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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
qUzxZj13RWY | ICLR.cc/2021/Conference | 2021 | AdaS: Adaptive Scheduling of Stochastic Gradients | ["Mahdi S. Hosseini", "Konstantinos N Plataniotis"] | The choice of learning rate has been explored in many stochastic optimization frameworks to adaptively tune the step-size of gradients in iterative training of deep neural networks. While adaptive optimizers (e.g. AdaM, AdaGrad, RMSProp, AdaBound) offer fast convergence, they exhibit poor generalization characteristics... | ["Adaptive Stochastic Optimization", "Deep Convolution Neural Network", "Low-Rank Factorization"] | ABSTRACTThe choice of learning rate has been explored in many stochastic optimizationframeworks to adaptively tune the step-size of gradients in iterative training ofdeep neural networks. While adaptive optimizers (e.g. Adam, AdaGrad, RM-SProp, AdaBound) offer fast convergence, they exhibit poor generalization char-act... | 3HCDAO8ObLv | Interesting theories but limited applicability. | 4: Ok but not good enough - rejection | # Summary
This paper proposes Adaptive Scheduling (AdaS) for momentum SGD. The algorithm operates in roughly three stages:
1. Optimize the parameters for $K$ timesteps using standard momentum.
2. For each convolutional block, perform low-rank factorization of the block to compute the average "knowledge gain".
3. Deca... | 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
AdaS: Adaptive Scheduling of Stochastic Gradients
### Paper Abstract
The choice of learning rate has been explored in many stochastic optimization frameworks to adaptive... |
_sCOYXNwaI | ICLR.cc/2021/Conference | 2021 | Diffeomorphic Template Transformers | ["Tycho F.A. van der Ouderaa", "Ivana Isgum", "Wouter B. Veldhuis", "Bob D. De Vos", "Pim Moeskops"] | In this paper we propose a spatial transformer network where the spatial transformations are limited to the group of diffeomorphisms. Diffeomorphic transformations are a kind of homeomorphism, which by definition preserve topology, a compelling property in certain applications.
We apply this diffemorphic spatial transf... | ["output", "prior shape", "spatial transformer network", "spatial transformations", "group", "diffeomorphisms", "diffeomorphic transformations", "kind", "homeomorphism"] | ABSTRACTIn this paper we propose a spatial transformer network where the spatial transforma-tions are limited to the group of diffeomorphisms. Diffeomorphic transformationsare a kind of homeomorphism, which by definition preserve topology, a compellingproperty in certain applications. We apply this diffemorphic spatial... | KWotcs72mb | Incremental methodological contribution and performance improvement | 5: Marginally below acceptance threshold | This submission proposes a diffeomorphic spatial transform network, which considers the specific transformation, i.e., diffeomorphism, in the data. The research topic is quite interesting, but the submission is at the preliminary stage and needs more work before publishing.
1) The writing of the paper could be improve... | 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
Diffeomorphic Template Transformers
### Paper Abstract
In this paper we propose a spatial transformer network where the spatial transformations are limited to the group ... |
aaI4jKANEH4 | ICML.cc/2020/Workshop/SAS | 2020 | Revisiting Representation Learning for Singing Voice Separation with Sinkhorn Distances | ["Stylianos Ioannis Mimilakis", "Konstantinos Drossos", "Gerald Schuller"] | In this work we present a method for unsupervised learning of audio representations, focused on the task of singing voice separation. We
build upon a previously proposed method for learning representations of time-domain music signals with a re-parameterized denoising autoencoder, extending it by using the family of Si... | ["Representation learning", "unsupervised learning", "music source separation"] | Revisiting Representation Learning for Singing Voice Separation withSinkhorn DistancesStylianos I. Mimilakis* 1Konstantinos Drossos* 2Gerald Schuller3AbstractIn this work we present a method for unsupervisedlearning of audio representations, focused on thetask of singing voice separation. We build upona previously prop... | YQugV-puMT | Review | 6: Marginally above acceptance threshold | The authors extend a previous work based on denoising-autoencoders to a new specific loss with the overall goal to separate singing voices from music. While the idea is only incremental with respect to previous works, results are interesting. More precisely, it is important to explore models that are able to directly d... | 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
Revisiting Representation Learning for Singing Voice Separation with Sinkhorn Distances
### Paper Abstract
In this work we present a method for unsupervised learning of ... |
SJ3rcZcxl | ICLR.cc/2017/conference | 2017 | Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic | ["Shixiang Gu", "Timothy Lillicrap", "Zoubin Ghahramani", "Richard E. Turner", "Sergey Levine"] | Model-free deep reinforcement learning (RL) methods have been successful in a wide variety of simulated domains. However, a major obstacle facing deep RL in the real world is their high sample complexity. Batch policy gradient methods offer stable learning, but at the cost of high variance, which often requires large b... | ["Deep learning", "Reinforcement Learning"] | ABSTRACTModel-free deep reinforcement learning (RL) methods have been successful in awide variety of simulated domains. However, a major obstacle facing deep RLin the real world is their high sample complexity. Batch policy gradient methodsoffer stable learning, but at the cost of high variance, which often requires la... | rJVKwJMSe | An interesting approach for using control variables to improve stability of deep RL control | 7: Good paper, accept | The paper proposes using a first-order Taylor expansion as a control variate in policy gradient-style methods. Empirical results in dynamical control tasks suggest that this algorithm reduces the sample complexity, while the theoretical results presented suggest the algorithm is unbiased but of lower variance.
The use ... | 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
Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic
### Paper Abstract
Model-free deep reinforcement learning (RL) methods have been successful in a wide ... |
HKbgd3zmh0t | ML_Reproducibility_Challenge/2021/Fall | 2021 | [Re] Learning to count everything | ["Matija Ter\u0161ek", "Domen Vre\u0161", "Ma\u0161a Kljun"] | Scope of Reproducibility
The core finding of the paper is a novel architecture FamNet for handling the few-shot counting task. We examine its implementation in the provided code on GitHub and compare it to the theory in the original paper. The authors also introduce a data set with 147 visual categories FSC-147, which ... | ["few-shot learning", "few-shot counting", "CNN", "counting data set"] | [Re] Learning to count everythingAnonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2The core finding of the paper is a novel architecture FamNet for handling the few-shot counting task. We examine its 3implementation in the provided code on GitHub and compare it to the theory i... | B0lxgZR0e-9 | Confirms Limited Reproducibility | 8: Top 50% of accepted papers, clear accept | The review claims that the training could not be reproduced because of limited information in the paper, and supplementary information provided by the authors. This is valid.
Otherwise, the results of the paper seem to have been reproduced approximately, because the proposed model analogously outperforms the baseline.... | 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
[Re] Learning to count everything
### Paper Abstract
Scope of Reproducibility The core finding of the paper is a novel architecture FamNet for handling the few-shot coun... |
SJMO2iCct7 | ICLR.cc/2019/Conference | 2019 | A NOVEL VARIATIONAL FAMILY FOR HIDDEN NON-LINEAR MARKOV MODELS | ["Daniel Hernandez Diaz", "Antonio Khalil Moretti", "Ziqiang Wei", "Shreya Saxena", "John Cunningham", "Liam Paninski"] | Latent variable models have been widely applied for the analysis and visualization of large datasets. In the case of sequential data, closed-form inference is possible when the transition and observation functions are linear. However, approximate inference techniques are usually necessary when dealing with nonlinear ev... | ["variational inference", "time series", "nonlinear dynamics", "neuroscience"] | ABSTRACTLatent variable models have been widely applied for the analysis and visualizationof large datasets. In the case of sequential data, closed-form inference is possiblewhen the transition and observation functions are linear. However, approximateinference techniques are usually necessary when dealing with nonline... | Bygvx6sc2X | Excellent method and results but need more comparisons and better writing | 8: Top 50% of accepted papers, clear accept | I'll start with a disclaimer: I have reviewed the NIPS 2019 submission of this paper which was eventually rejected. Compared to the NIPS version, this manuscript had significantly improved in its completeness. However, the writing still can be improved for rigor, consistency, typos, completeness, and readability.
Auth... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A NOVEL VARIATIONAL FAMILY FOR HIDDEN NON-LINEAR MARKOV MODELS
### Paper Abstract
Latent variable models have been widely applied for the analysis and visualization of l... |
HkgSk2A9Y7 | ICLR.cc/2019/Conference | 2019 | Stochastic Gradient Push for Distributed Deep Learning | ["Mahmoud Assran", "Nicolas Loizou", "Nicolas Ballas", "Mike Rabbat"] | Large mini-batch parallel SGD is commonly used for distributed training of deep networks. Approaches that use tightly-coupled exact distributed averaging based on AllReduce are sensitive to slow nodes and high-latency communication. In this work we show the applicability of Stochastic Gradient Push (SGP) for distribute... | ["optimization", "distributed", "large scale", "deep learning"] | ABSTRACTLarge mini-batch parallel SGD is commonly used for distributed training of deepnetworks. Approaches that use tightly-coupled exact distributed averaging basedon AllReduce are sensitive to slow nodes and high-latency communication. Inthis work we show the applicability of Stochastic Gradient Push (SGP) for dis-t... | rkxdTUvqhX | An interesting work that needs some clarification | 6: Marginally above acceptance threshold | This paper demonstrates the benefit of stochastic gradient push (SGP) in the distributed training of neural networks. The contributions are twofold: (1) the paper proves the convergence of SGP for nonconvex smooth functions and gives a reasonable estimation of the convergence rate; (2) the paper did many experiments an... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Stochastic Gradient Push for Distributed Deep Learning
### Paper Abstract
Large mini-batch parallel SGD is commonly used for distributed training of deep networks. Appro... |
H1lC8o0cKX | ICLR.cc/2019/Conference | 2019 | Unsupervised Emergence of Spatial Structure from Sensorimotor Prediction | ["Alban Laflaqui\u00e8re", "Michael Garcia Ortiz"] | Despite its omnipresence in robotics application, the nature of spatial knowledge and the mechanisms that underlie its emergence in autonomous agents are still poorly understood. Recent theoretical work suggests that the concept of space can be grounded by capturing invariants induced by the structure of space in an ag... | ["spatial perception", "grounding", "sensorimotor prediction", "unsupervised learning", "representation learning"] | ABSTRACTDespite its omnipresence in robotics application, the nature of spatial knowledgeand the mechanisms that underlie its emergence in autonomous agents are stillpoorly understood. Recent theoretical work suggests that the concept of spacecan be grounded by capturing invariants induced by the structure of space in ... | BkxVXXd5hQ | Interesting topic, contribution questionable, better write-up required | 6: Marginally above acceptance threshold | The paper proposes that agents can extract the spatial structure of the world if consistently explored solely from sensorimotor predictions.
The topic of the paper is relevant and fits the conference. However, I have doubts about the generality of the claims made.
Strong aspects:
Abstract makes the reader curious and... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Unsupervised Emergence of Spatial Structure from Sensorimotor Prediction
### Paper Abstract
Despite its omnipresence in robotics application, the nature of spatial knowl... |
oLltLS5F9R | ICLR.cc/2021/Conference | 2021 | Learning Graph Normalization for Graph Neural Networks | ["Yihao Chen", "Xin Tang", "Xianbiao Qi", "Chun-Guang Li", "Rong Xiao"] | Graph Neural Networks (GNNs) have emerged as a useful paradigm to process graph-structured data. Usually, GNNs are stacked to multiple layers and the node representations in each layer are computed through propagating and aggre- gating the neighboring node features with respect to the graph. To effectively train a GNN ... | ["Graph Neural Network", "Normalization", "Graph Normalization"] | xdZmEKlllYc | Attention-based integration of different batch/layer normalization approaches in graph neural networks with an articulated empirical benchmarking | 4: Ok but not good enough - rejection | The paper provides a unified view over different layer/batch normalization mechanisms for graph neural networks and formalizes an attention-driven approach to adaptively integrate the different forms of normalizations. An extensive empirical analysis is provided, spanning graph classification, link prediction and node ... | 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 Graph Normalization for Graph Neural Networks
### Paper Abstract
Graph Neural Networks (GNNs) have emerged as a useful paradigm to process graph-structured data... | |
SJZ2Mf-0- | ICLR.cc/2018/Conference | 2018 | Adaptive Memory Networks | ["Daniel Li", "Asim Kadav"] | Real-world Question Answering (QA) tasks consist of thousands of words that often represent many facts and entities. Existing models based on LSTMs require a large number of parameters to support external memory and do not generalize well for long sequence inputs. Memory networks attempt to address these limitations by... | ["Memory Networks", "Dynamic Networks", "Faster Inference", "Reasoning", "QA"] | ABSTRACTReal-world Question Answering (QA) tasks often consist of thousands of wordsthat represent many facts and entities. Existing models based on LSTMs requirea large number of parameters to support external memory and do not generalizeefficiently for long sequence inputs. Memory networks address these limitationsby... | By5ZMrqxG | Review | 4: Ok but not good enough - rejection | The authors propose a model for QA that given a question and a story adaptively determines the number of entity groups (banks). The paper is rather hard to follow as many task specific terms are not explained. For instance, it would benefit the paper if the authors introduced the definitions of a bank and a story. Thi... | 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
Adaptive Memory Networks
### Paper Abstract
Real-world Question Answering (QA) tasks consist of thousands of words that often represent many facts and entities. Existing... |
SJx94o0qYX | ICLR.cc/2019/Conference | 2019 | Precision Highway for Ultra Low-precision Quantization | ["Eunhyeok Park", "Dongyoung Kim", "Sungjoo Yoo", "Peter Vajda"] | Quantization of a neural network has an inherent problem called accumulated quantization error, which is the key obstacle towards ultra-low precision, e.g., 2- or 3-bit precision. To resolve this problem, we propose precision highway, which forms an end-to-end high-precision information flow while performing the ultra-... | ["neural network", "quantization", "optimization", "low-precision", "convolutional network", "recurrent network"] | ABSTRACTNeural network quantization has an inherent problem called accumulated quan-tization error , which is the key obstacle towards ultra-low precision, e.g., 2- or3-bit precision. To resolve this problem, we propose precision highway , whichforms an end-to-end high-precision information flow while performing the ul... | BJeYeCw-aQ | An OK paper but need more evaluation. | 6: Marginally above acceptance threshold | This paper investigates the problem of neural network quantization. The main idea is to employ an end-to-end precision highway to reduce the accumulated quantization error and meanwhile enable ultra-low precision in deep neural networks. The experimental results on the 3- and 2-bit quantizations of ResNet-18/50 and 2-... | 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
Precision Highway for Ultra Low-precision Quantization
### Paper Abstract
Quantization of a neural network has an inherent problem called accumulated quantization error,... |
TTUVg6vkNjK | ICLR.cc/2021/Conference | 2021 | RODE: Learning Roles to Decompose Multi-Agent Tasks | ["Tonghan Wang", "Tarun Gupta", "Anuj Mahajan", "Bei Peng", "Shimon Whiteson", "Chongjie Zhang"] | Role-based learning holds the promise of achieving scalable multi-agent learning by decomposing complex tasks using roles. However, it is largely unclear how to efficiently discover such a set of roles. To solve this problem, we propose to first decompose joint action spaces into restricted role action spaces by cluste... | ["Multi-Agent Reinforcement Learning", "Role-Based Learning", "Hierarchical Multi-Agent Learning", "Multi-Agent Transfer Learning"] | ABSTRACTRole-based learning holds the promise of achieving scalable multi-agent learningby decomposing complex tasks using roles. However, it is largely unclear howto efficiently discover such a set of roles. To solve this problem, we propose tofirst decompose joint action spaces into restricted role action spaces by c... | 5g1ohWEpSDe | Official Blind Review | 7: Good paper, accept | This paper introduces a bi-level hierarchical framework for achieving scalable multi-agent learning. In this framework, the high-level policy (role selector) coordinates role assignments in a smaller role space and at a lower temporal resolution. And the low-level policies (role policies) explore strategies in reduced ... | 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
RODE: Learning Roles to Decompose Multi-Agent Tasks
### Paper Abstract
Role-based learning holds the promise of achieving scalable multi-agent learning by decomposing co... |
S1xXiREKDB | ICLR.cc/2020/Conference | 2020 | Adversarial training with perturbation generator networks | ["Hyeungill Lee", "Sungyeob Han", "Jungwoo Lee"] | Despite the remarkable development of recent deep learning techniques, neural networks are still vulnerable to adversarial attacks, i.e., methods that fool the neural networks with perturbations that are too small for human eyes to perceive. Many adversarial training methods were introduced as to solve this problem, us... | ["Adversarial training", "Generative model", "Adaptive perturbation generator", "Robust optimization"] | ABSTRACTDespite the remarkable development of recent deep learning techniques, neuralnetworks are still vulnerable to adversarial attacks, i.e., methods that fool the neu-ral networks with perturbations that are too small for human eyes to perceive.Many adversarial training methods were introduced as to solve this prob... | HklZG5TYFB | Official Blind Review #1 | 3: Weak Reject | This paper introduces a new adversarial training approach, where a generator is used to generate the most challenging adversarial examples and the classifier is trained to correctly classify the generated adversarial examples. In this way, the robustness of the classifier is expected to be improved. For the generator, ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Adversarial training with perturbation generator networks
### Paper Abstract
Despite the remarkable development of recent deep learning techniques, neural networks are s... | |
sr89orrDo-o | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | AnimationPak: Packing Elements with Scripted Animations | ["Reza Adhitya Saputra", "Craig S. Kaplan", "Paul Asente"] | We present AnimationPak, a technique to
create animated packings by arranging animated two-dimensional
elements inside a static container.
