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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
B1x9ITVYDr | ICLR.cc/2020/Conference | 2020 | Compressive Recovery Defense: A Defense Framework for $\ell_0, \ell_2$ and $\ell_\infty$ norm attacks. | ["Jasjeet Dhaliwal", "Kyle Hambrook"] | We provide recovery guarantees for compressible signals that have been corrupted with noise and extend the framework introduced in \cite{bafna2018thwarting} to defend neural networks against $\ell_0$, $\ell_2$, and $\ell_{\infty}$-norm attacks. In the case of $\ell_0$-norm noise, we provide recovery guarantees for Iter... | ["adversarial input", "adversarial machine learning", "neural networks", "compressive sensing."] | Under review as a conference paper at ICLR 2020COMPRESSIVE RECOVERY DEFENSE : A D EFENSEFRAMEWORK FOR `0; `2,AND`1NORM ATTACKS .Anonymous authorsPaper under double-blind review1 A BSTRACTWe provide recovery guarantees for compressible signals that have been corrupted with noise andextend the framework introduced in Baf... | SylPwe8AKB | Official Blind Review #1 | 3: Weak Reject | This paper extends the compressive sensing framework introduced in Bafna et al. to handle l1 and l2 attacks. The authors provide theoretical analysis for several recovery algorithms (IHT, BP, DS) and provide experimental result on CIFAR-10, MNIST and Fashion-MNIST.
My major concern is how significant the provided res... | <|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 Recovery Defense: A Defense Framework for $\ell_0, \ell_2$ and $\ell_\infty$ norm attacks.
### Paper Abstract
We provide recovery guarantees for compressible... | |
HkgYmhR9KX | ICLR.cc/2019/Conference | 2019 | AD-VAT: An Asymmetric Dueling mechanism for learning Visual Active Tracking | ["Fangwei Zhong", "Peng Sun", "Wenhan Luo", "Tingyun Yan", "Yizhou Wang"] | Visual Active Tracking (VAT) aims at following a target object by autonomously controlling the motion system of a tracker given visual observations. Previous work has shown that the tracker can be trained in a simulator via reinforcement learning and deployed in real-world scenarios. However, during training, such a me... | ["Active tracking", "reinforcement learning", "adversarial learning", "multi agent"] | ABSTRACTVisual Active Tracking (V AT) aims at following a target object by autonomouslycontrolling the motion system of a tracker given visual observations. Previous workhas shown that the tracker can be trained in a simulator via reinforcement learningand deployed in real-world scenarios. However, during training, suc... | Hyxga8D52m | novel reward function in adversarial VAT appliation | 6: Marginally above acceptance threshold | This is in a visual active tracking application. The paper proposes a novel reward function - "partial zero sum", which only encourages the tracker-target competition when they are close and penalizes whey they are too far.
This is a very interesting problem and I see why their contribution could improve the system pe... | 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
AD-VAT: An Asymmetric Dueling mechanism for learning Visual Active Tracking
### Paper Abstract
Visual Active Tracking (VAT) aims at following a target object by autonomo... |
SJeUAj05tQ | ICLR.cc/2019/Conference | 2019 | DADAM: A consensus-based distributed adaptive gradient method for online optimization | ["Parvin Nazari", "Davoud Ataee Tarzanagh", "George Michailidis"] | Online and stochastic optimization methods such as SGD, ADAGRAD and ADAM are key algorithms in solving large-scale machine learning problems including deep learning. A number of schemes that are based on communications of nodes with a central server have been recently proposed in the literature to parallelize them. A b... | ["dadam", "distributed adaptive gradient", "adam", "online optimization dadam", "online optimization online", "stochastic optimization methods", "sgd", "adagrad", "key algorithms", "machine"] | ABSTRACTOnline and stochastic optimization methods such as SGD, ADAGRAD andADAM are key algorithms in solving large-scale machine learning problems in-cluding deep learning. A number of schemes that are based on communicationsof nodes with a central server have been recently proposed in the literature toparallelize the... | ryeU9QY627 | This paper proposes a consensus-based distributed method, namely DADAM, for online optimization. The technical details are well presented and the empirical results are convincing. | 8: Top 50% of accepted papers, clear accept | The proposed DADAM is a sophisticated combination of decentralized optimization and the adaptive moment estimation. DADAM enables data parallelization as well as decentralized computation, hence suitable for large scale machine learning problems.
Corollary 10 shows better performance of DADAM. Besides the detailed de... | 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
DADAM: A consensus-based distributed adaptive gradient method for online optimization
### Paper Abstract
Online and stochastic optimization methods such as SGD, ADAGRAD ... |
Hkxzx0NtDB | ICLR.cc/2020/Conference | 2020 | Your classifier is secretly an energy based model and you should treat it like one | ["Will Grathwohl", "Kuan-Chieh Wang", "Joern-Henrik Jacobsen", "David Duvenaud", "Mohammad Norouzi", "Kevin Swersky"] | We propose to reinterpret a standard discriminative classifier of p(y|x) as an energy based model for the joint distribution p(x, y). In this setting, the standard class probabilities can be easily computed as well as unnormalized values of p(x) and p(x|y). Within this framework, standard discriminative architectures m... | ["energy based models", "adversarial robustness", "generative models", "out of distribution detection", "outlier detection", "hybrid models", "robustness", "calibration"] | ABSTRACTWe propose to reinterpret a standard discriminative classifier of p(yjx)as an en-ergy based model for the joint distribution p(x;y). In this setting, the standardclass probabilities can be easily computed as well as unnormalized values of p(x)andp(xjy). Within this framework, standard discriminative architectur... | Skglf_h_9H | Official Blind Review #4 | 6: Weak Accept | This paper introduces the idea of energy based model to the traditional classifier, and proposes a new framework to improve the performances of the model in multiple aspects. The idea of reinterpreting the traditional classifier is very interesting, and the experiments show some good results of the proposed 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
Your classifier is secretly an energy based model and you should treat it like one
### Paper Abstract
We propose to reinterpret a standard discriminative classifier of p... | |
rkMnHjC5YQ | ICLR.cc/2019/Conference | 2019 | Improved Learning of One-hidden-layer Convolutional Neural Networks with Overlaps | ["Simon S. Du", "Surbhi Goel"] | We propose a new algorithm to learn a one-hidden-layer convolutional neural network where both the convolutional weights and the outputs weights are parameters to be learned. Our algorithm works for a general class of (potentially overlapping) patches, including commonly used structures for computer vision tasks. Our a... | ["deep learning", "parameter recovery", "convolutional neural networks", "non-convex optimization"] | ABSTRACTWe propose a new algorithm to learn a one-hidden-layer convolutional neural net-work where both the convolutional weights and the outputs weights are parametersto be learned. Our algorithm works for a general class of (potentially overlapping)patches, including commonly used structures for computer vision tasks... | SJlH_D8F37 | review | 6: Marginally above acceptance threshold | This paper gives a new algorithm for learning a two layer neural network which involves a single convolutional filter and a weight vector for different locations. The algorithm works on any symmetric input data. The techniques in this paper combines two previous approaches: 1. the algorithm Convotron for learning a sin... | 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
Improved Learning of One-hidden-layer Convolutional Neural Networks with Overlaps
### Paper Abstract
We propose a new algorithm to learn a one-hidden-layer convolutional... |
Bk_faUosM | MIDL.amsterdam/2018/Conference | 2018 | Morphological Detection Of Helicobector Pyloric Organisms On Gastric Mucosa Using Deep Learning Of The Artificial Intelligence | ["Shuanlong Che", "Chao Li", "Pifu Luo", "Longsen Chen"] | Since discovered, Helicobacter pylori (H. pylori) is acknoledged as one of the major causes of gastric ulcer, duodenal ulcer, gastritis, and gastric cancer. Detecting the infection of H. pylori in human body is of great significance for the treatment of various gastrointestinal diseases caused by Helicobacter pylori. T... | ["Helicobacter pylori", "whole slide image", "convolutional neural network", "CIFAR-10", "Tensorflow"] | Morphological Detection Of Helicobector PyloricOrganisms On Gastric Mucosa Using Deep LearningOf The Artificial IntelligenceShuanlong CheKingMed Diagnostics, Guangzhou, China.shuanlong2008@sina.comChao LiSemptian Co., Ltd., Shenzhen, China.landlee.work@gmail.comPifu LuoKingMed Diagnostics, Guangzhou, China.Opedac001@qq... | rkl3R-xCM | This paper presents the detection method of Helicobacter pylori (H. pylori) on whole slide images from gastric biopsy. Although they have defined an interesting problem, there is a lack of experiments results for evaluating algorithm and there is no technical novelty. | 2: Marginally below acceptance threshold |
Quality & Clarity
#1. This paper is well organized for description of proposed system except for some typos in sub-titles.
- “2.3 Data augment” -> Data augmentation
- What is meaning of “2.5 Machine detection”.
#2. There is a lack of experimental results. The results of proposed system should be shown step by ste... | 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
Morphological Detection Of Helicobector Pyloric Organisms On Gastric Mucosa Using Deep Learning Of The Artificial Intelligence
### Paper Abstract
Since discovered, Helic... |
0XXpJ4OtjW | ICLR.cc/2021/Conference | 2021 | Evolving Reinforcement Learning Algorithms | ["John D Co-Reyes", "Yingjie Miao", "Daiyi Peng", "Esteban Real", "Quoc V Le", "Sergey Levine", "Honglak Lee", "Aleksandra Faust"] | We propose a method for meta-learning reinforcement learning algorithms by searching over the space of computational graphs which compute the loss function for a value-based model-free RL agent to optimize. The learned algorithms are domain-agnostic and can generalize to new environments not seen during training. Our m... | ["reinforcement learning", "evolutionary algorithms", "meta-learning", "genetic programming"] | ABSTRACTWe propose a method for meta-learning reinforcement learning algorithms bysearching over the space of computational graphs which compute the loss functionfor a value-based model-free RL agent to optimize. The learned algorithms aredomain-agnostic and can generalize to new environments not seen during train-ing.... | Ccu-qH-9ZJE | Impressive approach to meta-learning RL algorithms with interesting analysis | 7: Good paper, accept | This paper proposes a method of discovering, through evolutionary search, programatically defined RL algorithms.
Pros:
-- Novel (to my knowledge) in the search language it uses, which renders the resulting learned algorithms very interpretable and generalizable (as compared to, say, using a meta-learned loss funct... | 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
Evolving Reinforcement Learning Algorithms
### Paper Abstract
We propose a method for meta-learning reinforcement learning algorithms by searching over the space of comp... |
r1YNw6sxg | ICLR.cc/2017/conference | 2017 | Learning Visual Servoing with Deep Features and Fitted Q-Iteration | ["Alex X. Lee", "Sergey Levine", "Pieter Abbeel"] | Visual servoing involves choosing actions that move a robot in response to observations from a camera, in order to reach a goal configuration in the world. Standard visual servoing approaches typically rely on manually designed features and analytical dynamics models, which limits their generalization capability and of... | ["Computer vision", "Deep learning", "Reinforcement Learning"] | ABSTRACTVisual servoing involves choosing actions that move a robot in response to ob-servations from a camera, in order to reach a goal configuration in the world.Standard visual servoing approaches typically rely on manually designed fea-tures and analytical dynamics models, which limits their generalization capabili... | SkC1OWMNl | Good approach for visual servoing, but could use better experimentation | 7: Good paper, accept | The paper proposes a novel approach for learning visual servoing based on Q-iteration. The main contributions of the paper are:
1. Bilinear dynamics model for predicting next frame (features) based on action and current frame
2. Formulation of servoing with a Q-function that learns weights for different feature channe... | 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 Visual Servoing with Deep Features and Fitted Q-Iteration
### Paper Abstract
Visual servoing involves choosing actions that move a robot in response to observat... |
BJgZBxBYPB | ICLR.cc/2020/Conference | 2020 | Learning Underlying Physical Properties From Observations For Trajectory Prediction | ["Ekaterina Nikonova", "Jochen Renz"] | In this work we present an approach that combines deep learning together with
laws of Newton’s physics for accurate trajectory predictions in physical games.
Our model learns to estimate physical properties and forces that generated given
observations, learns the relationships between available player’s actions and est... | ["Physical Games", "Deep Learning", "Physical Reasoning", "Transfer of Knowledge"] | ABSTRACTIn this work we present an approach that combines deep learning together withlaws of Newton’s physics for accurate trajectory predictions in physical games.Our model learns to estimate physical properties and forces that generated givenobservations, learns the relationships between available player’s actions an... | HkxD3tnBKH | Official Blind Review #1 | 3: Weak Reject | This paper presents a method for predicting the trajectories of objects in video (or phone) games by learning the physical parameters underlying the movements. To do so, the authors used some multi-layer perceptrons on the (x,y) trajectory in order to i) estimate the physical values of the equation (equation supposed t... | <|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 Underlying Physical Properties From Observations For Trajectory Prediction
### Paper Abstract
In this work we present an approach that combines deep learning to... | |
Hkz6aNqle | ICLR.cc/2017/conference | 2017 | Deep Error-Correcting Output Codes | ["Guoqiang Zhong", "Yuchen Zheng", "Peng Zhang", "Mengqi Li", "Junyu Dong"] | Existing deep networks are generally initialized with unsupervised methods, such as random assignments and greedy layerwise pre-training. This may result in the whole training process (initialization/pre-training + fine-tuning) to be very time consuming. In this paper, we combine the ideas of ensemble learning and deep... | ["deepecoc", "weights", "deep", "output codes", "feature learning", "outputs", "hand", "traditional ecoc", "output codes deep", "deep networks"] | ABSTRACTExisting deep networks are generally initialized with unsupervised methods, suchas random assignments and greedy layerwise pre-training. This may result in thewhole training process (initialization/pre-training + fine-tuning) to be very time-consuming. In this paper, we combine the ideas of ensemble learning an... | HkhDsHr4l | 3: Clear rejection | The paper proposes a strategy to pre-train the successive layers of a multi-layer perceptron for classification tasks. It consists in training successively each layer to predict an ECOC corresponding to the classification problem. The first layer predicts this ECOC from the input data and each successive layer takes as... | 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
Deep Error-Correcting Output Codes
### Paper Abstract
Existing deep networks are generally initialized with unsupervised methods, such as random assignments and greedy l... | |
gJYlaqL8i8 | ICLR.cc/2021/Conference | 2021 | Learning to Sample with Local and Global Contexts in Experience Replay Buffer | ["Youngmin Oh", "Kimin Lee", "Jinwoo Shin", "Eunho Yang", "Sung Ju Hwang"] | Experience replay, which enables the agents to remember and reuse experience from the past, has played a significant role in the success of off-policy reinforcement learning (RL). To utilize the experience replay efficiently, the existing sampling methods allow selecting out more meaningful experiences by imposing prio... | ["reinforcement learning", "experience replay buffer", "off-policy RL"] | ABSTRACTExperience replay, which enables the agents to remember and reuse experiencefrom the past, has played a significant role in the success of off-policy reinforcementlearning (RL). To utilize the experience replay efficiently, the existing samplingmethods allow selecting out more meaningful experiences by imposing... | oWLlXXhr0u | A sound idea and evaluation | 7: Good paper, accept | EDIT: The statements about ERO clarify the contribution considerably. 6-->7
The authors propose an adaptively sampling mechanism optimized for policy improvement (NERS). By incorporating minibatch-wide information into the sampling score (while maintaining permutation invariance), they are able to out-perform reasonab... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Sample with Local and Global Contexts in Experience Replay Buffer
### Paper Abstract
Experience replay, which enables the agents to remember and reuse exper... |
BJxV6A4FDH | ICLR.cc/2020/Conference | 2020 | DCTD: Deep Conditional Target Densities for Accurate Regression | ["Fredrik K. Gustafsson", "Martin Danelljan", "Goutam Bhat", "Thomas B. Sch\u00f6n"] | While deep learning-based classification is generally addressed using standardized approaches, a wide variety of techniques are employed for regression. In computer vision, one particularly popular such technique is that of confidence-based regression, which entails predicting a confidence value for each input-target p... | ["Computer vision", "deep learning", "regression", "object detection", "visual tracking"] | ABSTRACTWhile deep learning-based classification is generally addressed using standard-ized approaches, a wide variety of techniques are employed for regression. Incomputer vision, one particularly popular such technique is that of confidence-based regression, which entails predicting a confidence value for each input-... | Hyg75sK3tB | Official Blind Review #2 | 3: Weak Reject | This paper proposes Deep Conditional Target Densities (DCTD) for confidence-based regression problems. Under the core of the framework, DCTD adopts the Monte Carlo approximations to calculate the integral in the negative log-likelihood Eqn(2). The authors perform comprehensive experiments on four computer vision regres... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
DCTD: Deep Conditional Target Densities for Accurate Regression
### Paper Abstract
While deep learning-based classification is generally addressed using standardized app... | |
Hkbd5xZRb | ICLR.cc/2018/Conference | 2018 | Spherical CNNs | ["Taco S. Cohen", "Mario Geiger", "Jonas K\u00f6hler", "Max Welling"] | Convolutional Neural Networks (CNNs) have become the method of choice for learning problems involving 2D planar images. However, a number of problems of recent interest have created a demand for models that can analyze spherical images. Examples include omnidirectional vision for drones, robots, and autonomous cars, mo... | ["deep learning", "equivariance", "convolution", "group convolution", "3D", "vision", "omnidirectional", "shape recognition", "molecular energy regression"] | ABSTRACTConvolutional Neural Networks (CNNs) have become the method of choice forlearning problems involving 2D planar images. However, a number of problems ofrecent interest have created a demand for models that can analyze spherical images.Examples include omnidirectional vision for drones, robots, and autonomous car... | SJ3LYkFez | Spherical CNNs | 8: Top 50% of accepted papers, clear accept | Summary:
The paper proposes a framework for constructing spherical convolutional networks (ConvNets) based on a novel synthesis of several existing concepts. The goal is to detect patterns in spherical signals irrespective of how they are rotated on the sphere. The key is to make the convolutional architecture rotat... | 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
Spherical CNNs
### Paper Abstract
Convolutional Neural Networks (CNNs) have become the method of choice for learning problems involving 2D planar images. However, a numb... |
SyqShMZRb | ICLR.cc/2018/Conference | 2018 | Syntax-Directed Variational Autoencoder for Structured Data | ["Hanjun Dai", "Yingtao Tian", "Bo Dai", "Steven Skiena", "Le Song"] | Deep generative models have been enjoying success in modeling continuous data. However it remains challenging to capture the representations for discrete structures with formal grammars and semantics, e.g., computer programs and molecular structures. How to generate both syntactically and semantically correct data stil... | ["generative model for structured data", "syntax-directed generation", "molecule and program optimization", "variational autoencoder"] | ABSTRACTDeep generative models have been enjoying success in modeling continuous data.However it remains challenging to capture the representations for discrete struc-tures with formal grammars and semantics, e.g., computer programs and molec-ular structures. How to generate both syntactically and semantically correct ... | SkUs6e5lG | Strong paper presents state-of-the-art results | 7: Good paper, accept | NOTE:
Would the authors kindly respond to the comment below regarding Kekulisation of the Zinc dataset? Fair comparison of the data is a serious concern. I have listed this review as a good for publication due to the novelty of ideas presented, but the accusation of misrepresentation below is a serious one and I woul... | 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
Syntax-Directed Variational Autoencoder for Structured Data
### Paper Abstract
Deep generative models have been enjoying success in modeling continuous data. However it ... |
p0sMj8oH2O | logconference.io/LOG/2022/Conference | 2022 | Neural Graph Databases | ["Maciej Besta", "Patrick Iff", "Florian Scheidl", "Kazuki Osawa", "Nikoli Dryden", "Michal Podstawski", "Tiancheng Chen", "Torsten Hoefler"] | Graph databases (GDBs) enable processing and analysis of unstructured, complex,
rich, and usually vast graph datasets. Despite the large significance of GDBs
in both academia and industry, little effort has been made into integrating
them with the predictive power of graph neural networks (GNNs). In this work,
we show... | ["gdbs", "neural graph databases", "predictive power", "gnns", "systems", "lpg", "properties", "neural graph", "databases"] | Neural Graph DatabasesMaciej Besta1;yPatrick Iff1Florian Scheidl1Kazuki Osawa1Nikoli Dryden1Michal Podstawski2;3Tiancheng Chen1Torsten Hoefler1;y1Department of Computer Science, ETH Zurich2Warsaw University of Technology, Warsaw, Poland3TCL Research Europe, Warsaw, PolandyCorresponding authors:{maciej.besta, torsten.ho... | FUKelYzao3- | Contributions of the work
=========================
* The paper introduces LPG2vec, which is basically an encoder that converts LPG labels
and properties into vertex and edge embeddings for subsequent use in graph learning models
such as GCN, GAT, etc.
