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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
HmAhqnu3qu | ICLR.cc/2021/Conference | 2021 | Graph Representation Learning for Multi-Task Settings: a Meta-Learning Approach | ["Davide Buffelli", "Fabio Vandin"] | Graph Neural Networks (GNNs) have become the state-of-the-art method for many applications on graph structured data. GNNs are a framework for graph representation learning, where a model learns to generate low dimensional node embeddings that encapsulate structural and feature-related information. GNNs are usually trai... | ["Graph Representation Learning", "Multi-Task Learning", "Meta-Learning", "Graph Neural Networks"] | ABSTRACTGraph Neural Networks (GNNs) have become the state-of-the-art method formany applications on graph structured data. GNNs are a framework for graphrepresentation learning , where a model learns to generate low dimensional nodeembeddings that encapsulate structural and feature-related information. GNNsare usually... | 4nMX3y5g8J | Graph Representation Multi-Task Learning. | 7: Good paper, accept | This paper presents a multi-task framework to represent the node embedding for transferred knowledge. The methodology is based on the meta-learning, which is capable of producing multi-task node embedding. This paper is well-motivated and well-written. The experimental results illustrate the effectiveness of the model.... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Graph Representation Learning for Multi-Task Settings: a Meta-Learning Approach
### Paper Abstract
Graph Neural Networks (GNNs) have become the state-of-the-art method f... |
H1e3HlSFDr | ICLR.cc/2020/Conference | 2020 | Variational Constrained Reinforcement Learning with Application to Planning at Roundabout | ["Yuan Tian", "Minghao Han", "Lixian Zhang", "Wulong Liu", "Jun Wang", "Wei Pan"] | Planning at roundabout is crucial for autonomous driving in urban and rural environments. Reinforcement learning is promising not only in dealing with complicated environment but also taking safety constraints into account as a as a constrained Markov Decision Process. However, the safety constraints should be explicit... | ["Safe reinforcement learning", "Autonomous driving", "obstacle avoidance"] | ABSTRACTPlanning at roundabout is crucial for autonomous driving in urban and rural environ-ments. Reinforcement learning is promising not only in dealing with complicatedenvironment but also taking safety constraints into account as a as a constrainedMarkov Decision Process. However, the safety constraints should be e... | HJl-v-KAYH | Official Blind Review #2 | 1: Reject | This paper proposes a reinforcement learning model to solve a task relevant to autonomous driving, and evaluates that model in a traffic simulator called Simulation of Urban Mobility (SUMO).
The paper describes related work but the connection to the exact problem they are solving wasn’t 100% clear to me.
The descript... | <|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 Constrained Reinforcement Learning with Application to Planning at Roundabout
### Paper Abstract
Planning at roundabout is crucial for autonomous driving in ... | |
HyxpNnRcFX | ICLR.cc/2019/Conference | 2019 | Modulating transfer between tasks in gradient-based meta-learning | ["Erin Grant", "Ghassen Jerfel", "Katherine Heller", "Thomas L. Griffiths"] | Learning-to-learn or meta-learning leverages data-driven inductive bias to increase the efficiency of learning on a novel task. This approach encounters difficulty when transfer is not mutually beneficial, for instance, when tasks are sufficiently dissimilar or change over time. Here, we use the connection between grad... | ["meta-learning", "clustering", "learning-to-learn", "mixture", "hierarchical Bayes", "hierarchical model", "gradient-based meta-learning"] | ABSTRACTLearning-to-learn ormeta-learning leverages data-driven inductive bias to increasethe efficiency of learning on a novel task. This approach encounters difficultywhen transfer is not mutually beneficial, for instance, when tasks are sufficientlydissimilar or change over time. Here, we use the connection between ... | Skl0gf6On7 | Promising, but more work needed | 4: Ok but not good enough - rejection | This paper proposes a mixture of MAMLs (Finn et al., 2017) by exploiting the interpretation of MAML as a hierarchical Bayesian model (Grant et al. 2018). They propose an EM algorithm for joint training of parameter initializations and assignment of tasks to initializations. They further propose a non-parametric approac... | 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
Modulating transfer between tasks in gradient-based meta-learning
### Paper Abstract
Learning-to-learn or meta-learning leverages data-driven inductive bias to increase ... |
HJF3iD9xe | ICLR.cc/2017/conference | 2017 | Deep Learning with Sets and Point Clouds | ["Siamak Ravanbakhsh", "Jeff Schneider", "Barnabas Poczos"] | We introduce a simple permutation equivariant layer for deep learning with set structure. This type of layer, obtained by parameter-sharing, has a simple implementation and linear-time complexity in the size of each set. We use deep permutation-invariant networks to perform point-could classification and MNIST digit su... | ["Deep learning", "Structured prediction", "Computer vision", "Supervised Learning", "Semi-Supervised Learning"] | ABSTRACTWe introduce a simple permutation equivariant layer for deep learning with setstructure. This type of layer, obtained by parameter-sharing, has a simple im-plementation and linear-time complexity in the size of each set. We use deeppermutation-invariant networks to perform point-could classification and MNIST-d... | S1cGMcb4g | A promising work! | 5: Marginally below acceptance threshold | Pros :
- New and clear formalism for invariance on signals with known structure
- Good numerical results
Cons :
- The structure must be specified.
- The set structure dataset is too simple
- There is a gap between the large (and sometimes complex) theory introduced and the numerical experiments ; consequently a new r... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deep Learning with Sets and Point Clouds
### Paper Abstract
We introduce a simple permutation equivariant layer for deep learning with set structure. This type of layer,... |
WA7I-Fm4tmP | NeurIPS.cc/2022/Workshop/Federated_Learning | 2022 | LightVeriFL: Lightweight and Verifiable Secure Federated Learning | ["Baturalp Buyukates", "Jinhyun So", "Hessam Mahdavifar", "Salman Avestimehr"] | Secure aggregation protocols are implemented in federated learning to protect the local models of the participating users so that the server does not obtain any information beyond the aggregate model at each iteration. However, existing secure aggregation schemes fail to protect the integrity, i.e., correctness, of the... | ["federated learning", "verifiable learning", "secure aggregation", "trustworthiness", "privacy-preserving FL."] | LightVeriFL : Lightweight and Verifiable SecureFederated LearningBaturalp BuyukatesUniversity of Southern Californiabuyukate@usc.eduJinhyun SoUniversity of Southern Californiajinhyuns@usc.eduHessam MahdavifarUniversity of Michiganhessam@umich.eduSalman AvestimehrUniversity of Southern Californiaavestime@usc.eduAbstract... | YWPtMfWNNkm | Writing is not very clear, approach seems to have merits for others to consider and to improve on. | Paper proposes a new verification mechanism for secure aggregation in FL based on HE.
The paper does not describe the setting very clearly, reader needs to refer to Appendix to understand some of the concepts. Also there seems to be typos in formulas here and there. A good paper needs to be succinct in the main manusc... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
LightVeriFL: Lightweight and Verifiable Secure Federated Learning
### Paper Abstract
Secure aggregation protocols are implemented in federated learning to protect the lo... | ||
ryh9pmcee | ICLR.cc/2017/conference | 2017 | Energy-based Generative Adversarial Networks | ["Junbo Zhao", "Michael Mathieu", "Yann LeCun"] | We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce con... | ["Deep learning", "Unsupervised Learning", "Semi-Supervised Learning"] | ABSTRACTWe introduce the “Energy-based Generative Adversarial Network” model(EBGAN) which views the discriminator as an energy function that attributeslow energies to the regions near the data manifold and higher energies to otherregions. Similar to the probabilistic GANs, a generator is seen as being trainedto produce... | HJJ437wEg | 8: Top 50% of accepted papers, clear accept | This paper is a parallel work to Improving Generative Adversarial Networks with Denoising Feature Matching. The main solution of both papers is introducing autoencoder into discriminator to improve the stability and quality of GAN. Different to Denoising Feature Matching, EBGAN uses encoder-decoder instead of denoising... | 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
Energy-based Generative Adversarial Networks
### Paper Abstract
We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminato... | |
sp3Z1jiS2vn | ICLR.cc/2021/Conference | 2021 | Azimuthal Rotational Equivariance in Spherical CNNs | ["Carl Toft", "Georg B\u00f6kman", "Fredrik Kahl"] | In this work, we analyze linear operators from $L^2(S^2) \rightarrow L^2(S^2)$ which are equivariant to azimuthal rotations, that is, rotations around the z-axis. Several high-performing neural networks defined on the sphere are indeed equivariant to azimuthal rotations, but not $SO(3)$. Our main result is to show that... | ["rotations", "equivariant", "sphere", "azimuthal rotational equivariance", "spherical cnns", "work", "linear operators", "several", "neural networks"] | ABSTRACTIn this work, we analyze linear operators from L2pS2q ÑL2pS2qwhich areequivariant to azimuthal rotations, that is, rotations around the z-axis. Severalhigh-performing neural networks defined on the sphere are equivariant to az-imuthal rotations, but not to full SO(3) rotations. Our main result is to showthat a... | XmHQgLO7StT | An interesting theoretical paper, can be made stronger with better experiments | 5: Marginally below acceptance threshold | The authors present CNNs which are equivariant to azimuthal rotations, rather than the previously studied SO(3) rotations. The authors argue correctly that in many real-world settings, data have a canonical orientation in which they are recorded. In such cases, only equivariance to azimuthal rotations needs to be enfor... | 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
Azimuthal Rotational Equivariance in Spherical CNNs
### Paper Abstract
In this work, we analyze linear operators from $L^2(S^2) \rightarrow L^2(S^2)$ which are equivaria... |
S1lF8xHYwS | ICLR.cc/2020/Conference | 2020 | Unsupervised Domain Adaptation through Self-Supervision | ["Yu Sun", "Eric Tzeng", "Trevor Darrell", "Alexei A. Efros"] | This paper addresses unsupervised domain adaptation, the setting where labeled training data is available on a source domain, but the goal is to have good performance on a target domain with only unlabeled data. Like much of previous work, we seek to align the learned representations of the source and target domains wh... | ["unsupervised domain adaptation"] | ABSTRACTThis paper addresses unsupervised domain adaptation, the setting where labeledtraining data is available on a source domain, but the goal is to have good perfor-mance on a target domain with only unlabeled data. Like much of previous work,we seek to align the learned representations of the source and target dom... | rklFudcVqB | Official Blind Review #3 | 6: Weak Accept | This paper describes an approach to domain adaptation that uses
self-supervised losses to encourage source/target domain alignment for
unsupervised domain adaptation. The authors propose to use four
self-supervised tasks (variants of tasks used in the self-supervised
representation learning for object recognition liter... | <|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 Domain Adaptation through Self-Supervision
### Paper Abstract
This paper addresses unsupervised domain adaptation, the setting where labeled training data i... | |
H1MjAnqxg | ICLR.cc/2017/conference | 2017 | Intelligible Language Modeling with Input Switched Affine Networks | ["Jakob Foerster", "Justin Gilmer", "Jan Chorowski", "Jascha Sohl-dickstein", "David Sussillo"] | The computational mechanisms by which nonlinear recurrent neural networks (RNNs) achieve their goals remains an open question. There exist many problem domains where intelligibility of the network model is crucial for deployment. Here we introduce a recurrent architecture composed of input-switched affine transformati... | ["Natural language processing", "Deep learning", "Supervised Learning"] | ABSTRACTThe computational mechanisms by which nonlinear recurrent neural networks(RNNs) achieve their goals remains an open question. There exist many problemdomains where intelligibility of the network model is crucial for deployment. Herewe introduce a recurrent architecture composed of input-switched affine transfor... | SyblM4BEe | Original and creative work - hesitation about results. | 7: Good paper, accept | Summary: The authors propose an input switched affine network to do character-level language modeling, a kind of RNN without pointwise nonlinearity, but with switching the transition matrix & bias based on the input character. This is motivated by intelligibility, since it allows decomposition of output contribution in... | 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
Intelligible Language Modeling with Input Switched Affine Networks
### Paper Abstract
The computational mechanisms by which nonlinear recurrent neural networks (RNNs) ac... |
cnOvvf5Qe6h | NoDaLiDa/2023/Conference | 2023 | Spelling Correction for Estonian Learner Language | ["Kais Allkivi-Metsoja", "Jaagup Kippar"] | Second and foreign language (L2) learners often make specific spelling errors compared to native speakers. Language-independent spell-checking algorithms that rely on n-gram models can offer a simple solution for improving learner error detection and correction due to context-sensitivity. As the open-source speller pre... | ["spelling correction", "context-sensitive spell-checking", "learner language", "Estonian"] | Spelling Correction for Estonian Learner LanguageKais Allkivi-MetsojaSchool of Digital TechnologiesTallinn Universitykais@tlu.eeJaagup KipparSchool of Digital TechnologiesTallinn Universityjaagup@tlu.eeAbstractSecond and foreign language (L2) learn-ers tend to make specific spelling errorscompared to native speakers. L... | gfqMiH_M3l9 | The paper would need considerable re-writing in order to be published. The evaluation data would require major changes in the prose of the article also. I would recommend the authors revise their paper and re-submit it somewhere else; their work has lots of merit and it deserves to be published. | 4: Ok but not good enough - rejection |
Strong points of the paper
The paper is about an interesting question: what would be the best approach and/or tool for correcting texts written by learners of Estonian as a second language. The paper describes how a machine learning tool Jamspell was trained for Estonian, and compares its performance with other ML ap... | 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
Spelling Correction for Estonian Learner Language
### Paper Abstract
Second and foreign language (L2) learners often make specific spelling errors compared to native spe... |
T3RyQtRHebj | ICLR.cc/2021/Conference | 2021 | Slot Machines: Discovering Winning Combinations of Random Weights in Neural Networks | ["Maxwell Mbabilla Aladago", "Lorenzo Torresani"] | In contrast to traditional weight optimization in a continuous space, we demonstrate the existence of effective random networks whose weights are never updated. By selecting a weight among a fixed set of random values for each individual connection, our method uncovers combinations of random weights that match the perf... | ["initialization", "optimization"] | ABSTRACTIn contrast to traditional weight optimization in a continuous space, we demon-strate the existence of effective random networks whose weights are never up-dated. By selecting a weight among a fixed set of random values for each in-dividual connection, our method uncovers combinations of random weights thatmatc... | 2PjiHeEhDmd | Simple idea, well executed, well presented, and with good results - but lacks a bit in novelty | 7: Good paper, accept | **Update after authors' response**
I want to thank the authors for their response, and I am happy to see additional results on shared sets of weight values (which allows to easily relate the work to methods for training low-bit-width networks). To further increase impact and significance of the work it would be necessa... | 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
Slot Machines: Discovering Winning Combinations of Random Weights in Neural Networks
### Paper Abstract
In contrast to traditional weight optimization in a continuous sp... |
HNeeXUiQjkc | humanrobotinteraction.org/HRI/2022/Workshop/VAM-HRI | 2022 | Designing a User Study for Comparing 2D and VR Human-in-the-Loop Robot Planning Interfaces | ["Gregory LeMasurier", "Jordan Allspaw", "Murphy Wonsick", "James Tukpah", "Taskin Padir", "Holly Yanco", "Elizabeth Phillips"] | Human-in-the-loop robot teleoperation interfaces enable operators to control robots to complete complex tasks, as seen by the success of teams in the DARPA Robotics Challenge. In our prior work we discussed our virtual reality planning interface for performing dexterous robot teleoperation. In this work, we discuss pla... | ["Human-robot interaction (HRI)", "Robot Teleoperation", "Virtual Reality (VR)", "Mobile Robots"] | Designing a User Study for Comparing 2D and VRHuman-in-the-Loop Robot Planning InterfacesGregory LeMasurierUniversity of Massachusetts Lowellgregory lemasurier@student.uml.eduJordan AllspawUniversity of Massachusetts LowellJordan Allspaw@uml.eduMurphy WonsickNortheastern Universitywonsick.m@northeastern.eduJames Tukpah... | ScgnZY3Dxc | Thorough and relevant study proposed, accept | 9: Top 15% of accepted papers, strong accept | This paper describes a proposed study to compare a 2D interface using mouse and keyboard control with a VR interface using associated VR controls in a between-subjects study. Two tasks will be assigned to each participant, a manipulation task and a navigation task. There is strong motivation for this study with signifi... | 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
Designing a User Study for Comparing 2D and VR Human-in-the-Loop Robot Planning Interfaces
### Paper Abstract
Human-in-the-loop robot teleoperation interfaces enable ope... |
zsKWh2pRSBK | ICLR.cc/2021/Conference | 2021 | Poisoned classifiers are not only backdoored, they are fundamentally broken | ["Mingjie Sun", "Siddhant Agarwal", "J Zico Kolter"] | Under a commonly-studied “backdoor” poisoning attack against classification models, an attacker adds a small “trigger” to a subset of the training data, such that the presence of this trigger at test time causes the classifier to always predict some target class. It is often implicitly assumed that the poisoned classif... | ["Backdoor Attacks", "Denoised Smoothing", "Perceptually-Aligned Gradients"] | ABSTRACTUnder a commonly-studied “backdoor” poisoning attack against classificationmodels, an attacker adds a small “trigger” to a subset of the training data, suchthat the presence of this trigger at test time causes the classifier to always predictsome target class. It is often implicitly assumed that the poisoned cl... | besoquBw0jD | It is an interesting finding that backdoor-poisoned models are also vulnerable to alternative "triggers", although the reason behind it is unknown | 5: Marginally below acceptance threshold | Summary:
This paper demonstrates that backdoor-poisoned machine learning models can also be vulnerable to alternative triggers. Specifically, adversarial samples that are generated against models robustified with Denoised Smoothing often show backdoor patterns. Therefore, these adversarial samples can be used to create... | 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
Poisoned classifiers are not only backdoored, they are fundamentally broken
### Paper Abstract
Under a commonly-studied “backdoor” poisoning attack against classificatio... |
rylxrsR9Fm | ICLR.cc/2019/Conference | 2019 | Neuron Hierarchical Networks | ["Han Yue", "De-An Wu", "Lei Wu", "Ji Xie"] | In this paper, we propose a neural network framework called neuron hierarchical network (NHN), that evolves beyond the hierarchy in layers, and concentrates on the hierarchy of neurons. We observe mass redundancy in the weights of both handcrafted and randomly searched architectures. Inspired by the development of huma... | ["neural network", "architecture search", "evolution strategy"] | ABSTRACTIn this paper, we propose a neural network framework called neuron hierarchicalnetwork (NHN), that evolves beyond the hierarchy in layers, and concentrates onthe hierarchy of neurons. We observe mass redundancy in the weights of bothhandcrafted and randomly searched architectures. Inspired by the developmentof ... | BygYDUs93X | addressing an interesting problem but current contribution borderline | 5: Marginally below acceptance threshold | The paper proposes a method for neural architecture search. It is observed that in standard layer-wise neural networks, there is mass redundancy in the model weights. The authors propose to go beyond the hierarchy of layers, prune low-sensitivity neurons and add new neurons to the network. The concept of layer is weake... | 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
Neuron Hierarchical Networks
### Paper Abstract
In this paper, we propose a neural network framework called neuron hierarchical network (NHN), that evolves beyond the hi... |
ykvm7OLh7B | ACM.org/ICMI/2021/Workshop/GENEA | 2021 | Probabilistic Human-like Gesture Synthesis from Speech using GRU-based WGAN | ["Bowen Wu", "chaoran liu", "Carlos Ishi", "Hiroshi Ishiguro"] | Gestures are crucial for increasing the human-likeness of agents and robots to achieve smoother interactions with humans. The realization of an effective system to model human gestures, which are matched with the speech utterances, is necessary to be embedded in these agents. In this work, we propose a GRU-based autore... | ["gesture synthesis", "speech", "agents", "model", "probabilistic", "wgan probabilistic", "wgan gestures", "crucial", "robots", "smoother interactions"] | Probabilistic Human-like Gesture Synthesis from Speech using GRU-basedWGANBOWEN WU, Osaka University, JapanCHAORAN LIU, ATR, JapanCARLOS T. ISHI, Guardian Robot Project, RIKEN, JapanHIROSHI ISHIGURO, Osaka University, JapanGestures are crucial for increasing the human-likeness of agents and robots to achieve smoother i... | WIXdzfkO4G | Probabilistic Gesture Generation using WGANs | 6: Marginally above acceptance threshold | The paper proposes an approach for speech conditioned gesture synthesis that uses WGANs with Gradient penalty over a recurrent GRU layer. A contribution of this paper is to learn to generate a distribution of gestures for the same speech input.