We represent animated elements in a three-dimensional spacetime
domain, and view the animated packing problem as a three-dimensional
packing in that domain. Every element is repre... | ["Element Distribution", "Packing", "Physical Simulation", "Shape Deformation", "Animation", "Spacetime domain"] | ABSTRACTWe present AnimationPak, a technique to create animated packingsby arranging animated two-dimensional elements inside a staticcontainer. We represent animated elements in a three-dimensionalspacetime domain, and view the animated packing problem as athree-dimensional packing in that domain. Every element is rep... | mQW2MK_auX_ | nice work, should accept | 7: Good paper, accept | This paper extends the work of Saputra et al (GI 2018, TVCG 2019) on static packing to animated packings where each element potentially has some sort of scripted behaviour. The method creates a static optimization over a timespace cube, which then can be interpreted as an 2D animation by taking cross-sections perpendic... | 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
AnimationPak: Packing Elements with Scripted Animations
### Paper Abstract
We present AnimationPak, a technique to create animated packings by arranging animated two-dim... |
HJxB5sRcFQ | ICLR.cc/2019/Conference | 2019 | LayoutGAN: Generating Graphic Layouts with Wireframe Discriminators | ["Jianan Li", "Jimei Yang", "Aaron Hertzmann", "Jianming Zhang", "Tingfa Xu"] | Layout is important for graphic design and scene generation. We propose a novel Generative Adversarial Network, called LayoutGAN, that synthesizes layouts by modeling geometric relations of different types of 2D elements. The generator of LayoutGAN takes as input a set of randomly-placed 2D graphic elements and uses se... | ["layoutgan", "wireframe discriminators", "layouts", "generating graphic layouts", "graphic layouts", "important", "graphic design", "scene generation", "geometric relations"] | ABSTRACTLayout is important for graphic design and scene generation. We propose a novelGenerative Adversarial Network, called LayoutGAN, that synthesizes layouts bymodeling geometric relations of different types of 2D elements. The generatorof LayoutGAN takes as input a set of randomly-placed 2D graphic elements anduse... | rJxm6KDi2X | Interesting application of GAN on Layout rendering | 7: Good paper, accept | Summary:
The paper proposed to use GAN to synthesize graphical layouts. The generator takes a random input and generates class probabilities and geometric parameters based on a self-attention module. The discriminator is based on a differentiable wireframe rendering component (proposed by the paper) to allow back propa... | 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
LayoutGAN: Generating Graphic Layouts with Wireframe Discriminators
### Paper Abstract
Layout is important for graphic design and scene generation. We propose a novel Ge... |
HyKZyYlRZ | ICLR.cc/2018/Conference | 2018 | Large Scale Multi-Domain Multi-Task Learning with MultiModel | ["Lukasz Kaiser", "Aidan N. Gomez", "Noam Shazeer", "Ashish Vaswani", "Niki Parmar", "Llion Jones", "Jakob Uszkoreit"] | Deep learning yields great results across many fields,
from speech recognition, image classification, to translation.
But for each problem, getting a deep model to work well involves
research into the architecture and a long period of tuning.
We present a single model that yields good results on a number
of problems s... | ["multi-task learning", "transfer learning"] | ABSTRACTDeep learning yields great results across many fields, from speech recognition,image classification, to translation. But for each problem, getting a deep modelto work well involves research into the architecture and a long period of tuning.We present a single model that yields good results on a number of proble... | rJ6r_eqlf | Review | 6: Marginally above acceptance threshold | The paper presents a multi-task, multi-domain model based on deep neural networks. The proposed model is able to take inputs from various domains (image, text, speech) and solves multiple tasks, such as image captioning, machine translation or speech recognition. The proposed model is composed of several features learn... | 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
Large Scale Multi-Domain Multi-Task Learning with MultiModel
### Paper Abstract
Deep learning yields great results across many fields, from speech recognition, image cla... |
Skxd6gSYDS | ICLR.cc/2020/Conference | 2020 | Query-efficient Meta Attack to Deep Neural Networks | ["Jiawei Du", "Hu Zhang", "Joey Tianyi Zhou", "Yi Yang", "Jiashi Feng"] | Black-box attack methods aim to infer suitable attack patterns to targeted DNN models by only using output feedback of the models and the corresponding input queries. However, due to lack of prior and inefficiency in leveraging the query and feedback information, existing methods are mostly query-intensive for obtainin... | ["Adversarial attack", "Meta learning"] | ABSTRACTBlack-box attack methods aim to infer suitable attack patterns to targeted DNNmodels by only using output feedback of the models and the corresponding inputqueries. However, due to lack of prior and inefficiency in leveraging the query andfeedback information, existing methods are mostly query-intensive for obt... | SyxMU8GpYr | Official Blind Review #3 | 6: Weak Accept | Summary:
The authors of this paper propose a novel approach for query-efficient black-box adversarial attacks by leveraging meta-learning to approximate the gradient of given clean images.
Paper strength:
1. The overall idea of this paper is interesting to improve the transferability of adversarial examples through ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Query-efficient Meta Attack to Deep Neural Networks
### Paper Abstract
Black-box attack methods aim to infer suitable attack patterns to targeted DNN models by only usin... | |
WqXAKcwfZtI | ICLR.cc/2021/Conference | 2021 | f-Domain-Adversarial Learning: Theory and Algorithms for Unsupervised Domain Adaptation with Neural Networks | ["David Acuna", "Guojun Zhang", "Marc T Law", "Sanja Fidler"] | The problem of unsupervised domain adaptation arises in a variety of practical applications where the distribution of the training samples differs from those used at test time. The existing theory of domain adaptation derived generalization bounds based on divergence measures that are hard to optimize in practice. This... | ["Domain Adaptation", "Adversarial Learning", "Transfer Learning", "Neural Networks", "Deep Learning"] | ABSTRACTThe problem of unsupervised domain adaptation arises in a variety of practical ap-plications where the distribution of the training samples differs from those used attest time. The existing theory of domain adaptation derived generalization boundsbased on divergence measures that are hard to optimize in practic... | XqpmtgiABr | Experiments and writing need to be significantly improved. | 5: Marginally below acceptance threshold | Update after reading authors' response.
The authors didn't address my question on the statistical significance of their results compared with baselines.
The authors can address this question by "perform a double-sided student-t test to verify whether the performance difference between your method and the baselines are... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
f-Domain-Adversarial Learning: Theory and Algorithms for Unsupervised Domain Adaptation with Neural Networks
### Paper Abstract
The problem of unsupervised domain adapta... |
KwgQn_Aws3_ | ICLR.cc/2021/Conference | 2021 | Interpretable Sequence Classification Via Prototype Trajectory | ["Dat Hong", "Stephen Baek", "Tong Wang"] | We propose a novel interpretable recurrent neural network (RNN) model, called ProtoryNet, in which we introduce a new concept of prototype trajectories. Motivated by the prototype theory in modern linguistics, ProtoryNet makes a prediction by finding the most similar prototype for each sentence in a text sequence and f... | ["interpretable", "RNN", "prototypes"] | ABSTRACTWe propose a novel interpretable recurrent neural network (RNN) model, calledProtoryNet, in which we introduce a new concept of prototype trajectories. Moti-vated by the prototype theory in modern linguistics, ProtoryNet makes a predictionby finding the most similar prototype for each sentence in a text sequenc... | C6c9EC6_jdM | A framework for text data that classifies that explains through prototypes' results | 7: Good paper, accept | This paper presents ProtoryNet, a framework for text data that classifies and explains the prototypes' results. The key concept, that is the novelty of the work, is that this framework is based on sentence prototypes, called prototype trajectory in the paper. In particular, instead of working at the entity of the tex... | 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 Sequence Classification Via Prototype Trajectory
### Paper Abstract
We propose a novel interpretable recurrent neural network (RNN) model, called ProtoryNe... |
OifRuTHyQU | ICLR.cc/2021/Conference | 2021 | Deep Manifold Computing and Visualization Using Elastic Locally Isometric Smoothness | ["Stan Z. Li", "Zelin Zang", "Lirong Wu"] | The ability to preserve local geometry of highly nonlinear manifolds in high dimensional spaces and properly unfold them into lower dimensional hyperplanes is the key to the success of manifold computing, nonlinear dimensionality reduction (NLDR) and visualization. This paper proposes a novel method, called elastic lo... | ["manifold learning", "dimensionality reduction", "visualization", "data generation"] | ABSTRACTThe ability to preserve local geometry of highly nonlinear manifolds in highdimensional spaces and properly unfold them into lower dimensional hyperplanesis the key to the success of manifold computing, nonlinear dimensionality reduction(NLDR) and visualization. This paper proposes a novel method, called elasti... | ei15WixVurD | What is the problem you are solving? | 4: Ok but not good enough - rejection | This is yet another NLDR paper. But I don't see any shining point here. It is a marginal modification of UMAP, parameterized by neural networks.
What is the problem you are solving? You must be aware of tens of existing NLDR methods. Have you identified a severe problem among them and does your method solve the proble... | 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 Manifold Computing and Visualization Using Elastic Locally Isometric Smoothness
### Paper Abstract
The ability to preserve local geometry of highly nonlinear manifo... |
HVxMCn4uP-c | ICLR.cc/2022/Workshop/DGM4HSD | 2022 | Semi-Discrete Normalizing Flows through Differentiable Voronoi Tessellation | ["Ricky T. Q. Chen", "Brandon Amos", "Maximilian Nickel"] | Mapping between discrete and continuous distributions is a difficult task and many have had to resort to approximate or heuristical approaches. We propose a tessellation-based approach that directly learns quantization boundaries on a continuous space, complete with exact likelihood evaluations. This is done through co... | ["normalizing", "discrete", "differentiable voronoi tessellation", "continuous distributions", "difficult task", "many", "heuristical approaches", "quantization boundaries", "continuous space"] | ABSTRACTMapping between discrete and continuous distributions is a difficult task andmany have had to resort to approximate or heuristical approaches. We proposea tessellation-based approach that directly learns quantization boundaries on acontinuous space, complete with exact likelihood evaluations. This is done throu... | SYWgypmXcMq | Very high quality paper for a workshop, clear accept | 8: Top 50% of accepted papers, clear accept | The authors propose to use Voronoi tesselations to cleanly and intuitively model discrete data using normalizing flow methods. This approach provides a more natural method for dequantization; a simpler, less data-dependent approach is typically used instead when modelling discrete data with normalizing flows. This work... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Semi-Discrete Normalizing Flows through Differentiable Voronoi Tessellation
### Paper Abstract
Mapping between discrete and continuous distributions is a difficult task ... |
jnRqf0CzBK | ICLR.cc/2021/Conference | 2021 | Hierarchical Probabilistic Model for Blind Source Separation via Legendre Transformation | ["Simon Luo", "Lamiae Azizi", "Mahito Sugiyama"] | We present a novel blind source separation (BSS) method, called information geometric blind source separation (IGBSS). Our formulation is based on the log-linear model equipped with a hierarchically structured sample space, which has theoretical guarantees to uniquely recover a set of source signals by minimizing the K... | ["blind source separation", "log-linear model", "energy-based model", "information geometry"] | ABSTRACTWe present a novel blind source separation (BSS) method, called informationgeometric blind source separation (IGBSS). Our formulation is based on the log-linear model equipped with a hierarchically structured sample space, which hastheoretical guarantees to uniquely recover a set of source signals by minimizing... | f-HH1jyx3hZ | An interesting use of information geometry in blind source separation | 4: Ok but not good enough - rejection | Additional comments: I have read the authors' response and the other reviews. While my initial concerns have been mostly addressed, there are still concerns from the other reviews and I have revised my score accordingly.