Strong points
=============
1) The work is novel (modulo inspir... | <|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 Graph Databases
### Paper Abstract
Graph databases (GDBs) enable processing and analysis of unstructured, complex, rich, and usually vast graph datasets. Despite ... | |||
BbNIbVPJ-42 | ICLR.cc/2021/Conference | 2021 | The Risks of Invariant Risk Minimization | ["Elan Rosenfeld", "Pradeep Kumar Ravikumar", "Andrej Risteski"] | Invariant Causal Prediction (Peters et al., 2016) is a technique for out-of-distribution generalization which assumes that some aspects of the data distribution vary across the training set but that the underlying causal mechanisms remain constant. Recently, Arjovsky et al. (2019) proposed Invariant Risk Minimization (... | ["out-of-distribution generalization", "causality", "representation learning", "deep learning"] | ABSTRACTInvariant Causal Prediction (Peters et al., 2016) is a technique for out-of-distributiongeneralization which assumes that some aspects of the data distribution vary acrossthe training set but that the underlying causal mechanisms remain constant. Re-cently, Arjovsky et al. (2019) proposed Invariant Risk Minimiz... | ThTgSL_Nqme | The work provides notable theoretical understandings on the invariant risk minimization scheme. | 7: Good paper, accept | Pros:
* The work gives extended theoretical analysis on the effect of invariant risk minimization scheme, which is an increasingly popular framework for robust prediction. The work considerably extends the results in the original IRM paper. The results seem reasonable, and clarify implausible beliefs on the framework. ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The Risks of Invariant Risk Minimization
### Paper Abstract
Invariant Causal Prediction (Peters et al., 2016) is a technique for out-of-distribution generalization which... |
rJxRJeStvB | ICLR.cc/2020/Conference | 2020 | Learning scalable and transferable multi-robot/machine sequential assignment planning via graph embedding | ["Hyunwook Kang", "Aydar Mynbay", "James R. Morrison", "Jinkyoo Park"] | Can the success of reinforcement learning methods for simple combinatorial optimization problems be extended to multi-robot sequential assignment planning? In addition to the challenge of achieving near-optimal performance in large problems, transferability to an unseen number of robots and tasks is another key challen... | ["reinforcement learning", "multi-robot/machine", "scheduling", "planning", "scalability", "transferability", "mean-field inference", "graph embedding"] | ABSTRACTThere has recently been some success in the use of reinforcement learning meth-ods for single robot combinatorial optimization problems. In this paper, we de-velop the first learning-based method for multi robot/machine planning problemswith combinatorial nature. One real-world concern is the capability to achi... | BkxFKINy5S | Official Blind Review #2 | 3: Weak Reject | Paper addresses the problem of centralized multi-machine task assignment in an RL setting ("multi-robot reward collection"). Claim is that this has not been successfully done in a RL setting before, so a new problem is proposed (multi-agent pac-man) and results are presented on this problem. Approach proposed extends p... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning scalable and transferable multi-robot/machine sequential assignment planning via graph embedding
### Paper Abstract
Can the success of reinforcement learning me... | |
HJf7ts0cFm | ICLR.cc/2019/Conference | 2019 | State-Regularized Recurrent Networks | ["Cheng Wang", "Mathias Niepert"] | Recurrent networks are a widely used class of neural architectures. They have, however, two shortcomings. First, it is difficult to understand what exactly they learn. Second, they tend to work poorly on sequences requiring long-term memorization, despite having this capacity in principle. We aim to address both short... | ["recurrent network", "finite state machines", "state-regularized", "interpretability and explainability"] | ABSTRACTRecurrent networks are a widely used class of neural architectures. They have,however, two shortcomings. First, it is difficult to understand what exactly theylearn. Second, they tend to work poorly on sequences requiring long-term mem-orization, despite having this capacity in principle. We aim to address both... | Hyel1qtb67 | Good paper on regularizing the hidden state of LSTM to make sure it uses the cell state properly. | 6: Marginally above acceptance threshold |
Summary:
This paper is based on the observation that LSTMs use the hidden state to memorize information and the cell state (memory) is not fully utilized. To encourage the LSTM to utilize the cell state, authors constraint the hidden state to a set of centroid states and learn to transition between these centroids in... | 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
State-Regularized Recurrent Networks
### Paper Abstract
Recurrent networks are a widely used class of neural architectures. They have, however, two shortcomings. First,... |
djS_CaOq2F | ICLR.cc/2023/BlogPosts | 2023 | Thinking Like Transformers | ["Alexander M Rush", "Gail Weiss"] | Thinking like Transformers proposes a computational framework for Transformer-like calculations. The framework uses discrete computation to simulate Transformer computations. The resulting language RASP is a programming language where every program compiles down to a specific Transformer. In this blog post, we reimplem... | ["Transformers"] | 2CcsmcvORj | Thinking Like Transformers blogpost review | 8: Top 50% of accepted papers, clear accept | Summary:
Overall this is a good overview of the "Thinking Like Transformers" paper, and provides a nice selection of worked examples for a variant of the RASP language presented in that paper. The worked examples build to a 3 digit addition, via a series of well explained steps.
Comments:
- The description of the 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
Thinking Like Transformers
### Paper Abstract
Thinking like Transformers proposes a computational framework for Transformer-like calculations. The framework uses discret... | |
oyZxhRI2RiE | ICLR.cc/2021/Conference | 2021 | SCoRe: Pre-Training for Context Representation in Conversational Semantic Parsing | ["Tao Yu", "Rui Zhang", "Alex Polozov", "Christopher Meek", "Ahmed Hassan Awadallah"] | Conversational Semantic Parsing (CSP) is the task of converting a sequence of natural language queries to formal language (e.g., SQL, SPARQL) that can be executed against a structured ontology (e.g. databases, knowledge bases). To accomplish this task, a CSP system needs to model the relation between the ... | ["score", "context representation", "conversational semantic", "task", "structured ontology", "lms", "conversational semantic parsing", "csp", "sequence", "natural language queries"] | ABSTRACTConversational Semantic Parsing (CSP) is the task of converting a sequence ofnatural language queries to formal language (e.g., SQL, SPARQL) that can beexecuted against a structured ontology (e.g. databases, knowledge bases). Toaccomplish this task, a CSP system needs to model the relation between theunstructur... | rxQbNWIQsdA | Review (Edited after comments) | 4: Ok but not good enough - rejection | [Summary]
In this paper, the authors proposed a pre-training strategy for Conversational Semantic Parsing (CSP) tasks. The pre-training is run on top of any existing LM (i.e., in this work RoBERTA has been used), and uses three additional loss functions to inject the CSP inductive bias into the LM: Column Contextual Se... | 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
SCoRe: Pre-Training for Context Representation in Conversational Semantic Parsing
### Paper Abstract
Conversational Semantic Parsing (CSP) is the task of converting a se... |
uELnyih9gqb | ICLR.cc/2021/Conference | 2021 | WAVEQ: GRADIENT-BASED DEEP QUANTIZATION OF NEURAL NETWORKS THROUGH SINUSOIDAL REGULARIZATION | ["Ahmed T. Elthakeb", "Prannoy Pilligundla", "Tarek Elgindi", "Fatemehsadat Mireshghallah", "Charles-Alban Deledalle", "Hadi Esmaeilzadeh"] | Deep quantization of neural networks below eight bits can lead to superlinear benefits in storage and compute efficiency. However, homogeneously quantizing all the layers to the same level does not account for the distinction of the layers and their individual properties. Heterogenous assignment of bitwidths to individ... | ["waveq", "layers", "sinusoidal regularizer", "neural networks", "bitwidth", "period", "deep quantization", "quantized weights", "minima", "quantization levels"] | ABSTRACTDeep quantization of neural networks below eight bits can lead to superlinear ben-efits in storage and compute efficiency. However, homogeneously quantizing allthe layers to the same level does not account for the distinction of the layers andtheir individual properties. Heterogeneous assignment of bitwidths to... | P5FD1Wdg5e | Good paper, suggests a useful method for quantization-aware training | 7: Good paper, accept | The paper proposes using a sinusoidal regularizer for neural network quantization. The regularizer “WaveQ” (sin^2) pushes floating-point parameters towards quantized values. Because the period of the function is highly related to the required bit-width, it can be used to determine the bit-width while keeping good chara... | 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
WAVEQ: GRADIENT-BASED DEEP QUANTIZATION OF NEURAL NETWORKS THROUGH SINUSOIDAL REGULARIZATION
### Paper Abstract
Deep quantization of neural networks below eight bits can... |
ByeMB3Act7 | ICLR.cc/2019/Conference | 2019 | Learning to Screen for Fast Softmax Inference on Large Vocabulary Neural Networks | ["Patrick Chen", "Si Si", "Sanjiv Kumar", "Yang Li", "Cho-Jui Hsieh"] | Neural language models have been widely used in various NLP tasks, including machine translation, next word prediction and conversational agents. However, it is challenging to deploy these models on mobile devices due to their slow prediction speed, where the bottleneck is to compute top candidates in the softmax layer... | ["fast inference", "softmax computation", "natural language processing"] | ABSTRACTNeural language models have been widely used in various NLP tasks, includingmachine translation, next word prediction and conversational agents. However, itis challenging to deploy these models on mobile devices due to their slow pre-diction speed, where the bottleneck is to compute top candidates in the softma... | HyehHMVFnm | Fast and accurate approximation to softmax, but more in-depth analysis results would be required | 6: Marginally above acceptance threshold | This paper presents an approximation to the softmax function to reduce the computational cost at inference time and the proposed approach is evaluated on language modeling and machine translation tasks. The main idea of the proposed approach is to pick a subset of the most probable outputs on which exact softmax is per... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning to Screen for Fast Softmax Inference on Large Vocabulary Neural Networks
### Paper Abstract
Neural language models have been widely used in various NLP tasks, ... |
kHromd7SNA | ICLR.cc/2021/Conference | 2021 | Motion Representations for Articulated Animation | ["Aliaksandr Siarohin", "Oliver J. Woodford", "Jian Ren", "Menglei Chai", "Sergey Tulyakov"] | We propose novel motion representations for animating articulated objects consisting of distinct parts. In a completely unsupervised manner, our method identifies meaningful object parts, tracks them in a driving video, and infers their motions by considering their principal axes. In contrast to the previous keypoint-b... | ["deep learning", "imaga animation", "generative modeling"] | ABSTRACTWe propose novel motion representations for animating articulated objects consist-ing of distinct parts. In a completely unsupervised manner, our method identifiesmeaningful object parts, tracks them in a driving video, and infers their motions byconsidering their principal axes. In contrast to the previous key... | nLNUGzqhgG | Articulated motion transfer | 5: Marginally below acceptance threshold | The paper takes a still photograph of an individual and generates animation that follows video of another individual moving. The key idea is to generalize the FOMM paper to segment and transform nearly-rigid regions rather than key points, thus leading to more accurate flow fields for articulated motion. The results lo... | 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
Motion Representations for Articulated Animation
### Paper Abstract
We propose novel motion representations for animating articulated objects consisting of distinct part... |
WcyZqS5o9C | ICLR.cc/2021/Conference | 2021 | On Sparse Critical Paths of Neural Response | ["Ashkan Khakzar", "Soroosh Baselizadeh", "Saurabh Khanduja", "Christian Rupprecht", "Seong Tae Kim", "Nassir Navab"] | Is critical input information encoded in specific sparse paths within the network? The pruning objective --- finding a subset of neurons for which the response remains unchanged --- has been used to discover such paths. However, we show that paths obtained from this objective do not necessarily encode the input feature... | ["Interpretability", "Feature Attribution", "Neural Network Pruning", "Critical Paths"] | ABSTRACTIs critical input information encoded in specific sparse paths within the network?The pruning objective — finding a subset of neurons for which the response re-mains unchanged — has been used to discover such paths. However, we showthat paths obtained from this objective do not necessarily encode the input fea-... | WIbfFXuSefe | Results sound intriguing, but work lacks sufficient clarity to be evaluated | 4: Ok but not good enough - rejection | The authors examine “critical paths” in sparse, rectified deep networks. They develop an algorithm to establish that solutions to the pruning objective can include neurons that were once inactive (i.e. “dead”) and therefore not in fact a critical path, as well as examine other algorithms for this objective and arrive a... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On Sparse Critical Paths of Neural Response
### Paper Abstract
Is critical input information encoded in specific sparse paths within the network? The pruning objective -... |
0jPp4dKp3PL | ICLR.cc/2021/Conference | 2021 | Integrating linguistic knowledge into DNNs: Application to online grooming detection | ["Jay Morgan", "Adeline Paiement", "Nuria Lorenzo-Dus", "Anina Kinzel", "Matteo Di Cristofaro"] | Online grooming (OG) of children is a pervasive issue in an increasingly interconnected world. We explore various complementary methods to incorporate Corpus Linguistics (CL) knowledge into accurate and interpretable Deep Learning (DL) models. They provide an implicit text normalisation that adapts embedding spaces to ... | ["Machine Learning", "Corpus Linguistics"] | ABSTRACTOnline grooming (OG) of children is a pervasive issue in an increasingly intercon-nected world. We explore various complementary methods to incorporate CorpusLinguistics (CL) knowledge into accurate and interpretable Deep Learning (DL)models. They provide an implicit text normalisation that adapts embedding spa... | nmxzn-nfFVl | Combining deep networks and corpus linguistics for online grooming detection | 6: Marginally above acceptance threshold | This paper presents an approach to natural language processing which integrates corpus linguistics knowledge within deep neural networks (namely, an LSTM-based architecture with attention). The approach is tailored and evaluated on a specific application, namely online grooming detection.