### Strengths
- A potential advantage of this approach is that the generat... | 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
Probabilistic Human-like Gesture Synthesis from Speech using GRU-based WGAN
### Paper Abstract
Gestures are crucial for increasing the human-likeness of agents and robot... |
HkxAisC9FQ | ICLR.cc/2019/Conference | 2019 | Improved robustness to adversarial examples using Lipschitz regularization of the loss | ["Chris Finlay", "Adam M. Oberman", "Bilal Abbasi"] | We augment adversarial training (AT) with worst case adversarial training
(WCAT) which improves adversarial robustness by 11% over the current state-
of-the-art result in the `2-norm on CIFAR-10. We interpret adversarial training as
Total Variation Regularization, which is a fundamental tool in mathematical im-
age pro... | ["Adversarial training", "adversarial examples", "deep neural networks", "regularization", "Lipschitz constant"] | ABSTRACTWe augment adversarial training (AT) with worst case adversarial training(WCAT) which improves adversarial robustness by 11% over the current state-of-the-art result in the `2norm on CIFAR-10. We obtain verifiable average caseand worst case robustness guarantees, based on the expected and maximum valuesof the n... | H1e9KG-ch7 | Interesting idea -- could be significantly strengthened | 6: Marginally above acceptance threshold | Summary: this paper uses a trick to simplify the adversarial loss by one in which the adversarial perturbation appears in closed form.
pros:
- interesting idea
- experiments are interesting
cons:
- formal results are either trivial or could be improved in their statements
- experimental guarantees only, up to what... | 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
Improved robustness to adversarial examples using Lipschitz regularization of the loss
### Paper Abstract
We augment adversarial training (AT) with worst case adversaria... |
BygKZkBtDH | ICLR.cc/2020/Conference | 2020 | Balancing Cost and Benefit with Tied-Multi Transformers | ["Raj Dabre", "Raphael Rubino", "Atsushi Fujita"] | This paper proposes a novel procedure for training multiple Transformers with tied parameters which compresses multiple models into one enabling the dynamic choice of the number of encoder and decoder layers during decoding. In sequence-to-sequence modeling, typically, the output of the last layer of the N-layer encode... | ["tied models", "encoder-decoder", "multi-layer softmaxing", "depth prediction", "model compression"] | ABSTRACTThis paper proposes a novel procedure for training multiple Transformers withtied parameters which compresses multiple models into one enabling the dynamicchoice of the number of encoder and decoder layers during decoding. In sequence-to-sequence modeling, typically, the output of the last layer of the N-layer ... | Syl23DhcFr | Official Blind Review #1 | 1: Reject | ** Summary **
In this paper, the authors propose a new variant of Transformer called Tied-multi Transformer. Given such a model with an N-layer encoder and an M-layer decoder, it is trained with M*N loss functions, where each combination of the nth-layer of the encoder and the mth-layer of the decoder is used to train ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Balancing Cost and Benefit with Tied-Multi Transformers
### Paper Abstract
This paper proposes a novel procedure for training multiple Transformers with tied parameters ... | |
-H0Nxkvr_Z | eswc-conferences.org/ESWC/2021/Conference/Poster_and_Demo_Track | 2021 | Inbox - a messaging app based on Linked Data Notifications and Solid | ["Anton\u00edn Karola", "Jakub Kl\u00edmek"] | In the context of the recent web re-decentralization and true data ownership initiative, several W3C Recommendations building on top of linked data were published.
These include the Linked Data Platform for basic read-write capabilities, Linked Data Notifications (LDN) as a simple messaging protocol, and Activity Strea... | ["Solid", "Linked Data Notifications", "ActivityStreams", "messaging"] | Inbox - a messaging app based on Linked DataNotifications and SolidAntonín Karola and Jakub Klímek[0000000172343051]Department of Software Engineering, Faculty of Information Technology,Czech Technical University in Prague, Thákurova 9, 160 00 Praha 6, Czech Republickarolan1@fit.cvut.cz ,klimejak@fit.cvut.czAbstract... | -tW7pG3cIYY | Interesting paper of a Solid-based application | 7: Good paper, accept | This paper presents a description of Inbox: a Solid-based messaging application focused on end-user usability. It describes different features which are aligned with what is expected from these kind of applications.
In my opinion these kind of papers are long overdue from the Solid community side. It is important to s... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Inbox - a messaging app based on Linked Data Notifications and Solid
### Paper Abstract
In the context of the recent web re-decentralization and true data ownership init... |
uC7QRdX5OjI | ICLR.cc/2021/Conference | 2021 | How to Motivate Your Dragon: Teaching Goal-Driven Agents to Speak and Act in Fantasy Worlds | ["Prithviraj Ammanabrolu", "Jack Urbanek", "Margaret Li", "Arthur Szlam", "Tim Rockt\u00e4schel", "Jason E Weston"] | We seek to create agents that both act and communicate with other agents in pursuit of a goal. Towards this end, we extend LIGHT (Urbanek et al. 2019)---a large-scale crowd-sourced fantasy text-game---with a dataset of quests. These contain natural language motivations paired with in-game goals and human demonstrations... | ["reinforcement learning", "text-based games", "goal-driven dialogue", "natural language processing"] | ABSTRACTWe seek to create agents that both act and communicate with other agents in pursuitof a goal. Towards this end, we extend LIGHT (Urbanek et al., 2019)—a large-scale crowd-sourced fantasy text-game—with a dataset of “quests”. These containnatural language motivations paired with in-game goals and human demonstra... | JJb_InAHSG2 | Dialog systems paper is somewhat unsurprising from representation learning standpoint | 4: Ok but not good enough - rejection | The paper intends to contribute three items: an extended dataset, an approach to applying reinforcement learning with that dataset, and evaluation highlighting the role of combined reinforcement learning and pre-training (representation learning) in the new tasks. Considering this paper as an ICLR submission, the prima... | 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
How to Motivate Your Dragon: Teaching Goal-Driven Agents to Speak and Act in Fantasy Worlds
### Paper Abstract
We seek to create agents that both act and communicate wit... |
60avttW0Mv | logconference.io/LOG/2022/Conference | 2022 | Continuous Neural Algorithmic Planners | ["Yu He", "Petar Veli\u010dkovi\u0107", "Pietro Lio", "Andreea Deac"] | Neural algorithmic reasoning studies the problem of learning algorithms with neural networks, especially using graph architectures. A recent proposal, XLVIN, reaps the benefits of using a graph neural network that simulates the value iteration algorithm in deep reinforcement learning agents. It allows model-free planni... | ["xlvin", "problem", "algorithms", "neural networks", "graph architectures", "recent proposal", "benefits", "graph neural network", "value iteration algorithm"] | Continuous Neural Algorithmic PlannersYu HeUniversity of Cambridgeyh441@cam.ac.ukPetar Veli ˇckovi ́cDeepMindpetarv@google.comPietro LiòUniversity of Cambridgepl219@cam.ac.ukAndreea DeacMila, Université de Montréaldeacandr@mila.quebecAbstractNeural algorithmic reasoning studies the problem of learning algorithms wit... | 4pHEL_LLHTA | Summary. This paper proposes a framework namely CNAP to solve algorithmic reasoning in continuous planning using GNNs. CNAP extends XLVIN to deal with the continuous action space through discretization. It applies factorized joint policy to reduce the exponential policy space to a linear space. It further adopts four n... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Continuous Neural Algorithmic Planners
### Paper Abstract
Neural algorithmic reasoning studies the problem of learning algorithms with neural networks, especially using ... | |||
Skx5txzb0W | ICLR.cc/2018/Conference | 2018 | A Boo(n) for Evaluating Architecture Performance | ["Ondrej Bajgar", "Rudolf Kadlec", "and Jan Kleindienst"] | We point out important problems with the common practice of using the best single model performance for comparing deep learning architectures, and we propose a method that corrects these flaws. Each time a model is trained, one gets a different result due to random factors in the training process, which include random ... | ["evaluation", "methodology"] | A Boo(n) for Evaluating Architecture PerformanceOndrej Bajgar1Rudolf Kadlec1Jan Kleindienst1AbstractWe point out important problems with the com-mon practice of using the best single model perfor-mance for comparing deep learning architectures,and we propose a method that corrects these flaws.Each time a model is train... | rynGrnpeM | This paper introduces a metric designed to address the inadequacy of best model performance numbers but, since this metric requires many runs on the test data and also claims to normalize different experimental processes without a solid basis for this normalization, it may actually compound the problems highlighted. | 4: Ok but not good enough - rejection | This paper addresses multiple issues arising from the fact that commonly reported best model performance numbers are a single sample from a performance distribution. These problems are very real, and they deserve significant attention from the ML community. However, I feel that the proposed solution may actually compo... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Boo(n) for Evaluating Architecture Performance
### Paper Abstract
We point out important problems with the common practice of using the best single model performance f... |
UkxdauhUYnu | ICLR.cc/2021/Conference | 2021 | No Feature Is An Island: Adaptive Collaborations Between Features Improve Adversarial Robustness | ["Yufeng Zhang", "Yunan Zhang", "ChengXiang Zhai"] | To classify images, neural networks extract features from raw inputs and then sum them up with fixed weights via the fully connected layer. However, the weights are fixed despite the input types. Such fixed prior limits networks' flexibility in adjusting feature reliance, which in turn enables attackers to flip network... | ["robustness", "adversarial attack", "defense", "representation learning", "cooperative game", "feature selection", "adversarial robustness", "reliable machine learning"] | ABSTRACTTo classify images, neural networks extract features from raw inputs and then sumthem up with fixed weights via the fully connected layer. However, the weights arefixed despite the input types. Such fixed prior limits networks’ flexibility in adjust-ing feature reliance, which in turn enables attackers to flip ... | 0NFOgKe3nhv | Review report | 4: Ok but not good enough - rejection | This paper extends TRADES and aims to improve both standard accuracy and robustness by leveraging explainable feature disentanglement. I have several concerns about this submission.
1) The presentation of this paper should further be improved.
1-a) Informative figures and illustrations are required. For example, it... | 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
No Feature Is An Island: Adaptive Collaborations Between Features Improve Adversarial Robustness
### Paper Abstract
To classify images, neural networks extract features ... |
SFIx1eHodWc | kg-construct.github.io/KGCW/2022/Workshop | 2022 | What is needed in a Knowledge Graph Management Platform? A survey and a proposal | ["Samira Babalou"] | Knowledge Graphs (KGs) play a significant and growing role for semantics-based support of a wide variety of applications. Until recently, creating and maintaining such knowledge graphs was done in a one-off manner requiring significant manual effort and expertise. Over the last few years, the first KG management platfo... | ["Semantic Web", "Knowledge Graph", "Knowledge Graph Platform", "Data Services and Functionality"] | What is needed in a Knowledge GraphManagement Platform? A survey and a proposalSamira Babalou*ID1 2, Franziska ZanderID1 2, Erik KleinsteuberID1, Badr ElHaouniID1, David Schellenberger CostaID2, Jens KattgeID2 4, BirgittaK onig-RiesID1 2 31Heinz-Nixdorf Chair for Distributed Information SystemsInstitute for Computer Sc... | tjtF6jpc2iS | Review for What is needed in a Knowledge Graph Management Platform? A survey and a proposal | 5: Marginally below acceptance threshold | The paper proposes 2 main contributions. The first one is to survey 11 recent KG platforms by examining functionalities broadly in 4 different categories. The other one is proposing a KG platform, named iKNOW, tailored for biodiversity research.
Although the 2 contributions are related to each other, I would suggest 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
What is needed in a Knowledge Graph Management Platform? A survey and a proposal
### Paper Abstract
Knowledge Graphs (KGs) play a significant and growing role for semant... |
B1n4ZKsjG | MIDL.amsterdam/2018/Conference | 2018 | Radiomics Enabler®, an ETL (Extract-Transform-Load) for biomedical imaging in big-data projects | ["Karine Seymour", "Pierre Payoux"] | Clinical data warehouses are now routinely used by large hospitals. They allow researchers to pool medical data from millions of patients to create unified and harmonized patient cohorts for clinical trials and epidemiology studies. In bio-medical imaging, although there is a need to extract large amounts of data to he... | ["ETL", "DICOM", "PACS"] | Radiomics Enabler R, an ETL(Extract-Transform-Load) for biomedical imaging inbig-data projectsKarine SeymourMedexprim, Toulouse, Francekseymour@medexprim.comhttp://www.medexprim.comPierre PayouxDépartement de médecine nucléaire,Hôpital de Purpan,Place du Docteur Baylac,TSA 40031,31059 Toulouse Cedex 9payoux.p@chu-t... | r1GBh9g0z | no deep learning | 1: Strong rejection | The paper describes an open source tool, “Radiomics Enabler”, which makes it easier for researchers to perform a multi-criteria search on the PACS, filter the results and select the relevant image studies, anonymize the images and save them to an external medium. There is also a small experiment that estimates the tim... | 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
Radiomics Enabler®, an ETL (Extract-Transform-Load) for biomedical imaging in big-data projects
### Paper Abstract
Clinical data warehouses are now routinely used by lar... |
wuwiQtZaiTk | MICCAI.org/2022/Challenge/FLARE | 2022 | Abdominal CT organ segmentation by accelerated nnUNet with a coarse to fine strategy | ["Shoujin Huang", "Feng Ye", "Jingyu Li", "Ziran Chen", "Yue Zhang", "Tan Zhang", "Xin Nie", "Kairen Deng", "Mengye Lyu"] | Abdominal CT organ segmentation is known to be challenging. The segmentation of multiple abdominal organs enables quantitative analysis of different organs, providing invaluable input for computer-aided diagnosis (CAD) systems. Based on nnUNet, we develop an abdominal organ segmentation method applicable to both abdomi... | ["CT segmentation FLARE2022 Deep Learning"] | Abdominal CT organ segmentation byaccelerated nnUNet with a coarse to fine strategyShoujin Huang1,2[0000−0001−6094−129X], Lifeng Mei1, Jingyu Li3, Ziran Chen1,Yue Zhang2, Tan Zhang4, Xin Nie2, Kairen Deng5, andMengye Lyu1[0000−0001−5548−8136]1College of Health Science and Environmental Engineering, Shenzhen TechnologyU... | FvNj_EUyD1 | Great network design and SSL method | 9: Top 15% of accepted papers, strong accept | Advantages: 1. Great two-phase network with outstanding performance and efficiency. 2. Rational SSL method with performance improvement.