In addition, I am not completely satisfied with the authors' response concerning higher-order inte... | 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
Hierarchical Probabilistic Model for Blind Source Separation via Legendre Transformation
### Paper Abstract
We present a novel blind source separation (BSS) method, call... |
TyA5AyU_tSv | MIDL.io/2023/Short_Paper_Track | 2023 | Shape Equivariant Learning for Robust MRI Segmentation | ["Ainkaran Santhirasekaram", "Mathias Winkler", "Andrea G. Rockall", "Ben Glocker"] | The reliability of deep learning based segmentation models is essential to the safe translation of these models into clinical practise. Unfortunately, these models are sensitive to distributional shifts. This is particularly notable in MRI, where there is a large variation of acquisition protocols across different doma... | ["Shape Equivariance", "Robustness", "Segmentation", "MRI"] | Medical Imaging with Deep Learning – Under Review 2023 Short Paper – MIDL 2023 submissionShape Equivariant Learning for Robust MRI SegmentationAinkaran Santhirasekaram1a.santhirasekaram19@imperial.ac.ukMathias Winkler2m.winkler@imperial.ac.ukAndrea Rockall2a.rockall@imperial.ac.ukBen Glocker1b.glocker@imperial.ac.uk1De... | 2ArOBjKXTq | Shape equivariant learning to improve domain generalisability of segmentation | 7: Good paper, accept | The goal of this work is to improve the domain generalisability of a segmentation model by training it to learn texture invariant and shape equivariant features. The model is trained using multiple MRI contrasts (T2w and ADC map) to learn shape equivariant features by constraining the latent space to a finite set 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
Shape Equivariant Learning for Robust MRI Segmentation
### Paper Abstract
The reliability of deep learning based segmentation models is essential to the safe translation... |
Kz42iQirPJI | ICLR.cc/2021/Conference | 2021 | Towards Learning to Remember in Meta Learning of Sequential Domains | ["Zhenyi Wang", "Tiehang Duan", "Donglin Zhan", "Changyou Chen"] | Meta-learning has made rapid progress in past years, with recent extensions made to avoid catastrophic forgetting in the learning process, namely continual meta learning. It is desirable to generalize the meta learner’s ability to continuously learn in sequential domains, which is largely unexplored to-date. We found t... | ["Meta learning", "Continual Learning", "Sequential Domain Learning"] | ABSTRACTMeta-learning has made rapid progress in past years, with recent extensionsmade to avoid catastrophic forgetting in the learning process, namely contin-ual meta learning. It is desirable to generalize the meta learner’s ability tocontinuously learn in sequential domains , which is largely unexplored to-date.We ... | uous3Lj6VKZ | An interesting contribution that requires some clarifications | 6: Marginally above acceptance threshold | Summary
At the heart of this paper are two separate contributions. The first is a new online meta-learning problem setting where the meta-learner acts on a sequence of few-shot learning *domains*, as opposed to tasks within a single domain. The second is a method for meta-learning with this form of domain shift.
As f... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Learning to Remember in Meta Learning of Sequential Domains
### Paper Abstract
Meta-learning has made rapid progress in past years, with recent extensions made t... |
r1DPFCyA- | ICLR.cc/2018/Conference | 2018 | Discriminative k-shot learning using probabilistic models | ["Matthias Bauer", "Mateo Rojas-Carulla", "Jakub Bart\u0142omiej \u015awi\u0105tkowski", "Bernhard Sch\u00f6lkopf", "Richard E. Turner"] | This paper introduces a probabilistic framework for k-shot image classification. The goal is to generalise from an initial large-scale classification task to a separate task comprising new classes and small numbers of examples. The new approach not only leverages the feature-based representation learned by a neural ne... | ["discriminative k-shot learning", "probabilistic inference"] | ABSTRACTThis paper introduces a probabilistic framework for k-shot image classification.The goal is to generalise from an initial large-scale classification task to a separatetask comprising new classes and small numbers of examples. The new approach notonly leverages the feature-based representation learned by a neura... | SknsYOMZf | A simple probabilistic method to transfer knowledge from large datasets to smaller datasets (k-shot per class). | 5: Marginally below acceptance threshold | The authors introduce a probabilistic k-shot learning model based on previous training of a CNN on a large dataset. The weights of the softmax layer of the CNN are then used as the MAP solution in a concept learning scenario to put a prior over the soft-max weights of the classifier (logistic regression) for the datase... | 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
Discriminative k-shot learning using probabilistic models
### Paper Abstract
This paper introduces a probabilistic framework for k-shot image classification. The goal is... |
0oabwyZbOu | ICLR.cc/2021/Conference | 2021 | Mastering Atari with Discrete World Models | ["Danijar Hafner", "Timothy P Lillicrap", "Mohammad Norouzi", "Jimmy Ba"] | Intelligent agents need to generalize from past experience to achieve goals in complex environments. World models facilitate such generalization and allow learning behaviors from imagined outcomes to increase sample-efficiency. While learning world models from image inputs has recently become feasible for some tasks, m... | ["Atari", "world models", "model-based reinforcement learning", "reinforcement learning", "planning", "actor critic"] | ABSTRACTIntelligent agents need to generalize from past experience to achieve goals incomplex environments. World models facilitate such generalization and allowlearning behaviors from imagined outcomes to increase sample-efficiency. Whilelearning world models from image inputs has recently become feasible for sometask... | 0e3V9V8fpLF | In my opinion, not enough technical novelty to merit acceptance | 5: Marginally below acceptance threshold | This paper proposes DreamerV2, a set of modifications to the existing
model-based RL system Dreamer, and shows via an ablation study that
these modifications improve performance on the standard Atari
benchmark over Dreamer. It is shown that DreamerV2 performs well
compared to model-free RL methods as well, especially w... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Mastering Atari with Discrete World Models
### Paper Abstract
Intelligent agents need to generalize from past experience to achieve goals in complex environments. World ... |
mCLVeEpplNE | ICLR.cc/2021/Conference | 2021 | NBDT: Neural-Backed Decision Tree | ["Alvin Wan", "Lisa Dunlap", "Daniel Ho", "Jihan Yin", "Scott Lee", "Suzanne Petryk", "Sarah Adel Bargal", "Joseph E. Gonzalez"] | Machine learning applications such as finance and medicine demand accurate and justifiable predictions, barring most deep learning methods from use. In response, previous work combines decision trees with deep learning, yielding models that (1) sacrifice interpretability for accuracy or (2) sacrifice accuracy for inter... | ["explainability", "computer vision", "interpretability"] | ABSTRACTMachine learning applications such as finance and medicine demand accurate andjustifiable predictions, barring most deep learning methods from use. In response,previous work combines decision trees with deep learning, yielding models that(1) sacrifice interpretability for accuracy or (2) sacrifice accuracy for ... | mlTIpDngMw0 | Compelling interpretability methodological work but minor flaws in motivation, lack of discussion about practical limitations | 7: Good paper, accept | The authors did a fantastic job of answering questions, revising their manuscript in accord with reviewer feedback (Sec 3.4 title), and even adding new experimental results based on reviewer suggestions (mid-training hierarchy) and reflecting best practices in interpretability research. I was really impressed by their ... | 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
NBDT: Neural-Backed Decision Tree
### Paper Abstract
Machine learning applications such as finance and medicine demand accurate and justifiable predictions, barring most... |
VP_340AAQ- | ICLR.cc/2022/Track/Blog | 2022 | A visual exploration of Gaussian Processes and Infinite Neural Networks | ["Anonymous"] | Gaussian Processes generalize Gaussian distributions to functons as random variables (stochastic processes). As such, they are a powerful tool for regression and Bayesian inference. In the last years, they have received increased attention due to the interest in machine learning. An interesting connection between Gauss... | ["Infinite Neural Networks", "Gaussian Processes", "Bayesian Inference", "Machine Learning"] | B-4eO46wNZ5 | Not appropriate level | 4: Ok but not good enough - rejection | This blog post reads well as an introduction to Gaussian Processes and its connection to wide neural networks, as well as an introduction to Julia. However, the target audience seems to be advanced undergraduates or beginning PhD students, rather than at a research level, and I am concerned whether this is inappropriat... | 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A visual exploration of Gaussian Processes and Infinite Neural Networks
### Paper Abstract
Gaussian Processes generalize Gaussian distributions to functons as random var... | |
r1eh30NFwB | ICLR.cc/2020/Conference | 2020 | Variational Autoencoders with Normalizing Flow Decoders | ["Rogan Morrow", "Wei-Chen Chiu"] | Recently proposed normalizing flow models such as Glow (Kingma & Dhariwal, 2018) have been shown to be able to generate high quality, high dimensional images with relatively fast sampling speed. Due to the inherently restrictive design of architecture , however, it is necessary that their model are excessively deep in ... | ["model", "glow", "order", "variational autoencoders", "flow decoders", "flow models", "able", "high quality", "high dimensional images"] | ABSTRACTRecently proposed normalizing flow models such as Glow (Kingma & Dhariwal,2018) have been shown to be able to generate high quality, high dimensionalimages with relatively fast sampling speed. Due to their inherently restrictivearchitecture, however, it is necessary that they are excessively deep in order to tr... | HJex9qu3tH | Official Blind Review #1 | 3: Weak Reject | The paper proposes a combination of a conditioned flow-based model with a VAE. The main contribution of the paper is a two-phase training that allows to train the model. Unfortunately, a joint training of the model failed. In general, combining VAEs with flow-based models is an important research direction. Unfortunate... | <|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 Autoencoders with Normalizing Flow Decoders
### Paper Abstract
Recently proposed normalizing flow models such as Glow (Kingma & Dhariwal, 2018) have been sho... | |
HkNGsseC- | ICLR.cc/2018/Conference | 2018 | On the Expressive Power of Overlapping Architectures of Deep Learning | ["Or Sharir", "Amnon Shashua"] | Expressive efficiency refers to the relation between two architectures A and B, whereby any function realized by B could be replicated by A, but there exists functions realized by A, which cannot be replicated by B unless its size grows significantly larger. For example, it is known that deep networks are exponentially... | ["Deep Learning", "Expressive Efficiency", "Overlapping", "Receptive Fields"] | ABSTRACTExpressive efficiency refers to the relation between two architectures A and B,whereby any function realized by B could be replicated by A, but there existsfunctions realized by A, which cannot be replicated by B unless its size grows sig-nificantly larger. For example, it is known that deep networks are expone... | HkopHcseG | On the Expressive Power of Overlapping Architectures of Deep Learning | 8: Top 50% of accepted papers, clear accept | The paper analyzes the expressivity of convolutional arithmetic circuits (ConvACs), where neighboring neurons in a single layer have overlapping receptive fields. To compare the expressivity of overlapping networks with non-overlapping networks, the paper employs grid tensors computed from the output of the ConvACs. T... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Expressive Power of Overlapping Architectures of Deep Learning
### Paper Abstract
Expressive efficiency refers to the relation between two architectures A and B, ... |
rJe1DTNYPH | ICLR.cc/2020/Conference | 2020 | Towards Disentangling Non-Robust and Robust Components in Performance Metric | ["Yujun Shi", "Benben Liao", "Guangyong Chen", "Yun Liu", "Ming-ming Cheng", "Jiashi Feng"] | The vulnerability to slight input perturbations is a worrying yet intriguing property of deep neural networks (DNNs). Though some efforts have been devoted to investigating the reason behind such adversarial behavior, the relation between standard accuracy and adversarial behavior of DNNs is still little understood. In... | ["adversarial examples", "robust machine learning"] | ABSTRACTThe vulnerability to slight input perturbations is a worrying yet intriguing propertyof deep neural networks (DNNs). Though some efforts have been devoted toinvestigating the reason behind such adversarial behavior, the relation betweenstandard accuracy and adversarial behavior of DNNs is still little understoo... | B1x0RyRZcB | Official Blind Review #1 | 3: Weak Reject | The work explores the problem of robustness and adversarial
attacks in NN. In a multiclass prediction setting the idea
is to use a taylor expansion of a loss coined CCKL which
is the KL divergence between predictions for pairs of samples
from different classes.
The papers seems to find a convoluted route to arrive
to ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Towards Disentangling Non-Robust and Robust Components in Performance Metric
### Paper Abstract
The vulnerability to slight input perturbations is a worrying yet intrigu... | |
Bygka64KPH | ICLR.cc/2020/Conference | 2020 | Semi-Supervised Few-Shot Learning with Prototypical Random Walks | ["Ahmed Ayyad", "Nassir Navab", "Mohamed Elhoseiny", "Shadi Albarqouni"] | Learning from a few examples is a key characteristic of human intelligence that inspired machine learning researchers to build data-efficient AI models. Recent progress has shown that few-shot learning can be improved with access to unlabelled data, known as semi-supervised few-shot learning(SS-FSL). We introduce an S... | ["Few-Shot Learning", "Semi-Supervised Learning", "Random Walks"] | ABSTRACTLearning from a few examples is a key characteristic of human intelligence thatinspired machine learning researchers to build data efficient AI models. Recentprogress has shown that few-shot learning can be improved with access to unla-belled data, known as semi-supervised few-shot learning(SS-FSL). We introduc... | Syg4K6RjFS | Official Blind Review #2 | 3: Weak Reject | This submission introduces a version of the random-walk regularizer introduced in Haeusser et al. 2017, in the setting of a Prototypical Network for semi-supervised few-shot learning. The authors show that using this regularizer, SOTA results can be obtained, notably on the popular miniImageNet. They also show that the... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Semi-Supervised Few-Shot Learning with Prototypical Random Walks
### Paper Abstract
Learning from a few examples is a key characteristic of human intelligence that inspi... | |
ry-Se9kvG | ICLR.cc/2018/Workshop | 2018 | STOCHASTIC GRADIENT LANGEVIN DYNAMICS THAT EXPLOIT NEURAL NETWORK STRUCTURE | ["Zachary Nado", "Jasper Snoek", "Roger Grosse", "David Duvenaud", "Bowen Xu", "James Martens"] | Tractable approximate Bayesian inference for deep neural networks remains challenging. Stochastic Gradient Langevin Dynamics (SGLD) offers a tractable approximation to the gold standard of Hamiltonian Monte Carlo. We improve on existing methods for SGLD by incorporating a recently-developed tractable approximation of... | ["monte carlo", "Bayesian deep networks"] | ABSTRACTTractable approximate Bayesian inference for deep neural networks remains chal-lenging. Stochastic Gradient Langevin Dynamics (SGLD) offers a tractable approx-imation to the gold standard of Hamiltonian Monte Carlo. We improve on existingmethods for SGLD by incorporating a recently-developed tractable approxima... | BkPDD1-KM | Promising early result | 7: Good paper, accept | The authors propose to use the Kronecker-Factored Approximate Curvature (K-FAC) approximation of the Fisher information as a preconditioner of SGLD. The experiment result shows that it's a promising direction. | 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
STOCHASTIC GRADIENT LANGEVIN DYNAMICS THAT EXPLOIT NEURAL NETWORK STRUCTURE
### Paper Abstract
Tractable approximate Bayesian inference for deep neural networks remains... |
ByJbJwxCW | ICLR.cc/2018/Conference | 2018 | Relational Multi-Instance Learning for Concept Annotation from Medical Time Series | ["Sanjay Purushotham", "Zhengping Che", "Bo Jiang", "Tanachat Nilanon", "Yan Liu"] | Recent advances in computing technology and sensor design have made it easier to collect longitudinal or time series data from patients, resulting in a gigantic amount of available medical data. Most of the medical time series lack annotations or even when the annotations are available they could be subjective and pron... | ["Multi-instance learning", "Medical Time Series", "Concept Annotation"] | ABSTRACTRecent advances in computing technology and sensor design have made it easier tocollect longitudinal or time series data from patients, resulting in a gigantic amountof available medical data. Most of the medical time series lack annotations or evenwhen the annotations are available they could be subjective and... | Hk2mNy-gG | Overall, this is a reasonable paper with no obvious major flaws. The novelty and impact may be greater on the application side than on the methodology side. | 6: Marginally above acceptance threshold | The paper addresses the classification of medical time-series data by formulating the problem as a multi-instance learning (MIL) task, where there is an instance for each timestep of each time series, labels are observed at the time-series level (i.e. for each bag), and the goal is to perform instance-level and series-... | 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
Relational Multi-Instance Learning for Concept Annotation from Medical Time Series
### Paper Abstract
Recent advances in computing technology and sensor design have made... |
IKQYmATn893 | ijcai.org/IJCAI/2021/Workshop/NSNLI | 2021 | Exploring Multi-hop Reasoning Process in NLU from the View of Bayesian Probability | ["Yitian Li", "Jidong Tian", "Hao HE", "Yaohui Jin"] | Emerging pre-trained language models (PTLMs), such as BERT and RoBERTa, have already achieved great success on many natural language understanding (NLU) tasks, spurring widespread interest for their potential in scientific and social areas, with accompanying criticism on their ambiguousness of reasoning, especially mul... | ["Reasoning", "Bayesian probability"] | Exploring Multi-hop Reasoning Process in NLU from the View of BayesianProbabilityYitian Li1;2,Jidong Tian1;2,Hao He1;2andYaohui Jin1;21MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University2State Key Lab of Advanced Optical Communication System and Network, Shanghai Jiao TongUniversityfyiti... | kr-VHGSkRA | Review of ‚Multi-hop Reasoning Analysis Based on The Bayesian Probability‘ | 6: Marginally above acceptance threshold | The paper ‚Multi-hop Reasoning Analysis Based on The Bayesian Probability‘ investigates whether the estimated reasoning process of a pretrained transformer (PTLM) is consistent with the probabilistic reasoning process of a Bayesian network.