The approach is based on (1) ... | 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
Integrating linguistic knowledge into DNNs: Application to online grooming detection
### Paper Abstract
Online grooming (OG) of children is a pervasive issue in an incre... |
jwgZh4Y4U7 | ICLR.cc/2021/Conference | 2021 | Temporal and Object Quantification Nets | ["Jiayuan Mao", "Zhezheng Luo", "Chuang Gan", "Joshua B. Tenenbaum", "Jiajun Wu", "Leslie Pack Kaelbling", "Tomer Ullman"] | We aim to learn generalizable representations for complex activities by quantifying over both entities and time, as in “the kicker is behind all the other players,” or “the player controls the ball until it moves toward the goal.” Such a structural inductive bias of object relations, object quantification, and tempora... | ["Temporal Modeling", "Object-Centric Representations"] | ABSTRACTWe aim to learn generalizable representations for complex activities by quantifyingover both entities and time, as in “the kicker is behind all the other players,” or“the player controls the ball until it moves toward the goal.” Such a structuralinductive bias of object relations, object quantification, and tem... | LVaJ2Ryf6CH | Interesting application of logic formulation in complex event understanding | 6: Marginally above acceptance threshold | This paper overall presents a model that defines the multi-person activities using logic expressions and uses neural logic models to generate recognitions and predictions over events. Specifically, the model follows the neural logic machines (Dong et al.) to define the operations in the networks. Authors demonstrate 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
Temporal and Object Quantification Nets
### Paper Abstract
We aim to learn generalizable representations for complex activities by quantifying over both entities and tim... |
BkeC_J-R- | ICLR.cc/2018/Conference | 2018 | Combination of Supervised and Reinforcement Learning For Vision-Based Autonomous Control | ["Dmitry Kangin", "Nicolas Pugeault"] | Reinforcement learning methods have recently achieved impressive results on a wide range of control problems. However, especially with complex inputs, they still require an extensive amount of training data in order to converge to a meaningful solution. This limitation largely prohibits their usage for complex input ... | ["Reinforcement learning", "deep learning", "autonomous control"] | ABSTRACTReinforcement learning methods have recently achieved impressive results on awide range of control problems. However, especially with complex inputs, theystill require an extensive amount of training data in order to converge to a meaning-ful solution. This limitation largely prohibits their usage for complex i... | SJmcBU_ez | Benefit of RL portion is unclear | 5: Marginally below acceptance threshold | This paper proposes to combine reinforcement learning with supervised learning to speed up learning. Unlike their claim in the paper, the idea of combining supervised and RL is not new. A good example of this is a supervised actor-critic by Barto (2004). I think even alphaGo uses some form of supervision. However, if I... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Combination of Supervised and Reinforcement Learning For Vision-Based Autonomous Control
### Paper Abstract
Reinforcement learning methods have recently achieved impres... |
O7ms4LFdsX | ICLR.cc/2021/Conference | 2021 | Disentangled Recurrent Wasserstein Autoencoder | ["Jun Han", "Martin Renqiang Min", "Ligong Han", "Li Erran Li", "Xuan Zhang"] | Learning disentangled representations leads to interpretable models and facilitates data generation with style transfer, which has been extensively studied on static data such as images in an unsupervised learning framework. However, only a few works have explored unsupervised disentangled sequential representation lea... | ["Sequential Representation Learning", "Disentanglement", "Recurrent Generative Model"] | ABSTRACTLearning disentangled representations leads to interpretable models and facilitatesdata generation with style transfer, which has been extensively studied on staticdata such as images in an unsupervised learning framework. However, only afew works have explored unsupervised disentangled sequential representatio... | NkTLPf0DCy0 | The proposed method has theoretical foundations and shows promising results. | 7: Good paper, accept | Summary:
This paper proposes R-WAE to learn disentangled representations. Benefit from the characteristics of WAE, this paper shows that R-WAE can better disentangle the sequential data into content space and motion space. R-WAE achieves state-of-the-art performance in both disentanglement and unconditional generation.... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Disentangled Recurrent Wasserstein Autoencoder
### Paper Abstract
Learning disentangled representations leads to interpretable models and facilitates data generation wi... |
U7-FJu0iE3t | ICLR.cc/2021/Conference | 2021 | Succinct Explanations with Cascading Decision Trees | ["JIALU ZHANG", "Mark Santolucito", "Ruzica Piskac"] | Classic decision tree learning is a binary classification algorithm that constructs models with first-class transparency - every classification has a directly derivable explanation. However, learning decision trees on modern datasets generates large trees, which in turn generate decision paths of excessive depth, obscu... | ["Decision Trees", "Explainability", "Interpretability"] | ABSTRACTClassic decision tree learning is a binary classification algorithm that constructsmodels with first-class transparency - every classification has a directly derivableexplanation. However, learning decision trees on modern datasets generates largetrees, which in turn generate decision paths of excessive depth, ... | O98YplCR3ne | Interesting idea with a too limited empirical evaluation | 5: Marginally below acceptance threshold | *Summary*
This paper introduces the Cascading Decision Tree, a novel variant of decision trees with permits to extract short explanations for a class of interest. The idea is to realize a cascade of small decision trees: at a certain level, the tree is built using all points except the positive ones correctly classifie... | 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
Succinct Explanations with Cascading Decision Trees
### Paper Abstract
Classic decision tree learning is a binary classification algorithm that constructs models with fi... |
B1VWtsA5tQ | ICLR.cc/2019/Conference | 2019 | PPO-CMA: Proximal Policy Optimization with Covariance Matrix Adaptation | ["Perttu H\u00e4m\u00e4l\u00e4inen", "Amin Babadi", "Xiaoxiao Ma", "Jaakko Lehtinen"] | Proximal Policy Optimization (PPO) is a highly popular model-free reinforcement learning (RL) approach. However, in continuous state and actions spaces and a Gaussian policy -- common in computer animation and robotics -- PPO is prone to getting stuck in local optima. In this paper, we observe a tendency of PPO to prem... | ["Continuous Control", "Reinforcement Learning", "Policy Optimization", "Policy Gradient", "Evolution Strategies", "CMA-ES", "PPO"] | ABSTRACTProximal Policy Optimization (PPO) is a highly popular model-free reinforce-ment learning (RL) approach. However, in continuous state and actions spacesand a Gaussian policy – common in computer animation and robotics – PPO isprone to getting stuck in local optima. In this paper, we observe a tendency ofPPO to ... | rkeNAG762Q | Review | 4: Ok but not good enough - rejection | This paper proposes an improvement of the PPO algorithm inspired by some components of the CMA-ES black-box optimization method. The authors evaluate the proposed method on a few Mujoco domains and compare it with PPO method using simpler exploration strategies. The results show that PPO-CMA less likely to getting stuc... | 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
PPO-CMA: Proximal Policy Optimization with Covariance Matrix Adaptation
### Paper Abstract
Proximal Policy Optimization (PPO) is a highly popular model-free reinforcemen... |
oyWkrG-LD5 | robot-learning.org/CoRL/2023/Conference | 2023 | Geometry Matching for Multi-Embodiment Grasping | ["Maria Attarian", "Muhammad Adil Asif", "Jingzhou Liu", "Ruthrash Hari", "Animesh Garg", "Igor Gilitschenski", "Jonathan Tompson"] | While significant progress has been made on the problem of generating grasps, many existing learning-based approaches still concentrate on a single embodiment, provide limited generalization to higher DoF end-effectors and cannot capture a diverse set of grasp modes. In this paper, we tackle the problem of grasping mul... | ["Multi-Embodiment", "Dexterous Grasping", "Graph Neural Networks"] | Geometry Matching for Multi-Embodiment GraspingMaria Attarian1,2,Muhammad Adil Asif2,Jingzhou Liu2,Ruthrash Hari2,Animesh Garg2,3,Igor Gilitschenski2,Jonathan Tompson11Google DeepMind,2University of Toronto,3Georgia Institute of TechnologyFigure 1: GeoMatch : Our method enables multi-embodiment grasping by conditioning... | r7ASCAq-FX | Strengths:
1. The paper addresses an essential issue in dexterous grasping. The suggested method is novel and demonstrates impressive experimental results.
2. The concept of an autoregressive model to predict the contact of each keypoint sequentially is appealing.
Weaknesses:
1. While the paper emphasizes the need for... | 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
Geometry Matching for Multi-Embodiment Grasping
### Paper Abstract
While significant progress has been made on the problem of generating grasps, many existing learning-b... | ||
hLr7JzDFQBn | ICLR.cc/2021/Conference | 2021 | Mobile Construction Benchmark | ["Wenyu Han", "Chen Feng", "Haoran Wu", "Alexander Gao", "Armand Jordana", "Dongdong Liu", "Lerrel Pinto", "Ludovic Righetti"] | We need intelligent robots to perform mobile construction, the process of moving in an environment and modifying its geometry according to a design plan. Without GPS or similar techniques, carefully engineered and learning-based methods face challenges to exactly achieve the plan due to the difficulty of accurately loc... | ["localization", "dynamic environment", "deep reinforcement learning", "benchmark"] | ABSTRACTWe need intelligent robots to perform mobile construction, the process of movingin an environment and modifying its geometry according to a design plan. WithoutGPS or similar techniques, carefully engineered and learning-based methods facechallenges to exactly achieve the plan due to the difficulty of accuratel... | ZsTe3dj4jF | The main contribution is a collection of construction tasks a robot could do, and an experimental analaysis of different algortithms. | 4: Ok but not good enough - rejection | The paper addresses the problem of robots doing mobile constructions in 1/2/3D grids. Doing mobile constructions is difficult because the robots need to move blocks for the construction while localizing in space and planning how to do the construction. The problem is addressed only in simulation with relatively simpl... | 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
Mobile Construction Benchmark
### Paper Abstract
We need intelligent robots to perform mobile construction, the process of moving in an environment and modifying its geo... |
HyVxPsC9tm | ICLR.cc/2019/Conference | 2019 | DynCNN: An Effective Dynamic Architecture on Convolutional Neural Network for Surveillance Videos | ["De-Qin Gao", "Ping-Chen Tsai", "Shanq-Jang Ruan"] | The large-scale surveillance video analysis becomes important as the development of intelligent city. The heavy computation resources neccessary for state-of-the-art deep learning model makes the real-time processing hard to be implemented. This paper exploits the characteristic of high scene similarity generally exist... | ["CNN optimization", "Reduction on convolution calculation", "dynamic convolution", "surveillance video"] | ABSTRACTThe large-scale surveillance video analysis becomes important as the developmentof intelligent city. The heavy computation resources neccessary for state-of-the-art deep learning model makes the real-time processing hard to be implemented.This paper exploits the characteristic of high scene similarity generally... | BJglEsgqnX | Review for DynCNN: An Effective Dynamic Architecture on Convolutional Neural Network for Surveillance Videos | 4: Ok but not good enough - rejection | The paper addresses the problem of computational inefficiency in video surveillance understanding approaches. It suggests an approach called Dynamic Convolution consists of Frame differencing, Prediction, and Dyn-Convolution steps. The idea is to reuse some of the convolutional feature maps, and frame features particul... | 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
DynCNN: An Effective Dynamic Architecture on Convolutional Neural Network for Surveillance Videos
### Paper Abstract
The large-scale surveillance video analysis becomes ... |
Bkel6ertwS | ICLR.cc/2020/Conference | 2020 | Learning DNA folding patterns with Recurrent Neural Networks | ["Michal Rozenwald", "Aleksandra Galitsyna", "Ekaterina Khrameeva", "Grigory Sapunov", "Mikhail S. Gelfand"] |
The recent expansion of machine learning applications to molecular biology proved to have a significant contribution to our understanding of biological systems, and genome functioning in particular. Technological advances enabled the collection of large epigenetic datasets, including information about various DNA bind... | ["Machine Learning", "Recurrent Neural Networks", "3D chromatin structure", "topologically associating domains", "computational biology."] | ABSTRACTThe recent expansion of Machine Learning applications to molecular biologyproved to have a significant contribution to our understanding of biological sys-tems, and genome functioning in particular. Technological advances enabled thecollection of large epigenetic datasets, including information about various DN... | ryg8ROfaYH | Official Blind Review #1 | 3: Weak Reject | The authors utilise DNA spatial structures called Topologically-associative domains (TADs) from Hi-C data (of the Drosophila fly) and epigenetic marks (such as binding factors) from Chiq-seq data, thereby making use of physical proximity, to predict DNA folding patterns. The authors use a bidirectional LSTM RNN further... | <|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 DNA folding patterns with Recurrent Neural Networks
### Paper Abstract
The recent expansion of machine learning applications to molecular biology proved to ha... | |
B1WLiDJvM | ICLR.cc/2018/Workshop | 2018 | Generative Adversarial Network-Based Virtual Try-On with Clothing Region | ["Shizuma Kubo", "Yusuke Iwasawa", "Yutaka Matsuo"] | We propose a virtual try-on method based on generative adversarial networks (GANs). By considering clothing regions, this method enables us to reflect the pattern of clothes better than Conditional Analogy GAN (CAGAN), an existing virtual try-on method based on GANs. Our method first obtains the clothing region on a pe... | ["Generative Adversarial Networks", "Deep Learning", "Fashion", "E-commerce"] | ABSTRACTWe propose a virtual try-on method based on generative adversarial networks(GANs). By considering clothing regions, this method enables us to reflect thepattern of clothes better than Conditional Analogy GAN (CAGAN), an existingvirtual try-on method based on GANs. Our method first obtains the clothingregion on ... | Hy2XEI-Kf | Incremental work for unsatisfying results | 3: Clear rejection | The GAN approach has been compared to CAGAN for inpainting different clothes to an input image of a person in the foreground. The approach amounts to changing the loss function injecting some new terms, the newer one is a perceptual base one.
The results are not satisfying, even if better than CAGAN. In fact, the new... | 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
Generative Adversarial Network-Based Virtual Try-On with Clothing Region
### Paper Abstract
We propose a virtual try-on method based on generative adversarial networks (... |
B1e9csRcFm | ICLR.cc/2019/Conference | 2019 | The Importance of Norm Regularization in Linear Graph Embedding: Theoretical Analysis and Empirical Demonstration | ["Yihan Gao", "Chao Zhang", "Jian Peng", "Aditya Parameswaran"] | Learning distributed representations for nodes in graphs is a crucial primitive in network analysis with a wide spectrum of applications. Linear graph embedding methods learn such representations by optimizing the likelihood of both positive and negative edges while constraining the dimension of the embedding vectors. ... | ["Graph Embedding", "Generalization Analysis", "Matrix Factorization"] | ABSTRACTLearning distributed representations for nodes in graphs is a crucial primitive innetwork analysis with a wide spectrum of applications. Linear graph embeddingmethods learn such representations by optimizing the likelihood of both positiveand negative edges while constraining the dimension of the embedding vect... | BygctM2I3Q | The paper needs more clarification on the implication from the theoretical results as well as empirical results | 4: Ok but not good enough - rejection | The main contribution of this paper is that it shows both theoretically and empirically that in linear graph embeddings, the generalization error is bounded by the norms of the embedding vectors rather than the dimension of the embedding vectors.