Disadvantages: Fig.5 and typography should be checked. | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Abdominal CT organ segmentation by accelerated nnUNet with a coarse to fine strategy
### Paper Abstract
Abdominal CT organ segmentation is known to be challenging. The s... |
SygInj05Fm | ICLR.cc/2019/Conference | 2019 | Physiological Signal Embeddings (PHASE) via Interpretable Stacked Models | ["Hugh Chen", "Scott Lundberg", "Gabe Erion", "Su-In Lee"] | In health, machine learning is increasingly common, yet neural network embedding (representation) learning is arguably under-utilized for physiological signals. This inadequacy stands out in stark contrast to more traditional computer science domains, such as computer vision (CV), and natural language processing (NLP)... | ["Representation learning", "transfer learning", "health", "machine learning", "physiological signals", "interpretation", "feature attributions", "shapley values", "univariate embeddings", "LSTMs", "XGB", "neural networks", "stacked models", "model pipelines", "interpretable stacked models"] | ABSTRACTIn health, machine learning is increasingly common, yet neural network embed-ding (representation) learning is arguably under-utilized for physiological signals.This inadequacy stands out in stark contrast to more traditional computer sciencedomains, such as computer vision (CV), and natural language processing... | rkg-azY9hX | This paper presents an approach to produce embeddings for physiological signals that are interpretable. | 4: Ok but not good enough - rejection | The authors claim contributions in three areas:
1) Learning representations on physiological signals. The proposed approach uses LSTMS with a loss function that aims at predicting the next five minutes of the physiological signals. Based on their experiments, using this criteria outperforms
LSTM autoencoder approach... | 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
Physiological Signal Embeddings (PHASE) via Interpretable Stacked Models
### Paper Abstract
In health, machine learning is increasingly common, yet neural network embedd... |
Skg7VAEKDS | ICLR.cc/2020/Conference | 2020 | When Do Variational Autoencoders Know What They Don't Know? | ["Bin Dai", "David Wipf"] | Recently, the ability of deep generative models to detect outliers has been called into question because of the demonstration that they frequently assign higher probability density to samples from completely different data sets than were used for training. For example, a model trained on CIFAR-10 may counter-intuitive... | ["variational autoencoder", "generative model"] | ABSTRACTRecently, the ability of deep generative models to detect outliers has been calledinto question because of the demonstration that they frequently assign higher prob-ability density to samples from completely different data sets than were used fortraining. For example, a model trained on CIFAR-10 may counter-int... | rJlyrw7iYr | Official Blind Review #3 | 8: Accept | This paper provides an in-depth analysis of the recently observed phenomenon [7] that deep generative models may assign higher likelihood to out-of-distribution (OOD) examples for the specific model class of Variational Autoencoders (VAEs). The paper demonstrates that if appropriately designed in terms of network capac... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
When Do Variational Autoencoders Know What They Don't Know?
### Paper Abstract
Recently, the ability of deep generative models to detect outliers has been called into q... | |
iAmZUo0DxC0 | ICLR.cc/2021/Conference | 2021 | Unlearnable Examples: Making Personal Data Unexploitable | ["Hanxun Huang", "Xingjun Ma", "Sarah Monazam Erfani", "James Bailey", "Yisen Wang"] | The volume of "free" data on the internet has been key to the current success of deep learning. However, it also raises privacy concerns about the unauthorized exploitation of personal data for training commercial models. It is thus crucial to develop methods to prevent unauthorized data exploitation. This paper raises... | ["Unlearnable Examples", "Data Protection", "Adversarial Machine Learning"] | ABSTRACTThe volume of “free” data on the internet has been key to the current success ofdeep learning. However, it also raises privacy concerns about the unauthorizedexploitation of personal data for training commercial models. It is thus crucial todevelop methods to prevent unauthorized data exploitation. This paper r... | 5s5PTFho-qu | The idea presented in this paper is very interesting | 8: Top 50% of accepted papers, clear accept | Summary:
The authors proposed the idea of using invisible noise to make personal data unusable to authorized deep learning models. To achieve this goal, the authors proposed the idea of error-minimizing noise crafted by a min-min optimization method. The error-minimizing noise is then added to training examples to make... | 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
Unlearnable Examples: Making Personal Data Unexploitable
### Paper Abstract
The volume of "free" data on the internet has been key to the current success of deep learnin... |
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... | LDk_3oOVHSp | Dynamics of some weight combinations of real world DNN can be predicted | 7: Good paper, accept | The current work studies the implications of continuous symmetries of a DNN on its weight dynamics. Specifically, they consider linearly realized symmetries (such as a shift/translation to the logits), and use the fact that the loss/hessian/mini-batches, are insensitive to such shift to decouple their dynamics. They do... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Neural Mechanics: Symmetry and Broken Conservation Laws in Deep Learning Dynamics
### Paper Abstract
Understanding the dynamics of neural network parameters during train... |
SyW4Gjg0W | ICLR.cc/2018/Conference | 2018 | Kernel Graph Convolutional Neural Nets | ["Giannis Nikolentzos", "Polykarpos Meladianos", "Antoine J-P Tixier", "Konstantinos Skianis", "Michalis Vazirgiannis"] | Graph kernels have been successfully applied to many graph classification problems. Typically, a kernel is first designed, and then an SVM classifier is trained based on the features defined implicitly by this kernel. This two-stage approach decouples data representation from learning, which is suboptimal. On the other... | ["kernel", "features", "graphs", "convolutional neural nets", "kernels", "svm classifier", "decouples data representation", "learning"] | ABSTRACTGraph kernels have been successfully applied to many graph classification prob-lems. Typically, a kernel is first designed, and then an SVM classifier is trainedbased on the features defined implicitly by this kernel. This two-stage approachdecouples data representation from learning, which is suboptimal. On th... | rywsxcOxG | Clearly written but the evaluation is not thorough | 5: Marginally below acceptance threshold | This paper proposes a graph classification method by integrating three techniques, community detection, graph kernels, and CNNs.
* This paper is clearly written and easy to follow. Thus the clarity is high.
* The originality is not high as the application of neural networks for graph classification has already been s... | 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
Kernel Graph Convolutional Neural Nets
### Paper Abstract
Graph kernels have been successfully applied to many graph classification problems. Typically, a kernel is firs... |
BJxVT3EKDH | ICLR.cc/2020/Conference | 2020 | Corpus Based Amharic Sentiment Lexicon Generation | ["Girma Neshir", "Andeas Rauber", "and Solomon Atnafu"] | Sentiment classification is an active research area with several applications including analysis of political opinions, classifying comments, movie reviews, news reviews and product reviews. To employ rule based sentiment classification, we require sentiment lexicons. However, manual construction of sentiment lexicon i... | ["Amharic sentiment lexicon", "Amharic sentiment classification", "seed words"] | ABSTRACTSentiment classification is an active research area with several applications in-cludinganalysisofpoliticalopinions,classifyingcomments,moviereviews,newsreviews and product reviews. To employ rule based sentiment classification, werequiresentimentlexicons. However,manualconstructionofsentimentlexiconistime cons... | Hkec2hOaYB | Official Blind Review #1 | 1: Reject | This paper introduces a corpus-based approach to build sentiment lexicon for Amharic. In order to save time and costs for the resource-limited language, the lexicon is generated from an Amharic news corpus by the following steps: manually preparing polarized seed words lists (strongly positive and strongly negative), c... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Corpus Based Amharic Sentiment Lexicon Generation
### Paper Abstract
Sentiment classification is an active research area with several applications including analysis of ... | |
HJe62s09tX | ICLR.cc/2019/Conference | 2019 | Unsupervised Hyper-alignment for Multilingual Word Embeddings | ["Jean Alaux", "Edouard Grave", "Marco Cuturi", "Armand Joulin"] | We consider the problem of aligning continuous word representations, learned in multiple languages, to a common space. It was recently shown that, in the case of two languages, it is possible to learn such a mapping without supervision. This paper extends this line of work to the problem of aligning multiple languages ... | ["problem", "multiple languages", "common space", "languages", "unsupervised", "multilingual word embeddings", "multilingual word", "continuous word representations", "case", "possible"] | ABSTRACTWe consider the problem of aligning continuous word representations, learned inmultiple languages, to a common space. It was recently shown that, in the caseof two languages, it is possible to learn such a mapping without supervision. Thispaper extends this line of work to the problem of aligning multiple langu... | SyesiPQrhQ | Good work for multilingual embedding alignment | 7: Good paper, accept | This is a work regarding the alignment of word embedding for multiple languages.Though there are existing works similar to this one, most of them are only considering a pair of two languages, resulting in the composition issue mentioned in this work. The authors proposed a way of using a regularization term to reduce s... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Unsupervised Hyper-alignment for Multilingual Word Embeddings
### Paper Abstract
We consider the problem of aligning continuous word representations, learned in multiple... |
BJ4prNx0W | ICLR.cc/2018/Conference | 2018 | Learning what to learn in a neural program | ["Richard Shin", "Dawn Song"] | Learning programs with neural networks is a challenging task, addressed by a long line of existing work. It is difficult to learn neural networks which will generalize to problem instances that are much larger than those used during training. Furthermore, even when the learned neural program empirically works on all te... | ["oracle", "neural networks", "work", "difficult", "training", "verification set", "neural program", "challenging task", "long line"] | ABSTRACTLearning programs with neural networks is a challenging task, addressed by along line of existing work. It is difficult to learn neural networks which will gen-eralize to problem instances that are much larger than those used during training.Furthermore, even when the learned neural program empirically works on... | HJ0ww5VbG | Issue with scalability makes me not like the work | 5: Marginally below acceptance threshold | Previous work by Cai et al. (2017) shows how to use Neural Programmer-Interpreter (NPI) framework to prove correctness of a learned neural network program by introducing recursion. It requires generation of a diverse training set consisting of execution traces which describe in detail the role of each function in solvi... | 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 what to learn in a neural program
### Paper Abstract
Learning programs with neural networks is a challenging task, addressed by a long line of existing work. It... |
ng262VIrK08 | NeurIPS.cc/2021/Workshop/SVRHM | 2021 | Benchmarking human visual search computational models in natural scenes: models comparison and reference datasets | ["Ferm\u00edn Travi", "Gonzalo Ruarte", "Gaston Bujia", "Juan E Kamienkowski"] | Visual search is an essential part of almost any everyday human goal-directed interaction with the environment.
Nowadays, several algorithms are able to predict gaze positions during simple observation, but few models attempt to simulate human behavior during visual search in natural scenes. Furthermore, these models ... | ["visual search", "eye movements", "computational models", "ideal bayesian observer"] | Benchmarking human visual search computationalmodels in natural scenes: models comparison andreference datasetsFermin Travi1fermintravi@gmail.comGonzalo Ruarte1*gonzalorpg@gmail.comGaston Bujia1gbujia@dc.uba.arJuan E. Kamienkowski1,2juank@dc.uba.ar1Laboratorio de Inteligencia Artificial Aplicada, Instituto de Ciencias ... | _xtQk-ngdL | Nice attempt at highlighting some of the problems and working towards comparative benchmarking | 7: Good paper, accept | This paper discusses the problem of modelling targeted visual search tasks, and particularly looks at the problem of comparing how humans search against computational methods in terms of the gaze path. The paper highlights that existing work has often used different datasets and task formulations that make comparison h... | 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
Benchmarking human visual search computational models in natural scenes: models comparison and reference datasets
### Paper Abstract
Visual search is an essential part o... |
H1gR5iR5FX | ICLR.cc/2019/Conference | 2019 | Analysing Mathematical Reasoning Abilities of Neural Models | ["David Saxton", "Edward Grefenstette", "Felix Hill", "Pushmeet Kohli"] | Mathematical reasoning---a core ability within human intelligence---presents some unique challenges as a domain: we do not come to understand and solve mathematical problems primarily on the back of experience and evidence, but on the basis of inferring, learning, and exploiting laws, axioms, and symbol manipulation ru... | ["mathematics", "dataset", "algebraic", "reasoning"] | ABSTRACTMathematical reasoning—a core ability within human intelligence—presents someunique challenges as a domain: we do not come to understand and solve mathemat-ical problems primarily on the back of experience and evidence, but on the basis ofinferring, learning, and exploiting laws, axioms, and symbol manipulation... | SJeDlWsjj7 | Review of "Analysing Mathematical Reasoning Abilities of Neural Models" | 6: Marginally above acceptance threshold | This paper develops a framework for evaluating the ability of neural models on answering free-form mathematical problems. The contributions are i) a publicly available dataset, and ii) an evaluation of two existing model families, recurrent networks and the Transformer.
I think that this paper makes a good contributi... | 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
Analysing Mathematical Reasoning Abilities of Neural Models
### Paper Abstract
Mathematical reasoning---a core ability within human intelligence---presents some unique c... |
HJjvxl-Cb | ICLR.cc/2018/Conference | 2018 | Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor | ["Tuomas Haarnoja", "Aurick Zhou", "Pieter Abbeel", "Sergey Levine"] | Model-free deep reinforcement learning (RL) algorithms have been demonstrated on a range of challenging decision making and control tasks. However, these methods typically suffer from two major challenges: very high sample complexity and brittle convergence properties, which necessitate meticulous hyperparameter tuning... | ["deep reinforcement learning", "maximum entropy learning", "stochastic actor-critic"] | ABSTRACTModel-free deep reinforcement learning (RL) algorithms have been demonstratedon a range of challenging decision making and control tasks. However, thesemethods typically suffer from two major challenges: very high sample complexityand brittle convergence properties, which necessitate meticulous hyperparametertu... | SJ1QyNJZf | The paper considers the problem of Deep RL in continuous-action domains. It implements the well-studied idea of RL with Entropy bonus. The results on the control suit looks very promising, though the paper does not compare with the state-of-the-art variants of baseline. Also the implementation details of the algorithm ... | 5: Marginally below acceptance threshold | Quality and clarity:
It seems that the authors can do a better job to improve the readability of the paper and its conciseness. The current structure of paper seems a bit suboptimal to me. The first 6.5 page of the paper is used to explain the idea of RL with entropy reward and how it can be extended to the case of ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
### Paper Abstract
Model-free deep reinforcement learning (RL) algorith... |
BylEpe9ppX | AKBC.ws/2019/Conference | 2019 | Answering Visual-Relational Queries in Web-Extracted Knowledge Graphs | ["Daniel O\u00f1oro-Rubio", "Mathias Niepert", "Alberto Garc\u00eda-Dur\u00e1n", "Roberto Gonz\u00e1lez-S\u00e1nchez", "Roberto J. L\u00f3pez-Sastre"] | A visual-relational knowledge graph (KG) is a multi-relational graph whose entities are associated with images. We explore novel machine learning approaches for answering visual-relational queries in web-extracted knowledge graphs. To this end, we have created ImageGraph, a KG with 1,330 relation types, 14,870 entities... | ["queries", "knowledge graph", "entities", "images", "knowledge graphs", "imagegraph", "methods", "knowledge", "graph"] | Automated Knowledge Base Construction (2019) Conference paperAnswering Visual-Relational Queries inWeb-Extracted Knowledge GraphsDaniel Oñoro-Rubio daniel.onoro@neclab.euMathias Niepert mathias.niepert@neclab.euAlberto García-Durán alberto.duran@neclab.euRoberto González-Sánchez roberto.gonzalez@neclab.euNEC Labs ... | rJxQ2N6mfE | Interesting approach to a new task + new data set | 7: Good paper, accept | This work aims to address the problem of answering visual-relational queries in knowledge graphs where the entities are associated with web-extracted images.
The paper introduces a newly constructed large scale visual-relational knowledge graph built by scraping the web. Going beyond previous data sets like VisualGeno... | 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
Answering Visual-Relational Queries in Web-Extracted Knowledge Graphs
### Paper Abstract
A visual-relational knowledge graph (KG) is a multi-relational graph whose entit... |
ryeG924twB | ICLR.cc/2020/Conference | 2020 | Learning Expensive Coordination: An Event-Based Deep RL Approach | ["Zhenyu Shi*", "Runsheng Yu*", "Xinrun Wang*", "Rundong Wang", "Youzhi Zhang", "Hanjiang Lai", "Bo An"] | Existing works in deep Multi-Agent Reinforcement Learning (MARL) mainly focus on coordinating cooperative agents to complete certain tasks jointly. However, in many cases of the real world, agents are self-interested such as employees in a company and clubs in a league. Therefore, the leader, i.e., the manager of the c... | ["Multi-Agent Deep Reinforcement Learning", "Deep Reinforcement Learning", "Leader\u2013Follower Markov Game", "Expensive Coordination"] | ABSTRACTExisting works in deep Multi-Agent Reinforcement Learning (MARL) mainly fo-cus on coordinating cooperative agents to complete certain tasks jointly. However,in many cases of the real world, agents are self-interested such as employees ina company and clubs in a league. Therefore, the leader, i.e., the manager o... | BylAQBSCFS | Official Blind Review #2 | 6: Weak Accept | 1. Summary
The authors apply MARL to principal-agent / mechanism design problems where selfish agents need to be incentivized to coordinate towards a leader's (collective) goal.