For this, the authors perform experiments on the ‚Soft Reasoner‘ multi-hop data... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Exploring Multi-hop Reasoning Process in NLU from the View of Bayesian Probability
### Paper Abstract
Emerging pre-trained language models (PTLMs), such as BERT and RoB... |
SJlM0JSFDr | ICLR.cc/2020/Conference | 2020 | A Theoretical Analysis of Deep Q-Learning | ["Zhuoran Yang", "Yuchen Xie", "Zhaoran Wang"] | Despite the great empirical success of deep reinforcement learning, its theoretical foundation is less well understood. In this work, we make the first attempt to theoretically understand the deep Q-network (DQN) algorithm (Mnih et al., 2015) from both algorithmic and statistical perspectives. In specific, we focus on ... | ["reinforcement learning", "deep Q network", "minimax-Q learning", "zero-sum Markov Game"] | ABSTRACTDespite the great empirical success of deep reinforcement learning, its theoreticalfoundation is less well understood. In this work, we make the first attempt totheoretically understand the deep Q-network (DQN) algorithm (Mnih et al., 2015)from both algorithmic and statistical perspectives. In specific, we focu... | SJlXXDdAcr | Official Blind Review #4 | 3: Weak Reject | Overview:
This paper provides an analysis of fitted Q-iteration in the off-policy reinforcement learning setting, for the setup where the value function class of interest is a class of neural networks. The provide bounds on the rate at which fitted Q iteration converges to a near-optimal policy under the assumption th... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Theoretical Analysis of Deep Q-Learning
### Paper Abstract
Despite the great empirical success of deep reinforcement learning, its theoretical foundation is less well... | |
HklKEUUY_E | ICLR.cc/2019/Workshop/DeepGenStruct | 2019 | On the relationship between Normalising Flows and Variational- and Denoising Autoencoders | ["Alexey A. Gritsenko", "Jasper Snoek", "Tim Salimans"] | Normalising Flows (NFs) are a class of likelihood-based generative models that have recently gained popularity. They are based on the idea of transforming a simple density into that of the data. We seek to better understand this class of models, and how they compare to previously proposed techniques for generative mode... | ["variational autoencoders", "denoising variational autoencoders", "normalizing flows", "generative modelling", "image synthesis", "denoising autoencoders", "VAE", "DAE", "VDAE", "NF"] | ABSTRACTNormalising Flows (NFs) are a class of likelihood-based generative models thathave recently gained popularity. They are based on the idea of transforming asimple density into that of the data. We seek to better understand this class ofmodels, and how they compare to previously proposed techniques for generative... | BygFOVRG5E | Interesting exposition of unifying NFs, VAEs, and DAEs | 3: Marginally above acceptance threshold | This paper proposes a model family that unifies NFs, VAEs and DAEs. It also introduces an extension of this model that allows for using non-invertible encoders (e.g. projection to a smaller dimensionality) and discrete data. Overall, the idea is promising, but the empirical results are not strong enough to warrant a st... | 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
On the relationship between Normalising Flows and Variational- and Denoising Autoencoders
### Paper Abstract
Normalising Flows (NFs) are a class of likelihood-based gene... |
HNA0kUAFdbv | ICLR.cc/2021/Conference | 2021 | CANVASEMB: Learning Layout Representation with Large-scale Pre-training for Graphic Design | ["Yuxi Xie", "Danqing Huang", "Jinpeng Wang", "Chin-Yew Lin"] | Layout representation, which models visual elements in a canvas and their inter-relations, plays a crucial role in graphic design intelligence.
With a large variety of layout designs and the unique characteristic of layouts that visual elements are defined as a list of categorical (e.g. shape type) and numerical (e.g. ... | ["Layout Representation", "Pre-training"] | ABSTRACTLayout representation, which models visual elements in a canvas and their inter-relations, plays a crucial role in graphic design intelligence. With a large varietyof layout designs and the unique characteristic of layouts that visual elements aredefined as a list of categorical (e.g. shape type) and numerical ... | 4Ppf_MqM7OK | Different Approach on the non-trivial task of Layout representation, but unsatisfactory evaluation | 5: Marginally below acceptance threshold | Summary of the work:
This work presents a Transformer-based framework to learn Graphical Layout representation/embedding, taking inspiration from transformer-based models in the area of NLP. To train their model, named CANVASEMB, the paper contributes a dataset of Powerpoint Slides, with more than 1 million slides (t... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
CANVASEMB: Learning Layout Representation with Large-scale Pre-training for Graphic Design
### Paper Abstract
Layout representation, which models visual elements in a ca... |
X9LHtgR4vq | ICLR.cc/2021/Conference | 2021 | Bractivate: Dendritic Branching in Medical Image Segmentation Neural Architecture Search | ["Leila Abdelrahman"] | Researchers manually compose most neural networks through painstaking experimentation.
This process is taxing and explores only a limited subset of possible
architecture. Researchers design architectures to address objectives ranging from low
space complexity to high accuracy through hours of experimentation. Neural ar... | ["Neural Architecture Search", "Segmentation", "Computer Vision"] | ABSTRACTResearchers manually compose most neural networks through painstaking exper-imentation. This process is taxing and explores only a limited subset of possiblearchitecture. Researchers design architectures to address objectives ranging fromlow space complexity to high accuracy through hours of experimentation. Ne... | zX76m061gA1 | Unclear contribution and insufficient empirical comparison | 3: Clear rejection | The authors propose a neural architecture search (NAS) algorithm inspired by brain physiology. In particular, they propose a NAS algorithm based on neural dendritic branching, and apply it to three different segmentation tasks (namely cell nuclei, electron microscopy, and chest X-ray lung segmentation). The authors sha... | 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
Bractivate: Dendritic Branching in Medical Image Segmentation Neural Architecture Search
### Paper Abstract
Researchers manually compose most neural networks through pai... |
HNlzT3G720t | ML_Reproducibility_Challenge/2021/Fall | 2021 | Reproducibility Study of “Counterfactual Generative Networks” | ["Piyush Bagad", "Jesse Maas", "Paul Hilders", "Danilo de Goede"] | Scope of Reproducibility
In this work, we study the reproducibility of the paper Counterfactual Generative Networks (CGN) by Sauer and Geiger to verify their main claims, which state that (i) their proposed model can reliably generate high-quality counterfactual images by disentangling the shape, texture and background... | ["Reproducibility", "Deep Generative Models", "Causality", "Counterfactuals", "Robustness", "Explainability"] | Reproducibility Study of “Counterfactual Generative Networks”Anonymous Author(s)AffiliationAddressemailReproducibility Summary 1Scope of Reproducibility 2In this work, we study the reproducibility of the paper Counterfactual Generative Networks (CGN) by Sauer and Geiger 3to verify their main claims, which state that (i... | S__lQMFKoe9 | The paper is really well-written and should be accepted definitely! | 9: Top 15% of accepted papers, strong accept | Thanks for your great paper!
Summary: The authors re-implemented the codes for most experiments and managed to validate the 3 claims of the original paper: high-quality counterfactuals, inductive bias requirements, and out-of-distribution robustness. Besides, the authors also improved CGN training on the MNIST dataset... | 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
Reproducibility Study of “Counterfactual Generative Networks”
### Paper Abstract
Scope of Reproducibility In this work, we study the reproducibility of the paper Counter... |
B1lLw6EYwB | ICLR.cc/2020/Conference | 2020 | Gap-Aware Mitigation of Gradient Staleness | ["Saar Barkai", "Ido Hakimi", "Assaf Schuster"] | Cloud computing is becoming increasingly popular as a platform for distributed training of deep neural networks. Synchronous stochastic gradient descent (SSGD) suffers from substantial slowdowns due to stragglers if the environment is non-dedicated, as is common in cloud computing. Asynchronous SGD (ASGD) methods are i... | ["distributed", "asynchronous", "large scale", "gradient staleness", "staleness penalization", "sgd", "deep learning", "neural networks", "optimization"] | ABSTRACTCloud computing is becoming increasingly popular as a platform for distributedtraining of deep neural networks. Synchronous stochastic gradient descent (SSGD)suffers from substantial slowdowns due to stragglers if the environment is non-dedicated, as is common in cloud computing. Asynchronous SGD (ASGD) meth-od... | Hyejd9jotH | Official Blind Review #1 | 3: Weak Reject | This paper proposes a simple idea to mitigate gradient staleness in asynchronized distributed systems. Instead of scaling the staleness as in SA method, the authors propose to scale with the GAP, which is defined as the distance between current parameter and the staled parameter.
The paper is fairly well written, and ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Gap-Aware Mitigation of Gradient Staleness
### Paper Abstract
Cloud computing is becoming increasingly popular as a platform for distributed training of deep neural netw... | |
SkxXCi0qFX | ICLR.cc/2019/Conference | 2019 | ProMP: Proximal Meta-Policy Search | ["Jonas Rothfuss", "Dennis Lee", "Ignasi Clavera", "Tamim Asfour", "Pieter Abbeel"] | Credit assignment in Meta-reinforcement learning (Meta-RL) is still poorly understood. Existing methods either neglect credit assignment to pre-adaptation behavior or implement it naively. This leads to poor sample-efficiency during meta-training as well as ineffective task identification strategies.
This paper provide... | ["Meta-Reinforcement Learning", "Meta-Learning", "Reinforcement-Learning"] | ABSTRACTCredit assignment in Meta-reinforcement learning (Meta-RL) is still poorly un-derstood. Existing methods either neglect credit assignment to pre-adaptation be-havior or implement it naively. This leads to poor sample-efficiency during meta-training as well as ineffective task identification strategies. This pap... | HklUqmw637 | Review | 7: Good paper, accept | In this paper, the author proposed an efficient surrogate loss for estimating Hessian in the setting of Meta-reinforcement learning (Finn.et al, 2017), which significantly reduce the variance while introducing small bias. The author verified their proposed method with other meta-learning algorithms on the Mujoco bench... | 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
ProMP: Proximal Meta-Policy Search
### Paper Abstract
Credit assignment in Meta-reinforcement learning (Meta-RL) is still poorly understood. Existing methods either negl... |
B1guLAVFDB | ICLR.cc/2020/Conference | 2020 | Span Recovery for Deep Neural Networks with Applications to Input Obfuscation | ["Rajesh Jayaram", "David P. Woodruff", "Qiuyi Zhang"] | The tremendous success of deep neural networks has motivated the need to better understand the fundamental properties of these networks, but many of the theoretical results proposed have only been for shallow networks. In this paper, we study an important primitive for understanding the meaningful input space of a deep... | ["Span recovery", "low rank neural networks", "adversarial attack"] | ABSTRACTThe tremendous success of deep neural networks has motivated the need to better understand thefundamental properties of these networks, but many of the theoretical results proposed have onlybeen for shallow networks. In this paper, we study an important primitive for understanding themeaningful input space of a... | BketSBj2cB | Official Blind Review #4 | 3: Weak Reject | The paper studies the problem of span recovery for deep neural networks, that is
recovering the space of inputs that affect the output of the network.