The list of my concerns or cons of this paper is:
- For the main theor... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The Importance of Norm Regularization in Linear Graph Embedding: Theoretical Analysis and Empirical Demonstration
### Paper Abstract
Learning distributed representations... |
rkgHY0NYwr | ICLR.cc/2020/Conference | 2020 | Discovering Motor Programs by Recomposing Demonstrations | ["Tanmay Shankar", "Shubham Tulsiani", "Lerrel Pinto", "Abhinav Gupta"] | In this paper, we present an approach to learn recomposable motor primitives across large-scale and diverse manipulation demonstrations. Current approaches to decomposing demonstrations into primitives often assume manually defined primitives and bypass the difficulty of discovering these primitives. On the other hand,... | ["Learning from Demonstration", "Imitation Learning", "Motor Primitives"] | ABSTRACTIn this paper, we present an approach to learn recomposable motor primitivesacross large-scale and diverse manipulation demonstrations. Current approaches todecomposing demonstrations into primitives often assume manually defined primi-tives and bypass the difficulty of discovering these primitives. On the othe... | rylbhjAZ5H | Official Blind Review #1246 | 6: Weak Accept | The paper aims to learn middle-level motor task primitives from unlabeled actions. The main insight is that the decomposition of motor tasks can be learned using a set of LSTMs with a loss function that minimizes the differences between the original task and the recomposed task. They evaluate their approach on MIME dat... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Discovering Motor Programs by Recomposing Demonstrations
### Paper Abstract
In this paper, we present an approach to learn recomposable motor primitives across large-sca... | |
SzjyTIc5qMP | ICLR.cc/2021/Conference | 2021 | Information Lattice Learning | ["Haizi Yu", "James Evans", "Lav R. Varshney"] | Information Lattice Learning (ILL) is a general framework to learn decomposed representations, called rules, of a signal such as an image or a probability distribution. Each rule is a coarsened signal used to gain some human-interpretable insight into what might govern the nature of the original signal. To summarize th... | ["ill", "rules", "information lattice", "signal", "information lattice learning", "general framework", "decomposed representations", "image", "probability distribution"] | ABSTRACTInformation Lattice Learning (ILL) is a general framework to learn decomposedrepresentations, called rules, of a signal such as an image or a probability distri-bution. Each rule is a coarsened signal used to gain some human-interpretableinsight into what might govern the nature of the original signal. To summa... | afIQKoSffg | Interesting new direction | 7: Good paper, accept | This paper proposes a novel learning framework called information lattice learning. It is formulated as an optimization problem that finds decomposed hierarchical representations that are efficient in explaining data using a two-phased approach. ILL generalizes Shannon's information lattice and authors demonstrate ILL ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Information Lattice Learning
### Paper Abstract
Information Lattice Learning (ILL) is a general framework to learn decomposed representations, called rules, of a signal ... |
AJY3fGPF1DC | ICLR.cc/2021/Conference | 2021 | Selecting Treatment Effects Models for Domain Adaptation Using Causal Knowledge | ["Trent Kyono", "Ioana Bica", "Zhaozhi Qian", "Mihaela van der Schaar"] | Selecting causal inference models for estimating individualized treatment effects (ITE) from observational data presents a unique challenge since the counterfactual outcomes are never observed. The problem is challenged further in the unsupervised domain adaptation (UDA) setting where we only have access to labeled sam... | ["causal inference", "treatment effects", "healthcare"] | ABSTRACTSelecting causal inference models for estimating individualized treatment effects(ITE) from observational data presents a unique challenge since the counterfactualoutcomes are never observed. The problem is challenged further in the unsuperviseddomain adaptation (UDA) setting where we only have access to labele... | yWm2kx4fb0f | Very interesting, novel idea yet maybe needs to be linked more to other methods and evaluated accordingly | 4: Ok but not good enough - rejection | Summary: the paper attacks the problem of model selection for individual treatement effect (ITE) models when the domain of learning and prediction differ. Proposal is to use causal consistency as an additional "regularizer" in existing domain adaptation (DA) model selection methods. The "regularizer" would be scoring t... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Selecting Treatment Effects Models for Domain Adaptation Using Causal Knowledge
### Paper Abstract
Selecting causal inference models for estimating individualized treatm... |
BJ8c3f-0b | ICLR.cc/2018/Conference | 2018 | Auto-Encoding Sequential Monte Carlo | ["Tuan Anh Le", "Maximilian Igl", "Tom Rainforth", "Tom Jin", "Frank Wood"] | We build on auto-encoding sequential Monte Carlo (AESMC): a method for model and proposal learning based on maximizing the lower bound to the log marginal likelihood in a broad family of structured probabilistic models. Our approach relies on the efficiency of sequential Monte Carlo (SMC) for performing inference in st... | ["Variational Autoencoders", "Inference amortization", "Model learning", "Sequential Monte Carlo", "ELBOs"] | ABSTRACTWe build on auto-encoding sequential Monte Carlo ( AESMC ):1a method for modeland proposal learning based on maximizing the lower bound to the log marginallikelihood in a broad family of structured probabilistic models. Our approachrelies on the efficiency of sequential Monte Carlo ( SMC) for performing inferen... | BkaCH2KlG | Interesting extension of IWAEs using SMC | 7: Good paper, accept | [After author feedback]
I think the approach is interesting and warrants publication. However, I think some of the counter-intuitive claims on the proposal learning are overly strong, and not supported well enough by the experiments. In the paper the authors also need to describe the differences between their work and ... | 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
Auto-Encoding Sequential Monte Carlo
### Paper Abstract
We build on auto-encoding sequential Monte Carlo (AESMC): a method for model and proposal learning based on maxim... |
lwx5gi4MIh | logconference.io/LOG/2022/Conference | 2022 | Learning Distributed Geometric Koopman Operator for Sparse Networked Dynamical Systems | ["Sayak Mukherjee", "Sai Pushpak Nandanoori", "Sheng Guan", "Khushbu Agarwal", "Subhrajit Sinha", "Soumya Kundu", "Seemita Pal", "Yinghui Wu", "Draguna L Vrabie", "Sutanay Choudhury"] | The Koopman operator theory provides an alternative to studying nonlinear networked dynamical systems (NDS) by mapping the state space to an abstract higher dimensional space where the system evolution is linear. The recent works show the application of graph neural networks (GNNs) to learn state to object-centric embe... | ["koopman operator", "geometric koopman operator", "nds", "koopman operator theory", "alternative", "state space"] | Learning Distributed Geometric Koopman Operator forSparse Networked Dynamical SystemsSayak Mukherjee1, Sai Pushpak Nandanoori1, Sheng Guan2, Khushbu Agarwal1, Subhrajit Sinha1,Soumya Kundu1, Seemita Pal1, Yinghui Wu2, Draguna L Vrabie1, Sutanay Choudhury11Pacific Northwest National Laboratory, Richland, Washington, USA... | ghsEueCs_eS | ### **Summary**:
This paper presents a learnable simulation algorithm for Networked Dynamical Systems (NDS) based on Koopman operators. In particular, Koopman operator theory asserts that a non-linear dynamical system can be reparametrised via a set of “observables”, the dynamics of which are linear (and are governed b... | <|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 Distributed Geometric Koopman Operator for Sparse Networked Dynamical Systems
### Paper Abstract
The Koopman operator theory provides an alternative to studying... | |||
rk4Wf30qKQ | ICLR.cc/2019/Conference | 2019 | Security Analysis of Deep Neural Networks Operating in the Presence of Cache Side-Channel Attacks | ["Sanghyun Hong", "Michael Davinroy", "Yigitcan Kaya", "Stuart Nevans Locke", "Ian Rackow", "Kevin Kulda", "Dana Dachman-Soled", "Tudor Dumitra\u0219"] | Recent work has introduced attacks that extract the architecture information of deep neural networks (DNN), as this knowledge enhances an adversary’s capability to conduct attacks on black-box networks. This paper presents the first in-depth security analysis of DNN fingerprinting attacks that exploit cache side-channe... | ["DNN Security Analysis", "Fingerprinting Attacks", "Cache Side-Channel"] | ABSTRACTRecent work has introduced attacks that extract the architecture information ofdeep neural networks (DNN), as this knowledge enhances an adversary’s capabil-ity to conduct attacks on black-box networks. This paper presents the first in-depthsecurity analysis of DNN fingerprinting attacks that exploit cache side... | rygfvh7h2X | Unclear whether this paper surpasses prior research | 4: Ok but not good enough - rejection | The paper describes a cache side-channel attack on a deep learning model. In a cache side-channel attack, the attacker sets up a process on the same machine where the victim process (that is running the training or evaluation job for the DNN model) is running. It is assumed that the victim process uses a common shared ... | 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
Security Analysis of Deep Neural Networks Operating in the Presence of Cache Side-Channel Attacks
### Paper Abstract
Recent work has introduced attacks that extract the ... |
B1eCk1StPH | ICLR.cc/2020/Conference | 2020 | The Generalization-Stability Tradeoff in Neural Network Pruning | ["Brian R. Bartoldson", "Ari S. Morcos", "Adrian Barbu", "Gordon Erlebacher"] | Pruning neural network parameters is often viewed as a means to compress models, but pruning has also been motivated by the desire to prevent overfitting. This motivation is particularly relevant given the perhaps surprising observation that a wide variety of pruning approaches increase test accuracy despite sometimes ... | ["pruning", "generalization", "stability", "dynamics", "regularization"] | ABSTRACTPruning neural network parameters is often viewed as a means to compress models,but pruning has also been motivated by the desire to prevent overfitting. Thismotivation is particularly relevant given the perhaps surprising observation thata wide variety of pruning approaches increase test accuracy despite somet... | H1gBba7itH | Official Blind Review #1 | 1: Reject | The paper is an empirical study that looks into the effect of neural network pruning on both the model accuracy as well as the generalization risk (defined as the difference between the training error and the test error). It concludes that while some pruning methods work, others fail. The authors argue that such discre... | <|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 Generalization-Stability Tradeoff in Neural Network Pruning
### Paper Abstract
Pruning neural network parameters is often viewed as a means to compress models, but p... | |
rJzLciCqKm | ICLR.cc/2019/Conference | 2019 | Learning from Positive and Unlabeled Data with a Selection Bias | ["Masahiro Kato", "Takeshi Teshima", "Junya Honda"] | We consider the problem of learning a binary classifier only from positive data and unlabeled data (PU learning). Recent methods of PU learning commonly assume that the labeled positive data are identically distributed as the unlabeled positive data. However, this assumption is unrealistic in many instances of PU learn... | ["PU learning", "deep learning", "machine learning", "anomaly detection", "sampling bias"] | ABSTRACTWe consider the problem of learning a binary classifier only from positive data andunlabeled data ( PU learning ). Recent methods of PU learning commonly assumethat the labeled positive data are identically distributed as the unlabeled positivedata. However, this assumption is unrealistic in many instances of P... | rJgSzZK33Q | New technique for positive-unlabeled learning focussing on addressing selection bias. | 7: Good paper, accept | In this paper, the authors present a new technique to learn from positive and unlabeled data. Specifically they are addressing the issues that arise when the positive and unlabeled data do not come from the same distribution. The way to achieve this is to learn a scoring function which preserves -the order- of the labe... | 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
Learning from Positive and Unlabeled Data with a Selection Bias
### Paper Abstract
We consider the problem of learning a binary classifier only from positive data and un... |
Pwsm7d0iWJD | robot-learning.org/CoRL/2023/Conference | 2023 | Learning Lyapunov-Stable Polynomial Dynamical Systems Through Imitation | ["Amin Abyaneh", "Hsiu-Chin Lin"] | Imitation learning is a paradigm to address complex motion planning problems by learning a policy to imitate an expert's behavior. However, relying solely on the expert's data might lead to unsafe actions when the robot deviates from the demonstrated trajectories. Stability guarantees have previously been provided util... | ["Imitation learning", "Safe learning", "Motion planning", "Dynamical system", "Semidefinite programming", "Robotic manipulation"] | Learning Lyapunov-Stable Polynomial DynamicalSystems Through ImitationAmin AbyanehDepartment of Electrical and Computer EngineeringMcGill Universityamin.abyaneh@mail.mcgill.caHsiu-Chin LinSchool of Computer ScienceMcGill Universityhsiu-chin.lin@cs.mcgill.caAbstract: Imitation learning is a paradigm to address complex m... | -IZS31hdL2z | **Strengths**
- The paper is well-written and enjoyable to read.
- I went through all the proofs in the appendix and the math seems sound and refreshingly very well explained.
- This approach can be seen as is a very interesting application of "polynomial regression", which is having a resurgence thanks to recent com... | 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 Lyapunov-Stable Polynomial Dynamical Systems Through Imitation
### Paper Abstract
Imitation learning is a paradigm to address complex motion planning problems b... | ||
WgYphAbJZS- | MICCAI.org/2022/Challenge/FLARE | 2022 | Coarse to Fine Automatic Segmentation of Abdominal Multiple Organs Based on Semi-Supervised Network | ["Yi Lv", "Yu Ning", "Junchen Wang"] | Abdominal multi-organ segmentation is fast becoming a key instrument in preoperative diagnosis. Using the results of abdominal CT image segmentation for three-dimensional reconstruction is an intuitive and accurate method for surgical planning. In this paper, we propose a stable three-stage fast automatic segmentation ... | ["Medical Image Segmentation", "Deep Learning", "Neural Network"] | Coarse to Fine Automatic Segmentation ofAbdominal Multiple OrgansYi Lv1,2[0000 −0002−3778−430X], Yu Ning2[0000 −0001−5319−3588], and JunchenWang2,3[0000 −0002−0916−9932]∗1North China Research Institute of Electro-optics, Beijing, 100015, China2School of Mechanical Engineering and Automation, Beihang University, Beijing... | 3P1NWlgw2ED | MICCAI-FLARE | 7: Good paper, accept | Advance:
'a stable three-stage fast automatic segmentation method for abdominal 13 organs ' is a good idea.
Weakness:
Lack of DSC and NSC analysis at each stage.
Advise:
In terms of time, it can be accelerated using TensorRT. | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Coarse to Fine Automatic Segmentation of Abdominal Multiple Organs Based on Semi-Supervised Network
### Paper Abstract
Abdominal multi-organ segmentation is fast becomin... |
Byj54-bAW | ICLR.cc/2018/Conference | 2018 | A Tensor Analysis on Dense Connectivity via Convolutional Arithmetic Circuits | ["Emilio Rafael Balda", "Arash Behboodi", "Rudolf Mathar"] | Several state of the art convolutional networks rely on inter-connecting different layers to ease the flow of information and gradient between their input and output layers. These techniques have enabled practitioners to successfully train deep convolutional networks with hundreds of layers. Particularly, a novel way o... | ["DenseNets", "Tensor Analysis", "Convolutional Arithmetic Circuits"] | ABSTRACTSeveral state of the art convolutional networks rely on inter-connecting differ-ent layers to ease the flow of information and gradient between their input andoutput layers. These techniques have enabled practitioners to successfully traindeep convolutional networks with hundreds of layers. Particularly, a nove... | rkoPNE9gG | Paper attempts to provide theoretic justification for 'DenseNet' by looking at an arithmetic circuit variant. There are many red flags and issues with writing, as explained below. | 4: Ok but not good enough - rejection | The paper attempts to provide a theoretical justification for "DenseNet", a neural net architecture proposed by Huang et al. that contains connections between non-successive layers. The general goal is to look at "arithmetic circuit" (AC) variants of DenseNets, in which ReLUs are replaced by linear combinations and po... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Tensor Analysis on Dense Connectivity via Convolutional Arithmetic Circuits
### Paper Abstract
Several state of the art convolutional networks rely on inter-connecting... |
HJlSmC4FPS | ICLR.cc/2020/Conference | 2020 | Robust And Interpretable Blind Image Denoising Via Bias-Free Convolutional Neural Networks | ["Sreyas Mohan", "Zahra Kadkhodaie", "Eero P. Simoncelli", "Carlos Fernandez-Granda"] | We study the generalization properties of deep convolutional neural networks for image denoising in the presence of varying noise levels. We provide extensive empirical evidence that current state-of-the-art architectures systematically overfit to the noise levels in the training set, performing very poorly at new nois... | ["denoising", "overfitting", "generalization", "robustness", "interpretability", "analysis of neural networks"] | ABSTRACTWe study the generalization properties of deep convolutional neural networks forimage denoising in the presence of varying noise levels. We provide extensiveempirical evidence that current state-of-the-art architectures systematically overfitto the noise levels in the training set, performing very poorly at new... | SJeF98JCtS | Official Blind Review #1 | 6: Weak Accept | This paper studies the generalization properties of convolutional neural networks for image denoising. This paper shows that removing constant terms from CNN models provides strong generalization across noise levels. Also, this paper provides the interpretability of the denoising model based on a linear-algebra 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
Robust And Interpretable Blind Image Denoising Via Bias-Free Convolutional Neural Networks
### Paper Abstract
We study the generalization properties of deep convolutiona... | |
5IxMM3wSLDm | ICLR.cc/2021/Conference | 2021 | Novelty Detection with Rotated Contrastive Predictive Coding | ["Dong Huk Park", "Trevor Darrell"] | The current dominant paradigm for novelty detection relies on a learned model’s capability to recover the regularities. To this end, reconstruction-based learning is often used in which the normality of an observation is expressed in how well it can be reconstructed. However, this can be limiting as anomalous data can ... | ["novelty detection", "rotated contrastive predictive", "normality", "anomalous data", "current dominant paradigm", "novelty detection relies", "learned model", "capability", "regularities", "end"] | ABSTRACTThe current dominant paradigm for novelty detection relies on a learned model’scapability to recover the regularities. To this end, reconstruction-based learningis often used in which the normality of an observation is expressed in how wellit can be reconstructed. However, this can be limiting as anomalous data... | Y5Ra6xST3HQ | Review | 3: Clear rejection | Summary:
The paper proposes a novelty detection algorithm which uses contrastive predictive coding (CPC) and a pretext task to obtain a novelty score of a sample.
Strength:
The paper provides a fairly extensive summary of related literature and attempts to include important baselines in the experiments.
