require models to recognize novel classes with only a few supported samples. However, it remains challenging for the model to generalize novel classes with such limited samples. Driven by human behavior, researchers introduced semantic information (e.g. novel categories descriptions, label na... | ["few-shot learning", "cross-modal", "image classification"] | ABSTRACTFew-shot learning problems require models to recognize novel classes with only afew supported samples. However, it remains challenging for the model to gener-alize novel classes with such limited samples. Driven by human behavior, re-searchers introduced semantic information (e.g. novel categories descriptions,... | xWe75ojCk8 | The basic idea is well motivated, while some details are not clearly described. | 5: Marginally below acceptance threshold | Traditional meta-learning FSL assumes that the semantic information during inference is accurate. This paper proposes a cross-modal knowledge enhancement mechanism, which can loose the accurate semantic information restriction by indexing relevant information from auxiliary semantic knowledge. Experiments on two few-sh... | 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
cross-modal knowledge enhancement mechanism for few-shot learning
### Paper Abstract
Few-shot learning problems require models to recognize novel classes with only a few... |
ByOnmlWC- | ICLR.cc/2018/Conference | 2018 | Policy Optimization by Genetic Distillation | ["Tanmay Gangwani", "Jian Peng"] | Genetic algorithms have been widely used in many practical optimization problems.
Inspired by natural selection, operators, including mutation, crossover
and selection, provide effective heuristics for search and black-box optimization.
However, they have not been shown useful for deep reinforcement learning, possibly
... | ["Genetic algorithms", "deep reinforcement learning", "imitation learning"] | ABSTRACTGenetic algorithms have been widely used in many practical optimization prob-lems. Inspired by natural selection, operators, including mutation, crossoverand selection, provide effective heuristics for search and black-box optimization.However, they have not been shown useful for deep reinforcement learning, po... | By_t2wVlG | novel but incomplete | 8: Top 50% of accepted papers, clear accept | This is a highly interesting paper that proposes a set of methods that combine ideas from imitation learning, evolutionary computation and reinforcement learning in a novel way. It combines the following ingredients:
a) a population-based setup for RL
b) a pair-selection and crossover operator
c) a policy-gradient base... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Policy Optimization by Genetic Distillation
### Paper Abstract
Genetic algorithms have been widely used in many practical optimization problems. Inspired by natural sel... |
EQtwFlmq7mx | ICLR.cc/2021/Conference | 2021 | Stochastic Proximal Point Algorithm for Large-scale Nonconvex Optimization: Convergence, Implementation, and Application to Neural Networks | ["Aysegul Bumin", "Kejun Huang"] | We revisit the stochastic proximal point algorithm (SPPA) for large-scale nonconvex optimization problems. SPPA has been shown to converge faster and more stable than the celebrated stochastic gradient descent (SGD) algorithm, and its many variations, for convex problems. However, the per-iteration update of SPPA is de... | ["sppa", "convergence", "nonconvex optimization", "implementation", "application", "sgd", "algorithm", "neural networks"] | ABSTRACTWe revisit the stochastic proximal point algorithm (SPPA) for large-scale nonconvexoptimization problems. SPPA has been shown to converge faster and more stablethan the celebrated stochastic gradient descent (SGD) algorithm, and its manyvariations, for convex problems. However, the per-iteration update of SPPA ... | 1VbaktTPQzJ | the novelty is not enough | 4: Ok but not good enough - rejection | This paper revisits the stochastic proximal point algorithm (SPPA) and apply SPPA to solve nonconvex optimization problems with efficient subproblem solvers.
Firstly, there is a work [Chen et. al 2020] which also provides the convergence result of SPPA on manifold problem and it is not the weakly convex setting.