The leader is modeled as a semi-MDP with event-based policy gradients and modules to model/predict followers' actions. The leader sends mess... | <|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 Expensive Coordination: An Event-Based Deep RL Approach
### Paper Abstract
Existing works in deep Multi-Agent Reinforcement Learning (MARL) mainly focus on coor... | |
HyM8V2A9Km | ICLR.cc/2019/Conference | 2019 | ACTRCE: Augmenting Experience via Teacher’s Advice | ["Yuhuai Wu", "Harris Chan", "Jamie Kiros", "Sanja Fidler", "Jimmy Ba"] | Sparse reward is one of the most challenging problems in reinforcement learning (RL). Hindsight Experience Replay (HER) attempts to address this issue by converting a failure experience to a successful one by relabeling the goals. Despite its effectiveness, HER has limited applicability because it lacks a compact and u... | ["language goals", "task generalization", "hindsight experience replays", "language grounding"] | ABSTRACTSparse reward is one of the most challenging problems in reinforcement learning(RL). Hindsight Experience Replay (HER) attempts to address this issue by con-verting a failure experience to a successful one by relabeling the goals. Despiteits effectiveness, HER has limited applicability because it lacks a compac... | BkxtS7Ko2X | Simple and intuitive idea but evaluation seems to be quite lacking | 5: Marginally below acceptance threshold | This paper considers the assumption implicit in hindsight experience replay (HER), namely that we have access to a mapping from states to goals. Rather than satisfying this requirement by defining goals as states, which involves great redundancy, the paper proposes a natural language goal representation. Concretely, fo... | 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
ACTRCE: Augmenting Experience via Teacher’s Advice
### Paper Abstract
Sparse reward is one of the most challenging problems in reinforcement learning (RL). Hindsight Exp... |
RGeQOjc58d | ICLR.cc/2021/Conference | 2021 | Improved Gradient based Adversarial Attacks for Quantized Networks | ["Kartik Gupta", "Thalaiyasingam Ajanthan"] | Neural network quantization has become increasingly popular due to efficient memory consumption and faster computation resulting from bitwise operations on the quantized networks. Even though they exhibit excellent generalization capabilities, their robustness properties are not well-understood. In this work, we system... | ["binary neural network", "gradient masking", "fake robustness", "temperature scaling", "adversarial attack", "signal propagation"] | ABSTRACTNeural network quantization has become increasingly popular due to efficientmemory consumption and faster computation resulting from bitwise operationson the quantized networks. Even though they exhibit excellent generalizationcapabilities, their robustness properties are not well-understood. In this work,we sy... | s442b6v_ki | Official Blind Review | 5: Marginally below acceptance threshold | This paper studies the robustness of quantized neural networks against adversarial attacks. The authors use some slight modification of existing methods to successfully increase the attack success rate. In general, I think the idea is interesting. But I have some concerns that need to be addressed:
1. I am not fully co... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Improved Gradient based Adversarial Attacks for Quantized Networks
### Paper Abstract
Neural network quantization has become increasingly popular due to efficient memory... |
tcxy7vRVKlg | NoDaLiDa/2023/Conference | 2023 | Training and Evaluating Norwegian Sentence Embedding Models | ["Bernt Ivar Utst\u00f8l N\u00f8dland"] | We train and evaluate Norwegian sentence embedding models using the contrastive learning methodology SimCSE. We start from pre-trained Norwegian encoder models and train both unsupervised and supervised models. The models are evaluated on a machine-translated version of semantic textual similarity datasets, as well as ... | ["sentence embeddings", "contrastive learning", "norwegian"] | Training and Evaluating Norwegian Sentence Embedding ModelsBernt Ivar Utstøl NødlandNorwegian Defence Research EstablishmentInstituttveien 202007 Kjellerbernt-ivar-utstol.nodland@ffi.noAbstractWe train and evaluate Norwegian sentenceembedding models using the contrastivelearning methodology SimCSE. We startfrom pre-tra... | 4ayQAwkfFB9 | The research presents and evaluates several sentence transformer models trained for the Norwegian (sometimes together with English) language. | 8: Top 50% of accepted papers, clear accept | The research is interesting and clearly described. Moreover, the novelty of the research is emphasized. My remarks:
1. The Related Work section focuses only on what is done for the Norwegian language and on the authors’ research. It would be interesting to see the broader picture here (i.e., to have a summary of what ... | 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
Training and Evaluating Norwegian Sentence Embedding Models
### Paper Abstract
We train and evaluate Norwegian sentence embedding models using the contrastive learning m... |
H1oyRlYgg | ICLR.cc/2017/conference | 2017 | On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima | ["Nitish Shirish Keskar", "Dheevatsa Mudigere", "Jorge Nocedal", "Mikhail Smelyanskiy", "Ping Tak Peter Tang"] | The stochastic gradient descent (SGD) method and its variants are algorithms of choice for many Deep Learning tasks. These methods operate in a small-batch regime wherein a fraction of the training data, say $32$--$512$ data points, is sampled to compute an approximation to the gradient. It has been observed in practi... | ["Deep learning", "Optimization"] | ABSTRACTThe stochastic gradient descent (SGD) method and its variants are algorithms ofchoice for many Deep Learning tasks. These methods operate in a small-batchregime wherein a fraction of the training data, say 32–512data points, is sampledto compute an approximation to the gradient. It has been observed in practice... | SJVFyZfNe | Little novelty but valuable empirical evidence | 10: Top 5% of accepted papers, seminal paper | The paper is an empirical study to justify that: 1. SGD with smaller batch sizes converges to flatter minima, 2. flatter minima have better generalization ability.
Pros and Cons:
Although there is little novelty in the paper, I think the work is of great value in shedding light into some interesting questions around ... | 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 Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima
### Paper Abstract
The stochastic gradient descent (SGD) method and its variants are algor... |
KTEde38blNB | ICLR.cc/2021/Conference | 2021 | Intervention Generative Adversarial Nets | ["Jiadong Liang", "Liangyu Zhang", "Cheng Zhang", "Zhihua Zhang"] | In this paper we propose a novel approach for stabilizing the training process of Generative Adversarial Networks as well as alleviating the mode collapse problem. The main idea is to incorporate a regularization term that we call intervention into the objective. We refer to the resulting generative model as Interventi... | ["generative adversarial nets", "novel", "training process", "generative adversarial networks", "mode collapse problem", "main idea", "regularization term", "intervention", "objective"] | ABSTRACTIn this paper we propose a novel approach for stabilizing the training process ofGenerative Adversarial Networks as well as alleviating the mode collapse problem.The main idea is to incorporate a regularization term that we call intervention intothe objective. We refer to the resulting generative model as Inter... | P7HFrvRZOes | Clear Reject | 3: Clear rejection | Intervention Generative Adversarial Nets
Summary
This paper proposes a GAN training procedure where images are encoded into latents, which are perturbed and passed through a generator/decoder, and the discriminator objective is augmented to encourage it to classify the perturbation. Results are presented for generati... | 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
Intervention Generative Adversarial Nets
### Paper Abstract
In this paper we propose a novel approach for stabilizing the training process of Generative Adversarial Netw... |
BJgPCveAW | ICLR.cc/2018/Conference | 2018 | Characterizing Sparse Connectivity Patterns in Neural Networks | ["Sourya Dey", "Kuan-Wen Huang", "Peter A. Beerel", "Keith M. Chugg"] | We propose a novel way of reducing the number of parameters in the storage-hungry fully connected layers of a neural network by using pre-defined sparsity, where the majority of connections are absent prior to starting training. Our results indicate that convolutional neural networks can operate without any loss of acc... | ["Machine learning", "Neural networks", "Sparse neural networks", "Pre-defined sparsity", "Scatter", "Connectivity patterns", "Adjacency matrix", "Parameter Reduction", "Morse code"] | ABSTRACTWe propose a novel way of reducing the number of parameters in the storage-hungry fully connected layers of a neural network by using pre-defined sparsity,where the majority of connections are absent prior to starting training. Our re-sults indicate that convolutional neural networks can operate without any los... | S1951kYef | hard to follow, confused about many points | 4: Ok but not good enough - rejection | This paper examines sparse connection patterns in upper layers of convolutional image classification networks. Networks with very few connections in the upper layers are experimentally determined to perform almost as well as those with full connection masks. Heuristics for distributing connections among windows/group... | 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
Characterizing Sparse Connectivity Patterns in Neural Networks
### Paper Abstract
We propose a novel way of reducing the number of parameters in the storage-hungry fully... |
SkyQWDcex | ICLR.cc/2017/conference | 2017 | A Context-aware Attention Network for Interactive Question Answering | ["Huayu Li", "Martin Renqiang Min", "Yong Ge", "Asim Kadav"] | We develop a new model for Interactive Question Answering (IQA), using Gated-Recurrent-Unit recurrent networks (GRUs) as encoders for statements and questions, and another GRU as a decoder for outputs. Distinct from previous work, our approach employs context-dependent word-level attention for more accurate statement r... | ["Deep learning", "Natural language processing", "Applications"] | ABSTRACTWe develop a new model for Interactive Question Answering (IQA), using Gated-Recurrent-Unit recurrent networks (GRUs) as encoders for statements and ques-tions, and another GRU as a decoder for outputs. Distinct from previous work,our approach employs context-dependent word-level attention for more accuratestat... | r1PJRDf4g | review | 4: Ok but not good enough - rejection | This paper proposes an "interactive" version of the bAbI dataset by adding supporting questions/answers to the dataset in cases where there is not enough information to answer the question. Interactive QA is certainly an interesting problem and is well-motivated by the paper. However, I don't feel like the bAbI extensi... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Context-aware Attention Network for Interactive Question Answering
### Paper Abstract
We develop a new model for Interactive Question Answering (IQA), using Gated-Recu... |
VLihM67Wdi6 | robot-learning.org/CoRL/2023/Conference | 2023 | STERLING: Self-Supervised Terrain Representation Learning from Unconstrained Robot Experience | ["Haresh Karnan", "Elvin Yang", "Daniel Farkash", "Garrett Warnell", "Joydeep Biswas", "Peter Stone"] | Terrain awareness, i.e., the ability to identify and distinguish different types of terrain, is a critical ability that robots must have to succeed at autonomous off-road navigation. Current approaches that provide robots with this awareness either rely on labeled data which is expensive to collect, engineered features... | ["Vision-based Navigation", "Representation Learning", "Learning from Experience"] | STERLING: Self-Supervised Terrain RepresentationLearning from Unconstrained Robot ExperienceHaresh Karnan1haresh.miriyala@utexas.eduElvin Yang1elvin.yang@utexas.eduDaniel Farkash2dmf248@cornell.eduGarett Warnell1,3garrett.a.warnell.civ@army.milJoydeep Biswas1joydeepb@cs.utexas.eduPeter Stone1,4pstone@cs.utexas.eduAbstr... | tpcg7-JvOTn | Strengths
- Using VICReg for learning terrain representation is clever and makes a lot of sense, given that the strength of VICReg is that it can learn from multi-modal inputs.
- The experiment is thorough with comparison to multiple baselines.
- The paper is well-written and easy to follow.
Weaknesses
- The paper m... | 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
STERLING: Self-Supervised Terrain Representation Learning from Unconstrained Robot Experience
### Paper Abstract
Terrain awareness, i.e., the ability to identify and dis... | ||
NibHms070zC | ICLR.cc/2021/Conference | 2021 | Faster Federated Learning with Decaying Number of Local SGD Steps | ["Jed Mills", "Jia Hu", "Geyong Min"] | In Federated Learning (FL), client devices collaboratively train a model without sharing the private data present on the devices. Federated Stochastic Gradient Descent (FedSGD) is a recent generalisation of the popular Federated Averaging algorithm. Recent works show that when client data is distributed heterogeneously... | ["Distributed Machine Learning", "Federated Learning"] | ABSTRACTIn Federated Learning (FL), client devices collaboratively train a model withoutsharing the private data present on the devices. Federated Stochastic Gradient De-scent (FedSGD) is a recent generalisation of the popular Federated Averaging al-gorithm. Recent works show that when client data is distributed hetero... | s9Rl1cB-80 | Theory and experiments need improvement. | 5: Marginally below acceptance threshold |
# Contributions
1. This paper proposes a practical and effective way to reduce computations and accelerate FedSGD.
2. This paper provides theoretical analysis of the convergence of the proposed algorithm.
3. This paper provides numerical experiments to prove the performance in practice.
# Strong points
1. Decayin... | 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
Faster Federated Learning with Decaying Number of Local SGD Steps
### Paper Abstract
In Federated Learning (FL), client devices collaboratively train a model without sha... |
B1KIbeJwG | ICLR.cc/2018/Workshop | 2018 | Deep Semiparametric Learning | ["Mika Sarkin Jain", "Jack Lindsey"] | We introduce a semiparametric approach to deep learning. Inspired by complementary learning systems theory in cognitive neuroscience, our approach combines elements of parametric and nonparametric learning by giving a neural network access to a differentiable nearest-neighbor algorithm. Analysis of our model suggests ... | ["deep learning", "nonparametric", "episodic learning", "nearest neighbors", "complementary learning systems"] | ABSTRACTWe introduce a semiparametric approach to deep learning. Inspired by comple-mentary learning systems theory in cognitive neuroscience, our approach com-bines elements of parametric and nonparametric learning by giving a neural net-work access to a differentiable nearest-neighbor algorithm. Analysis of our model... | HJte_VWFG | Interesting work but not good enough for acceptance | 4: Ok but not good enough - rejection | This paper proposes a semiparametric deep model which concatenates the nearest neighbors embedding (non-parametric) and instance embedding (parametric) for classification tasks. The experiments in domain adaption tried to argue that the proposed semiparametric model may take advantages of both parts, and converge quick... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deep Semiparametric Learning
### Paper Abstract
We introduce a semiparametric approach to deep learning. Inspired by complementary learning systems theory in cognitive n... |
SyeyF0VtDr | ICLR.cc/2020/Conference | 2020 | Recurrent Event Network : Global Structure Inference Over Temporal Knowledge Graph | ["Woojeong Jin", "He Jiang", "Meng Qu", "Tong Chen", "Changlin Zhang", "Pedro\u00a0Szekely", "Xiang Ren"] | Modeling dynamically-evolving, multi-relational graph data has received a surge of interests with the rapid growth of heterogeneous event data. However, predicting future events on such data requires global structure inference over time and the ability to integrate temporal and structural information, which are not yet... | ["Temporal Knowledge Graphs", "Representation Learning", "Graph Sequence Inference", "Knowledge Graph Completion"] | ABSTRACTModeling dynamically-evolving, multi-relational graph data has received a surgeof interests with the rapid growth of heterogeneous event data. However, predictingfuture events on such data requires global structure inference over time and theability to integrate temporal and structural information, which are no... | HJl_y3iTFS | Official Blind Review #2 | 6: Weak Accept | This paper presents the Re-Net model which sequentially generates a temporal knowledge graph (TKG) in an autoregressive fashion by taking both global and local information into account.