The core id... | 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
Novelty Detection with Rotated Contrastive Predictive Coding
### Paper Abstract
The current dominant paradigm for novelty detection relies on a learned model’s capabilit... |
o3iritJHLfO | ICLR.cc/2021/Conference | 2021 | Bidirectional Variational Inference for Non-Autoregressive Text-to-Speech | ["Yoonhyung Lee", "Joongbo Shin", "Kyomin Jung"] | Although early text-to-speech (TTS) models such as Tacotron 2 have succeeded in generating human-like speech, their autoregressive architectures have several limitations: (1) They require a lot of time to generate a mel-spectrogram consisting of hundreds of steps. (2) The autoregressive speech generation shows a lack o... | ["text-to-speech", "speech synthesis", "non-autoregressive", "VAE"] | ABSTRACTAlthough early text-to-speech (TTS) models such as Tacotron 2 have succeededin generating human-like speech, their autoregressive architectures have severallimitations: (1) They require a lot of time to generate a mel-spectrogram con-sisting of hundreds of steps. (2) The autoregressive speech generation lacks r... | 5QX06tv9adT | Potentially valuable contribution to parallel TTS, with some concerns. | 6: Marginally above acceptance threshold | Summary:
This paper presents BVAE-TTS, which applies hierarchical VAEs (using an approach motivated by NVAE and Ladder VAEs) to the problem of parallel TTS. The main components of the system are a dot product-based attention mechanism that is used during training to produce phoneme duration targets for the parallel du... | 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
Bidirectional Variational Inference for Non-Autoregressive Text-to-Speech
### Paper Abstract
Although early text-to-speech (TTS) models such as Tacotron 2 have succeeded... |
7nfCtKep-v | ICLR.cc/2021/Conference | 2021 | EXPLORING VULNERABILITIES OF BERT-BASED APIS | ["Xuanli He", "Lingjuan Lyu", "Lichao Sun", "Xiaojun Chang", "Jun Zhao"] | Natural language processing (NLP) tasks, ranging from text classification to text
generation, have been revolutionised by pretrained BERT models. This allows
corporations to easily build powerful APIs by encapsulating fine-tuned BERT
models. These BERT-based APIs are often designed to not only provide reliable
service ... | ["BERT-based models", "vulnerabilities", "attribute inference", "transferability"] | ABSTRACTNatural language processing (NLP) tasks, ranging from text classification to textgeneration, have been revolutionised by pretrained BERT models. This allowscorporations to easily build powerful APIs by encapsulating fine-tuned BERTmodels. These BERT-based APIs are often designed to not only provide reliableserv... | 9JMsuQ5pNvG | Expanded Discussion of Impact and Limitations | 4: Ok but not good enough - rejection | ################################
Summary:
This paper presents a model extraction attack (MEA) for BERT-based models that are hosted behind an API. Using the model obtained in this step, the work aims to subsequently demonstrate attribute inference attacks (AIA) to expose sensitive information of the underlying data ... | 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
EXPLORING VULNERABILITIES OF BERT-BASED APIS
### Paper Abstract
Natural language processing (NLP) tasks, ranging from text classification to text generation, have been r... |
S1j4RqYxg | ICLR.cc/2017/conference | 2017 | Efficient Calculation of Polynomial Features on Sparse Matrices | ["Andrew Nystrom", "John Hughes"] | We provide an algorithm for polynomial feature expansion that both operates on
and produces a compressed sparse row matrix without any densification. For a
vector of dimension D, density d, and degree k the algorithm has time complexity
O(d^k * D^k) where k is the polynomial-feature order; this is an improvement by a f... | ["Supervised Learning"] | ABSTRACTWe provide an algorithm for polynomial feature expansion that both operates onand produces a compressed sparse row matrix without any densification. For avector of dimension D, density d, and degree kthe algorithm has time complexityO(dkDk)where kis the polynomial-feature order; this is an improvement by afacto... | HJsy68EEx | This is not quite related with ICLR's field of interests | 3: Clear rejection | The paper is beyond my expertise. I cannot give any solid review comments regarding the techniques that are better than an educated guess.
However, it seems to me that the topic is not very relevant to the focus of ICLR. Also the quality of writing requires improvement, especially literature review and experiment anal... | 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
Efficient Calculation of Polynomial Features on Sparse Matrices
### Paper Abstract
We provide an algorithm for polynomial feature expansion that both operates on and pro... |
BJlS634tPr | ICLR.cc/2020/Conference | 2020 | PC-DARTS: Partial Channel Connections for Memory-Efficient Architecture Search | ["Yuhui Xu", "Lingxi Xie", "Xiaopeng Zhang", "Xin Chen", "Guo-Jun Qi", "Qi Tian", "Hongkai Xiong"] | Differentiable architecture search (DARTS) provided a fast solution in finding effective network architectures, but suffered from large memory and computing overheads in jointly training a super-net and searching for an optimal architecture. In this paper, we present a novel approach, namely Partially-Connected DARTS,... | ["Neural Architecture Search", "DARTS", "Regularization", "Normalization"] | ABSTRACTDifferentiable architecture search (DARTS) provided a fast solution in findingeffective network architectures, but suffered from large memory and computingoverheads in jointly training a super-network and searching for an optimal archi-tecture. In this paper, we present a novel approach, namely, Partially-Conne... | ryg_OAC6YB | Official Blind Review #3 | 6: Weak Accept | ---
revised score. rebuttal clears my concerns.
---
Summary:
The paper proposes a partially connected differential architecture search (PC-DARTS) technique, that uses a variant of channel dropout for each node's output feature maps, and a weighted summation of concatenating all previous nodes. Searched architecture on... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
PC-DARTS: Partial Channel Connections for Memory-Efficient Architecture Search
### Paper Abstract
Differentiable architecture search (DARTS) provided a fast solution in ... | |
rqUGZQ-0XZ5 | ICLR.cc/2022/Workshop/EmeCom | 2022 | Emergent Communication for Understanding Human Language Evolution: What's Missing? | ["Lukas Galke", "Yoav Ram", "Limor Raviv"] | Emergent communication protocols among humans and artificial neural network agents do not yet share the same properties and show some critical mismatches in results. We describe three important phenomena with respect to the emergence and benefits of compositionality: ease-of-learning, generalization, and group size eff... | ["Emergent Communication", "Language Evolution"] | ABSTRACTEmergent communication protocols among humans and artificial neural networkagents do not yet share the same properties and show some critical mismatchesin results. We describe three important phenomena with respect to the emergenceand benefits of compositionality: ease-of-learning, generalization, and group siz... | r0femQPPvfq | Well-written discussion on the need of incorporating realistic constraints in EC simulations in order to reduce the gap between neural EC results and sociolinguistics evidences | Accept | **Summary**
This paper notes that results obtained in neural emergent communication often do not fit broader linguistic literature evidences. In particular, it notices that (1) the impact of group size on compositionality and (2) the gain of generalization due to compositionality are not fully replicated with artifici... | 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
Emergent Communication for Understanding Human Language Evolution: What's Missing?
### Paper Abstract
Emergent communication protocols among humans and artificial neural... |
SJxTroR9F7 | ICLR.cc/2019/Conference | 2019 | Supervised Policy Update for Deep Reinforcement Learning | ["Quan Vuong", "Yiming Zhang", "Keith W. Ross"] | We propose a new sample-efficient methodology, called Supervised Policy Update (SPU), for deep reinforcement learning. Starting with data generated by the current policy, SPU formulates and solves a constrained optimization problem in the non-parameterized proximal policy space. Using supervised regression, it then con... | ["Deep Reinforcement Learning"] | ABSTRACTWe propose a new sample-efficient methodology, called Supervised Policy Update(SPU), for deep reinforcement learning. Starting with data generated by the cur-rent policy, SPU formulates and solves a constrained optimization problem in thenon-parameterized proximal policy space. Using supervised regression, it t... | Sylkwyddn7 | SUPERVISED POLICY UPDATE | 6: Marginally above acceptance threshold | Overall this paper is ok. The algorithm seems novel, but is clearly very closely related to other things in the literature. The paper is also let down by poor exposition in several areas. The numerical results seem reasonably strong, at least against relatively old baselines.
Equation 8 is crucial to the final algorit... | 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
Supervised Policy Update for Deep Reinforcement Learning
### Paper Abstract
We propose a new sample-efficient methodology, called Supervised Policy Update (SPU), for dee... |
S13wCE9xx | ICLR.cc/2017/conference | 2017 | Riemannian Optimization for Skip-Gram Negative Sampling | ["Alexander Fonarev", "Alexey Grinchuk", "Gleb Gusev", "Pavel Serdyukov", "Ivan Oseledets"] | Skip-Gram Negative Sampling (SGNS) word embedding model, well known by its implementation in "word2vec" software, is usually optimized by stochastic gradient descent. It can be shown that optimizing for SGNS objective can be viewed as an optimization problem of searching for a good matrix with the low-rank constraint. ... | ["Natural language processing", "Unsupervised Learning"] | ABSTRACTSkip-Gram Negative Sampling (SGNS) word embedding model, well known byits implementation in “word2vec” software, is usually optimized by stochastic gra-dient descent. It can be shown that optimizing for SGNS objective can be viewedas an optimization problem of searching for a good matrix with the low-rank con-s... | SJmfRhWEg | still somewhat confused | 5: Marginally below acceptance threshold | Dear authors,
The authors' response clarified some of my confusion. But I still have the following question:
-- The response said a first contribution is a different formulation: you divide the word embedding learning into two steps, step 1 looks for a low-rank X (by Riemannian optimization), step 2 factorizes X into... | 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
Riemannian Optimization for Skip-Gram Negative Sampling
### Paper Abstract
Skip-Gram Negative Sampling (SGNS) word embedding model, well known by its implementation in "... |
BUlyHkzjgmA | ICLR.cc/2021/Conference | 2021 | Improved Estimation of Concentration Under $\ell_p$-Norm Distance Metrics Using Half Spaces | ["Jack Prescott", "Xiao Zhang", "David Evans"] | Concentration of measure has been argued to be the fundamental cause of adversarial vulnerability. Mahloujifar et al. (2019) presented an empirical way to measure the concentration of a data distribution using samples, and employed it to find lower bounds on intrinsic robustness for several benchmark datasets. However,... | ["Adversarial Examples", "Concentration of Measure", "Gaussian Isoperimetric Inequality"] | ABSTRACTConcentration of measure has been argued to be the fundamental cause of adversar-ial vulnerability. Mahloujifar et al. (2019b) presented an empirical way to measurethe concentration of a data distribution using samples, and employed it to find lowerbounds on intrinsic robustness for several benchmark datasets. ... | -4FEQtjybFl | Generalized Gaussian Concentration for Adversarial Robustness | 6: Marginally above acceptance threshold | Review
The authors generalized the Gaussian Isoperimetric Inequality to non-spherical Gaussian measures with $\ell_p$ metric structures $p\geq 2$. Building on the generalized inequality, they propose a sample-based algorithm to estimate the concentration of measure using half-spaces. The main contribution is Theorem 3.... | 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
Improved Estimation of Concentration Under $\ell_p$-Norm Distance Metrics Using Half Spaces
### Paper Abstract
Concentration of measure has been argued to be the fundame... |
qiAxL3Xqx1o | ICLR.cc/2021/Conference | 2021 | GG-GAN: A Geometric Graph Generative Adversarial Network | ["Igor Krawczuk", "Pedro Abranches", "Andreas Loukas", "Volkan Cevher"] | We study the fundamental problem of graph generation. Specifically, we treat graph generation from a geometric perspective by associating each node with a position in space and then connecting the edges based on a similarity function. We then provide new solutions to the key challenges that prevent the widespread appli... | ["GAN", "generative adversarial network", "WGAN", "GNN", "graph neural network", "generative model", "graph"] | ABSTRACTWe study the fundamental problem of graph generation. Specifically, we treatgraph generation from a geometric perspective by associating each node witha position in space and then connecting edges in-between based on a similarityfunction. We then provide new solutions to the key challenges that prevent thewides... | lnCIq70Fc8D | Interesting problem and a neat idea, but some essential parts are missing. | 5: Marginally below acceptance threshold | In general, this paper deals with an interesting and essential problem to generate geometric graphs under several standards. The whole algorithm seems easy to implement or reproduce. It seems with minor modifications to traditional autoregressor based generative graph models, the proposed framework can effectively mode... | 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
GG-GAN: A Geometric Graph Generative Adversarial Network
### Paper Abstract
We study the fundamental problem of graph generation. Specifically, we treat graph generation... |
I3zV6igAT9 | ICLR.cc/2021/Conference | 2021 | Quantile Regularization : Towards Implicit Calibration of Regression Models | ["Saiteja Utpala", "Piyush Rai"] | Recent works have shown that most deep learning models are often poorly calibrated, i.e., they may produce overconfident
predictions that are wrong, implying that their uncertainty estimates are unreliable. While a number of approaches have been proposed recently to calibrate classification models, relatively little ... | ["Calibration", "Reliable Uncertainty Quantification", "Probabilistic Deep Learning"] | ABSTRACTDeep learning models are often poorly calibrated, i.e., they may produce over-confident predictions that are wrong, implying that their uncertainty estimates areunreliable. While a number of approaches have been proposed recently to cali-brate classification models, relatively little work exists on calibrating ... | 4tyZRo2rsZh | Review | 6: Marginally above acceptance threshold | The authors consider the problem of learning quantile calibrated regressions model. A probabilistic regression model is a model that, given an input, outputs a distribution over possible scalar values. A quantile calibrated model is one is such that, for all quantiles $p$, the probability over $X, Y$ that the model pre... | 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
Quantile Regularization : Towards Implicit Calibration of Regression Models
### Paper Abstract
Recent works have shown that most deep learning models are often poorly c... |
BklS6ANFDH | ICLR.cc/2020/Conference | 2020 | Semi-Supervised Semantic Dependency Parsing Using CRF Autoencoders | ["Zixia Jia", "Youmi Ma", "Jiong Cai", "Kewei Tu"] | Semantic dependency parsing, which aims to find rich bi-lexical relationships, allows words to have multiple dependency heads, resulting in graph-structured representations. We propose an approach to semi-supervised learning of semantic dependency parsers based on the CRF autoencoder framework. Our encoder is a discrim... | ["Semi-Supervised Learning", "Semantic Dependency Parsing", "CRF Autoencoder", "Natural Language Processing"] | ABSTRACTSemantic dependency parsing, which aims to find rich bi-lexical relationships,allows words to have multiple dependency heads, resulting in graph-structuredrepresentations. We propose an approach to semi-supervised learning of semanticdependency parsers based on the CRF autoencoder framework. Our encoder isa dis... | Hkg0Vmn8YS | Official Blind Review #1393 | 3: Weak Reject | This paper focuses on semi-supervised semantic dependency parsing, using the CRF-autoencoder for integrating unlabeled data to train the model in a semi-supervised style. The paper has obtained a certain improvement in the results, indicating the effectiveness of semi-supervised training on low resource labeled data ta... | <|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 Semantic Dependency Parsing Using CRF Autoencoders
### Paper Abstract
Semantic dependency parsing, which aims to find rich bi-lexical relationships, allo... | |
SJx9ngStPH | ICLR.cc/2020/Conference | 2020 | NAS-Bench-1Shot1: Benchmarking and Dissecting One-shot Neural Architecture Search | ["Arber Zela", "Julien Siems", "Frank Hutter"] | One-shot neural architecture search (NAS) has played a crucial role in making
NAS methods computationally feasible in practice. Nevertheless, there is still a
lack of understanding on how these weight-sharing algorithms exactly work due
to the many factors controlling the dynamics of the process. In order to allow
a sc... | ["Neural Architecture Search", "Deep Learning", "Computer Vision"] | ABSTRACTOne-shot neural architecture search (NAS) has played a crucial role in makingNAS methods computationally feasible in practice. Nevertheless, there is still alack of understanding on how these weight-sharing algorithms exactly work dueto the many factors controlling the dynamics of the process. In order to allow... | BkeRiU01qB | Official Blind Review #2 | 8: Accept |
This paper proposes a benchmark dataset for evaluating One-Short Neural Architecture Search models. It extends on the idea of the NASBench-101 (Ying et. al., 2019) into One-Shot models.