Seco... | 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
Stochastic Proximal Point Algorithm for Large-scale Nonconvex Optimization: Convergence, Implementation, and Application to Neural Networks
### Paper Abstract
We revisit... |
BygXFkSYDH | ICLR.cc/2020/Conference | 2020 | Target-Embedding Autoencoders for Supervised Representation Learning | ["Daniel Jarrett", "Mihaela van der Schaar"] | Autoencoder-based learning has emerged as a staple for disciplining representations in unsupervised and semi-supervised settings. This paper analyzes a framework for improving generalization in a purely supervised setting, where the target space is high-dimensional. We motivate and formalize the general framework of ta... | ["autoencoders", "supervised learning", "representation learning", "target-embedding", "label-embedding"] | ABSTRACTAutoencoder-based learning has emerged as a staple for disciplining representationsin unsupervised and semi-supervised settings. This paper analyzes a frameworkfor improving generalization in a purely supervised setting, where the target spaceis high-dimensional. We motivate and formalize the general framework ... | rkecJZjucH | Official Blind Review #3 | 8: Accept | This is an extremely well-written and well-motivated paper. The idea of target-embedding autoencoders is extremely relevant for problems where the dimension of the label space is as large (or larger) than the dimension of the input features. The experiments are thorough, the theoretical guarantees are extremely well 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
Target-Embedding Autoencoders for Supervised Representation Learning
### Paper Abstract
Autoencoder-based learning has emerged as a staple for disciplining representatio... | |
HJePy3RcF7 | ICLR.cc/2019/Conference | 2019 | Rethinking learning rate schedules for stochastic optimization | ["Rong Ge", "Sham M. Kakade", "Rahul Kidambi", "Praneeth Netrapalli"] | There is a stark disparity between the learning rate schedules used in the practice of large scale machine learning and what are considered admissible learning rate schedules prescribed in the theory of stochastic approximation. Recent results, such as in the 'super-convergence' methods which use oscillating learning r... | ["SGD", "learning rate", "step size schedules", "stochastic approximation", "stochastic optimization", "deep learning", "non-convex optimization", "stochastic gradient descent"] | ABSTRACTThere is a stark disparity between the learning rate schedules used in the prac-tice of large scale machine learning and what are considered admissible learningrate schedules prescribed in the theory of stochastic approximation. Recent re-sults, such as in the ’super-convergence’ methods which use oscillating l... | Byg6z7GqhX | neat technical results, but misleading narrative | 4: Ok but not good enough - rejection | The paper studies the effect of learning-rate choices for stochastic optimization, focusing on least-mean-squares with decaying stepsizes. The main result is showing that exponentially decaying stepsizes can yield improved rates of convergence of the final iterate in terms of dependence on the condition number. The 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
Rethinking learning rate schedules for stochastic optimization
### Paper Abstract
There is a stark disparity between the learning rate schedules used in the practice of ... |
rJEgeXFex | ICLR.cc/2017/conference | 2017 | Predicting Medications from Diagnostic Codes with Recurrent Neural Networks | ["Jacek M. Bajor", "Thomas A. Lasko"] | It is a surprising fact that electronic medical records are failing at one of their primary purposes, that of tracking the set of medications that the patient is actively taking. Studies estimate that up to 50% of such lists omit active drugs, and that up to 25% of all active medications do not appear on the appropriat... | ["Deep learning", "Supervised Learning", "Applications"] | ABSTRACTIt is a surprising fact that electronic medical records are failing at one of their pri-mary purposes, that of tracking the set of medications that the patient is activelytaking. Studies estimate that up to 50% of such lists omit active drugs, and thatup to 25% of all active medications do not appear on the app... | rJEGyBz4g | Strong application work, very important problem | 8: Top 50% of accepted papers, clear accept | In light of the detailed author responses and further updates to the manuscript, I am raising my score to an 8 and reiterating my support for this paper. I think it will be among the strongest non-traditional applied deep learning work at ICLR and will receive a great deal of interest and attention from attendees.