The generation of TKG is motivated via a joint distribution problem which is then parametrized by the usage of a recurrent event encod... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Recurrent Event Network : Global Structure Inference Over Temporal Knowledge Graph
### Paper Abstract
Modeling dynamically-evolving, multi-relational graph data has rece... | |
Hyepjh4FwB | ICLR.cc/2020/Conference | 2020 | ProtoAttend: Attention-Based Prototypical Learning | ["Sercan O. Arik", "Tomas Pfister"] | We propose a novel inherently interpretable machine learning method that bases decisions on few relevant examples that we call prototypes. Our method, ProtoAttend, can be integrated into a wide range of neural network architectures including pre-trained models. It utilizes an attention mechanism that relates the encode... | ["Interpretability", "sample-based explanations", "prototypes", "confidence estimation"] | ABSTRACTWe propose a novel inherently interpretable machine learning method that basesdecisions on few relevant examples that we call prototypes . Our method, ProtoAt-tend, can be integrated into a wide range of neural network architectures includingpre-trained models. It utilizes an attention mechanism that relates th... | Byx38twAYH | Official Blind Review #1 | 6: Weak Accept | The aim of this work is to make deep learning classifiers more interpretable by "projecting" each input sample into a small collection of prototype examples (with some weighting over those) and then basing the decision on a combination of the latent representations of the chosen prototypes. In this way, the chosen cate... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
ProtoAttend: Attention-Based Prototypical Learning
### Paper Abstract
We propose a novel inherently interpretable machine learning method that bases decisions on few rel... | |
HJMN-xWC- | ICLR.cc/2018/Conference | 2018 | Learning Parsimonious Deep Feed-forward Networks | ["Zhourong Chen", "Xiaopeng Li", "Nevin L. Zhang"] | Convolutional neural networks and recurrent neural networks are designed with network structures well suited to the nature of spacial and sequential data respectively. However, the structure of standard feed-forward neural networks (FNNs) is simply a stack of fully connected layers, regardless of the feature correlatio... | ["Parsimonious Deep Feed-forward Networks", "structure learning", "classification", "overfitting", "fewer parameters", "high interpretability"] | ABSTRACTConvolutional neural networks and recurrent neural networks are designed withnetwork structures well suited to the nature of spacial and sequential data respec-tively. However, the structure of standard feed-forward neural networks (FNNs)is simply a stack of fully connected layers, regardless of the feature cor... | BJ25hzHWf | Needs improvement | 5: Marginally below acceptance threshold | The main strengths of the paper are the supporting experimental results in comparison to plain feed-forward networks (FNNs). The proposed method is focused on discovering sparse neural networks. The experiments show that sparsity is achieved and still the discovered sparse networks have comparable or better performan... | 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 Parsimonious Deep Feed-forward Networks
### Paper Abstract
Convolutional neural networks and recurrent neural networks are designed with network structures well... |
S1xD6xHKDr | ICLR.cc/2020/Conference | 2020 | Building Hierarchical Interpretations in Natural Language via Feature Interaction Detection | ["Hanjie Chen", "Guangtao Zheng", "Yangfeng Ji"] | The interpretability of neural networks has become crucial for their applications in real world with respect to the reliability and trustworthiness. Existing explanation generation methods usually provide important features by scoring their individual contributions to the model prediction and ignore the interactions be... | ["Hierarchical Interpretations", "Natural Language Processing", "Feature Interaction"] | ABSTRACTThe interpretability of neural networks has become crucial for their applicationsin real-world with respect to reliability and trustworthiness. Existing explanationgeneration methods usually provide important features by scoring their individualcontributions to the model prediction and ignore the interactions b... | SylViVGJ5H | Official Blind Review #1 | 3: Weak Reject |
Summary: The paper describes a method for providing explanations for the predictions made by a text classifier. The method is novel in that it aims to explicitly model interactions between words so that e.g. the phrase "not bad" appears as an explanation itself, rather than as separate words "not" and "bad". They prod... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Building Hierarchical Interpretations in Natural Language via Feature Interaction Detection
### Paper Abstract
The interpretability of neural networks has become crucial... | |
SJCPLLpaW | ICLR.cc/2018/Conference | 2018 | Exploring the Hidden Dimension in Accelerating Convolutional Neural Networks | ["Zhihao Jia", "Sina Lin", "Charles R. Qi", "Alex Aiken"] | DeePa is a deep learning framework that explores parallelism in all parallelizable dimensions to accelerate the training process of convolutional neural networks. DeePa optimizes parallelism at the granularity of each individual layer in the network. We present an elimination-based algorithm that finds an optimal paral... | ["Parallelism of Convolutional Neural Networks", "Accelerating Convolutional Neural Networks"] | ABSTRACTDeePa is a deep learning framework that explores parallelism in all parallelizabledimensions to accelerate the training process of convolutional neural networks.DeePa optimizes parallelism at the granularity of each individual layer in the net-work. We present an elimination-based algorithm that finds an optima... | BJ_nVijxM | Potential for accelerating CNNs | 5: Marginally below acceptance threshold | The paper proposes an approach that offers speedup on common convolutional neural networks. It presents the approach well and shows results comparing with other popular frameworks used in the field.
Originality
- The automation of parallelism across the different dimensions in each of the layers appears somewhat new. ... | 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
Exploring the Hidden Dimension in Accelerating Convolutional Neural Networks
### Paper Abstract
DeePa is a deep learning framework that explores parallelism in all paral... |
HJgEe1SKPr | ICLR.cc/2020/Conference | 2020 | GAN-based Gaussian Mixture Model Responsibility Learning | ["Wanming Huang", "Shuai Jiang", "Xuan Liang", "Ian Oppermann", "Richard Yi Da Xu"] | Mixture Model (MM) is a probabilistic framework which allows us to define a dataset containing K different modes. When each of the modes is associated with a Gaussian distribution, we refer it as Gaussian MM, or GMM. Given a data point x, GMM may assume the existence of a random index k ∈ {1, . . . , K } identifying wh... | ["Generative Adversarial Networks"] | ABSTRACTMixture Model (MM) is a probabilistic framework which allows us to define adataset containing Kdifferent modes. When each of the modes is associated witha Gaussian distribution, we refer it as Gaussian MM, or GMM. Given a data pointx, GMM may assume the existence of a random index k2f1;:::;Kgidentify-ing which ... | rylsua2Q9S | Official Blind Review #3 | 1: Reject | This paper proposes to use GMM as the latent prior distribution of GAN. The model adds the means, covariances and the discrete priors of the GMM as learnable parameters to the GAN, which are jointly optimized with other GAN parameters. The model also adds a discrete classifier to the training process. During training, ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
GAN-based Gaussian Mixture Model Responsibility Learning
### Paper Abstract
Mixture Model (MM) is a probabilistic framework which allows us to define a dataset containin... | |
3K4-U_5cRw | robot-learning.org/CoRL/2022/Workshop/LangRob | 2022 | ProgPrompt: Generating Situated Robot Task Plans using Large Language Models | ["Ishika Singh", "Valts Blukis", "Arsalan Mousavian", "Ankit Goyal", "Danfei Xu", "Jonathan Tremblay", "Dieter Fox", "Jesse Thomason", "Animesh Garg"] | Task planning can require defining myriad domain knowledge about the world in which a robot needs to act. To ameliorate that effort, large language models (LLMs) can be used to score potential next actions during task planning, and even generate action sequences directly, given an instruction in natural language with n... | ["task planning", "instruction following", "LLM planner", "task plans as programs"] | PROG PROMPT : Generating Situated Robot Task Plansusing Large Language ModelsIshika Singh1, Valts Blukis2, Arsalan Mousavian2, Ankit Goyal2, Danfei Xu2,Jonathan Tremblay2, Dieter Fox2, Jesse Thomason1, Animesh Garg2Abstract — Task planning can require defining myriad do-main knowledge about the world in which a robot n... | 5DNxCRMqY8w | Extensive experiments, well written study on LLMs for robotic task planning | 7: Good paper, accept | **Summary**:
This work introduces ProgPrompt, a programmatic prompt format that provides a structured template for LLMs to perform semantic task planning. The main contribution is to express the prompt template in a very structured way, explicitly providing available objects and action affordances. They showcase plan ... | 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
ProgPrompt: Generating Situated Robot Task Plans using Large Language Models
### Paper Abstract
Task planning can require defining myriad domain knowledge about the worl... |
ByeUBANtvB | ICLR.cc/2020/Conference | 2020 | Learning to solve the credit assignment problem | ["Benjamin James Lansdell", "Prashanth Ravi Prakash", "Konrad Paul Kording"] | Backpropagation is driving today's artificial neural networks (ANNs). However, despite extensive research, it remains unclear if the brain implements this algorithm. Among neuroscientists, reinforcement learning (RL) algorithms are often seen as a realistic alternative: neurons can randomly introduce change, and use un... | ["biologically plausible deep learning", "node perturbation", "REINFORCE", "synthetic gradients", "feedback alignment"] | ABSTRACTBackpropagation is driving today’s artificial neural networks (ANNs). However,despite extensive research, it remains unclear if the brain implements this algo-rithm. Among neuroscientists, reinforcement learning (RL) algorithms are oftenseen as a realistic alternative: neurons can randomly introduce change, and... | rylITWMqKS | Official Blind Review #2 | 6: Weak Accept | This paper proposes a method that addresses the "weight transport" problem [1] (not cited by the authors) emphasizing the biological infeasibility of artificial neural networks (ANNs) which are trained by gradients computed by the backpropagation algorithm [2a, 2b]. It is arguably the most eminent criticism (among many... | <|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 solve the credit assignment problem
### Paper Abstract
Backpropagation is driving today's artificial neural networks (ANNs). However, despite extensive resea... | |
HkxjqxBYDB | ICLR.cc/2020/Conference | 2020 | Episodic Reinforcement Learning with Associative Memory | ["Guangxiang Zhu*", "Zichuan Lin*", "Guangwen Yang", "Chongjie Zhang"] | Sample efficiency has been one of the major challenges for deep reinforcement learning. Non-parametric episodic control has been proposed to speed up parametric reinforcement learning by rapidly latching on previously successful policies. However, previous work on episodic reinforcement learning neglects the relationsh... | ["Deep Reinforcement Learning", "Episodic Control", "Episodic Memory", "Associative Memory", "Non-Parametric Method", "Sample Efficiency"] | ABSTRACTSample efficiency has been one of the major challenges for deep reinforcementlearning. Non-parametric episodic control has been proposed to speed up para-metric reinforcement learning by rapidly latching on previously successful poli-cies. However, previous work on episodic reinforcement learning neglects there... | HkgePooptS | Official Blind Review #3 | 6: Weak Accept | The paper proposes to combine DQN with a nonparametric estimate of the optimal Q function based on the graph of all observed transitions in the buffer. Specifically, they use the nonparametric estimate as a regularizer in the DQN loss. They show that this regularizer facilitates learning, and compare to other nonparame... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Episodic Reinforcement Learning with Associative Memory
### Paper Abstract
Sample efficiency has been one of the major challenges for deep reinforcement learning. Non-pa... | |
BJsD7L1vz | ICLR.cc/2018/Workshop | 2018 | Diversity-Driven Exploration Strategy for Deep Reinforcement Learning | ["Zhang-Wei Hong", "Tzu-Yun Shann", "Shih-Yang Su", "Yi-Hsiang Chang", "Chun-Yi Lee"] | Efficient exploration remains a challenging research problem in reinforcement learning, especially when an environment contains large state spaces, deceptive local optima, or sparse rewards. To tackle this problem, we present a diversity-driven approach for exploration, which can be easily combined with both off- and o... | ["Deep Reinforcement Learning", "Exploration"] | ABSTRACTEfficient exploration remains a challenging research problem in reinforcementlearning, especially when an environment contains large state spaces, deceptivelocal optima, or sparse rewards. To tackle this problem, we present a diversity-driven approach for exploration, which can be easily combined with both off-... | H14mqu3uf | Review | 7: Good paper, accept | # Summary
This paper proposes a new method for exploration in reinforcement learning. The idea is to encourage the policy to be different from its recent policies by introducing KL-divergence penalty between the current policy and the past policies. The result shows that the proposed method (applied to DQN and A2C) per... | 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
Diversity-Driven Exploration Strategy for Deep Reinforcement Learning
### Paper Abstract
Efficient exploration remains a challenging research problem in reinforcement le... |
BkeK-nRcFX | ICLR.cc/2019/Conference | 2019 | The Nonlinearity Coefficient - Predicting Generalization in Deep Neural Networks | ["George Philipp", "Jaime G. Carbonell"] | For a long time, designing neural architectures that exhibit high performance was considered a dark art that required expert hand-tuning. One of the few well-known guidelines for architecture design is the avoidance of exploding or vanishing gradients. However, even this guideline has remained relatively vague and circ... | ["deep learning", "neural networks", "nonlinearity", "activation functions", "exploding gradients", "vanishing gradients", "neural architecture search"] | ABSTRACTFor a long time, designing neural architectures that exhibit high performance wasconsidered a dark art that required expert hand-tuning. One of the few well-knownguidelines for architecture design is the avoidance of exploding or vanishing gra-dients. However, even this guideline has remained relatively vague a... | Bkx3LR333Q | Solid contribution | 7: Good paper, accept | This paper proposes a metric to measure the "nonlinearity" of neural network, and presents evidence that the value of this metric at initialization time is predictive of generalization performance.
Apart from a few problems I think this paper is well written and thorough. The contribution is solid, although not earth ... | 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 Nonlinearity Coefficient - Predicting Generalization in Deep Neural Networks
### Paper Abstract
For a long time, designing neural architectures that exhibit high per... |
rJgzz3Y4qV | MIDL.io/2019/Conference/Abstract | 2019 | Guiding 3D U-nets with signed distance fi elds for creating 3D models from images | ["Kristine Aavild Juhl", "Rasmus Reinhold Paulsen", "Anders Bjorholm Dahl", "Vedrana Andersen Dahl", "Ole De Backer", "Klaus Fuglsang Kofoed", "Oscar Camara"] | Morphological analysis of the left atrial appendage is an important tool to assess risk of
ischemic stroke. Most deep learning approaches for 3D segmentation is guided by binary
labelmaps, which results in voxelized segmentations unsuitable for morphological analysis.
We propose to use signed distance fields to guide ... | ["Signed distance fi\felds", "pixel-wise regression", "left atrial appendage"] | Medical Imaging with Deep Learning 2019 MIDL 2019 { Extended Abstract TrackGuiding 3D U-nets with signed distance elds for creating3D models from imagesKristine Aavild Juhl kajul@dtu.dkRasmus Reinhold Paulsen rapa@dtu.dkAnders Bjorholm Dahl abda@dtu.dkVedrana Andersen Dahl vand@dtu.dkTechnical University of Denmark, Ko... | HyepJjxSjE | Review | 3: Accept | The interesting part in this work is to use signed distance fields to seek morphologically consistent 3D models. Synthetic data and real data demonstrated promising results. | 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
Guiding 3D U-nets with signed distance fi elds for creating 3D models from images
### Paper Abstract
Morphological analysis of the left atrial appendage is an important ... |
_eQoI06U3zF | auai.org/UAI/2022/Workshop/CRL | 2022 | Structure by Architecture: Disentangled Representations without Regularization | ["Felix Leeb", "Giulia Lanzillotta", "Yashas Annadani", "Michel Besserve", "Stefan Bauer", "Bernhard Sch\u00f6lkopf"] | We study the problem of self-supervised structured representation learning using autoencoders for downstream tasks such as generative modeling. Unlike most methods which rely on matching an arbitrary, relatively unstructured, prior distribution for sampling, we propose a sampling technique that relies solely on the ind... | ["Hierarchical Representation Learning", "Disentanglement", "Structured Representation Learning", "Deep Autoencoders"] | Structure by Architecture: Disentangled Representations without RegularizationAbstractWe study the problem of self-supervised struc-tured representation learning using autoencodersfor downstream tasks such as generative modeling.Unlike most methods which rely on matching anarbitrary, relatively unstructured, prior dist... | 0W7B6QjZGJ2 | review for Structure by Architecture: Disentangled Representations without Regularization | 6: Marginally above acceptance threshold | Summary: This paper proposes the Structural Autoencoder (SAE) architecture for self-supervised structure representation learning. Compared to prior work, the key change in the architecture is the introduction of a sequence of independent random variables that are fed into the decoder at different levels of NN. The arch... | 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
Structure by Architecture: Disentangled Representations without Regularization
### Paper Abstract
We study the problem of self-supervised structured representation learn... |
VdWaMgaTKtX | JSYS/2021/Mar_Papers | 2021 | SoK: Function-as-a-Service: From An Application Developer's Perspective | ["Ali Raza", "Ibrahim Matta", "Nabeel Akhtar", "Vasiliki Kalavri", "Vatche Isahagian"] | In the past few years, FaaS has gained significant popularity and became a go-to choice for deploying cloud applications and micro-services. FaaS with its unique 'pay as you go' pricing model and key performance benefits over other cloud services, offers an easy and intuitive programming model to build cloud applicatio... | ["serverless", "faas", "lambda", "gcf", "cloud"] | Journal of Systems Research (JSys) 2021Function -as-a-Service : From an Application Developer ’sPerspectiveAli RazaBoston Universityaraza@bu.eduIbrahim MattaBoston Universitymatta@bu.eduNabeel AkhtarAkamai Technologies Inc.nakhtar@akamai.comVasiliki KalavriBoston Universityvkalavri@bu.eduVatche IsahagianIBM Researchvat... | vbexX5bqWup | Good first part, improvement potential for second part | Paper summary
-------------
This SoK paper in the area of serverless computing systematizes measurement studies, summarizes a serverless economic model, discusses challenges for different classes of applications, and outlines future research directions. It primarily takes the perspective of an application developer (o... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
SoK: Function-as-a-Service: From An Application Developer's Perspective
### Paper Abstract
In the past few years, FaaS has gained significant popularity and became a go-... | ||
rJccgk3iM | MIDL.amsterdam/2018/Conference | 2018 | Deep learning based automatic liver tumor segmentation in CT with shape-based post-processing | ["Grzegorz Chlebus", "Andrea Schenk", "Jan Hendrik Moltz", "Bram van Ginneken", "Horst Karl Hahn", "Hans Meine"] | Accurate automatic liver tumor segmentation would have a big impact on liver therapy planning procedures and follow-up reporting, thanks to automation, standardization and incorporation of full volumetric information. In this work, we propose a fully automatic method for liver tumor segmentation in CT images based on a... | ["liver", "tumor", "segmentation", "CT", "deep learning", "convolutional neural network"] | Deep learning based automatic liver tumorsegmentation in CT with shape-based post-processingGrzegorz Chlebus, Andrea Schenk, Jan Hendrik MoltzFraunhofer MEVIS, Bremen, Germany{grzegorz.chlebus,andrea.schenk,jan.moltz}@mevis.fraunhofer.deBram van GinnekenRadboud UMC, Nijmegen, The NetherlandsFraunhofer MEVIS, Bremen, Ge... | BJyYfgCTf | Nice manuscript, well described method, public data used. | 4: Top 50% of accepted papers, clear accept | The authors present a method for detecting and segmenting liver tumors from CT. They evaluate their method on the publicly available LiTS dataset. The method is well described. Contributions of the paper are made clear. Clear description of the evaluation metric for the lesion matching. Overall and interesting and high... | 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
Deep learning based automatic liver tumor segmentation in CT with shape-based post-processing
### Paper Abstract
Accurate automatic liver tumor segmentation would have a... |
62qGdDGSyqr | ecmlpkdd.org/ECMLPKDD/2021/Workshop/TeachML | 2021 | An Introduction to AI for GLAM | ["Daniel van Strien", "Mark Bell", "Nora Rose McGregor", "Michael Trizna"] | There is a growing interest in utilising Machine Learning (ML) techniques within Galleries, Libraries, Archives and Museums (GLAM), and a corresponding demand for training to enable practitioners to engage confidently in this area. Staff at these institutions are seeking practical knowledge and skills in ML concepts an... | ["Machine Learning", "ICML", "GLAM"] | An Introduction to AI for GLAMDaniel van Strien* 1Mark Bell* 2Nora McGregor* 1Michael Trizna* 3AbstractThere is a growing interest in utilising MachineLearning (ML) techniques within Galleries, Li-braries, Archives and Museums (GLAM), and acorresponding demand for training to enable prac-titioners to engage confidently... | Qg8SXuEM-RT | Conceptual ML Teaching workshop for GLAM employees | 6: Marginally above acceptance threshold | Pros:
+ the idea developing a workshop to advocate/promote ML education for staffs working within Galleries, Libraries, Archives and Museums (GLAM) is helpful to a wide community.