I think this is important work. With all the development of neural archictecture search methods, reproducible research is paramount... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
NAS-Bench-1Shot1: Benchmarking and Dissecting One-shot Neural Architecture Search
### Paper Abstract
One-shot neural architecture search (NAS) has played a crucial role ... | |
x6x7FWFNZpg | ICLR.cc/2021/Conference | 2021 | Decentralized SGD with Asynchronous, Local and Quantized Updates | ["Giorgi Nadiradze", "Amirmojtaba Sabour", "Peter Davies", "Ilia Markov", "Shigang Li", "Dan Alistarh"] | The ability to scale distributed optimization to large node counts has been one of the main enablers of recent progress in machine learning. To this end, several techniques have been explored, such as asynchronous, quantized and decentralized communication--which significantly reduce the impact of communication and sy... | ["distributed machine learning", "SGD", "decentralized algorithms", "quantization"] | ABSTRACTThe ability to scale distributed optimization to large node counts has been one ofthe main enablers of recent progress in machine learning. To this end, severaltechniques have been explored, such as asynchronous, decentralized, or quantizedcommunication–which significantly reduce the cost of synchronization, an... | 9UWnXgSHY-5 | simple and effective distributed SGD, with asynchronous, decentralized and reduced communication extensions | 7: Good paper, accept | ### Summary
The paper proposes and analyses a distributed learning algorithm for training with Stochastic Gradient Descent a global model on a regular graph, that allows for local and asynchronous gradient updates. Nodes continuously update their local models $X^i$ by gradient descent, while they communicate with their... | 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
Decentralized SGD with Asynchronous, Local and Quantized Updates
### Paper Abstract
The ability to scale distributed optimization to large node counts has been one of th... |
SkPoRg10b | ICLR.cc/2018/Conference | 2018 | Rethinking generalization requires revisiting old ideas: statistical mechanics approaches and complex learning behavior | ["Charles H. Martin", "Michael W. Mahoney"] | We describe an approach to understand the peculiar and counterintuitive generalization properties of deep neural networks. The approach involves going beyond worst-case theoretical capacity control frameworks that have been popular in machine learning in recent years to revisit old ideas in the statistical mechanics o... | ["generalization", "old ideas", "statistical mechanics approaches", "behavior", "deep neural networks", "model", "algorithms", "complex learning behavior", "complex", "peculiar"] | ABSTRACTWe describe an approach to understand the peculiar and counterintuitive general-ization properties of deep neural networks. The approach involves going beyondworst-case theoretical capacity control frameworks that have been popular in ma-chine learning in recent years to revisit old ideas in the statistical mec... | SyBLHu_gM | Interesting remarks, nice review, but maybe lack new results? | 7: Good paper, accept | I find myself having a very hard time making a review of this paper, because I mostly agree with the intro and discussion, and certainly agree that the "typical" versus "worse case" analysis is certainly an important point. The authors are making a strong case for the use of these models to understand overfitting and... | 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 generalization requires revisiting old ideas: statistical mechanics approaches and complex learning behavior
### Paper Abstract
We describe an approach to und... |
SkgpBJrtvS | ICLR.cc/2020/Conference | 2020 | Contrastive Representation Distillation | ["Yonglong Tian", "Dilip Krishnan", "Phillip Isola"] | Often we wish to transfer representational knowledge from one neural network to another. Examples include distilling a large network into a smaller one, transferring knowledge from one sensory modality to a second, or ensembling a collection of models into a single estimator. Knowledge distillation, the standard appr... | ["Knowledge Distillation", "Representation Learning", "Contrastive Learning", "Mutual Information"] | ABSTRACTOften we wish to transfer representational knowledge from one neural network toanother. Examples include distilling a large network into a smaller one, transferringknowledge from one sensory modality to a second, or ensembling a collection ofmodels into a single estimator. Knowledge distillation, the standard a... | SkeiTRlTYS | Official Blind Review #3 | 3: Weak Reject | I do not necessarily see something wrong with the paper, but I'm not convinced of the significance (or sufficient novelty) of the approach.
The way I understand it, an independent assumption on internal representation is relaxed by capturing correlations between them and higher order dependencies between them using a... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Contrastive Representation Distillation
### Paper Abstract
Often we wish to transfer representational knowledge from one neural network to another. Examples include di... | |
Hklo5RNtwS | ICLR.cc/2020/Conference | 2020 | Behavior-Guided Reinforcement Learning | ["Aldo Pacchiano", "Jack Parker-Holder", "Yunhao Tang", "Anna Choromanska", "Krzysztof Choromanski", "Michael I. Jordan"] | We introduce a new approach for comparing reinforcement learning policies, using Wasserstein distances (WDs) in a newly defined latent behavioral space. We show that by utilizing the dual formulation of the WD, we can learn score functions over trajectories that can be in turn used to lead policy optimization towards (... | ["Reinforcement Learning", "Optimal Transport", "Evolution Strategies"] | ABSTRACTWe introduce a new approach for comparing reinforcement learning policies, usingWasserstein distances (WDs) in a newly defined latent behavioral space. We showthat by utilizing the dual formulation of the WD, we can learn score functionsover trajectories that can be in turn used to lead policy optimization towa... | Bye86rHeqB | Official Blind Review #2 | 6: Weak Accept |
Summary: this paper proposes a new regularized policy optimization (PO) method which is based on Wasserstein distances. Its idea is to use SGD to optimize the dual form of the WD, then used in two different policy search approaches TRPO and Evolution Strategies. The evaluations are carried out on a variety of control ... | <|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-Guided Reinforcement Learning
### Paper Abstract
We introduce a new approach for comparing reinforcement learning policies, using Wasserstein distances (WDs) in... | |
SkEYojRqtm | ICLR.cc/2019/Conference | 2019 | Representation Degeneration Problem in Training Natural Language Generation Models | ["Jun Gao", "Di He", "Xu Tan", "Tao Qin", "Liwei Wang", "Tieyan Liu"] | We study an interesting problem in training neural network-based models for natural language generation tasks, which we call the \emph{representation degeneration problem}. We observe that when training a model for natural language generation tasks through likelihood maximization with the weight tying trick, especially... | ["Natural Language Processing", "Representation Learning"] | ABSTRACTWe study an interesting problem in training neural network-based models for nat-ural language generation tasks, which we call the representation degenerationproblem . We observe that when training a model for natural language genera-tion tasks through likelihood maximization with the weight tying trick, especia... | ryeWCxFd3X | A simple regularization to solve a representation degeneration problem | 7: Good paper, accept | This work proposes a simple regularization term which penalize cosine similarity of word embedding parameters in the loss function. The motivation comes from empirical studies of word embedding parameters in three tasks, translation, word2vec and classification, and showed that the parameters for the translation task a... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Representation Degeneration Problem in Training Natural Language Generation Models
### Paper Abstract
We study an interesting problem in training neural network-based mo... |
H1efEp4Yvr | ICLR.cc/2020/Conference | 2020 | Global Concavity and Optimization in a Class of Dynamic Discrete Choice Models | ["Yiding Feng", "Ekaterina Khmelnitskaya", "Denis Nekipelov"] | Discrete choice models with unobserved heterogeneity are commonly used Econometric models for dynamic Economic behavior which have been adopted in practice to predict behavior of individuals and firms from schooling and job choices to strategic decisions in market competition. These models feature optimizing agents wh... | ["Reinforcement learning", "Policy Gradient", "Global Concavity", "Dynamic Discrete Choice Model"] | ABSTRACTDiscrete choice models with unobserved heterogeneity are commonly used Econo-metric models for dynamic Economic behavior which have been adopted in practiceto predict behavior of individuals and firms from schooling and job choices tostrategic decisions in market competition. These models feature optimizing age... | ryg-vTdpKB | Official Blind Review #2 | 6: Weak Accept | The paper is consider dynamic discrete choice models. It shows in an important class of discrete choice models the value function is globally concave in the policy, implying that for example policy gradients are globally convergent and are likely to converge faster in practice compared to fix-point approaches.
The pa... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Global Concavity and Optimization in a Class of Dynamic Discrete Choice Models
### Paper Abstract
Discrete choice models with unobserved heterogeneity are commonly used ... | |
_GmNcCEAea3 | eswc-conferences.org/ESWC/2021/Conference/Resources_Track | 2021 | ParaQA: A Question Answering Dataset with Paraphrase Responses for Single-Turn Conversation | ["Endri Kacupaj", "Barshana Banerjee", "Kuldeep Singh", "Jens Lehmann"] | This paper presents ParaQA, a question answering (QA) dataset with multiple paraphrased responses for single-turn conversation over knowledge graphs (KG). The dataset was created using a semi-automated framework for generating diverse paraphrasing of the answers using techniques such as back-translation. The existing d... | ["Question Answering", "Paraphrase Responses", "Single-Turn Conversation", "Knowledge Graph", "Dataset"] | ParaQA: A Question Answering Dataset withParaphrase Responses for Single-Turn ConversationEndri Kacupaj1, Barshana Banerjee1, Kuldeep Singh2, and Jens Lehmann1;31University of Bonn, Bonn, Germanyfkacupaj,jens.lehmann g@cs.uni-bonn.de s6babane@uni-bonn.de2Zerotha Research and Cerence GmbH, Germanykuldeep.singh1@cerence.... | Bg0VKP9286 | The authors presents a question answering approach based on multiple paraphrased responses for conversations over knowledge graphs. | 2: Accept | The paper argues that answer paraphrasing improves the machine learning models’ performance for single-turn conversations. The authors proposed ParaQA as a question answering dataset with multiple paraphrase responses for single-turn conversation over KGs. | 2: The reviewer is willing to defend the evaluation but not sufficiently familiar with the state of the art or the specific topic 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
ParaQA: A Question Answering Dataset with Paraphrase Responses for Single-Turn Conversation
### Paper Abstract
This paper presents ParaQA, a question answering (QA) data... |
SJu63o10b | ICLR.cc/2018/Conference | 2018 | UNSUPERVISED METRIC LEARNING VIA NONLINEAR FEATURE SPACE TRANSFORMATIONS | ["Pin Zhang", "Bibo Shi", "JundongLiu"] | In this paper, we propose a nonlinear unsupervised metric learning framework to boost of the performance of clustering algorithms. Under our framework, nonlinear distance metric learning and manifold embedding are integrated and conducted simultaneously to increase the natural separations among data samples. The metric... | ["Metric Learning", "K-means", "CPD", "Clustering"] | ABSTRACTIn this paper, we propose a nonlinear unsupervised metric learning framework toboost of the performance of clustering algorithms. Under our framework, nonlineardistance metric learning and manifold embedding are integrated and conductedsimultaneously to increase the natural separations among data samples. Theme... | rkX795cez | The paper proposed a new | 4: Ok but not good enough - rejection | This paper proposed a nonlinear unsupervised metric learning framework. The authors combine Coherent Point Drifting and the k-means approaches under the trace minimization framework. However, I am afraid that the novelty and insight of this work is not good enough for acceptance.
Pros:
The paper is well written and ea... | 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
UNSUPERVISED METRIC LEARNING VIA NONLINEAR FEATURE SPACE TRANSFORMATIONS
### Paper Abstract
In this paper, we propose a nonlinear unsupervised metric learning framework ... |
xTJEN-ggl1b | ICLR.cc/2021/Conference | 2021 | LambdaNetworks: Modeling long-range Interactions without Attention | ["Irwan Bello"] | We present lambda layers -- an alternative framework to self-attention -- for capturing long-range interactions between an input and structured contextual information (e.g. a pixel surrounded by other pixels). Lambda layers capture such interactions by transforming available contexts into linear functions, termed lambd... | ["deep learning", "neural networks", "attention", "transformer", "vision", "image classification"] | ABSTRACTWe present lambda layers – an alternative framework to self-attention – for cap-turing long-range interactions between an input and structured contextual infor-mation (e.g. a pixel surrounded by other pixels). Lambda layers capture suchinteractions by transforming available contexts into linear functions, terme... | Pfi8mGS4AX1 | The ideas and results are interesting, but it lacks some important discussion and results. | 6: Marginally above acceptance threshold | Summary
1. This paper present a new method, the lambda layer, for capturing long-term dependency.
2. lambda layer summarizes the context into the fixed vector to reduce the computation burden
3. This paper validates the performance on image classification and object detection tasks.
Strengths
1. The lambda layer is a ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
LambdaNetworks: Modeling long-range Interactions without Attention
### Paper Abstract
We present lambda layers -- an alternative framework to self-attention -- for captu... |
SJgg_yHKvH | ICLR.cc/2020/Conference | 2020 | Molecule Property Prediction and Classification with Graph Hypernetworks | ["Eliya Nachmani", "Lior Wolf"] | Graph neural networks are currently leading the performance charts in learning-based molecule property prediction and classification. Computational chemistry has, therefore, become the a prominent testbed for generic graph neural networks, as well as for specialized message passing methods. In this work, we demonstrate... | ["classification", "molecule property prediction", "graph hypernetworks", "hypernetworks", "property prediction", "neural networks", "performance charts", "computational chemistry", "prominent testbed"] | ABSTRACTGraph neural networks are currently leading the performance charts in learning-based molecule property prediction and classification. Computational chemistryhas, therefore, become the a prominent testbed for generic graph neural networks,as well as for specialized message passing methods. In this work, we demon... | BkeygBpRFB | Official Blind Review #2 | 3: Weak Reject | The authors take four current graph neural network architectures: 1. The neural message passing networks of Jorgensen et al. 2. The invariant graph networks of Maron et al., 3. The graph isomorphism networks of Xu et al. and 4. The decoding hypernetwork of Nachmani & Wolf and extend them by adding extra layers to learn... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Molecule Property Prediction and Classification with Graph Hypernetworks
### Paper Abstract
Graph neural networks are currently leading the performance charts in learnin... | |
B1e-kxSKDH | ICLR.cc/2020/Conference | 2020 | Structured Object-Aware Physics Prediction for Video Modeling and Planning | ["Jannik Kossen", "Karl Stelzner", "Marcel Hussing", "Claas Voelcker", "Kristian Kersting"] | When humans observe a physical system, they can easily locate components, understand their interactions, and anticipate future behavior, even in settings with complicated and previously unseen interactions. For computers, however, learning such models from videos in an unsupervised fashion is an unsolved research probl... | ["self-supervised learning", "probabilistic deep learning", "structured models", "video prediction", "physics prediction", "planning", "variational auteoncoders", "model-based reinforcement learning", "VAEs", "unsupervised", "variational", "graph neural networks", "tractable probabilistic models", "attend-infer-repeat"... | ABSTRACTWhen humans observe a physical system, they can easily locate objects, under-stand their interactions, and anticipate future behavior. For computers, however,learning such models from videos in an unsupervised fashion is an unsolved re-search problem. In this paper, we present STOVE, a novel state-space modelfo... | rylOjQdDuS | Official Blind Review #1 | 6: Weak Accept | This paper introduces a structured deep generative model for video frame prediction, with an object recognition model based on the Attend, Infer, Repeat (AIR) model by Eslami et al. (2016) and a graph neural network as a latent dynamics model. The model is evaluated on two synthetic physics simulation datasets (N-body ... | <|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 Object-Aware Physics Prediction for Video Modeling and Planning
### Paper Abstract
When humans observe a physical system, they can easily locate components, u... | |
B1lMMx1CW | ICLR.cc/2018/Conference | 2018 | THE EFFECTIVENESS OF A TWO-LAYER NEURAL NETWORK FOR RECOMMENDATIONS | ["Oleg Rybakov", "Vijai Mohan", "Avishkar Misra", "Scott LeGrand", "Rejith Joseph", "Kiuk Chung", "Siddharth Singh", "Qian You", "Eric Nalisnick", "Leo Dirac", "Runfei Luo"] | We present a personalized recommender system using neural network for recommending
products, such as eBooks, audio-books, Mobile Apps, Video and Music.
It produces recommendations based on customer’s implicit feedback history such
as purchases, listens or watches. Our key contribution is to formulate recommendation
pro... | ["Recommender systems", "deep learning", "personalization"] | ABSTRACTWe present a personalized recommender system using neural network for recom-mending products, such as eBooks, audio-books, Mobile Apps, Video and Music.It produces recommendations based on customer’s implicit feedback history suchas purchases, listens or watches. Our key contribution is to formulate recommen-da... | SJqZdRi1f | Good industrial/empirical paper with some surprising findings. | 6: Marginally above acceptance threshold | The paper proposes a new neural network based method for recommendation.
The main finding of the paper is that a relatively simple method works for recommendation, compared to other methods based on neural networks that have been recently proposed.
This contribution is not bad for an empirical paper. There's certainl... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
THE EFFECTIVENESS OF A TWO-LAYER NEURAL NETWORK FOR RECOMMENDATIONS
### Paper Abstract
We present a personalized recommender system using neural network for recommending... |
QSMvGB5j5- | ICLR.cc/2021/Conference | 2021 | Higher-order Structure Prediction in Evolving Graph Simplicial Complexes | ["Manohar Kaul", "Masaaki Imaizumi"] | Dynamic graphs are rife with higher-order interactions, such as co-authorship relationships and protein-protein interactions in biological networks, that naturally arise between more than two nodes at once. In spite of the ubiquitous presence of such higher-order interactions, limited attention has been paid to the hig... | ["Higher-order", "graph simplicial complex", "link prediction"] | ABSTRACTDynamic graphs are rife with higher-order interactions, such as co-authorshiprelationships and protein-protein interactions in biological networks, that natu-rally arise between more than two nodes at once. In spite of the ubiquitouspresence of such higher-order interactions, limited attention has been paid to ... | Jrd2km6-jP_ | The paper develops a nonparametric kernel estimator based method for high-order structure prediction in evolving graphs. | 4: Ok but not good enough - rejection | This paper provide a method for high-order structure prediction problem. Specifically, the paper first defines a high-order structure on graphs named graph simplicial complex (GSC). Then the paper introduces a feature generation method used for the high-order structures. The features are also used in the proposed metho... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Higher-order Structure Prediction in Evolving Graph Simplicial Complexes
### Paper Abstract
Dynamic graphs are rife with higher-order interactions, such as co-authorship... |
Syx7w1hatV | vivoconference.org/VIVO/2019/Conference | 2019 | From CRIS to RPS - discovering research potential form research management systems; Omega-PSIR Experience | ["Jakub Koperwas", "\u0141ukasz Skonieczny", "Wac\u0142aw Struk", "Henryk Rybi\u0144ski"] | CRIS systems are becoming a mandatory element of University Information ecosystems. CRIS provides significant amenities for reporting routines for research councils, national authorities, employee evaluation, academic degrees communities and provide valuable insights for university management. Properly implemented (in... | ["Current Research Information System", "Research Profiling System", "OMEGA-PSIR", "Visualizations", "recommendation engines", "Ontology"] | ByxOGHl9aE | Intriguing intersection between text mining and ontological work | 8: Top 50% of accepted papers, clear accept | The use of VIVO as or within CRIS systems is a topic that is relevant to the VIVO community. The abstract describes the OMEGA-PSIR system which has had multiple deployments and so appears to be a working system in production. Furthermore, the abstract mentions the use of text mining and artificial intelligence along ... | 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
From CRIS to RPS - discovering research potential form research management systems; Omega-PSIR Experience
### Paper Abstract
CRIS systems are becoming a mandatory elemen... | |
mfJepDyIUcQ | ICLR.cc/2021/Conference | 2021 | Safety Verification of Model Based Reinforcement Learning Controllers | ["akshita gupta", "Inseok Hwang"] | Model-based reinforcement learning (RL) has emerged as a promising tool for developing controllers for real world systems (e.g., robotics, autonomous driving, etc.). However, real systems often have constraints imposed on their state space which must be satisfied to ensure the safety of the system and its environment.... | ["Reachable set", "state constraints", "safety verification", "model-based reinforcement learning"] | ABSTRACTModel-based reinforcement learning (RL) has emerged as a promising tool fordeveloping controllers for real world systems (e.g., robotics, autonomous driving,etc.). However, real systems often have constraints imposed on their state spacewhich must be satisfied to ensure the safety of the system and its environm... | uRqafS_kMB1 | Verification of model-based RL using reachable set analysis | 7: Good paper, accept | # Summary
The paper presents a method for safety verification of trained model-based RL agents using reachable set analysis.