---... | 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
Predicting Medications from Diagnostic Codes with Recurrent Neural Networks
### Paper Abstract
It is a surprising fact that electronic medical records are failing at one... |
r1eU1gHFvH | ICLR.cc/2020/Conference | 2020 | Under what circumstances do local codes emerge in feed-forward neural networks | ["Ella M. Gale", "Nicolas Martin"] | Localist coding schemes are more easily interpretable than the distributed schemes but generally believed to be biologically implausible. Recent results have found highly selective units and object detectors in NNs that are indicative of local codes (LCs). Here we undertake a constructionist study on feed-forward NNs a... | ["localist coding", "emergence", "contructionist science", "neural networks", "feed-forward", "learning representation", "distributed coding", "generalisation", "memorisation", "biological plausibility", "deep-NNs", "training conditions"] | ABSTRACTLocalist coding schemes are more easily interpretable than the distributed schemesbut generally believed to be biologically implausible. Recent results have foundhighly selective units and object detectors in NNs that are indicative of local codes(LCs). Here we undertake a constructionist study on feed-forward ... | BJeh3af0FH | Official Blind Review #1 | 3: Weak Reject | The authors studied the local codes in neural networks through a set of controlled experiments (by controlling the invariance in the input data, dimensions of the input and the hidden layers, etc.), and identified some common conditions under which local codes are more likely to emerge.
The fact that local codes tend ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Under what circumstances do local codes emerge in feed-forward neural networks
### Paper Abstract
Localist coding schemes are more easily interpretable than the distribu... | |
NsMLjcFaO8O | ICLR.cc/2021/Conference | 2021 | WaveGrad: Estimating Gradients for Waveform Generation | ["Nanxin Chen", "Yu Zhang", "Heiga Zen", "Ron J Weiss", "Mohammad Norouzi", "William Chan"] | This paper introduces WaveGrad, a conditional model for waveform generation which estimates gradients of the data density. The model is built on prior work on score matching and diffusion probabilistic models. It starts from a Gaussian white noise signal and iteratively refines the signal via a gradient-based sampler c... | ["vocoder", "diffusion", "score matching", "text-to-speech", "gradient estimation", "waveform generation"] | ABSTRACTThis paper introduces WaveGrad , a conditional model for waveform generationwhich estimates gradients of the data density. The model is built on prior workon score matching and diffusion probabilistic models. It starts from a Gaussianwhite noise signal and iteratively refines the signal via a gradient-based sam... | jo6d7RFDm-g | A very good paper studying diffusion probabilistic models for conditional audio synthesis task | 8: Top 50% of accepted papers, clear accept | The work uses diffusion probabilistic models for conditional speech synthesis tasks, specifically to convert mel-spectrogram to the raw audio waveform. Results from the proposed approach match the state-of-the-art WaveRNN model. The paper is very well-written and it is quite easy to follow. The study of the total numbe... | 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
WaveGrad: Estimating Gradients for Waveform Generation
### Paper Abstract
This paper introduces WaveGrad, a conditional model for waveform generation which estimates gra... |
BJlaG0VFDH | ICLR.cc/2020/Conference | 2020 | Decoupling Weight Regularization from Batch Size for Model Compression | ["Dongsoo Lee", "Se Jung Kwon", "Byeongwook Kim", "Yongkweon Jeon", "Baeseong Park", "Jeongin Yun", "Gu-Yeon Wei"] | Conventionally, compression-aware training performs weight compression for every mini-batch to compute the impact of compression on the loss function. In this paper, in order to study when would be the right time to compress weights during optimization steps, we propose a new hyper-parameter called Non-Regularization ... | ["Model compression", "Weight Regularization", "Batch Size", "Gradient Descent"] | ABSTRACTBatch size selection affects the generalization ability of gradient descent, andsmall batch size is usually preferred. Conventionally, weight regularization isperformed for every mini-batch without considering when would be the right timeto regularize weights during optimization steps. For model compression, wh... | rkeT1bLCFH | Official Blind Review #1 | 8: Accept | I think that the paper is very well written, I like it. The authors localized a phenomenon and demonstrated how to exploit it. I trust the results because I performed exactly the same experiments for CIFAR-10 with longer non-regularization periods and found that there is no effect (this is also that the authors show i... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Decoupling Weight Regularization from Batch Size for Model Compression
### Paper Abstract
Conventionally, compression-aware training performs weight compression for ever... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.