+ detailed lesson topics, and student performance evaluation schemes are provided.
+ the introduction of ML ethics in the workshop.
Cons:
-... | 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
An Introduction to AI for GLAM
### Paper Abstract
There is a growing interest in utilising Machine Learning (ML) techniques within Galleries, Libraries, Archives and Mus... |
bIrL42I_NF8 | ICLR.cc/2021/Conference | 2021 | On the Effect of Consensus in Decentralized Deep Learning | ["Tao Lin", "Lingjing Kong", "Anastasia Koloskova", "Martin Jaggi", "Sebastian U Stich"] | Decentralized training of deep learning models enables on-device learning over networks, as well as efficient scaling to large compute clusters. Experiments in earlier works revealed that decentralized training often suffers from generalization issues: the performance of models trained in a decentralized fashion is in ... | ["decentralized deep learning", "performance", "effect", "consensus", "training", "models", "generalization gap", "consensus distance", "deep learning models", "learning"] | ABSTRACTDecentralized training of deep learning models enables on-device learning overnetworks, as well as efficient scaling to large compute clusters. Experiments inearlier works revealed that decentralized training often suffers from generalizationissues: the performance of models trained in a decentralized fashion i... | _NdYak17A7a | Useful contributions on decentralized methods for training deep network, if somewhat incremental | 7: Good paper, accept | This paper studies decentralized gradient methods for training deep networks. It focuses on the so-called "critical consensus distance" and how disagreement during different stages of training ultimately effects optimization (training loss) and learning (generalization error). Theory is provided for the case of synchro... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Effect of Consensus in Decentralized Deep Learning
### Paper Abstract
Decentralized training of deep learning models enables on-device learning over networks, as ... |
SJl28R4YPr | ICLR.cc/2020/Conference | 2020 | Graph Neural Networks for Reasoning 2-Quantified Boolean Formulas | ["Fei Wang", "Zhanfu Yang", "Ziliang Chen", "Guannan Wei", "Tiark Rompf"] | It is valuable yet remains challenging to apply neural networks in logical reasoning tasks. Despite some successes witnessed in learning SAT (Boolean Satisfiability) solvers for propositional logic via Graph Neural Networks (GNN), there haven't been any successes in learning solvers for more complex predicate logic. I... | ["Graph Neural Networks", "2-Quantified Boolean Formula", "Symbolic Reasoning"] | ABSTRACTIt is valuable yet remains challenging to apply neural networks in logical reasoningtasks. Despite some successes witnessed in learning SAT (Boolean Satisfiability)solvers for propositional logic via Graph Neural Networks (GNN), there haven’tbeen any successes in learning solvers for more complex predicate logi... | B1xdwB9TKr | Official Blind Review #1 | 8: Accept | This paper explores how graph neural networks can be applied to test satisfiability of 2QBF logical formulas. They show that a straightforward extension of a GNN-based SAT solver to 2QBF fails to outperform random chance, and argue that this is because proving either satisfiability or unsatisfiability of 2QBF requires ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Graph Neural Networks for Reasoning 2-Quantified Boolean Formulas
### Paper Abstract
It is valuable yet remains challenging to apply neural networks in logical reasoning... | |
BJ6oOfqge | ICLR.cc/2017/conference | 2017 | Temporal Ensembling for Semi-Supervised Learning | ["Samuli Laine", "Timo Aila"] | In this paper, we present a simple and efficient method for training deep neural networks in a semi-supervised setting where only a small portion of training data is labeled. We introduce self-ensembling, where we form a consensus prediction of the unknown labels using the outputs of the network-in-training on differen... | ["labels", "unknown labels", "training", "temporal", "learning temporal", "learning", "simple", "efficient", "deep neural networks", "setting"] | ABSTRACTIn this paper, we present a simple and efficient method for training deep neuralnetworks in a semi-supervised setting where only a small portion of training datais labeled. We introduce self-ensembling, where we form a consensus predictionof the unknown labels using the outputs of the network-in-training on dif... | B1u6EURmg | 8: Top 50% of accepted papers, clear accept | This paper presents a semi-supervised technique for “self-ensembling” where the model uses a consensus prediction (computed from previous epochs) as a target to regress to, in addition to the usual supervised learning loss. This has connections to the “dark knowledge” idea, ladder networks work is shown in this paper 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
Temporal Ensembling for Semi-Supervised Learning
### Paper Abstract
In this paper, we present a simple and efficient method for training deep neural networks in a semi-s... | |
ccwT339SIu | ICLR.cc/2021/Conference | 2021 | Contrastive Video Textures | ["Medhini Narasimhan", "Shiry Ginosar", "Andrew Owens", "Alexei A Efros", "Trevor Darrell"] | Existing methods for video generation struggle to generate more than a short sequence of frames. We introduce a non-parametric approach for infinite video generation based on learning to resample frames from an input video. Our work is inspired by Video Textures, a classic method relying on pixel similarity to stitch s... | ["Video Textures", "Audio Conditioned Video Synthesis", "Self-Supervised learning", "Contrastive Learning"] | ABSTRACTExisting methods for video generation struggle to generate more than a short sequence offrames. We introduce a non-parametric approach for infinite video generation based onlearning to resample frames from an input video. Our work is inspired by Video Textures, aclassic method relying on pixel similarity to sti... | zLpFPhQFarT | Promising results, more discussions/comparisons are requested | 6: Marginally above acceptance threshold | In this paper, the authors proposed a non-parametric approach for video generation, i.e., video frame (un)conditional resampling. The proposed method is inspired by Video Textures (Sch¨odl et al., 2000), which synthesizes new videos by stitching together snippets of an existing video. Comparing to existing 'video textu... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Contrastive Video Textures
### Paper Abstract
Existing methods for video generation struggle to generate more than a short sequence of frames. We introduce a non-paramet... |
enhd0P_ERBO | ICLR.cc/2021/Conference | 2021 | Learning a Transferable Scheduling Policy for Various Vehicle Routing Problems based on Graph-centric Representation Learning | ["Inwook Kim", "Jinkyoo Park"] | Reinforcement learning has been used to learn to solve various routing problems. however, most of the algorithm is restricted to finding an optimal routing strategy for only a single vehicle. In addition, the trained policy under a specific target routing problem is not able to solve different types of routing problems... | ["Vehicle Routing Problem", "Multiple Traveling Salesmen Problem", "Capacitated Vehicle Routing Problem", "Reinforcement Learning", "Graph Neural Network"] | ABSTRACTReinforcement learning has been used to learn to solve various routing problems.however, most of the algorithm is restricted to finding an optimal routing strategyfor only a single vehicle. In addition, the trained policy under a specific target rout-ing problem is not able to solve different types of routing p... | __c7moOLpiT | An RL algorithm for mCVRP problem is proposed | 6: Marginally above acceptance threshold |
An end-to-end RL algorithm, Graph-centric RL-based Transferable Scheduler (GRLTS) is proposed to solve the capacitated multi-vehicle VRP problem.
The fuel capacity of vehicles is considered and vehicles can visit charging stations in the middle of their routes toward the customers, and if their charge is about to fin... | 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 a Transferable Scheduling Policy for Various Vehicle Routing Problems based on Graph-centric Representation Learning
### Paper Abstract
Reinforcement learning h... |
HJlzxgBtwH | ICLR.cc/2020/Conference | 2020 | Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack | ["Francesco Croce", "Matthias Hein"] | The evaluation of robustness against adversarial manipulations of neural networks-based classifiers is mainly tested with empirical attacks as the methods for the exact computation, even when available, do not scale to large networks. We propose in this paper a new white-box adversarial attack wrt the $l_p$-norms for $... | ["adversarial attacks", "adversarial robustness"] | Under review as a conference paper at ICLR 2020Minimally distorted Adversarial Exampleswith a Fast Adaptive Boundary AttackAnonymous authorsPaper under double-blind reviewAbstractThe evaluation of robustness against adversarial manipulations of neuralnetworks-based classifiers is mainly tested with empirical attacks as... | Hkgp5zksFS | Official Blind Review #2 | 6: Weak Accept | The authors propose a new gradient-based method (FAB) for constructing adversarial perturbations for deep neural networks. At a high level, the method repeatedly estimates the decision boundary based on the linearization of the classifier at a given point and projects to the closest "misclassified" example based on tha... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack
### Paper Abstract
The evaluation of robustness against adversarial manipulations of neural... | |
rkgv9oRqtQ | ICLR.cc/2019/Conference | 2019 | Compound Density Networks | ["Agustinus Kristiadi", "Asja Fischer"] | Despite the huge success of deep neural networks (NNs), finding good mechanisms for quantifying their prediction uncertainty is still an open problem. It was recently shown, that using an ensemble of NNs trained with a proper scoring rule leads to results competitive to those of Bayesian NNs. This ensemble method can b... | ["uncertainty in neural networks", "ensemble", "mixture model"] | ABSTRACTDespite the huge success of deep neural networks (NNs), finding good mecha-nisms for quantifying their prediction uncertainty is still an open problem. It wasrecently shown, that using an ensemble of NNs trained with a proper scoring ruleleads to results competitive to those of Bayesian NNs. This ensemble metho... | r1xXo7UF3m | potentially interesting ideas; could be better justified and validated. | 5: Marginally below acceptance threshold | The paper addresses the problem of producing sensible (high) uncertainties on out of distribution (OOD) data along with accurate predictions on in-distribution data. The authors consider a model wherein the weights of the network (\theta) are drawn from a matrix normal distribution whose parameters are in-turn a (non-l... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Compound Density Networks
### Paper Abstract
Despite the huge success of deep neural networks (NNs), finding good mechanisms for quantifying their prediction uncertainty... |
SkFEGHx0Z | ICLR.cc/2018/Conference | 2018 | Nearest Neighbour Radial Basis Function Solvers for Deep Neural Networks | ["Benjamin J. Meyer", "Ben Harwood", "Tom Drummond"] | We present a radial basis function solver for convolutional neural networks that can be directly applied to both distance metric learning and classification problems. Our method treats all training features from a deep neural network as radial basis function centres and computes loss by summing the influence of a featu... | ["classification problems", "radial basis function", "network", "deep neural networks", "convolutional neural networks", "distance metric learning"] | ABSTRACTWe present a radial basis function solver for convolutional neural networks thatcan be directly applied to both distance metric learning and classification problems.Our method treats all training features from a deep neural network as radial basisfunction centres and computes loss by summing the influence of a ... | BJ1FJQ5lG | The paper proposes a non-linear RBF kernel based layer on top of standard CNNs for improving classification. The presentation quality is not good, and the novelty is low. | 3: Clear rejection | - The paper proposes to use RBF kernel based neurons with each training data point as a center of
one of the RBF kernel neuron. (i) Kernel based neural networks have been explored before [A] and
(ii) ideas similar to the nearest neighbour based efficient but approximate learning for mixture
of Gaussians like sett... | 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
Nearest Neighbour Radial Basis Function Solvers for Deep Neural Networks
### Paper Abstract
We present a radial basis function solver for convolutional neural networks t... |
xgrZkRHliXR | robot-learning.org/CoRL/2023/Conference | 2023 | Learning to Design and Use Tools for Robotic Manipulation | ["Ziang Liu", "Stephen Tian", "Michelle Guo", "Karen Liu", "Jiajun Wu"] | When limited by their own morphologies, humans and some species of animals have the remarkable ability to use objects from the environment toward accomplishing otherwise impossible tasks. Robots might similarly unlock a range of additional capabilities through tool use. Recent techniques for jointly optimizing morpholo... | ["tool use", "manipulation", "design"] | Learning to Design and Use Toolsfor Robotic ManipulationZiang Liu∗, Stephen Tian∗, Michelle Guo, C. Karen Liu, Jiajun WuDepartment of Computer ScienceStanford University, United Statesziangliu,tians@stanford.eduAbstract: When limited by their own morphologies, humans and some speciesof animals have the remarkable abili... | xOsMgB63P2e | Quality of the paper:
The paper is of interesting topic as we know that general manipulation (say in a single cell scenario or home robots) may require robots to be able to change their tools on the fly. With that respect I find the paper addresses the right problem which will be of interest in the years to come.Never... | 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 to Design and Use Tools for Robotic Manipulation
### Paper Abstract
When limited by their own morphologies, humans and some species of animals have the remarkab... | ||
B1gIf305Ym | ICLR.cc/2019/Conference | 2019 | NSGA-Net: A Multi-Objective Genetic Algorithm for Neural Architecture Search | ["Zhichao Lu", "Ian Whalen", "Vishnu Boddeti", "Yashesh Dhebar", "Kalyanmoy Deb", "Erik Goodman", "Wolfgang Banzhaf"] | This paper introduces NSGA-Net, an evolutionary approach for neural architecture search (NAS). NSGA-Net is designed with three goals in mind: (1) a NAS procedure for multiple, possibly conflicting, objectives, (2) efficient exploration and exploitation of the space of potential neural network architectures, and (3) out... | ["neural architecture search", "evolutionary algorithms"] | ABSTRACTThis paper introduces NSGA-Net, an evolutionary approach for neural architecturesearch (NAS). NSGA-Net is designed with three goals in mind: (1) a NAS pro-cedure for multiple, possibly conflicting, objectives, (2) efficient exploration andexploitation of the space of potential neural network architectures, and ... | Hyl0mlAch7 | A combination of architecture search ideas | 6: Marginally above acceptance threshold | This paper proposes a search method for neural network architectures such that two (potentially) conflicting objectives: maximization of final performance and minimization of computational complexity can be pursued simultaneously. The motivation for the approach is that a principled multiobjective search procedure (NSG... | 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
NSGA-Net: A Multi-Objective Genetic Algorithm for Neural Architecture Search
### Paper Abstract
This paper introduces NSGA-Net, an evolutionary approach for neural archi... |
HMEiDPTOTmY | ICLR.cc/2021/Conference | 2021 | Later Span Adaptation for Language Understanding | ["Rongzhou Bao", "Zhuosheng Zhang", "hai zhao"] | Pre-trained contextualized language models (PrLMs) broadly use fine-grained tokens (words or sub-words) as minimal linguistic unit in pre-training phase. Introducing span-level information in pre-training has shown capable of further enhancing PrLMs. However, such methods require enormous resources and are lack of adap... | ["prlms", "information", "span adaptation", "language understanding", "tokens", "phase", "representation", "contextualized language models", "words", "minimal linguistic unit"] | ABSTRACTPre-trained contextualized language models (PrLMs) broadly use fine-grainedtokens (words or sub-words) as minimal linguistic units in the pre-training phase.Introducing span-level information in pre-training has shown capable of furtherenhancing PrLMs. However, such methods require enormous resources and lackad... | iBuyelfEcm | A simple, useful and easy to understand idea. However, the paper lacks novelty and the empirical results may not be convincing. | 4: Ok but not good enough - rejection | This paper introduces a new mechanism based on span-based representation grouping for better language model finetuning. Specifically, firstly, the paper proposes to utilize an n-gram splitter to segment the text into several segments. Secondly, a hierarchical network is built based on a three-level network (subword lev... | 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
Later Span Adaptation for Language Understanding
### Paper Abstract
Pre-trained contextualized language models (PrLMs) broadly use fine-grained tokens (words or sub-word... |
r1SBxSkvM | ICLR.cc/2018/Workshop | 2018 | Replacing Loss Functions And Target Representations For Adversarial Defense | ["Sean Saito", "Sujoy Roy"] | Recent works have shown that neural networks are susceptible to adversarial data, despite demonstrating high performance across various tasks. Hence, there is a growing need to develop techniques that make neural networks more robust against attacks given their increasingly frequent applications in real-life use cases.... | ["adversarial attacks", "target representation", "loss function"] | ABSTRACTRecent works have shown that neural networks are susceptible to adversarial data,despite demonstrating high performance across various tasks. Hence, there isa growing need to develop techniques that make neural networks more robustagainst attacks given their increasingly frequent applications in real-life use c... | rklMviodG | review | 4: Ok but not good enough - rejection | The paper presents three approaches to better defend an image model against adversarial attacks.