The authors propose two algorithm, one for the case of ReLU activations and one
for the case of differentiable activations, and theoretically prove that they can
recover... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Span Recovery for Deep Neural Networks with Applications to Input Obfuscation
### Paper Abstract
The tremendous success of deep neural networks has motivated the need to... | |
1k6uXBMy4fj | NeurIPS.cc/2022/Workshop/Federated_Learning | 2022 | The Interpolated MVU Mechanism For Communication-efficient Private Federated Learning | ["Chuan Guo", "Kamalika Chaudhuri", "Pierre Stock", "Michael Rabbat"] | We consider private federated learning (FL), where a server aggregates differentially private gradient updates from a large number of clients in order to train a machine learning model. The main challenge here is balancing privacy with both classification accuracy of the learned model as well as the amount of communica... | ["differential privacy", "federated learning", "data compression"] | The Interpolated MVU Mechanism ForCommunication-efficient Private Federated LearningAnonymous Author(s)AffiliationAddressemailAbstractWe consider private federated learning (FL), where a server aggregates differentially 1private gradient updates from a large number of clients in order to train a machine 2learning model... | HRTe4JyV5xM | review | This paper proposed the I-MVU mechanism that reduces communication cost while providing a differential privacy guarantee. This paper extends the previously introduced MVU mechanism to continuous-valued vectors. This paper could achieve a good tradeoff between utility and privacy. This paper conducted experiments to sup... | <|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 Interpolated MVU Mechanism For Communication-efficient Private Federated Learning
### Paper Abstract
We consider private federated learning (FL), where a server aggr... | ||
MBpHUFrcG2x | ICLR.cc/2021/Conference | 2021 | Projected Latent Markov Chain Monte Carlo: Conditional Sampling of Normalizing Flows | ["Chris Cannella", "Mohammadreza Soltani", "Vahid Tarokh"] | We introduce Projected Latent Markov Chain Monte Carlo (PL-MCMC), a technique for sampling from the exact conditional distributions learned by normalizing flows. As a conditional sampling method, PL-MCMC enables Monte Carlo Expectation Maximization (MC-EM) training of normalizing flows from incomplete data. Through exp... | ["Conditional Sampling", "Normalizing Flows", "Markov Chain Monte Carlo", "Missing Data Inference"] | ABSTRACTWe introduce Projected Latent Markov Chain Monte Carlo (PL-MCMC), a tech-nique for sampling from the exact conditional distributions learned by normaliz-ing flows. As a conditional sampling method, PL-MCMC enables Monte CarloExpectation Maximization (MC-EM) training of normalizing flows from incom-plete data. T... | y_YRoMR2zpN | Official Review 4 | 6: Marginally above acceptance threshold | Summary:
The submission introduces an approach for conditional sampling by running a Random walk Metropolis algorithm on the latent space as a pullback of normalizing flows. The normalizing flows are learnt via a MC-EM approach for incomplete training data. The method is illustrated on numerous experiments sampling mis... | 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
Projected Latent Markov Chain Monte Carlo: Conditional Sampling of Normalizing Flows
### Paper Abstract
We introduce Projected Latent Markov Chain Monte Carlo (PL-MCMC),... |
Skgy464Kvr | ICLR.cc/2020/Conference | 2020 | Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions | ["Yao Qin", "Nicholas Frosst", "Sara Sabour", "Colin Raffel", "Garrison Cottrell", "Geoffrey Hinton"] | Adversarial examples raise questions about whether neural network models are sensitive to the same visual features as humans. In this paper, we first detect adversarial examples or otherwise corrupted images based on a class-conditional reconstruction of the input. To specifically attack our detection mechanism, we pro... | ["Adversarial Examples", "Detection of adversarial attacks"] | ABSTRACTAdversarial examples raise questions about whether neural network models aresensitive to the same visual features as humans. In this paper, we first detectadversarial examples or otherwise corrupted images based on a class-conditionalreconstruction of the input. To specifically attack our detection mechanism, w... | ByeNI910FS | Official Blind Review #3 | 6: Weak Accept | This paper proposed a new defense method for capsule networks. For both white-box and black-box settings, the proposed CapNets has shown superior performance than two variants of CNNs. The visualizations of adversarial examples generated by the CapNets are more aligned with the human perception which is very insightful... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions
### Paper Abstract
Adversarial examples raise questions about whether neural n... | |
RepN5K31PT3 | ICLR.cc/2021/Conference | 2021 | On the Dynamic Regret of Online Multiple Mirror Descent | ["Nima Eshraghi", "and Ben Liang"] | We study the problem of online convex optimization, where a learner makes sequential decisions to minimize an accumulation of strongly convex costs over time. The quality of decisions is given in terms of the dynamic regret, which measures the performance of the learner relative to a sequence of dynamic minimizers. Pri... | ["Online learning", "Online Convex Optimization", "Mirror Descent"] | ABSTRACTWe study the problem of online convex optimization, where a learner makes se-quential decisions to minimize an accumulation of strongly convex costs over time.The quality of decisions is given in terms of the dynamic regret, which measuresthe performance of the learner relative to a sequence of dynamic minimize... | 31Ew_uNiVqp | Natural generalization of OMGD path-length regret; unclear justification of quantity G_T | 5: Marginally below acceptance threshold | Summary:
This paper studies the dynamic regret of online multiple mirror descent, which is online mirror descent with M repeated steps on each of T sequential loss functions. The authors show three bounds for the dynamic regret of OMMD, which generalizes OMGD [Zhang et al. '17]: C_T (the path length of the minimizer se... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Dynamic Regret of Online Multiple Mirror Descent
### Paper Abstract
We study the problem of online convex optimization, where a learner makes sequential decisions... |
S1LVSrcge | ICLR.cc/2017/conference | 2017 | Variable Computation in Recurrent Neural Networks | ["Yacine Jernite", "Edouard Grave", "Armand Joulin", "Tomas Mikolov"] | Recurrent neural networks (RNNs) have been used extensively and with increasing success to model various types of sequential data. Much of this progress has been achieved through devising recurrent units and architectures with the flexibility to capture complex statistics in the data, such as long range dependency or l... | ["Natural language processing", "Deep learning"] | ABSTRACTRecurrent neural networks (RNNs) have been used extensively and withincreasing success to model various types of sequential data. Much of thisprogress has been achieved through devising recurrent units and architectureswith the flexibility to capture complex statistics in the data, such as long rangedependency ... | rJ0I-lfEe | review | 4: Ok but not good enough - rejection | TLDR: The authors present Variable Computation in Recurrent Neural Networks (VCRNN). VCRNN is similar in nature to Adaptive Computation Time (Graves et al., 2016). Imagine a vanilla RNN, at each timestep only a subset (i.e., "variable computation") of the state is updated. Experimental results are not convincing, 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
Variable Computation in Recurrent Neural Networks
### Paper Abstract
Recurrent neural networks (RNNs) have been used extensively and with increasing success to model var... |
Cri3xz59ga | ICLR.cc/2021/Conference | 2021 | Deciphering and Optimizing Multi-Task Learning: a Random Matrix Approach | ["Malik Tiomoko", "Hafiz Tiomoko Ali", "Romain Couillet"] | This article provides theoretical insights into the inner workings of multi-task and transfer learning methods, by studying the tractable least-square support vector machine multi-task learning (LS-SVM MTL) method, in the limit of large ($p$) and numerous ($n$) data. By a random matrix analysis applied to a Gaussian mi... | ["Transfer Learning", "Multi Task Learning", "Random Matrix Theory"] | ABSTRACTThis article provides theoretical insights into the inner workings of multi-task andtransfer learning methods, by studying the tractable least-square support vectormachine multi-task learning (LS-SVM MTL) method, in the limit of large ( p) andnumerous (n) data. By a random matrix analysis applied to a Gaussian ... | Rfy-fDHZwEj | This paper studies the LS-SVM MTL method in multi-task learning in the large-p large-n setting. By utilizing mathematical tools in random matrix theory, it characterizes the asymptotic behavior of the classification score and thus proposes to choose the decision threshold and training labels in an optimal way. The theo... | 7: Good paper, accept | Overall, I vote for accepting this paper. I list the strong and weak points in the following.
### Strong points
- The theoretical result is sound and significant. It not only matches the simulation well, but also provides a way to optimally choose the decision threshold and training labels. The proposed method is simpl... | 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
Deciphering and Optimizing Multi-Task Learning: a Random Matrix Approach
### Paper Abstract
This article provides theoretical insights into the inner workings of multi-t... |
FTit3PiAw4 | ICLR.cc/2021/Conference | 2021 | Training Federated GANs with Theoretical Guarantees: A Universal Aggregation Approach | ["Yikai Zhang", "Hui Qu", "Huidong Liu", "Qi Chang", "Dimitris N. Metaxas", "Chao Chen"] | Recently, Generative Adversarial Networks (GANs) have demonstrated their potential in federated learning, i.e., learning a centralized model from data privately hosted by multiple sites. A federated GAN jointly trains a centralized generator and multiple private discriminators hosted at different sites. A major theoret... | ["Federated Learning", "GAN", "Deep Learning"] | ABSTRACTRecently, Generative Adversarial Networks (GANs) have demonstrated their po-tential in federated learning, i.e., learning a centralized model from data privatelyhosted by multiple sites. A federated GAN jointly trains a centralized generatorand multiple private discriminators hosted at different sites. A major ... | QGN9ClnHyO3 | An insufficiently clear paper | 3: Clear rejection |
This paper studies gederated Generative Adversarial Networks (federated GANs).
In particular, the authors propose a new method, UA-GAN, which is claimed to be better than earlier approaches.
I have several concerns. First, the writing can be improved significantly. Second, the statements in the text are often rat... | 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
Training Federated GANs with Theoretical Guarantees: A Universal Aggregation Approach
### Paper Abstract
Recently, Generative Adversarial Networks (GANs) have demonstra... |
PK2debCKaG | robot-learning.org/CoRL/2023/Conference | 2023 | Language Conditioned Traffic Generation | ["Shuhan Tan", "Boris Ivanovic", "Xinshuo Weng", "Marco Pavone", "Philipp Kraehenbuehl"] | Simulation forms the backbone of modern self-driving development. Simulators help develop, test, and improve driving systems without putting humans, vehicles, or their environment at risk. However, simulators face a major challenge: They rely on realistic, scalable, yet interesting content. While recent advances in ren... | ["Self-driving", "Content generation", "Large language model"] | Language Conditioned Traffic GenerationShuhan Tan1Boris Ivanovic2Xinshuo Weng2Marco Pavone2Philipp Krähenbühl11UT Austin2NVIDIAAbstract: Simulation forms the backbone of modern self-driving development.Simulators help develop, test, and improve driving systems without putting hu-mans, vehicles, or their environment a... | YDq-QO03Odq | Strength
1. The method that uses GPT-4 to generate traffic scenes is interesting and novel.
2. The performance improvement is decent.
3. The method can be used for language-conditioned traffic scene editing, which is never shown by previous methods before.
4. The paper writes well
Weakness
1. It would be interestin... | 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
Language Conditioned Traffic Generation
### Paper Abstract
Simulation forms the backbone of modern self-driving development. Simulators help develop, test, and improve d... | ||
r1l-e3Cqtm | ICLR.cc/2019/Conference | 2019 | Deep Probabilistic Video Compression | ["Jun Han", "Salvator Lombardo", "Christopher Schroers", "Stephan Mandt"] | We propose a variational inference approach to deep probabilistic video compression. Our model uses advances in variational autoencoders (VAEs) for sequential data and combines it with recent work on neural image compression. The approach jointly learns to transform the original video into a lower-dimensional represe... | ["variational inference", "video compression", "deep generative models"] | ABSTRACTWe propose a variational inference approach to deep probabilistic video compres-sion. Our model uses advances in variational autoencoders (V AEs) for sequentialdata and combines it with recent work on neural image compression. The ap-proach jointly learns to transform the original video into a lower-dimensional... | ByltbXgKnQ | Good approach to deep learning based video compression, but empirical section needs work | 5: Marginally below acceptance threshold | Summary
=======
This work on video compression extends the variational autoencoder of Balle et al. (2016; 2018) from images to videos. The latent space consists of a global part encoding information about the entire video, and a local part encoding information about each frame. Correspondingly, the encoder consists 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
Deep Probabilistic Video Compression
### Paper Abstract
We propose a variational inference approach to deep probabilistic video compression. Our model uses advances in ... |
UlIJS3dcMi | graphicsinterface.org/Graphics_Interface/2023/Conference_SD | 2023 | A Non-Contact Heart Rate Measurement Method Resistant to Illumination Changes Based on Fast Wavelet Transform and Second-Order Blind Identification in Far-Field Environments | ["Anonymous"] | Heart rate(HR) is a key parameter for evaluating a person’s physiological condition. In recent years, there have been many researches on remote heart rate measurement. However, these methods are mostly conducted in close-range scenarios, making them inapplicable in many scenarios. Remote photoplethysmography (rPPG) pro... | ["heart rate measurement", "environments", "illumination changes", "fast wavelet transform", "blind identification", "resistant", "environments heart rate", "key parameter", "person"] | ABSTRACTHeart rate(HR) is a key parameter for evaluating a person’s physio-logical condition. In recent years, there have been many researcheson remote heart rate measurement. However, these methods aremostly conducted in close-range scenarios, making them inappli-cable in many scenarios. Remote photoplethysmography (r... | k6Vj161yLNo | Out-of-scope and straightforward | 2: Strong rejection | The paper includes author information, which violates the double-blind policy. Nevertheless, I provide my comments below.
The paper presents a remote photoplethysmography (rPPG) framework for heart rate detection in far-field environments. The proposed method addresses the limitations of existing close-range heart rat... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Non-Contact Heart Rate Measurement Method Resistant to Illumination Changes Based on Fast Wavelet Transform and Second-Order Blind Identification in Far-Field Environme... |
rJlJF1SYPB | ICLR.cc/2020/Conference | 2020 | Universality Theorems for Generative Models | ["Valentin Khrulkov", "Ivan Oseledets"] | Despite the fact that generative models are extremely successful in practice, the theory underlying this phenomenon is only starting to catch up with practice. In this work we address the question of the universality of generative models: is it true that neural networks can approximate any data manifold arbitrarily wel... | ["generative models", "theory", "universality", "manifolds", "differential geometry"] | ABSTRACTDespite the fact that generative models are extremely successful in practice, thetheory underlying this phenomenon is only starting to catch up with practice. Inthis work we address the question of the universality of generative models: is ittrue that neural networks can approximate any data manifold arbitraril... | ryxRtxe4qS | Official Blind Review #3 | 1: Reject | Summary: The paper provides certain basic guarantees on when manifolds can be written (exactly or approximately) as the image of a map approximated by a neural net. This is couched as answering the question of when GANs can approximate certain distributions, though in fairness, distributions aren't really ever consider... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Universality Theorems for Generative Models
### Paper Abstract
Despite the fact that generative models are extremely successful in practice, the theory underlying this p... | |
rylf7QFUz | ICLR.cc/2018/Workshop | 2018 | More systematic than claimed: Insights on the SCAN tasks | ["Markus Kliegl", "Wei Xu"] | We show that some standard attention-based architectures widely used in Neural Machine
Translation as well as a pointer-based variant achieve results on some of the compositional
SCAN tasks that are far superior to those reported in Lake & Baroni (2018). We next show that
there is high variance in the test accuracy acr... | ["sequence-to-sequence recurrent networks", "pointer networks", "compositionality", "systematicity", "generalization"] | ABSTRACTWe show that some standard attention-based architectures widely used in NeuralMachine Translation as well as a pointer-based variant achieve results on some ofthe compositional SCAN tasks that are far superior to those reported in Lake &Baroni (2018). We next show that there is high variance in the test accurac... | HkcavUyKG | Some intriguing results but more a pile of puzzling experiments than a result | 4: Ok but not good enough - rejection | Summary:
This paper revisits the SCAN tasks introduced in a (rejected, invited to workshop track) ICLR 2018 paper, which convert simplified English to a simple formal semantics action sequence and suggests that sequence models can do them rather better than the results reported in the original paper (Lake and Baroni ... | 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
More systematic than claimed: Insights on the SCAN tasks
### Paper Abstract
We show that some standard attention-based architectures widely used in Neural Machine Transl... |
XbJiphOWXiU | ICLR.cc/2021/Conference | 2021 | Empirically Verifying Hypotheses Using Reinforcement Learning | ["Kenneth Marino", "Rob Fergus", "Arthur Szlam", "Abhinav Gupta"] | This paper formulates hypothesis verification as an RL problem. Specifically, we aim to build an agent that, given a hypothesis about the dynamics of the world, can take actions to generate observations which can help predict whether the hypothesis is true or false. Existing RL algorithms fail to solve this task, even ... | ["hypotheses", "reinforcement", "agent", "hypothesis", "task", "hypothesis verification", "rl problem", "dynamics", "world", "actions"] | ABSTRACTThis paper formulates hypothesis verification as an RL problem. Specifically, weaim to build an agent that, given a hypothesis about the dynamics of the world, cantake actions to generate observations which can help predict whether the hypothesisis true or false. Existing RL algorithms fail to solve this task, ... | a2kn9FoUxsp | review | 3: Clear rejection |
This paper presents the problem formulation for a hypothesis testing within a reinforcement learning paradigm. In particular, given a hypothesis represented as a formal, templated language, an agent needs to perform a sequence of actions and check whether the hypothesis holds true or false.