The method allows to both analyse whether any of the start states might end in an unsafe state, via the forward reachable tube, and to determine which start states do or do not end in an unsafe ... | 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
Safety Verification of Model Based Reinforcement Learning Controllers
### Paper Abstract
Model-based reinforcement learning (RL) has emerged as a promising tool for deve... |
BSllnh4uDZq | ICLR.cc/2022/Workshop/DGM4HSD | 2022 | Inkorrect: Digital Ink Spelling Correction | ["Andrii Maksai", "Henry A. Rowley", "Jesse Berent", "Claudiu Musat"] | We introduce Inkorrect, a digital ink (online handwriting) spelling correction approach. We show that existing metrics don’t capture the quality of spelling correction, and propose a new one. Our approach outperforms previous work in automated and human evaluation, while also being more data- and label-efficient. | ["digital ink", "correction", "inkorrect", "correction inkorrect", "online handwriting", "metrics", "capture", "quality", "new", "previous work"] | ABSTRACTWe introduce Inkorrect, a digital ink (online handwriting) spelling correction ap-proach. We show that existing metrics don’t capture the quality of spelling cor-rection, and propose a new one. Our approach outperforms previous work inautomated and human evaluation, while also being more data- and label-efficie... | rcIx81L7hW5 | work of somewhat novel ideas, but with poor presentation | 6: Marginally above acceptance threshold | The paper described a new method to auto-correct handwritten words.
Pros:
1. The paper provides a new way of evaluating the performance of the spell-correcting model. It argues that recognizability, often used as an evaluation for models in this field, is not sufficient. A new metric to take similarity between origin... | 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
Inkorrect: Digital Ink Spelling Correction
### Paper Abstract
We introduce Inkorrect, a digital ink (online handwriting) spelling correction approach. We show that exist... |
Hyl_vjC5KQ | ICLR.cc/2019/Conference | 2019 | Hierarchical Reinforcement Learning via Advantage-Weighted Information Maximization | ["Takayuki Osa", "Voot Tangkaratt", "Masashi Sugiyama"] | Real-world tasks are often highly structured. Hierarchical reinforcement learning (HRL) has attracted research interest as an approach for leveraging the hierarchical structure of a given task in reinforcement learning (RL). However, identifying the hierarchical policy structure that enhances the performance of RL is n... | ["Hierarchical reinforcement learning", "Representation learning", "Continuous control"] | ABSTRACTReal-world tasks are often highly structured. Hierarchical reinforcement learn-ing (HRL) has attracted research interest as an approach for leveraging the hier-archical structure of a given task in reinforcement learning (RL). However, iden-tifying the hierarchical policy structure that enhances the performance... | H1gOFYTK3m | Interesting ideas and analysis but somewhat unclear motivation and limited empirical evaluation | 7: Good paper, accept | Revision: The authors addressed most of my concerns and clearly put in effort to improve the paper. The paper explains the central idea better, is more precise in terminology in general, and the additional ablation gives more insight into the relative importance of the advantage weighting. I still think that the result... | 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
Hierarchical Reinforcement Learning via Advantage-Weighted Information Maximization
### Paper Abstract
Real-world tasks are often highly structured. Hierarchical reinfor... |
3T9iFICe0Y9 | ICLR.cc/2021/Conference | 2021 | The Recurrent Neural Tangent Kernel | ["Sina Alemohammad", "Zichao Wang", "Randall Balestriero", "Richard Baraniuk"] | The study of deep neural networks (DNNs) in the infinite-width limit, via the so-called neural tangent kernel (NTK) approach, has provided new insights into the dynamics of learning, generalization, and the impact of initialization. One key DNN architecture remains to be kernelized, namely, the recurrent neural network... | ["Neural Tangent Kernel", "Recurrent Neural Network", "Gaussian Process", "Overparameterization"] | ABSTRACTThe study of deep neural networks (DNNs) in the infinite-width limit, via theso-called neural tangent kernel (NTK) approach, has provided new insights into thedynamics of learning, generalization, and the impact of initialization. One key DNNarchitecture remains to be kernelized, namely, the recurrent neural ne... | 4OkzS-X4Flk | Theoretical results are valuable, empirical results are insufficient. | 6: Marginally above acceptance threshold | In this paper the neural tangent kernel for RNNs is derived. It is emphasized how this results in a proper kernel that can handle samples of different lengths.
The theoretical derivations are correct to the best of my knowledge and help to complete the picture of deep architectures for which the NTK has been derived.... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
The Recurrent Neural Tangent Kernel
### Paper Abstract
The study of deep neural networks (DNNs) in the infinite-width limit, via the so-called neural tangent kernel (NTK... |
SJe-HkBKDS | ICLR.cc/2020/Conference | 2020 | Amharic Text Normalization with Sequence-to-Sequence Models | ["Seifedin Shifaw Mohamed", "Solomon Teferra Abate (PhD)"] | All areas of language and speech technology, directly or indirectly, require handling of real text. In addition to ordinary words and names, the real text contains non-standard words (NSWs), including numbers, abbreviations, dates, currency, amounts, and acronyms. Typically, one cannot find NSWs in a dictionary, nor ca... | ["Text Normalization", "Sequence-to-Sequence Model", "Encoder-Decoder"] | Amharic Text Normalization with Sequence -to-Sequence Model s 1. Introduction Text normalization or the transformation of words from the written to the spoken form is an important problem. It is also quite a complex problem, due to the range of different kinds of non -standard words (NSW ’s), the special processing req... | HkgF8KJT9r | Official Blind Review #1 | 1: Reject | Text normalization or the transformation of words from the written to the spoken form is an important and realistic question in natural language processing. This paper aims to use sequence-to-sequence models to perform text normalization.
However, this paper does not use the official template and the content is too sh... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Amharic Text Normalization with Sequence-to-Sequence Models
### Paper Abstract
All areas of language and speech technology, directly or indirectly, require handling of r... | |
H1edEyBKDS | ICLR.cc/2020/Conference | 2020 | Plug and Play Language Models: A Simple Approach to Controlled Text Generation | ["Sumanth Dathathri", "Andrea Madotto", "Janice Lan", "Jane Hung", "Eric Frank", "Piero Molino", "Jason Yosinski", "Rosanne Liu"] | Large transformer-based language models (LMs) trained on huge text corpora have shown unparalleled generation capabilities. However, controlling attributes of the generated language (e.g. switching topic or sentiment) is difficult without modifying the model architecture or fine-tuning on attribute-specific data and en... | ["controlled text generation", "generative models", "conditional generative models", "language modeling", "transformer"] | ABSTRACTLarge transformer-based language models (LMs) trained on huge text corporahave shown unparalleled generation capabilities. However, controlling attributesof the generated language (e.g. switching topic or sentiment) is difficult withoutmodifying the model architecture or fine-tuning on attribute-specific data a... | H1x4jhziFS | Official Blind Review #2 | 6: Weak Accept | The authors describe a method for training plug and play language models, a way to incorporate control elements into pre-trained LMs. In contrast to existing work, which often trains conditioned upon the control element, the authors emphasize that their method does not require re-training the initial LM. This is exciti... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Plug and Play Language Models: A Simple Approach to Controlled Text Generation
### Paper Abstract
Large transformer-based language models (LMs) trained on huge text corp... | |
3rRgu7OGgBI | ICLR.cc/2021/Conference | 2021 | Bi-tuning of Pre-trained Representations | ["Jincheng Zhong", "Ximei Wang", "Zhi Kou", "Jianmin Wang", "Mingsheng Long"] | It is common within the deep learning community to first pre-train a deep neural network from a large-scale dataset and then fine-tune the pre-trained model to a specific downstream task. Recently, both supervised and unsupervised pre-training approaches to learning representations have achieved remarkable advances, wh... | ["Deep learning", "fine-tuning", "pre-training"] | ABSTRACTIt is common within the deep learning community to first pre-train a deep neuralnetwork from a large-scale dataset and then fine-tune the pre-trained model to aspecific downstream task. Recently, both supervised and unsupervised pre-trainingapproaches to learning representations have achieved remarkable advance... | PMQYOdoBgH | Using contrastive losses may improve fine-tuning | 5: Marginally below acceptance threshold | ### Summary
The paper suggests to extend fine-tuning of pre-trained representations by adding additional contrastive losses to leverage the intrinsic structure of the downstream training data. The authors call the presented method Bi-tuning and evaluate it by comparing it to other variants of fine-tuning.
### Quality... | 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
Bi-tuning of Pre-trained Representations
### Paper Abstract
It is common within the deep learning community to first pre-train a deep neural network from a large-scale d... |
q8qLAbQBupm | ICLR.cc/2021/Conference | 2021 | Neural Mechanics: Symmetry and Broken Conservation Laws in Deep Learning Dynamics | ["Daniel Kunin", "Javier Sagastuy-Brena", "Surya Ganguli", "Daniel LK Yamins", "Hidenori Tanaka"] | Understanding the dynamics of neural network parameters during training is one of the key challenges in building a theoretical foundation for deep learning. A central obstacle is that the motion of a network in high-dimensional parameter space undergoes discrete finite steps along complex stochastic gradients derived f... | ["learning dynamics", "symmetry", "loss landscape", "stochastic differential equation", "modified equation analysis", "conservation law", "hessian", "geometry", "physics", "gradient flow"] | ABSTRACTUnderstanding the dynamics of neural network parameters during training is one ofthe key challenges in building a theoretical foundation for deep learning. A centralobstacle is that the motion of a network in high-dimensional parameter spaceundergoes discrete finite steps along complex stochastic gradients deri... | mB9TwkPWk2J | Interesting theoretical perspective on dynamics of DNNs but with some confusions | 6: Marginally above acceptance threshold | This paper analyzes the learning dynamics of DNNs from the perspective of symmetry of some parameters. It is interesting to borrow ideas from physics, which I believe is a right way to go.
Specifically, the paper derives analytical form of parameters under the cases of translation, scale and inversion invariances, and... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Mechanics: Symmetry and Broken Conservation Laws in Deep Learning Dynamics
### Paper Abstract
Understanding the dynamics of neural network parameters during train... |
BkltNhC9FX | ICLR.cc/2019/Conference | 2019 | Posterior Attention Models for Sequence to Sequence Learning | ["Shiv Shankar", "Sunita Sarawagi"] | Modern neural architectures critically rely on attention for mapping structured inputs to sequences. In this paper we show that prevalent attention architectures do not adequately model the dependence among the attention and output tokens across a predicted sequence.
We present an alternative architecture called Poste... | ["posterior inference", "attention", "seq2seq learning", "translation"] | ABSTRACTModern neural architectures critically rely on attention for mapping structuredinputs to sequences. In this paper we show that prevalent attention architecturesdo not adequately model the dependence among the attention and output tokensacross a predicted sequence. We present an alternative architecture called P... | BJe8NreK2m | Posterior attention improves sequence to sequence learning | 9: Top 15% of accepted papers, strong accept | Originality: Existing attention models do not statistically express interactions among multiple attentions. The authors of this manuscript reformulate p(y|x) and define prior attention distribution (a_t depends on previous outputs y_<t) and posterior attention distribution (a_t depends on current output y_t as well), 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
Posterior Attention Models for Sequence to Sequence Learning
### Paper Abstract
Modern neural architectures critically rely on attention for mapping structured inputs to... |
BklhsgSFvB | ICLR.cc/2020/Conference | 2020 | Learning to Transfer via Modelling Multi-level Task Dependency | ["Haonan Wang", "Zhenbang Wu", "Ziniu Hu", "Yizhou Sun"] | Multi-task learning has been successful in modeling multiple related tasks with large, carefully curated labeled datasets. By leveraging the relationships among different tasks, multi-task learning framework can improve the performance significantly. However, most of the existing works are under the assumption that the... | ["multi-task learning", "attention mechanism"] | ABSTRACTMulti-task learning has been successful in leveraging the dependency among dif-ferent tasks to improve general performance. However, when applying to ‘dis-crete’ data (graph/text), most of the existing frameworks only leverage the generaltask dependency with the assumption that the task dependency remains the s... | SJxJYVNg9S | Official Blind Review #2 | 1: Reject | The submission argues for the modeling the relationships between different tasks and incorporating such relationships when training multi-task frameworks. Though the basic concept (usefulness of modeling and incorporating the relationships among tasks) is valid, the submission has a number of critical issues, namely mi... | <|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 Transfer via Modelling Multi-level Task Dependency
### Paper Abstract
Multi-task learning has been successful in modeling multiple related tasks with large, ... | |
kvhzKz-_DMF | ICLR.cc/2021/Conference | 2021 | Variational Information Bottleneck for Effective Low-Resource Fine-Tuning | ["Rabeeh Karimi mahabadi", "Yonatan Belinkov", "James Henderson"] | While large-scale pretrained language models have obtained impressive results when fine-tuned on a wide variety of tasks, they still often suffer from overfitting in low-resource scenarios. Since such models are general-purpose feature extractors, many of these features are inevitably irrelevant for a given target task... | ["Transfer learning", "NLP", "large-scale pre-trained language models", "over-fitting", "robust", "biases", "variational information bottleneck"] | ABSTRACTWhile large-scale pretrained language models have obtained impressive resultswhen fine-tuned on a wide variety of tasks, they still often suffer from overfittingin low-resource scenarios. Since such models are general-purpose feature extractors,many of these features are inevitably irrelevant for a given target... | MQZkybuGqSp | The paper proposes a method to avoid overfitting while finetuning the large pretrained models for downstream tasks on small scale datasets. | 7: Good paper, accept | The paper proposes a method to avoid overfitting while finetuning the large pretrained models for downstream tasks on small scale datasets. It has been shown that many SOTA models usually overfit w.r.t. spurious correlations in the data and as a result fail miserably when tested for generalization on the out of domain ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Variational Information Bottleneck for Effective Low-Resource Fine-Tuning
### Paper Abstract
While large-scale pretrained language models have obtained impressive result... |
-6vS_4Kfz0 | ICLR.cc/2021/Conference | 2021 | Optimizing Memory Placement using Evolutionary Graph Reinforcement Learning | ["Shauharda Khadka", "Estelle Aflalo", "Mattias Marder", "Avrech Ben-David", "Santiago Miret", "Shie Mannor", "Tamir Hazan", "Hanlin Tang", "Somdeb Majumdar"] | For deep neural network accelerators, memory movement is both energetically expensive and can bound computation. Therefore, optimal mapping of tensors to memory hierarchies is critical to performance. The growing complexity of neural networks calls for automated memory mapping instead of manual heuristic approaches; ye... | ["Reinforcement Learning", "Memory Mapping", "Device Placement", "Evolutionary Algorithms"] | ABSTRACTFor deep neural network accelerators, memory movement is both energeticallyexpensive and can bound computation. Therefore, optimal mapping of tensorsto memory hierarchies is critical to performance. The growing complexity ofneural networks calls for automated memory mapping instead of manual heuristicapproaches... | XlRlf2Pb_M | Well-written paper on memory mapping method that outperforms native NNP-I compiler by 28-78% | 5: Marginally below acceptance threshold | The paper describes a machine learning method for mapping computational nodes in a neural network onto different levels of the memory hierarchy (DRAM, LLC, and SRAM) to minimize latency of inference. The proposed approach builds on CERL, combining policy gradient, reinforcement learning, and graph neural network, achie... | 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 Memory Placement using Evolutionary Graph Reinforcement Learning
### Paper Abstract
For deep neural network accelerators, memory movement is both energeticall... |
K1L6rVKTUuh | MICCAI.org/2021/Challenge/KiTS | 2021 | kits21_short_paper_aladdin5_team | ["Aladdin Wu"] | Kidney cyst and tumor are very small compared to the whole volume of CT data, and so it’s difficult to achieve ideal precision for normal training and prediction. We propose a method to train and learn the data efficiently. | ["nnUNet", "ROI"] | Short Paper Submissions for the MICCAI KiTS21 Challenge Y. J. Wu1, Y.L. Zhao1 The aladdin5 team 7766963@qq.com Abstract. We use nnUNet and also tried several methods to process and train the data. Keywords: nnUNet, ROI 1 Introduction Kidney cyst and tumor are very small compared to the whole volume of CT data, and so i... | RiqG7yjMcYI | Official Review | 3: Clear rejection | The authors will need to consult the kits21 paper template for a list of prompts that should be addressed. The paper cannot be published in its current form and will need a substantial rewrite. This is fine for an initial intention-to-submit submission, but please rewrite and submit your revision as soon as possible so... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
kits21_short_paper_aladdin5_team