The first one is to replace the loss from cross entropy (CE) to mean squared error (MSE).
Results on adversarial attacks show better defense with this loss. Sadly, no results are shown to see the potential weakening of the ... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Replacing Loss Functions And Target Representations For Adversarial Defense
### Paper Abstract
Recent works have shown that neural networks are susceptible to adversaria... |
Qun8fv4qSby | ICLR.cc/2021/Conference | 2021 | Transient Non-stationarity and Generalisation in Deep Reinforcement Learning | ["Maximilian Igl", "Gregory Farquhar", "Jelena Luketina", "Wendelin Boehmer", "Shimon Whiteson"] | Non-stationarity can arise in Reinforcement Learning (RL) even in stationary environments. For example, most RL algorithms collect new data throughout training, using a non-stationary behaviour policy. Due to the transience of this non-stationarity, it is often not explicitly addressed in deep RL and a single neural ne... | ["Reinforcement Learning", "Generalization"] | ABSTRACTNon-stationarity can arise in Reinforcement Learning ( RL) even in stationaryenvironments. For example, most RLalgorithms collect new data throughouttraining, using a non-stationary behaviour policy. Due to the transience of thisnon-stationarity, it is often not explicitly addressed in deep RLand a single neura... | i5xYGGV8pO2 | This paper tackles the problem of distribution shift in the context of reinforcement learning, e.g. resulting from gradual exploration of the state-space or optimization of the critic. The main idea is that this kind of non-stationarity can have a lasting negative effect on generalization, even after the data distribut... | 7: Good paper, accept | The paper deals with a relevant issue. The simplified supervised learning setting is a good way of looking at the issue of non-stationarity in isolation and it makes a compelling case that neural networks optimized by SGD can have generalization issues in settings where the data distribution changes over time, even aft... | 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
Transient Non-stationarity and Generalisation in Deep Reinforcement Learning
### Paper Abstract
Non-stationarity can arise in Reinforcement Learning (RL) even in station... |
H1lxeRNYvB | ICLR.cc/2020/Conference | 2020 | Neural Operator Search | ["Wei Li", "Shaogang Gong", "Xiatian Zhu"] | Existing neural architecture search (NAS) methods explore a limited feature-transformation-only search space while ignoring other advanced feature operations such as feature self-calibration by attention and dynamic convolutions. This disables the NAS algorithms to discover more advanced network architectures. We addre... | ["deep learning", "autoML", "neural architecture search", "image classification", "attention learning", "dynamic convolution"] | ABSTRACTExisting neural architecture search (NAS) methods explore a limited feature-transformation-only search space, ignoring other advanced feature operationssuch as feature self-calibration by attention and dynamic convolutions. This dis-ables the NAS algorithms to discover more advanced network architectures. Weadd... | rylEwLw6Yr | Official Blind Review #3 | 3: Weak Reject | This paper proposes a complicated NAS approach, with a lot more ops, a larger search space and an extra teacher network. However, the results on CIFAR10 and ImageNet are both not competitive to other SOTA results.
My major concern is whether the extra complexity introduced by this paper is necessary:
1. The new searc... | <|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 Operator Search
### Paper Abstract
Existing neural architecture search (NAS) methods explore a limited feature-transformation-only search space while ignoring oth... | |
HylTBhA5tQ | ICLR.cc/2019/Conference | 2019 | The Limitations of Adversarial Training and the Blind-Spot Attack | ["Huan Zhang*", "Hongge Chen*", "Zhao Song", "Duane Boning", "Inderjit S. Dhillon", "Cho-Jui Hsieh"] | The adversarial training procedure proposed by Madry et al. (2018) is one of the most effective methods to defend against adversarial examples in deep neural net- works (DNNs). In our paper, we shed some lights on the practicality and the hardness of adversarial training by showing that the effectiveness (robustness on... | ["Adversarial Examples", "Adversarial Training", "Blind-Spot Attack"] | ABSTRACTThe adversarial training procedure proposed by Madry et al. (2018) is one of themost effective methods to defend against adversarial examples in deep neural net-works (DNNs). In our paper, we shed some lights on the practicality and thehardness of adversarial training by showing that the effectiveness (robustne... | rJe1e3jj3X | Clear and simple idea, insightful experiments. | 7: Good paper, accept | The paper is well written and the main contribution, a methodology to find “blind-spot attacks” well motivated and differences to prior work stated clearly.
The empirical results presented in Figure 1 and 2 are very convincing. The gain of using a sufficiently more complicated approach to assess the overall distance b... | 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
The Limitations of Adversarial Training and the Blind-Spot Attack
### Paper Abstract
The adversarial training procedure proposed by Madry et al. (2018) is one of the mos... |
H1faSn0qY7 | ICLR.cc/2019/Conference | 2019 | DL2: Training and Querying Neural Networks with Logic | ["Marc Fischer", "Mislav Balunovic", "Dana Drachsler-Cohen", "Timon Gehr", "Ce Zhang", "Martin Vechev"] | We present DL2, a system for training and querying neural networks with logical constraints. The key idea is to translate these constraints into a differentiable loss with desirable mathematical properties and to then either train with this loss in an iterative manner or to use the loss for querying the network for inp... | ["neural networks", "training with constraints", "querying networks", "semantic training"] | ABSTRACTWe present DL2, a system for training and querying neural networks with logicalconstraints. DL2 is more expressive than prior work and can capture a richer classof constraints on inputs, outputs and internals of models. Using DL2, one candeclaratively specify domain knowledge to be enforced during training or p... | HJx46y-q37 | Review | 6: Marginally above acceptance threshold | In this paper the authors propose DL2 a system for training and querying neural networks with logical constraints
The proposed approach is intriguing but in my humble opinion the presentation of the paper could be improved. Indeed I think that the paper is bit too hard to follow.
The example at page 2 is not clearly ... | 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
DL2: Training and Querying Neural Networks with Logic
### Paper Abstract
We present DL2, a system for training and querying neural networks with logical constraints. The... |
rkl_f6EFPS | ICLR.cc/2020/Conference | 2020 | The Probabilistic Fault Tolerance of Neural Networks in the Continuous Limit | ["El-Mahdi El-Mhamdi", "Rachid Guerraoui", "Andrei Kucharavy", "Sergei Volodin"] | The loss of a few neurons in a brain rarely results in any visible loss of function. However, the insight into what “few” means in this context is unclear. How many random neuron failures will it take to lead to a visible loss of function? In this paper, we address the fundamental question of the impact of the crash of... | ["Robustness", "theory of neural networks", "fault tolerance", "continuous limit", "Taylor expansion", "error bound", "neuromorphic computing", "continuous networks", "functional derivative"] | ABSTRACTThe loss of a few neurons in a brain rarely results in any visible loss of function.However, the insight into what “few” means in this context is unclear. How manyrandom neuron failures will it take to lead to a visible loss of function? In this paper,we address the fundamental question of the impact of the cra... | Skx2F789YH | Official Blind Review #1 | 1: Reject |
Review: This paper considers the problem of dropping neurons from a neural network. In the case where this is done randomly, this corresponds to the widely studied dropout algorithm. If the goal is to become robust to randomly dropped neurons during evaluation, then it seems sufficient to just train with dropout (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
The Probabilistic Fault Tolerance of Neural Networks in the Continuous Limit
### Paper Abstract
The loss of a few neurons in a brain rarely results in any visible loss o... | |
B1lKS2AqtX | ICLR.cc/2019/Conference | 2019 | Eidetic 3D LSTM: A Model for Video Prediction and Beyond | ["Yunbo Wang", "Lu Jiang", "Ming-Hsuan Yang", "Li-Jia Li", "Mingsheng Long", "Li Fei-Fei"] | Spatiotemporal predictive learning, though long considered to be a promising self-supervised feature learning method, seldom shows its effectiveness beyond future video prediction. The reason is that it is difficult to learn good representations for both short-term frame dependency and long-term high-level relations. W... | ["eidetic", "lstm", "video prediction", "model", "relations", "rnns", "network", "spatiotemporal predictive learning", "promising", "feature"] | ABSTRACTSpatiotemporal predictive learning, though long considered to be a promising self-supervised feature learning method, seldom shows its effectiveness beyond futurevideo prediction. The reason is that it is difficult to learn good representationsfor both short-term frame dependency and long-term high-level relati... | HJeJHrJc27 | The authors propose a futurue video prediction model based on recurrent 3D-CNNs and propose a novel memory mechanism (Eidetic Memory) to capture long term relationships inside the recurrent layer itself. They obtain surpass the state of the art on two commonly used, (relatively) simple benchmark video prediction datase... | 7: Good paper, accept | AFTER REBUTTAL:
This is an overall good work, and I do think proves its point. The results on the TaxiBJ dataset (not TatxtBJ, please correct the name in the paper) are compelling, and the concerns regarding some of the text explainations have been corrected.
-----
The proposed model uses a 3D-CNN with a new kind of... | 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
Eidetic 3D LSTM: A Model for Video Prediction and Beyond
### Paper Abstract
Spatiotemporal predictive learning, though long considered to be a promising self-supervised ... |
kOm3jWX8YN | robot-learning.org/CoRL/2023/Conference | 2023 | Learning to Discern: Imitating Heterogeneous Human Demonstrations with Preference and Representation Learning | ["Sachit Kuhar", "Shuo Cheng", "Shivang Chopra", "Matthew Bronars", "Danfei Xu"] | Practical Imitation Learning (IL) systems rely on large human demonstration datasets for successful policy learning. However, challenges lie in maintaining the quality of collected data and addressing the suboptimal nature of some demonstrations, which can compromise the overall dataset quality and hence the learning o... | ["Imitation Learning", "Preference Learning", "Manipulation"] | Learning to Discern: Imitating HeterogeneousHuman Demonstrations with Preference andRepresentation LearningSachit Kuhar∗, Shuo Cheng, Shivang Chopra, Matthew Bronars, Danfei XuGeorgia Institute of Technology{kuhar, shuocheng, shivangchopra11, mbronars, danfei }@gatech.eduAbstract: Practical Imitation Learning (IL) syst... | 8yIgCaG2GV | I think the paper was well organized and fairly easy to read and well motivated. While the individual concepts introduced by the paper are not particularly novel, the motivation behind putting them together (for the most part) to solve the problem of imitation learning from heterogenous distributions is explained convi... | 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 Discern: Imitating Heterogeneous Human Demonstrations with Preference and Representation Learning
### Paper Abstract
Practical Imitation Learning (IL) system... | ||
r1csyUJvG | ICLR.cc/2018/Workshop | 2018 | EXPLOITING LATENT INFORMATION IN RELATIONAL DATABASES VIA WORD EMBEDDING | ["Rajesh Bordawekar", "Oded Shmueli", "Bortik Bandyopadhyay"] | We propose Cognitive Databases, an approach for transparently enabling Artificial Intelligence (AI) capabilities in relational databases. A novel aspect of our design is to first view the structured data source as meaningful unstructured text, and then use the text to build a word embedding model. This model captures t... | ["Word Embedding", "Relational Databases", "Novel SQL queries"] | ABSTRACTWe propose Cognitive Databases, an approach for transparently enabling ArtificialIntelligence (AI) capabilities in relational databases. A novel aspect of our designis to first view the structured data source as meaningful unstructured text, and thenuse the text to build a word embedding model. This model captu... | SJAXPVGFz | Limited novelty and lacks experimental details | 4: Ok but not good enough - rejection | This paper applies word2vec model to relational databases and calls the resulting system Cognitive Databases. While the CI queries part is interesting, the abstract is shallow in terms of details. Moreover, apart from a few qualitative examples, there is no quantitative comparison to demonstrate utility of the proposed... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
EXPLOITING LATENT INFORMATION IN RELATIONAL DATABASES VIA WORD EMBEDDING
### Paper Abstract
We propose Cognitive Databases, an approach for transparently enabling Artifi... |
H1xD9sR5Fm | ICLR.cc/2019/Conference | 2019 | Minimum Divergence vs. Maximum Margin: an Empirical Comparison on Seq2Seq Models | ["Huan Zhang", "Hai Zhao"] | Sequence to sequence (seq2seq) models have become a popular framework for neural sequence prediction. While traditional seq2seq models are trained by Maximum Likelihood Estimation (MLE), much recent work has made various attempts to optimize evaluation scores directly to solve the mismatch between training and evaluati... | ["sequence to sequence", "training criteria"] | ABSTRACTSequence to sequence (seq2seq) models have become a popular framework forneural sequence prediction. While traditional seq2seq models are trained by Max-imum Likelihood Estimation (MLE), much recent work has made various attemptsto optimize evaluation scores directly to solve the mismatch between training andev... | rJxph-Uo3X | some interesting results and connections, but also some technical issues; revisions have improved the paper | 7: Good paper, accept | The authors have updated the paper and clarified some things, and now my impression of the paper has improved. It still feels a little incremental to me, but the potential application areas of these sorts of models are quite large and therefore incremental improvements are not insignificant. This paper suggests some na... | 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
Minimum Divergence vs. Maximum Margin: an Empirical Comparison on Seq2Seq Models
### Paper Abstract
Sequence to sequence (seq2seq) models have become a popular framework... |
BygdR0VKDr | ICLR.cc/2020/Conference | 2020 | Discrete Transformer | ["Jambay Kinley", "Yuntian Deng", "Alexander M. Rush"] | The transformer has become a central model for many NLP tasks from translation to language modeling to representation learning. Its success demonstrates the effectiveness of stacked attention as a replacement for recurrence for many tasks. In theory attention also offers more insights into the model’s internal decision... | ["transformer", "natural language processing", "attention"] | ABSTRACTThe transformer has become a central model for many NLP tasks from translationto language modeling to representation learning. Its success demonstrates the ef-fectiveness of stacked attention as a replacement for recurrence for many tasks.In theory attention also offers more insights into the model’s internal d... | SJlRIhKvtS | Official Blind Review #1 | 3: Weak Reject |
Summary:
This paper presents two modifications to the standard transformer architecture with the goal of improving interpretability while retaining performance in NLP tasks. The first modification is to switch from soft attention to hard attention using a Gumbel-Softmax, which forces the model to attend only to impor... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Discrete Transformer
### Paper Abstract
The transformer has become a central model for many NLP tasks from translation to language modeling to representation learning. I... | |
7Z29QbHxIL | ICLR.cc/2021/Conference | 2021 | FTSO: Effective NAS via First Topology Second Operator | ["Likang Wang", "Lei Chen"] | Existing one-shot neural architecture search (NAS) methods generally contain a giant supernet, which leads to heavy computational cost. Our method, named FTSO, separates the whole architecture search into two sub-steps. In the first step, we only search for the topology, and in the second step, we only search for the o... | ["Neural Architecture Search", "DARTS"] | ABSTRACTExisting one-shot neural architecture search (NAS) methods generally contain agiant super-net, which leads to heavy computational cost. Our method, namedFTSO, separates the whole architecture search into two sub-steps. In the first step,we only search for the topology, and in the second step, we only search for... | kBrCZZJrlbT | An interesting paper with very promising result | 5: Marginally below acceptance threshold | This work researches the issue of neural architecture search (NAS), which is of significance for practical applications of deep neural networks and has become an active research topic in the past several years. Many methods on NAS have been developed recently. The computational efficiency of search has been one of the ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
FTSO: Effective NAS via First Topology Second Operator
### Paper Abstract
Existing one-shot neural architecture search (NAS) methods generally contain a giant supernet, ... |
BJxsrgStvr | ICLR.cc/2020/Conference | 2020 | Drawing Early-Bird Tickets: Toward More Efficient Training of Deep Networks | ["Haoran You", "Chaojian Li", "Pengfei Xu", "Yonggan Fu", "Yue Wang", "Xiaohan Chen", "Richard G. Baraniuk", "Zhangyang Wang", "Yingyan Lin"] | (Frankle & Carbin, 2019) shows that there exist winning tickets (small but critical subnetworks) for dense, randomly initialized networks, that can be trained alone to achieve comparable accuracies to the latter in a similar number of iterations. However, the identification of these winning tickets still requires the c... | ["tickets", "eb tickets", "efficient training", "deep networks", "existence", "mask distance", "efficient", "frankle", "carbin", "small"] | ABSTRACT(Frankle & Carbin, 2019) shows that there exist winning tickets (small but crit-ical subnetworks) for dense, randomly initialized networks, that can be trainedalone to achieve a comparable accuracy to the latter in a similar number of it-erations. However, the identification of these winning tickets still requi... | Bkx7VRIhFH | Official Blind Review #3 | 6: Weak Accept | The authors further study the lottery ticket hypothesis formulated by Frankle and Carbin. They demonstrate that the sparsity pattern corresponding to a lottery ticket for a given initialization can be uncovered via low-cost training. By doing so, they propose a method to: 1) first identify the lottery ticket efficientl... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Drawing Early-Bird Tickets: Toward More Efficient Training of Deep Networks