The paper proposes multip... | 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
Empirically Verifying Hypotheses Using Reinforcement Learning
### Paper Abstract
This paper formulates hypothesis verification as an RL problem. Specifically, we aim to ... |
SkMON20ctX | ICLR.cc/2019/Conference | 2019 | On the Trajectory of Stochastic Gradient Descent in the Information Plane | ["Emilio Rafael Balda", "Arash Behboodi", "Rudolf Mathar"] | Studying the evolution of information theoretic quantities during Stochastic Gradient Descent (SGD) learning of Artificial Neural Networks (ANNs) has gained popularity in recent years.
Nevertheless, these type of experiments require estimating mutual information and entropy which becomes intractable for moderately lar... | ["Stochastic gradient descent", "Deep neural networks", "Entropy", "Information theory", "Markov chains", "Hidden Markov process."] | ABSTRACTStudying the evolution of information theoretic quantities during Stochastic Gra-dient Descent (SGD) learning of Artificial Neural Networks (ANNs) has gainedpopularity in recent years. Nevertheless, these type of experiments require es-timating mutual information and entropy which becomes intractable for moder-... | Byg9ksJZaQ | unclear motivation | 4: Ok but not good enough - rejection | In summary, this paper does the following:
- The initial problem is to analyze the trajectory of SGD in training ANNs in the space of P of probability measures on Y \times Y. This problem is interesting, but difficult.
- the paper constructs a Markov chain that follows a shortest path in TV metric on P
(the \alpha SM... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Trajectory of Stochastic Gradient Descent in the Information Plane
### Paper Abstract
Studying the evolution of information theoretic quantities during Stochastic... |
Sk8J83oee | ICLR.cc/2017/conference | 2017 | Generative Adversarial Parallelization | ["Daniel Jiwoong Im", "He Ma", "Chris Dongjoo Kim", "Graham Taylor"] | Generative Adversarial Networks (GAN) have become one of the most studied frameworks for unsupervised learning due to their intuitive formulation. They have also been shown to be capable of generating convincing examples in limited domains, such as low-resolution images. However, they still prove difficult to train in ... | ["Unsupervised Learning"] | ABSTRACTGenerative Adversarial Networks (GAN) have become one of the most studiedframeworks for unsupervised learning due to their intuitive formulation. Theyhave also been shown to be capable of generating convincing examples in limiteddomains, such as low-resolution images. However, they still prove difficult to trai... | S1IoaUXEx | Review | 4: Ok but not good enough - rejection | This paper proposes an extension of the GAN framework known as GAP whereby multiple generators and discriminators are trained in parallel. The generator/discriminator pairing is shuffled according to a periodic schedule.
Pros:
+ The proposed approach is simple and easy to replicate.
Cons:
- The paper is confusing to ... | 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
Generative Adversarial Parallelization
### Paper Abstract
Generative Adversarial Networks (GAN) have become one of the most studied frameworks for unsupervised learning ... |
HJeTo2VFwH | ICLR.cc/2020/Conference | 2020 | A Signal Propagation Perspective for Pruning Neural Networks at Initialization | ["Namhoon Lee", "Thalaiyasingam Ajanthan", "Stephen Gould", "Philip H. S. Torr"] | Network pruning is a promising avenue for compressing deep neural networks. A typical approach to pruning starts by training a model and then removing redundant parameters while minimizing the impact on what is learned. Alternatively, a recent approach shows that pruning can be done at initialization prior to training,... | ["neural network pruning", "signal propagation perspective", "sparse neural networks"] | ABSTRACTNetwork pruning is a promising avenue for compressing deep neural networks. Atypical approach to pruning starts by training a model and then removing redun-dant parameters while minimizing the impact on what is learned. Alternatively, arecent approach shows that pruning can be done at initialization prior to tr... | SJlF8AaTtB | Official Blind Review #1 | 6: Weak Accept | In this paper, the authors studied and formalized the effect of initialization to connection-sensitivity-based pruning. The authors first pointed out that a previously studied pruning criterion -- connection sensitivity (CS) -- is a normalized magnitude of gradients. Based on signal propagation theory, to achieve a 'fa... | <|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 Signal Propagation Perspective for Pruning Neural Networks at Initialization
### Paper Abstract
Network pruning is a promising avenue for compressing deep neural netwo... | |
SkgRW64twr | ICLR.cc/2020/Conference | 2020 | Deep Multi-View Learning via Task-Optimal CCA | ["Heather D. Couture", "Roland Kwitt", "J.S. Marron", "Melissa Troester", "Charles M. Perou", "Marc Niethammer"] | Canonical Correlation Analysis (CCA) is widely used for multimodal data analysis and, more recently, for discriminative tasks such as multi-view learning; however, it makes no use of class labels. Recent CCA methods have started to address this weakness but are limited in that they do not simultaneously optimize the C... | ["multi-view", "components analysis", "CCA", "representation learning", "deep learning"] | ABSTRACTMulti-view learning seeks to form better models by making use of multiple featuresets representing the same samples. Exploiting feature correlations during train-ing can enable better models. The traditional method for computing correlationsbetween feature sets is Canonical Correlation Analysis (CCA), which fin... | rkeDzTZ6KS | Official Blind Review #1 | 3: Weak Reject | This is a interesting paper on an important topic, but it was a main weakness: it assumes that the reader is deeply familiar with the CCA. In order to make the paper more accessible to a general audience, the authors should:
1) have at least one sentence in the abstract that explains in layman terms why is CCA importan... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Deep Multi-View Learning via Task-Optimal CCA
### Paper Abstract
Canonical Correlation Analysis (CCA) is widely used for multimodal data analysis and, more recently, for... | |
4mkxyuPcFt | ICLR.cc/2021/Conference | 2021 | Disentangling Adversarial Robustness in Directions of the Data Manifold | ["Jiancong Xiao", "Liusha Yang", "Zhi-Quan Luo"] | Using generative models (GAN or VAE) to craft adversarial examples, i.e. generative adversarial examples, has received increasing attention in recent years. Previous studies showed that the generative adversarial examples work differently compared to that of the regular adversarial examples in many aspects, such as att... | ["Adversarial Robustness", "Adversarial Training", "Generative Models"] | ABSTRACTUsing generative models (GAN or V AE) to craft adversarial examples, i.e. gen-erative adversarial examples, has received increasing attention in recent years.Previous studies showed that the generative adversarial examples work differentlycompared to that of the regular adversarial examples in many aspects, suc... | ceP75ed9eB | Disentangling Adversarial Robustness in Directions of the Data Manifold | 6: Marginally above acceptance threshold | The present paper proposes an interesting study about two different kinds of adversarials, depending on the area of the variance and manifold that is attacked. Although the results are promising, there is some lack of further experimentation to provide more meaningful results, so I would recommend providing more experi... | 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
Disentangling Adversarial Robustness in Directions of the Data Manifold
### Paper Abstract
Using generative models (GAN or VAE) to craft adversarial examples, i.e. gener... |
AEhp_Cqq-h | MIDL.io/2020/Conference | 2020 | Deep learning approach to describe and classify fungi microscopic images | ["Bartosz Zieli\u0144ski", "Agnieszka Sroka-Oleksiak", "Dawid Rymarczyk", "Adam Piekarczyk", "Monika Brzychczy-W\u0142och"] | Preliminary diagnosis of fungal infections can rely on microscopic examination. However, in many cases, it does not allow unambiguous identification of the species by microbiologist due to their visual similarity. Therefore, it is usually necessary to use additional biochemical tests. That involves additional costs and... | ["mycological diagnosis", "microscopic images", "deep learning", "bag-of-words"] | Medical Imaging with Deep Learning 2020 Short Paper { MIDL 2020 submissionDeep learning approach to describe and classify fungimicroscopic imagesBartosz Zieli nski1;2bartosz.zielinski@uj.edu.plAgnieszka Sroka-Oleksiak3;4agnieszka.sroka@uj.edu.plDawid Rymarczyk1;2dawid.rymarczyk@ii.uj.edu.plAdam Piekarczyk1adam.jan.piek... | du3w4oL6k | o In this paper, authors use a deep neural network to classify microscopic fungal images. They intend that their technique be used to cut costs and time of identification in cases of fungal infections. A combined bag-of-words and AlexNet approach achieved the best accuracy (82.4 path-based and 93.9 scan-based) on their... | 2: Weak reject | One particular strength of this paper was the authors’ choice of fungal types to train and test their models on. They specify that the chosen types overlap a decent amount with most common fungal infections. Another strength of their paper was the combination of multiple methods, AlexNet and bag-of-words to achieve a b... | 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 approach to describe and classify fungi microscopic images
### Paper Abstract
Preliminary diagnosis of fungal infections can rely on microscopic examinatio... |
HJxMYANtPH | ICLR.cc/2020/Conference | 2020 | The Local Elasticity of Neural Networks | ["Hangfeng He", "Weijie Su"] | This paper presents a phenomenon in neural networks that we refer to as local elasticity. Roughly speaking, a classifier is said to be locally elastic if its prediction at a feature vector x' is not significantly perturbed, after the classifier is updated via stochastic gradient descent at a (labeled) feature vector x ... | ["local elasticity", "neural networks", "phenomenon", "classifier", "feature vector x", "elastic", "prediction", "stochastic gradient descent", "dissimilar", "certain sense"] | ABSTRACTThis paper presents a phenomenon in neural networks that we refer to as localelasticity . Roughly speaking, a classifier is said to be locally elastic if its pre-diction at a feature vector x0isnotsignificantly perturbed, after the classifieris updated via stochastic gradient descent at a (labeled) feature vect... | ryxdtTj1qS | Official Blind Review #3 | 6: Weak Accept | This paper studies an interesting phenomenon in neural network models that the classifier's prediction at a one input will not be significantly perturbed after the classifier is updated via sgd at another input that is dissimilar from the former one. This phenomenon is termed as the local elasticity, which provides ano... | <|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 Local Elasticity of Neural Networks
### Paper Abstract
This paper presents a phenomenon in neural networks that we refer to as local elasticity. Roughly speaking, a ... | |
B1lLw6EYwB | ICLR.cc/2020/Conference | 2020 | Gap-Aware Mitigation of Gradient Staleness | ["Saar Barkai", "Ido Hakimi", "Assaf Schuster"] | Cloud computing is becoming increasingly popular as a platform for distributed training of deep neural networks. Synchronous stochastic gradient descent (SSGD) suffers from substantial slowdowns due to stragglers if the environment is non-dedicated, as is common in cloud computing. Asynchronous SGD (ASGD) methods are i... | ["distributed", "asynchronous", "large scale", "gradient staleness", "staleness penalization", "sgd", "deep learning", "neural networks", "optimization"] | ABSTRACTCloud computing is becoming increasingly popular as a platform for distributedtraining of deep neural networks. Synchronous stochastic gradient descent (SSGD)suffers from substantial slowdowns due to stragglers if the environment is non-dedicated, as is common in cloud computing. Asynchronous SGD (ASGD) meth-od... | B1lQL-d6YH | Official Blind Review #3 | 3: Weak Reject | The paper introduces a new variant of asynchronous SGD, GA-ASGD, for distributed training. The goal is to mitigate the gradient staleness issue caused by asynchronously applying gradients to an old version of parameters. Prior work addresses this issue by penalizing the learning step of a worker linearly to its missed ... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Gap-Aware Mitigation of Gradient Staleness
### Paper Abstract
Cloud computing is becoming increasingly popular as a platform for distributed training of deep neural netw... | |
B1xcLJrYwH | ICLR.cc/2020/Conference | 2020 | Lean Images for Geo-Localization | ["Moti Kadosh", "Yael Moses", "Ariel Shamir"] | Most computer vision tasks use textured images. In this paper we consider the geo-localization task - finding the pose of a camera in a large 3D scene from a single lean image, i.e. an image with no texture. We aim to experimentally explore whether texture and correlation between nearby images are necessary in a CNN-b... | ["Geo Localization", "Deep Learning", "Computer Vision", "Camera Localization"] | ABSTRACTMost computer vision tasks use textured images. In this paper we consider thegeo-localization task – finding the pose of a camera in a large 3D scene from asingle lean image , i.e. an image with no texture. We aim to experimentally explorewhether texture and correlation between nearby images are necessary in a ... | BJxZfI20KB | Official Blind Review #2 | 3: Weak Reject | The paper tackles the problem of geolocating an image from a 3D model of a city. This is an interesting and challenging problem since it involves solving the matching under 3D transformations at large scale.
The proposed approach is to use geometric features such as edges, faces and depths of the image, which is calle... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Lean Images for Geo-Localization
### Paper Abstract
Most computer vision tasks use textured images. In this paper we consider the geo-localization task - finding the pos... | |
qG4ZVCCyCB0 | ICLR.cc/2021/Conference | 2021 | How Important is the Train-Validation Split in Meta-Learning? | ["Yu Bai", "Minshuo Chen", "Pan Zhou", "Tuo Zhao", "Jason D. Lee", "Sham M. Kakade", "Huan Wang", "Caiming Xiong"] | Meta-learning aims to perform fast adaptation on a new task through learning a "prior" from multiple existing tasks. A common practice in meta-learning is to perform a train-validation split where the prior adapts to the task on one split of the data, and the resulting predictor is evaluated on another split. Despite i... | ["Learning theory", "meta-learning"] | ABSTRACTMeta-learning aims to perform fast adaptation on a new task through learning a“prior” from multiple existing tasks. A common practice in meta-learning is toperform a train-validation split where the prior adapts to the task on one split ofthe data, and the resulting predictor is evaluated on another split. Desp... | _AMKAu33Lb | In this paper, the meta-learning is formulated as a multiple ridge regression problems and the standard statistical asymptotic theory is used to derive some theoretical results. The impact of the results will be rather limited. | 5: Marginally below acceptance threshold | In this paper, the authors study the theoretical properties of meta-learning. In particular, the train-validation split to tackle the linear centroid meta-learning problem is investigated using statistical asymptotic theory. First, the authors proved that the train-validation method has statistical consistency, while t... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
How Important is the Train-Validation Split in Meta-Learning?