### Paper Abstract
Kidney cyst and tumor are very small compared to the whole volume of CT data, and so it’s difficult to achieve ideal ... | |
r1xN5oA5tm | ICLR.cc/2019/Conference | 2019 | Phrase-Based Attentions | ["Phi Xuan Nguyen", "Shafiq Joty"] | Most state-of-the-art neural machine translation systems, despite being different
in architectural skeletons (e.g., recurrence, convolutional), share an indispensable
feature: the Attention. However, most existing attention methods are token-based
and ignore the importance of phrasal alignments, the key ingredient for ... | ["neural machine translation", "natural language processing", "attention", "transformer", "seq2seq", "phrase-based", "phrase", "n-gram"] | ABSTRACTMost state-of-the-art neural machine translation systems, despite being differentin architectural skeletons ( e.g., recurrence, convolutional), share an indispensablefeature: the Attention. However, most existing attention methods are token-basedand ignore the importance of phrasal alignments, the key ingredien... | r1eHjtZq27 | This work starts from the premise that current models attends to unigram representation, which is wrong (keys and values already depends on multiple source/target positions). The empirical results are missing recent improvements. The reported empirical advantage compared to baseline is thin. | 5: Marginally below acceptance threshold | Phrase-Based Attention
Paper Summary:
Neural translation attention computes latent alignments which pairs input/target positions. Phrase-based systems used to align pairs of spans (n-grams) rather than individual positions, this work explores neural architectures to align spans. It does so by compositing attention an... | 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
Phrase-Based Attentions
### Paper Abstract
Most state-of-the-art neural machine translation systems, despite being different in architectural skeletons (e.g., recurrence... |
rkxt8oC9FQ | ICLR.cc/2019/Conference | 2019 | Perfect Match: A Simple Method for Learning Representations For Counterfactual Inference With Neural Networks | ["Patrick Schwab", "Lorenz Linhardt", "Walter Karlen"] | Learning representations for counterfactual inference from observational data is of high practical relevance for many domains, such as healthcare, public policy and economics. Counterfactual inference enables one to answer "What if...?" questions, such as "What would be the outcome if we gave this patient treatment $t_... | ["counterfactual inference", "representations", "neural networks", "perfect match", "simple", "observational data", "complex", "number"] | ABSTRACTLearning representations for counterfactual inference from observational data is ofhigh practical relevance for many domains, such as healthcare, public policy andeconomics. Counterfactual inference enables one to answer ”What if...?” ques-tions, such as ”What would be the outcome if we gave this patient treatm... | HygsMMPq2X | Review: Interesting paper and impressive results; why novel vs. standard PSM? | 5: Marginally below acceptance threshold | ========= Summary =========
The authors propose a novel method for counterfactual inference (i.e. individual/heterogeneous treatment effect, as well as average treatment effect) with neural networks. They perform propensity score matching within each minibatch in order to match the covariate distributions during train... | 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
Perfect Match: A Simple Method for Learning Representations For Counterfactual Inference With Neural Networks
### Paper Abstract
Learning representations for counterfact... |
mu0WNwWWWCE | ICLR.cc/2021/Conference | 2021 | Essentials for Class Incremental Learning | ["Sudhanshu Mittal", "Silvio Galesso", "Thomas Brox"] | Contemporary neural networks are limited in their ability to learn from evolving streams of training data. When trained sequentially on new or evolving tasks, their accuracy drops sharply, making them unsuitable for many real-world applications. In this work, we shed light on the causes of this well known yet unsolved ... | ["class-incremental learning", "catastrophic forgetting"] | ABSTRACTContemporary neural networks are limited in their ability to learn from evolvingstreams of training data. When trained sequentially on new or evolving tasks,their accuracy drops sharply, making them unsuitable for many real-world appli-cations. In this work, we shed light on the causes of this well known yet un... | 3Z9q_woqd6 | Review Essentials for Class IL | 7: Good paper, accept | Paper investigates essential parts of class IL methods. Based on their analysis they propose a simple IL method that obtains remarkably good results. The separate decisions are well motivated and the paper shows a genuine intention (and success) to understand the reasons for catastrophic forgetting and the working of 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
Essentials for Class Incremental Learning
### Paper Abstract
Contemporary neural networks are limited in their ability to learn from evolving streams of training data. W... |
ByxVpY5htN | MIDL.io/2019/Conference/Abstract | 2019 | Towards End-to-End Image-to-Tree for Vasculature Modeling | ["Manish Sharma", "Matthew C. H. Lee", "James Batten", "Michiel Schaap", "Ben Glocker"] | This work explores an end-to-end image-to-tree approach for extracting accurate representations of vessel structures which may be beneficial for diagnosis of stenosis (blockages) and modeling of blood flow. Current image segmentation approaches capture only an implicit representation, while this work utilizes a subscal... | ["Vasculature", "tree extraction", "retinal vessels", "diabetic retinopathy", "visualization"] | Proceedings of Machine Learning Research { Under Review:1{4, 2019 Full Paper { MIDL 2019 submissionTowards End-to-End Image-to-Tree for Vasculature ModelingManish Sharma skmanish@google.comMatthew C. H. Lee matthew.lee13@imperial.ac.ukJames Batten j.batten@imperial.ac.ukMichiel Schaap mschaap@heartflow.comBen Glocker b... | rylZ_tYrjN | Interesting idea, but concerns in method design, effectiveness and lack of baseline comparison | 2: Reject | A U-net is used to incrementally predict vessel tree branch points in each image patch, effectively traversing all vessels in an image and thereby building a vessel tree. Methodology is questionable in case of parallel and/or kissing vessels. It is evaluated on a rather outdated DRIVE dataset of color retinal images, ... | 2: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards End-to-End Image-to-Tree for Vasculature Modeling
### Paper Abstract
This work explores an end-to-end image-to-tree approach for extracting accurate representati... |
B1NT3TAIM | ICLR.cc/2018/Workshop | 2018 | PPP-Net: Platform-aware Progressive Search for Pareto-optimal Neural Architectures | ["Jin-Dong Dong", "An-Chieh Cheng", "Da-Cheng Juan", "Wei Wei", "Min Sun"] | Recent breakthroughs in Neural Architectural Search (NAS) have achieved state-of-the-art performances in many applications such as image recognition. However, these techniques typically ignore platform-related constrictions (e.g., inference time and power consumptions) that can be critical for portable devices with lim... | ["Architecture Search", "Platform-awared", "Pareto Optimality"] | ABSTRACTRecent breakthroughs in Neural Architectural Search (NAS) have achieved state-of-the-art performances in many applications such as image recognition. How-ever, these techniques typically ignore platform-related constrictions ( e.g., infer-ence time and power consumptions) that can be critical for portable devic... | rkQIJMlYG | An extension for efficient neural architecture search for CNN using the pareto optimality. | 7: Good paper, accept | The paper presents an extension of the method in Liu et at (2017a) by using a Pareto optimal search under multiple objectives fron efficient neural architecture search for CNNs. The paper is well written and clear, towards its objections and added value. Furthermore, the authors evaluated their proposed method using th... | 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
PPP-Net: Platform-aware Progressive Search for Pareto-optimal Neural Architectures
### Paper Abstract
Recent breakthroughs in Neural Architectural Search (NAS) have achi... |
HFUI1pfQnCF | ML_Reproducibility_Challenge/2021/Fall | 2021 | [Re] Differentiable Spatial Planning using Transformers | ["Rohit Ranjan", "Himadri Bhakta", "Animesh Jha", "Parv Maheshwari"] | Reproducibility Summary
Scope of Reproducibility
This report covers our reproduction effort of the paper `Differentiable Spatial Planning using Transformers' by Chaplot et al. In this paper, the problem of spatial path planning in a differentiable way is considered. They show that their proposed method of using Spati... | ["accuracy", "mapper module", "differentiable spatial planning", "transformers", "authors", "source code", "synthetic datasets", "size", "results"] | ML Reproducibility Challenge 2021[Re] Differentiable Spatial Planning using TransformersAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2This report covers our reproduction effort of the paper ‘Differentiable Spatial Planning using Transformers’ by Chaplot 3et al. [ 1]. In th... | rTMxOXJgVWq | Reproducibility Review - Differentiable Spatial Planning using Transformers | 9: Top 15% of accepted papers, strong accept | The original authors make three claims and the reproducibility report sets out to verify these. The authors were unable to reproduce the results of the original paper to a high accuracy (Results were different by 14.7% which is a large margin), though they observe similar drop-off accuracy in different model settings. ... | 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
[Re] Differentiable Spatial Planning using Transformers
### Paper Abstract
Reproducibility Summary Scope of Reproducibility This report covers our reproduction effort ... |
S1xJFREKvB | ICLR.cc/2020/Conference | 2020 | Amortized Nesterov's Momentum: Robust and Lightweight Momentum for Deep Learning | ["Kaiwen Zhou", "Yanghua Jin", "Qinghua Ding", "James Cheng"] | Stochastic Gradient Descent (SGD) with Nesterov's momentum is a widely used optimizer in deep learning, which is observed to have excellent generalization performance. However, due to the large stochasticity, SGD with Nesterov's momentum is not robust, i.e., its performance may deviate significantly from the expectatio... | ["momentum", "nesterov", "optimization", "deep learning", "neural networks"] | ABSTRACTStochastic Gradient Descent (SGD) with Nesterov’s momentum is a widely usedoptimizer in deep learning, which is observed to have excellent generalization per-formance. In this work, we propose Amortized Nesterov’s Momentum , which is aspecial variant of Nesterov’s momentum. Compared with Nesterov’s momentum,our... | H1gz-n-oKr | Official Blind Review #3 | 8: Accept | This paper provides a new simple method to incorporate Nesterov momentum into standard SGD for deep learning, with good empirical and theoretical results. Overall I think this paper should be accepted, some minor comments follow.
At no point does Polyak's heavy ball method get mentioned, even though the variant of Nes... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Amortized Nesterov's Momentum: Robust and Lightweight Momentum for Deep Learning
### Paper Abstract
Stochastic Gradient Descent (SGD) with Nesterov's momentum is a wide... | |
BJkjUrJwz | ICLR.cc/2018/Workshop | 2018 | Circumventing negative transfer via cross generative initialisation | ["Wenjun Bai", "Changqin Quan", "Zhi-Wei Luo"] |
Negative transfer – a special type of transfer learning – refers to the interference of the previous knowledge with new learning. In this research, through an empirical study, we demonstrate the futile defence to the negative transfer via conventional neural network based transfer techniques, i.e., mid-level feature e... | ["transfer learning", "negative transfer"] | ABSTRACTNegative transfer – a special type of transfer learning – refers to the interferenceof the previous knowledge with new learning. In this research, through an em-pirical study, we demonstrate the futile defence to the negative transfer via con-ventional neural network based transfer techniques, i.e., mid-level f... | B1BgjuxKG | Official Review | 4: Ok but not good enough - rejection | Summary: This paper studies the topical problem of transfer learning. Th focus is on proposing a transfer learning method that is more robust to negative transfer (transfer outcome being worse than train from scratch without any transfer), which is a risk in the case of many existing transfer learning methods. The prop... | 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
Circumventing negative transfer via cross generative initialisation
### Paper Abstract
Negative transfer – a special type of transfer learning – refers to the interfere... |
6puUoArESGp | ICLR.cc/2021/Conference | 2021 | Debiasing Concept-based Explanations with Causal Analysis | ["Mohammad Taha Bahadori", "David Heckerman"] | Concept-based explanation approach is a popular model interpertability tool because it expresses the reasons for a model's predictions in terms of concepts that are meaningful for the domain experts. In this work, we study the problem of the concepts being correlated with confounding information in the features. We pro... | ["Interpretability", "Concept-based Explanation"] | ABSTRACTConcept-based explanation approach is a popular model interpertability tool be-cause it expresses the reasons for a model’s predictions in terms of concepts thatare meaningful for the domain experts. In this work, we study the problem ofthe concepts being correlated with confounding information in the features.... | TwywvSjOV5N | Good paper, with a clever use of instrumental variables. I liked the results but I would have liked a clear description of the scope in which the method can be applied. | 7: Good paper, accept | Abstract: The paper proposed a way to learn unbiased (debiased) concept-based explainable models in the presence of unobserved confounders by the use of labels as instrumental variables. The proposed algorithm has 3 main steps: (1) regresses concept labels from the final labels (2) replace the original concepts with 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
Debiasing Concept-based Explanations with Causal Analysis
### Paper Abstract
Concept-based explanation approach is a popular model interpertability tool because it expre... |
PoAI5RrlUFj | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | Towards Easy Vocabulary Drafts with Neologism 2.0 | ["Johannes Lipp", "Lars Gleim", "Michael Cochez", "Iraklis Dimitriadis", "Hussain Ali", "Daniel Hoppe Alvarez", "Christoph Lange", "Stefan Decker"] | Shared vocabularies and ontologies are essential for many applications. Although standards and recommendations already cover many areas, adaptations are usually necessary to represent concrete use-cases properly. Domain experts are unfamiliar with ontology engineering, which creates special requirements for needed tool... | ["Vocabulary Creation", "Ontology Creation", "Vocabulary Drafts", "Knowledge Graph", "Semantic Web"] | Towards Easy Vocabulary Drafts with Neologism 2.0Johannes Lipp1;3, Lars Gleim1, Michael Cochez2, Iraklis Dimitriadis3,Hussain Ali3, Daniel Hoppe Alvarez3, Christoph Lange1;3, and Stefan Decker1;31Chair of Information Systems, RWTH Aachen University, Aachen, Germany2Department of Computer Science, Vrije Universiteit Ams... | 3zX1heh3bqs | It is not enough clear the novelty of the proposed system while I can understand the authors' motivations. | 5: Marginally below acceptance threshold | This paper presents a collaborative vocabulary creation tool called Neologism 2.0.
The authors’ motivation is providing a tool to create draft vocabulary not for ontology engineers but domain experts.
Therefore, the tool is designed under the idea that most of existing ontology editing tools are too complex for domai... | 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 Easy Vocabulary Drafts with Neologism 2.0
### Paper Abstract
Shared vocabularies and ontologies are essential for many applications. Although standards and recom... |
Hku9NK5lx | ICLR.cc/2017/conference | 2017 | Training Compressed Fully-Connected Networks with a Density-Diversity Penalty | ["Shengjie Wang", "Haoran Cai", "Jeff Bilmes", "William Noble"] | Deep models have achieved great success on a variety of challenging tasks. How- ever, the models that achieve great performance often have an enormous number of parameters, leading to correspondingly great demands on both computational and memory resources, especially for fully-connected layers. In this work, we propos... | ["Deep learning"] | ABSTRACTDeep models have achieved great success on a variety of challenging tasks. How-ever, the models that achieve great performance often have an enormous numberof parameters, leading to correspondingly great demands on both computationaland memory resources, especially for fully-connected layers. In this work, wepr... | SkFKTRbVx | Good results | 6: Marginally above acceptance threshold | The paper shows promising results but it is difficult to read and follow. It presents different things closely related and it is difficult to asses the performance of each one. Diversity, sparsity, regularization term, tying weights. Anyway results are good. | 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
Training Compressed Fully-Connected Networks with a Density-Diversity Penalty
### Paper Abstract
Deep models have achieved great success on a variety of challenging task... |
EGVxmJKLC2L | ICLR.cc/2021/Conference | 2021 | Learning not to learn: Nature versus nurture in silico | ["Robert Tjarko Lange", "Henning Sprekeler"] | Animals are equipped with a rich innate repertoire of sensory, behavioral and motor skills, which allows them to interact with the world immediately after birth. At the same time, many behaviors are highly adaptive and can be tailored to specific environments by means of learning. In this work, we use mathematical anal... | ["Meta-Learning", "Reinforcement Learning"] | ABSTRACTAnimals are equipped with a rich innate repertoire of sensory, behavioral and motorskills, which allows them to interact with the world immediately after birth. Atthe same time, many behaviors are highly adaptive and can be tailored to specificenvironments by means of learning. In this work, we use mathematical... | 9lldrgajfe | Creative proposal; analysis seems flawed | 5: Marginally below acceptance threshold | Summary:
This paper explores the effect of time horizon on meta-reinforcement learning agents. Using a recurrent meta-learner, it demonstrates that different strategies are learned based on the time horizon during meta-training.
Pros:
Creative proposal to study when incorporating new data to adapt to a task is warrant... | 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 not to learn: Nature versus nurture in silico
### Paper Abstract
Animals are equipped with a rich innate repertoire of sensory, behavioral and motor skills, whi... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.