### Paper Abstract
(Frankle & Carbin, 2019) shows that there exist winning tickets (small but... | |
S5S3eTEmouw | ICLR.cc/2021/Conference | 2021 | Offline Meta-Reinforcement Learning with Advantage Weighting | ["Eric Mitchell", "Rafael Rafailov", "Xue Bin Peng", "Sergey Levine", "Chelsea Finn"] | This paper introduces the offline meta-reinforcement learning (offline meta-RL) problem setting and proposes an algorithm that performs well in this setting. Offline meta-RL is analogous to the widely successful supervised learning strategy of pre-training a model on a large batch of fixed, pre-collected data (possibly... | ["offline", "meta-reinforcement learning", "meta-learning", "reinforcement learning", "maml"] | ABSTRACTThis paper introduces the offline meta-reinforcement learning (offline meta-RL)problem setting and proposes an algorithm that performs well in this setting. Offlinemeta-RL is analogous to the widely successful supervised learning strategy of pre-training a model on a large batch of fixed, pre-collected data (po... | xojCFeLv-bv | Review | 6: Marginally above acceptance threshold | The paper proposes the problem of fully offline meta-RL. Here, the idea is to leverage offline experience from multiple tasks to enable fast adaptation to new tasks. The paper distinguishes two settings of offline meta-RL, one where only the training data is collected offline and testing corresponds to sampling online ... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Offline Meta-Reinforcement Learning with Advantage Weighting
### Paper Abstract
This paper introduces the offline meta-reinforcement learning (offline meta-RL) problem s... |
Syx1DkSYwB | ICLR.cc/2020/Conference | 2020 | Variance Reduction With Sparse Gradients | ["Melih Elibol", "Lihua Lei", "Michael I. Jordan"] | Variance reduction methods such as SVRG and SpiderBoost use a mixture of large and small batch gradients to reduce the variance of stochastic gradients. Compared to SGD, these methods require at least double the number of operations per update to model parameters. To reduce the computational cost of these methods, we i... | ["optimization", "variance reduction", "machine learning", "deep neural networks"] | ABSTRACTVariance reduction methods such as SVRG (Johnson & Zhang, 2013) and Spi-derBoost (Wang et al., 2018) use a mixture of large and small batch gradientsto reduce the variance of stochastic gradients. Compared to SGD (Robbins &Monro, 1951), these methods require at least double the number of operations perupdate to... | rJx0y3HZYS | Official Blind Review #2 | 3: Weak Reject | Summary:
The author(s) provide a method which combines some property of SCGS method and SpiderBoost. Theoretical results are provided and achieve the state-of-the-art complexity, which match the one of SpiderBoost. Numerical experiments show some advantage compared to SpiderBoost on some deep neural network architectu... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Variance Reduction With Sparse Gradients
### Paper Abstract
Variance reduction methods such as SVRG and SpiderBoost use a mixture of large and small batch gradients to r... | |
SyxZOsA9tX | ICLR.cc/2019/Conference | 2019 | Accelerated Value Iteration via Anderson Mixing | ["Yujun Li", "Chengzhuo Ni", "Guangzeng Xie", "Wenhao Yang", "Shuchang Zhou", "Zhihua Zhang"] | Acceleration for reinforcement learning methods is an important and challenging theme. We introduce the Anderson acceleration technique into the value iteration, developing an accelerated value iteration algorithm that we call Anderson Accelerated Value Iteration (A2VI). We further apply our method to the Deep Q-learni... | ["Reinforcement Learning"] | ABSTRACTAcceleration for reinforcement learning methods is an important and challenging theme.We introduce the Anderson acceleration technique into the value iteration, developing anaccelerated value iteration algorithm that we call Anderson Accelerated Value Iteration(A2VI). We further apply our method to the Deep Q-l... | Syg7ehHc3X | Extension to the Approximate DP case needed | 4: Ok but not good enough - rejection | This paper introduces the "Anderson mixing" ideas from the broader literature on general fixed-point problems to the specific problem of finding the fixed-point to the Bellman optimality equations for a Markov Decision Processes. The general idea is to summarizes the history of previous iterates (value functions in thi... | 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
Accelerated Value Iteration via Anderson Mixing
### Paper Abstract
Acceleration for reinforcement learning methods is an important and challenging theme. We introduce th... |
-qh0M9XWxnv | ICLR.cc/2021/Conference | 2021 | Analyzing the Expressive Power of Graph Neural Networks in a Spectral Perspective | ["Muhammet Balcilar", "Guillaume Renton", "Pierre H\u00e9roux", "Benoit Ga\u00fcz\u00e8re", "S\u00e9bastien Adam", "Paul Honeine"] | In the recent literature of Graph Neural Networks (GNN), the expressive power of models has been studied through their capability to distinguish if two given graphs are isomorphic or not. Since the graph isomorphism problem is NP-intermediate, and Weisfeiler-Lehman (WL) test can give sufficient but not enough evidence ... | ["Graph Neural Networks", "Spectral Graph Filter", "Spectral Analysis"] | ABSTRACTIn the recent literature of Graph Neural Networks (GNN), the expressive power ofmodels has been studied through their capability to distinguish if two given graphsare isomorphic or not. Since the graph isomorphism problem is NP-intermediate,and Weisfeiler-Lehman (WL) test can give sufficient but not enough evid... | dmm11HAMPK | An interesting perspective for analyzing Graph Neural Network | 8: Top 50% of accepted papers, clear accept | ==== Summary ====
This work studies the performance of different Graph Neural Network (GNN) from a spectral perspective. In particular, it shows the kernel for all kinds of proposed GNN models can be expressed in a general form with a specific frequency response definition, which indicates the spectral property (spect... | 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
Analyzing the Expressive Power of Graph Neural Networks in a Spectral Perspective
### Paper Abstract
In the recent literature of Graph Neural Networks (GNN), the express... |
r1e74a4twH | ICLR.cc/2020/Conference | 2020 | CZ-GEM: A FRAMEWORK FOR DISENTANGLED REPRESENTATION LEARNING | ["Akash Srivastava", "Yamini Bansal", "Yukun Ding", "Bernhard Egger", "Prasanna Sattigeri", "Josh Tenenbaum", "David D. Cox", "Dan Gutfreund"] | Learning disentangled representations of data is one of the central themes in unsupervised learning in general and generative modelling in particular. In this work, we tackle a slightly more intricate scenario where the observations are generated from a conditional distribution of some known control variate and some... | ["disentangled representation learning", "gan", "generative model", "simulator"] | ABSTRACTLearning disentangled representations of data is one of the central themes inunsupervised learning in general and generative modeling in particular. In thiswork, we tackle a slightly more intricate scenario where the observations aregenerated from a conditional distribution of some known control variate and som... | ryeEcSeaYr | Official Blind Review #3 | 1: Reject | This paper proposes a hybrid technique for rendering “control-variate” and class-conditional image in two steps, first by generating an approximate rendering of the image (“Y”) conditional on the control variate and then filling in the details with a conditional GAN dependent on a latent noise variable Z (although I no... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
CZ-GEM: A FRAMEWORK FOR DISENTANGLED REPRESENTATION LEARNING
### Paper Abstract
Learning disentangled representations of data is one of the central themes in unsuper... | |
SkpSlKIel | ICLR.cc/2017/conference | 2017 | Why Deep Neural Networks for Function Approximation? | ["Shiyu Liang", "R. Srikant"] | Recently there has been much interest in understanding why deep neural networks are preferred to shallow networks. We show that, for a large class of piecewise smooth functions, the number of neurons needed by a shallow network to approximate a function is exponentially larger than the corresponding number of neurons n... | ["neurons", "deep neural networks", "function approximation", "networks", "depth", "results", "much interest", "large class", "piecewise smooth functions", "number"] | ABSTRACTRecently there has been much interest in understanding why deep neural networksare preferred to shallow networks. We show that, for a large class of piecewisesmooth functions, the number of neurons needed by a shallow network to ap-proximate a function is exponentially larger than the corresponding number ofneu... | rkrLAnRXl | 7: Good paper, accept | This paper shows:
1. Easy, constructive proofs to derive e-error upper-bounds on neural networks with O(log 1/e) layers and O(log 1/e) ReLU units.
2. Extensions of the previous results to more general function classes, such as smooth or vector-valued functions.
3. Lower bounds on the neural network size, as a fu... | 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
Why Deep Neural Networks for Function Approximation?
### Paper Abstract
Recently there has been much interest in understanding why deep neural networks are preferred to ... | |
BkxAUjRqY7 | ICLR.cc/2019/Conference | 2019 | An Information-Theoretic Metric of Transferability for Task Transfer Learning | ["Yajie Bao", "Yang Li", "Shao-Lun Huang", "Lin Zhang", "Amir R. Zamir", "Leonidas J. Guibas"] | An important question in task transfer learning is to determine task transferability, i.e. given a common input domain, estimating to what extent representations learned from a source task can help in learning a target task. Typically, transferability is either measured experimentally or inferred through task relatedne... | ["transfer learning", "task transfer learning", "H-score", "transferability"] | ABSTRACTAn important question in task transfer learning is to determine task transferabil-ity, i.e. given a common input domain, estimating to what extent representationslearned from a source task can help in learning a target task. Typically, transferabil-ity is either measured experimentally or inferred through task ... | BklF6M1c2X | new measure of transferability | 6: Marginally above acceptance threshold | The paper proposes the H score H(f), a quantity that measure the goodness of feature f(x) for predicting some target y. This heavily builds on Makur et al. (2015) who introduce information vectors, the error exponent, and the DTM matrix. The paper connects H(f) with these quantities to justify the proposal (e.g., H(f) ... | 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
An Information-Theoretic Metric of Transferability for Task Transfer Learning
### Paper Abstract
An important question in task transfer learning is to determine task tra... |
Qr0aRliE_Hb | ICLR.cc/2021/Conference | 2021 | Simple Augmentation Goes a Long Way: ADRL for DNN Quantization | ["Lin Ning", "Guoyang Chen", "Weifeng Zhang", "Xipeng Shen"] | Mixed precision quantization improves DNN performance by assigning different layers with different bit-width values. Searching for the optimal bit-width for each layer, however, remains a challenge. Deep Reinforcement Learning (DRL) shows some recent promise. It however suffers instability due to function approximation... | ["Reinforcement Learning", "Quantization", "mixed precision", "augmented deep reinforcement learning", "DNN"] | ABSTRACTMixed precision quantization improves DNN performance by assigning differentlayers with different bit-width values. Searching for the optimal bit-width for eachlayer, however, remains a challenge. Deep Reinforcement Learning (DRL) showssome recent promise. It however suffers instability due to function approxim... | SppWcOiYOAX | It modifies a step in DRL based method that quantizes each networks with different bit numbers. A good engineering trick. | 6: Marginally above acceptance threshold | This work attempts to improve DRL based mixed-precision network compression. Compared to the standard action-critic method, he proposed method inserts an additional candidate evaluation step on k proposed actions as shown in Algorithm 1. As the paper title suggests, the proposed augmentation is “simple” yet practically... | 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
Simple Augmentation Goes a Long Way: ADRL for DNN Quantization
### Paper Abstract
Mixed precision quantization improves DNN performance by assigning different layers wit... |
48goXfYCVFX | ICLR.cc/2021/Conference | 2021 | Interpretable Relational Representations for Food Ingredient Recommendation Systems | ["Kana Maruyama", "Michael Spranger"] | Supporting chefs with ingredient recommender systems to create new recipes is challenging, as good ingredient combinations depend on many factors like taste, smell, cuisine style, texture among others. There have been few attempts to address these issues using machine learning. Importantly, useful models do obvious... | ["Metric Learning", "Gastronomy", "Memory Network", "Knowledge Graph", "Interpretable"] | ABSTRACTSupporting chefs with ingredient recommender systems to create new recipes ischallenging, as good ingredient combinations depend on many factors like taste,smell, cuisine style, texture among others. There have been few attempts to ad-dress these issues using machine learning. Useful Machine Learning models doo... | wx_asc3GUk2 | Overall this paper addresses an interesting problem. However, the paper is poorly written and proposed approaches are not novel. | 3: Clear rejection | Reject.
Summary
This paper works on the problem of create new recipes. It uses recommendation approaches on it. The paper use two approaches for explainability and conducted experiments on two real-world datasets.
Stregthens
1. The explainability in recommendation is an important research problem.
2. The recommendatio... | 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
Interpretable Relational Representations for Food Ingredient Recommendation Systems
### Paper Abstract
Supporting chefs with ingredient recommender systems to create new... |
H113pWZRb | ICLR.cc/2018/Conference | 2018 | Topology Adaptive Graph Convolutional Networks | ["Jian Du", "Shanghang Zhang", "Guanhang Wu", "Jos\u00e9 M. F. Moura", "Soummya Kar"] | Convolution acts as a local feature extractor in convolutional neural networks (CNNs). However, the convolution operation is not applicable when the input data is supported on an irregular graph such as with social networks, citation networks, or knowledge graphs. This paper proposes the topology adaptive graph convolu... | ["graph convolutional neural networks", "graph-structured data", "semi-classification"] | ABSTRACTConvolution acts as a local feature extractor in convolutional neural networks (C-NNs). However, the convolution operation is not applicable when the input data issupported on an irregular graph such as with social networks, citation networks, orknowledge graphs. This paper proposes the topology adaptive graph ... | ry6GiiKlz | Reasonable technical contribution, solid evaluation, well written | 6: Marginally above acceptance threshold | The authors propose a new CNN approach to graph classification that generalizes previous work. Instead of considering the direct neighborhood of a vertex in the convolution step, a filter based on outgoing walks of increasing length is proposed. This incorporates information from more distant vertices in one propagatio... | 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
Topology Adaptive Graph Convolutional Networks
### Paper Abstract
Convolution acts as a local feature extractor in convolutional neural networks (CNNs). However, the co... |
ByJHuTgA- | ICLR.cc/2018/Conference | 2018 | On the State of the Art of Evaluation in Neural Language Models | ["G\u00e1bor Melis", "Chris Dyer", "Phil Blunsom"] | Ongoing innovations in recurrent neural network architectures have provided a steady influx of apparently state-of-the-art results on language modelling benchmarks. However, these have been evaluated using differing codebases and limited computational resources, which represent uncontrolled sources of experimental vari... | ["rnn", "language modelling"] | ABSTRACTOngoing innovations in recurrent neural network architectures have provided asteady influx of apparently state-of-the-art results on language modelling bench-marks. However, these have been evaluated using differing codebases and limitedcomputational resources, which represent uncontrolled sources of experiment... | rJTcBCtxG | With extensive tuning, LSTM beats other new models | 5: Marginally below acceptance threshold | The authors did extensive tuning of the parameters for several recurrent neural architectures. The results are interesting. However the corpus the authors choose are quite small, the variance of the estimate will be quite high, I suspect whether the same conclusions could be drawn.
It would be more convincing if there... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the State of the Art of Evaluation in Neural Language Models
### Paper Abstract
Ongoing innovations in recurrent neural network architectures have provided a steady i... |
SJgn464tPB | ICLR.cc/2020/Conference | 2020 | Stabilizing Off-Policy Reinforcement Learning with Conservative Policy Gradients | ["Chen Tessler", "Nadav Merlis", "Shie Mannor"] | In recent years, advances in deep learning have enabled the application of reinforcement learning algorithms in complex domains. However, they lack the theoretical guarantees which are present in the tabular setting and suffer from many stability and reproducibility problems \citep{henderson2018deep}. In this work, we ... | ["Deep Reinforcement Learning", "Variance Reduction", "Policy Gradient"] | ABSTRACTIn recent years, advances in deep learning have enabled the application of re-inforcement learning algorithms in complex domains. However, they lack thetheoretical guarantees which are present in the tabular setting and suffer frommany stability and reproducibility problems [10]. In this work, we suggest a sim-... | B1ey8nZ1cB | Official Blind Review #1 | 1: Reject | This paper proposes a simple method for stabilizing the off-policy deep reinforcement learning algorithm, which updates the target network only when the online network performs better than the target network in order to ensure the stability guarantees. More specifically, at every T time steps, they execute both the onl... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Stabilizing Off-Policy Reinforcement Learning with Conservative Policy Gradients
### Paper Abstract
In recent years, advances in deep learning have enabled the applicati... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.