### Paper Abstract
Meta-learning aims to perform fast adaptation on a new task through learning a "prior" f... |
HkCvZXbC- | ICLR.cc/2018/Conference | 2018 | 3C-GAN: AN CONDITION-CONTEXT-COMPOSITE GENERATIVE ADVERSARIAL NETWORKS FOR GENERATING IMAGES SEPARATELY | ["Yeu-Chern Harn", "Vladimir Jojic"] | We present 3C-GAN: a novel multiple generators structures, that contains one conditional generator that generates a semantic part of an image conditional on its input label, and one context generator generates the rest of an image. Compared to original GAN model, this model has multiple generators and gives control ove... | ["model", "image", "images", "generative adversarial networks", "label information", "object", "present", "conditional generator", "semantic part"] | ABSTRACTWe present 3C-GAN: a novel multiple generators structures, that contains oneconditional generator that generates a semantic part of an image conditional on itsinput label, and one context generator generates the rest of an image. Comparedto original GAN model, this model has multiple generators and gives contro... | ByfJW0vlf | Review from AnonReviewer1 | 4: Ok but not good enough - rejection | [Overview]
This paper proposed a new generative adversarial network, called 3C-GAN for generating images in a composite manner. In 3C-GAN, the authors exploited two generators, one (G1) is for generating context images, and the other one (G2) is for generating semantic contents. To generate the semantic contents, 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
3C-GAN: AN CONDITION-CONTEXT-COMPOSITE GENERATIVE ADVERSARIAL NETWORKS FOR GENERATING IMAGES SEPARATELY
### Paper Abstract
We present 3C-GAN: a novel multiple generators... |
2HqKwHaBwv | logconference.io/LOG/2022/Conference | 2022 | Graph-Time Convolutional Autoencoders | ["Mohammad Sabbaqi", "Riccardo Taormina", "Alan Hanjalic", "Elvin Isufi"] | We introduce graph-time convolutional autoencoder (GTConvAE), a novel spatiotemporal architecture tailored to learn unsupervised representations from multivariate time series on networks. The GTConvAE leverages product graphs to represent the spatiotemporal information and a principled joint convolution operation over ... | ["convolutional autoencoders", "gtconvae", "product graph", "encoder", "network structure", "decoder", "convolutional autoencoder", "novel spatiotemporal architecture", "unsupervised representations", "multivariate time series"] | Graph-Time Convolutional AutoencodersMohammad Sabbaqi Riccardo Taormina Alan Hanjalic Elvin IsufiDelft University of Technology{m.sabbaqi, r.taormina, a.hanjalic, e.isufi-1}@tudelft.nlAbstractWe introduce graph-time convolutional autoencoder (GTConvAE), a novel spa-tiotemporal architecture tailored to unsupervised lear... | HB7Iak2-qHs | This paper addresses the problem of defining architectures for graph-time convolutional autoencoders.
The key point of the article is the definition of (1), where a GSO is represented by means of product graphs, yielding a representation of the dynamic data on the graph as a static signal of the product graph. The othe... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Graph-Time Convolutional Autoencoders
### Paper Abstract
We introduce graph-time convolutional autoencoder (GTConvAE), a novel spatiotemporal architecture tailored to le... | |||
B1em8TVtPr | ICLR.cc/2020/Conference | 2020 | Discourse-Based Evaluation of Language Understanding | ["Damien Sileo", "Tim Van-De-Cruys", "Camille Pradel", "Philippe Muller"] | New models for natural language understanding have made unusual progress recently, leading to claims of universal text representations. However, current benchmarks are predominantly targeting semantic phenomena; we make the case that discourse and pragmatics need to take center stage in the evaluation of natural langu... | ["Natural Language Understanding", "Pragmatics", "Discourse", "Semantics", "Evaluation", "BERT", "Natural Language Processing"] | ABSTRACTNew models for natural language understanding have made unusual progressrecently, leading to claims of universal text representations. However, currentbenchmarks are predominantly targeting semantic phenomena; we make the casethat discourse and pragmatics need to take center stage in the evaluation of nat-ural ... | H1xiUtOFFS | Official Blind Review #3 | 6: Weak Accept |
The paper presents a new GLUE-like dataset collection called DiscEval, which focuses on discourse and pragmatics. Like GLUE and SuperGLUE, datasets from existing works are collated and formated. The tasks are all classification tasks.
The paper also evaluates several baselines including bag-of-words, BiLSTM encodings... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Discourse-Based Evaluation of Language Understanding
### Paper Abstract
New models for natural language understanding have made unusual progress recently, leading to cla... | |
rklEUjR5tm | ICLR.cc/2019/Conference | 2019 | SHE2: Stochastic Hamiltonian Exploration and Exploitation for Derivative-Free Optimization | ["Haoyi Xiong", "Wenqing Hu", "Zhanxing Zhu", "Xinjian Li", "Yunchao Zhang", "Jun Huan"] | Derivative-free optimization (DFO) using trust region methods is frequently used for machine learning applications, such as (hyper-)parameter optimization without the derivatives of objective functions known. Inspired by the recent work in continuous-time minimizers, our work models the common trust region methods wit... | ["derivative-free optimization"] | ABSTRACTDerivative-free optimization (DFO) using trust region methods is frequently usedfor machine learning applications, such as (hyper-)parameter optimization with-out the derivatives of objective functions known. Inspired by the recent workin continuous-time minimizers, our work models the common trust region meth-... | SkeDMj0n2X | Additional benchmarking is needed | 4: Ok but not good enough - rejection | Derivative-free optimization is not a novel domain and your work could benefit from some accepted benchmarking practices. For instance, you can consider the Black-Box Optimization Benchmarking (BBOB) Workshop and its COCO platform which was used to test many optimization algorithms including the ones mentioned in the p... | 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
SHE2: Stochastic Hamiltonian Exploration and Exploitation for Derivative-Free Optimization
### Paper Abstract
Derivative-free optimization (DFO) using trust region metho... |
IVwXaHpiO0 | ICLR.cc/2021/Conference | 2021 | SyncTwin: Transparent Treatment Effect Estimation under Temporal Confounding | ["Zhaozhi Qian", "Yao Zhang", "Ioana Bica", "Angela Wood", "Mihaela van der Schaar"] | Estimating causal treatment effects using observational data is a problem with few solutions when the confounder has a temporal structure, e.g. the history of disease progression might impact both treatment decisions and clinical outcomes. For such a challenging problem, it is desirable for the method to be transparent... | ["treatment effect", "interpretability", "healthcare", "causal inference"] | ABSTRACTEstimating causal treatment effects using observational data is a problem with fewsolutions when the confounder has a temporal structure, e.g. the history of diseaseprogression might impact both treatment decisions and clinical outcomes. For sucha challenging problem, it is desirable for the method to be transp... | RszXAePFYc6 | - This paper develops SyncTwin, a method for causal inference with time-series data. The main selling points of the paper are explainability, a small number of donors are used to construct the control, and trustworthiness, the method can identify individuals whose ITE cannot be reliably estimated. | 4: Ok but not good enough - rejection | - pros
- The problem is well motivated. Estimating the causal effect with time-series data is a practical and important problem. Transparency is a desirable property.
- The idea connects works in econometrics to modern ML.
- cons
- High level points
- The paper proposes to use the number of donor ... | 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
SyncTwin: Transparent Treatment Effect Estimation under Temporal Confounding
### Paper Abstract
Estimating causal treatment effects using observational data is a problem... |
HygUOoC5KX | ICLR.cc/2019/Conference | 2019 | Are Generative Classifiers More Robust to Adversarial Attacks? | ["Yingzhen Li", "John Bradshaw", "Yash Sharma"] | There is a rising interest in studying the robustness of deep neural network classifiers against adversaries, with both advanced attack and defence techniques being actively developed. However, most recent work focuses on discriminative classifiers, which only model the conditional distribution of the labels given the ... | ["generative models", "adversarial attack", "defence", "detection", "Bayes' rule"] | ABSTRACTThere is a rising interest in studying the robustness of deep neural network classi-fiers against adversaries, with both advanced attack and defence techniques beingactively developed. However, most recent work focuses on discriminative classi-fiers, which only model the conditional distribution of the labels g... | B1xvkiIZgN | Interesting work but confusing presentation and some missing areas | 4: Ok but not good enough - rejection | This paper explores the potential of generative models for adversarial robustness. It presents some interesting and well formulated findings but is lacking in some meaningful ways.
-It does a good job of summarizing the literature though it misses out on some very recent but relevant work
-It introduces three detecti... | 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
Are Generative Classifiers More Robust to Adversarial Attacks?
### Paper Abstract
There is a rising interest in studying the robustness of deep neural network classifier... |
oZIvHV04XgC | ICLR.cc/2021/Conference | 2021 | Wandering within a world: Online contextualized few-shot learning | ["Mengye Ren", "Michael Louis Iuzzolino", "Michael Curtis Mozer", "Richard Zemel"] | We aim to bridge the gap between typical human and machine-learning environments by extending the standard framework of few-shot learning to an online, continual setting. In this setting, episodes do not have separate training and testing phases, and instead models are evaluated online while learning novel classes. As ... | ["Few-shot learning", "continual learning", "lifelong learning"] | ABSTRACTWe aim to bridge the gap between typical human and machine-learning environ-ments by extending the standard framework of few-shot learning to an online,continual setting. In this setting, episodes do not have separate training and testingphases, and instead models are evaluated online while learning novel class... | W4Ku2cMDVPZ | New realistic learning paradigm | 6: Marginally above acceptance threshold | Summary:
This work aims to make a realistic learning setting by combining few-shot learning and continual learning in the online setting. Similar to few-shot learning, the model needs to adapt to new classes with a few samples (at least in the beginning). Similar to continual learning, the model needs to learn new cla... | 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
Wandering within a world: Online contextualized few-shot learning
### Paper Abstract
We aim to bridge the gap between typical human and machine-learning environments by ... |
SJx371HFvr | ICLR.cc/2020/Conference | 2020 | On the Distribution of Penultimate Activations of Classification Networks | ["Minkyo Seo", "Yoonho Lee", "Suha Kwak"] | This paper considers probability distributions of penultimate activations in deep classification networks. We first identify a dual relation between the activations and the weights of the final fully connected layer: learning the networks with the cross-entropy loss makes their (normalized) penultimate activations follow ... | ["penultimate activations", "distribution", "classification networks", "activations", "weights", "\ufb01nal", "layer", "class", "generative model", "probability distributions"] | ABSTRACTThis paper considers probability distributions of penultimate activations in deepclassification networks. We first identify a dual relation between the activationsand the weights of the final fully connected layer: learning the networks with thecross-entropy loss makes their (normalized) penultimate activations... | rJeZ1dKCOH | Official Blind Review #2 | 1: Reject | The authors observe that training with the softmax loss is (approximately) maximum log-likelihood for a model in which each class-conditional distribution on the last hidden layer is a von Mises-Fisher (vMF) distribution. They claim that this shows that the class-conditional distribution is a vMF distribution. I do 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
On the Distribution of Penultimate Activations of Classification Networks
### Paper Abstract
This paper considers probability distributions of penultimate activations in... | |
SJeeL04KvH | ICLR.cc/2020/Conference | 2020 | Robust Federated Learning Through Representation Matching and Adaptive Hyper-parameters | ["Hesham Mostafa"] | Federated learning is a distributed, privacy-aware learning scenario which trains a single model on data belonging to several clients. Each client trains a local model on its data and the local models are then aggregated by a central party. Current federated learning methods struggle in cases with heterogeneous clien... | ["federated learning", "hyper-parameter tuning", "regularization"] | ABSTRACTFederated learning is a distributed, privacy-aware learning scenario which trains asingle model on data belonging to several clients. Each client trains a local modelon its data and the local models are then aggregated by a central party. Currentfederated learning methods struggle in cases with heterogeneous cl... | S1lYvJP6tH | Official Blind Review #1 | 3: Weak Reject | In this paper, the authors propose a novel representation matching scheme to reduce the divergence of local models in federated learning. In addition, the authors propose an online hyper-parameter tuning scheme. The paper is well-written. The empirical results show good performance. In overall, I think the authors prop... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Robust Federated Learning Through Representation Matching and Adaptive Hyper-parameters
### Paper Abstract
Federated learning is a distributed, privacy-aware learning ... | |
BzjLaVvr955 | robot-learning.org/CoRL/2023/Conference | 2023 | Topology-Matching Normalizing Flows for Out-of-Distribution Detection in Robot Learning | ["Jianxiang Feng", "Jongseok Lee", "Simon Geisler", "Stephan G\u00fcnnemann", "Rudolph Triebel"] | To facilitate reliable deployments of autonomous robots in the real world, Out-of-Distribution (OOD) detection capabilities are often required. A powerful approach for OOD detection is based on density estimation with Normalizing Flows (NFs). However, we find that prior work with NFs attempts to match the complex targe... | ["Normalizing Flows", "Out-of-Distribution Detection", "Robotic Introspection"] | Topology-Matching Normalizing Flows forOut-of-Distribution Detection in Robot LearningJianxiang Feng∗,1Jongseok Lee2,3, Simon Geisler1, Stephan G ̈unnemann1, Rudolph Triebel2,31Department of Informatics, Technical University of Munich (TUM)2Institute of Robotics and Mechatronics, German Aerospace Center (DLR)3Departmen... | KGz-PQ0agp6 | Strengths:
- The proposed approach balances computational complexity and performance, yielding good density modeling yet remaining practical to deploy in real-time contexts.
- The paper is well written, and the authors provide a good amount of background on the prior advances the authors build on in this work.
Weaknes... | 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-Matching Normalizing Flows for Out-of-Distribution Detection in Robot Learning
### Paper Abstract
To facilitate reliable deployments of autonomous robots in the... | ||
SJgYpoX2WH | MICCAI.org/2019/Workshop/COMPAY | 2019 | Prior Consistent CNN with Multi-Task Learning for Colon Image Classification | ["Chaoyang Yan", "Chengfei Cai", "Jiawei Xie", "Yao Fu", "Hui Shuai", "Xiangshan Fan", "Jun Xu"] | As adenocarcinoma is the most common cancer, the pathology diagnoses for it is of great significance. In the field of digital pathology, although deep learning method has achieved good results, it is theorem agnostic and the accumulated pathology-level knowledge is ignored. Specifically, the degree of gland differentia... | ["Multi-Task Learning", "Gland Prior Attention", "Colorectal Cancer", "Convolutional Neural Network"] | Prior Consistent CNN with Multi-Task Learningfor Colon Image ClassicationChaoyang Yan1, Chengfei Cai1, Jiawei Xie1, Yao Fu2, Hui Shuai1, XiangshanFan2, and Jun Xu1)1Nanjing University of Information Science & Technology, Nanjing 210044, China2Gulou Hospital, Nanjing 210008, China.Abstract. As adenocarcinoma is the most... | HJlWJ-Tk4H | Attempt to use clinical prior knowledge in colon image classification | 5: Marginally below acceptance threshold | This paper proposes a CNN-based approach with multitask learning for simultaneous segmentation and classification of colon images. It aims to incorporate prior knowledge about clinical-pathological diagnostic criteria of adenocarcinomas in grading the images. The proposed approach uses the probability map produced by t... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Prior Consistent CNN with Multi-Task Learning for Colon Image Classification
### Paper Abstract
As adenocarcinoma is the most common cancer, the pathology diagnoses for ... |
r1lgTGL5DE | ICLR.cc/2019/Workshop/drlStructPred | 2019 | Buy 4 REINFORCE Samples, Get a Baseline for Free! | ["Wouter Kool", "Herke van Hoof", "Max Welling"] | REINFORCE can be used to train models in structured prediction settings to directly optimize the test-time objective. However, the common case of sampling one prediction per datapoint (input) is data-inefficient. We show that by drawing multiple samples (predictions) per datapoint, we can learn with significantly less ... | ["reinforce", "multiple samples", "baseline", "sequence generation", "structured prediction", "travelling salesman problem"] | ABSTRACTREINFORCE can be used to train models in structured prediction settings to di-rectly optimize the test-time objective. However, the common case of samplingone prediction per datapoint (input) is data-inefficient. We show that by drawingmultiple samples (predictions) per datapoint, we can learn with significantl... | SJlR4YwSFE | Multi-sample REINFORCE estimators w/ and w/o replacement | 4: Top 50% of accepted papers, clear accept | The authors derive a REINFORCE estimator based on sampling without replacement (building on the recent work of Kool et al.) and show improvements over existing techniques on structured prediction.
The paper is well-written and the experiments are informative. The paper could be improved by theoretically quantifying th... | 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
Buy 4 REINFORCE Samples, Get a Baseline for Free!
### Paper Abstract
REINFORCE can be used to train models in structured prediction settings to directly optimize the tes... |
rkeS1RVtPS | ICLR.cc/2020/Conference | 2020 | Cyclical Stochastic Gradient MCMC for Bayesian Deep Learning | ["Ruqi Zhang", "Chunyuan Li", "Jianyi Zhang", "Changyou Chen", "Andrew Gordon Wilson"] | The posteriors over neural network weights are high dimensional and multimodal. Each mode typically characterizes a meaningfully different representation of the data. We develop Cyclical Stochastic Gradient MCMC (SG-MCMC) to automatically explore such distributions. In particular, we propose a cyclical stepsize schedul... | ["bayesian deep", "mode", "posteriors", "neural network weights", "high dimensional", "multimodal", "different representation", "data", "distributions"] | ABSTRACTThe posteriors over neural network weights are high dimensional and multimodal.Each mode typically characterizes a meaningfully different representation of thedata. We develop Cyclical Stochastic Gradient MCMC (SG-MCMC) to auto-matically explore such distributions. In particular, we propose a cyclical step-size... | S1eVwgy15B | Official Blind Review #3 | 8: Accept | The paper propose a new MCMC scheme which is demonstrated to perform well for estimating Bayesian neural networks. The key idea is to not keep lowering the step sizes, but -- at pre-specified times -- go back to large step sizes.
The paper is timely, the proposed algorithm is novel, and the theoretical analysis also s... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Cyclical Stochastic Gradient MCMC for Bayesian Deep Learning
### Paper Abstract
The posteriors over neural network weights are high dimensional and multimodal. Each mode... | |
B1e7hs05Km | ICLR.cc/2019/Conference | 2019 | Efficient Exploration through Bayesian Deep Q-Networks | ["Kamyar Azizzadenesheli", "Animashree Anandkumar"] | We propose Bayesian Deep Q-Networks (BDQN), a principled and a practical Deep Reinforcement Learning (DRL) algorithm for Markov decision processes (MDP). It combines Thompson sampling with deep-Q networks (DQN). Thompson sampling ensures more efficient exploration-exploitation tradeoff in high dimensions. It is typical... | ["Deep RL", "Exploration Exploitation", "DQN", "Bayesian Regret", "Thompson Sampling"] | ABSTRACTWe propose Bayesian Deep Q-Networks ( BDQN ), a principled and a practicalDeep Reinforcement Learning (DRL) algorithm for Markov decision processes(MDP). It combines Thompson sampling with deep- Qnetworks ( DQN ). Thompsonsampling ensures efficient exploration-exploitation trade-off in high dimensions.It is typ... | H1gi8TCPhX | Major clarity issues | 4: Ok but not good enough - rejection | Update after feedback: I would like to thank the authors for huge work done on improving the paper. I appreciate the tight time constrains given during the discussion phase and big steps towards more clear paper, but at the current stage I keep my opinion that the paper is not ready for publication. Also variability of... | 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
Efficient Exploration through Bayesian Deep Q-Networks
### Paper Abstract
We propose Bayesian Deep Q-Networks (BDQN), a principled and a practical Deep Reinforcement Lea... |
B1eoyAVFwH | ICLR.cc/2020/Conference | 2020 | Feature Partitioning for Efficient Multi-Task Architectures | ["Alejandro Newell", "Lu Jiang", "Chong Wang", "Li-Jia Li", "Jia Deng"] | Multi-task learning promises to use less data, parameters, and time than training separate single-task models. But realizing these benefits in practice is challenging. In particular, it is difficult to define a suitable architecture that has enough capacity to support many tasks while not requiring excessive compute fo... | ["multi-task learning", "neural architecture search", "multi-task architecture search"] | ABSTRACTMulti-task learning promises to use less data, parameters, and time than trainingseparate single-task models. But realizing these benefits in practice is challeng-ing. In particular, it is difficult to define a suitable architecture that has enoughcapacity to support many tasks while not requiring excessive com... | rkxwhDbbqB | Official Blind Review #1 | 3: Weak Reject | This paper proposes a framework for learning multi-task convolutional neural networks. For each layer of the network, the proposed algorithm assigns a subset of the layer's channels to each of the tasks. This is in contrast to existing methods that assign whole layers to tasks. There are two key ideas here: (1) instead... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Feature Partitioning for Efficient Multi-Task Architectures
### Paper Abstract
Multi-task learning promises to use less data, parameters, and time than training separate... | |
rumv7QmLUue | ICLR.cc/2021/Conference | 2021 | A Gradient Flow Framework For Analyzing Network Pruning | ["Ekdeep Singh Lubana", "Robert P. Dick"] | Recent network pruning methods focus on pruning models early-on in training. To estimate the impact of removing a parameter, these methods use importance measures that were originally designed to prune trained models. Despite lacking justification for their use early-on in training, such measures result in surprisingly... | ["Network pruning", "Gradient flow", "Early pruning"] | ABSTRACTRecent network pruning methods focus on pruning models early-on in training.To estimate the impact of removing a parameter, these methods use importancemeasures that were originally designed to prune trained models. Despite lackingjustification for their use early-on in training, such measures result in surpris... | 1DY7M6s1lU2 | Use of gradient flow to derive interesting relationship and interpretations for pruning early | 7: Good paper, accept | The paper contributes to explaining why saliency measures used for pruning trained models may (or may not) also be effective for pruning untrained or minimally trained models, by developing the relationship between those saliency measures and different forms of the norm of model parameters based on the evolution of mod... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
A Gradient Flow Framework For Analyzing Network Pruning
### Paper Abstract
Recent network pruning methods focus on pruning models early-on in training. To estimate the i... |
rJfMusFll | ICLR.cc/2017/conference | 2017 | Batch Policy Gradient Methods for Improving Neural Conversation Models | ["Kirthevasan Kandasamy", "Yoram Bachrach", "Ryota Tomioka", "Daniel Tarlow", "David Carter"] | We study reinforcement learning of chat-bots with recurrent neural network
architectures when the rewards are noisy and expensive to
obtain. For instance, a chat-bot used in automated customer service support can
be scored by quality assurance agents, but this process can be expensive, time consuming
and noisy.
Previo... | ["Natural language processing", "Reinforcement Learning"] | ABSTRACTWe study reinforcement learning of chatbots with recurrent neural network archi-tectures when the rewards are noisy and expensive to obtain. For instance, a chat-bot used in automated customer service support can be scored by quality assuranceagents, but this process can be expensive, time consuming and noisy. ... | H1bSmrx4x | 7: Good paper, accept | The author propose to use a off-policy actor-critic algorithm in a batch-setting to improve chat-bots.
The approach is well motivated and the paper is well written, except for some intuitions for why the batch version outperforms the on-line version (see comments on "clarification regarding batch vs. online setting").
... | 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
Batch Policy Gradient Methods for Improving Neural Conversation Models
### Paper Abstract
We study reinforcement learning of chat-bots with recurrent neural network ar... | |
uELnyih9gqb | ICLR.cc/2021/Conference | 2021 | WAVEQ: GRADIENT-BASED DEEP QUANTIZATION OF NEURAL NETWORKS THROUGH SINUSOIDAL REGULARIZATION | ["Ahmed T. Elthakeb", "Prannoy Pilligundla", "Tarek Elgindi", "Fatemehsadat Mireshghallah", "Charles-Alban Deledalle", "Hadi Esmaeilzadeh"] | Deep quantization of neural networks below eight bits can lead to superlinear benefits in storage and compute efficiency. However, homogeneously quantizing all the layers to the same level does not account for the distinction of the layers and their individual properties. Heterogenous assignment of bitwidths to individ... | ["waveq", "layers", "sinusoidal regularizer", "neural networks", "bitwidth", "period", "deep quantization", "quantized weights", "minima", "quantization levels"] | ABSTRACTDeep quantization of neural networks below eight bits can lead to superlinear ben-efits in storage and compute efficiency. However, homogeneously quantizing allthe layers to the same level does not account for the distinction of the layers andtheir individual properties. Heterogeneous assignment of bitwidths to... | DnXIqb4p1S2 | Interesting Motivation but Unfortunately Unsatisfactory Empirical and Theoretical Results | 4: Ok but not good enough - rejection | The paper proposes WaveQ which proposes a sinusoidal regularization approach for quantizing Neural Networks. The motivation is to enable mixed-precision quantization, since homogeneously quantizing the model to low precision can lead to accuracy degradation.
The main problem with heterogeneous/mixed-precision bit-widt... | 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
WAVEQ: GRADIENT-BASED DEEP QUANTIZATION OF NEURAL NETWORKS THROUGH SINUSOIDAL REGULARIZATION
### Paper Abstract
Deep quantization of neural networks below eight bits can... |
J_pvI6ap5Mn | ICLR.cc/2021/Conference | 2021 | Transfer Learning of Graph Neural Networks with Ego-graph Information Maximization | ["Qi Zhu", "Yidan Xu", "Haonan Wang", "Chao Zhang", "Jiawei Han", "Carl Yang"] | Graph neural networks (GNNs) have been shown with superior performance in various applications, but training dedicated GNNs can be costly for large-scale graphs. Some recent work started to study the pre-training of GNNs. However, none of them provide theoretical insights into the design of their frameworks, or clear r... | ["Transfer learning", "graph neural networks"] | ABSTRACTGraph neural networks (GNNs) have been shown with superior performance invarious applications, but training dedicated GNNs can be costly for large-scalegraphs. Some recent work started to study the pre-training of GNNs. However,none of them provide theoretical insights into the design of their frameworks, orcle... | qG05RqeoeCq | interesting results but a limited coverage | 6: Marginally above acceptance threshold | The authors proposed a transfer learning scheme for graph neural networks. The proposed method ego-graph information maximization allows learning transferable models. The authors studied structure-respecting node features and provided a theoretical analysis of the transferability of GNNs. The proposed method significan... | 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
Transfer Learning of Graph Neural Networks with Ego-graph Information Maximization
### Paper Abstract
Graph neural networks (GNNs) have been shown with superior performa... |
Syzn9i05Ym | ICLR.cc/2019/Conference | 2019 | Learning Neural Random Fields with Inclusive Auxiliary Generators | ["Yunfu Song", "Zhijian Ou"] | Neural random fields (NRFs), which are defined by using neural networks to implement potential functions in undirected models, provide an interesting family of model spaces for machine learning. In this paper we develop a new approach to learning NRFs with inclusive-divergence minimized auxiliary generator - the inclus... | ["Neural random fields", "Deep generative models", "Unsupervised learning", "Semi-supervised learning"] | ABSTRACTNeural random fields (NRFs), which are defined by using neural networks toimplement potential functions in undirected models, provide an interesting familyof model spaces for machine learning. In this paper we develop a new approachto learning NRFs with inclusive-divergence minimized auxiliary generator - thein... | r1g2osJOnX | Incremental Contribution | 5: Marginally below acceptance threshold | The paper proposes the inclusive neural random field model. Compared the existing work, the model is different because of the use of the inclusive-divergence minimization for the generative model and the use of stochastic gradient Langevin dynamics (SGLD) and stochastic gradient Hamiltonian Monte Carlo (SGHMC) for sam... | 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 Neural Random Fields with Inclusive Auxiliary Generators
### Paper Abstract
Neural random fields (NRFs), which are defined by using neural networks to implement... |
5JnS8wROG9 | ICLR.cc/2021/Conference | 2021 | On the Inductive Bias of a CNN for Distributions with Orthogonal Patterns | ["Alon Brutzkus", "Amir Globerson"] | Training overparameterized convolutional neural networks with gradient based optimization is the most successful learning method for image classification. However, their generalization properties are far from understood. In this work, we consider a simplified image classification task where images contain orthogonal pa... | ["Deep learning theory", "generalization", "overparemeterization", "CNN"] | ABSTRACTTraining overparameterized convolutional neural networks with gradient basedoptimization is the most successful learning method for image classification.However, their generalization properties are far from understood. In this work,we consider a simplified image classification task where images contain orthog-o... | 9J7ST3q2a5 | Restricted model and complex measures make this work less attractive | 5: Marginally below acceptance threshold | This paper studies a new theoretical framework to understand the ability of
ConvNets to deal with pattern recognition tasks. The authors suggest a new property
of convents where filters have large dot products with patterns occurring in images classified as 1 and small dot products with patterns appearing with images... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Inductive Bias of a CNN for Distributions with Orthogonal Patterns
### Paper Abstract
Training overparameterized convolutional neural networks with gradient based... |
SkexAREFDH | ICLR.cc/2020/Conference | 2020 | Learning Sparsity and Quantization Jointly and Automatically for Neural Network Compression via Constrained Optimization | ["Haichuan Yang", "Shupeng Gui", "Yuhao Zhu", "Ji Liu"] | Deep Neural Networks (DNNs) are widely applied in a wide range of usecases. There is an increased demand for deploying DNNs on devices that do not have abundant resources such as memory and computation units. Recently, network compression through a variety of techniques such as pruning and quantization have been propos... | ["model compression", "pruning", "quantization", "autoML"] | WITHDRAWAnonymous authors | rkeZMIJ8YH | Official Blind Review #3 | 3: Weak Reject | This paper introduces an approach to jointly pruning and quantizing a deep network so as to satisfy a global budget constraint. This is formulated as an optimization problem with a constraint involving the number of nonzero elements in the network weights and the bitwidth to encode these nonzero elements. A solution to... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Learning Sparsity and Quantization Jointly and Automatically for Neural Network Compression via Constrained Optimization
### Paper Abstract
Deep Neural Networks (DNNs) a... | |
SDyj8aZBPrs | graphicsinterface.org/Graphics_Interface/2022/Conference | 2022 | An Empirical Study on the Effect of Quick and Careful Labeling Styles in Image Annotation | ["Chia-Ming Chang", "Xi Yang", "Takeo Igarashi"] | Assigning a label to difficult data requires a long time, particularly when non-expert annotators attempt to select the best possible label. However, there have been no detailed studies exploring a label selection style during annotation. This is very important and may affect the efficiency and quality of annotation. I... | ["Cognitive Psychology", "Labeling Style", "Non-Expert Data Annotation", "Data Collection", "Machine Learning"] | ABSTRACT Assigning a label to difficult data requires a long time, particularly when non-expert annotators attempt to select the best possible label. However, there have been no detailed studies exploring a label selection style during annotation. This is very important and may affect the efficiency and quality of anno... | WG8RVf0YK_ | An interesting research that is also an important in the field of non-experts data annotation. | 7: Good paper, accept | The paper discusses the human cognitive process in data labelling tasks for later use in machine learning. The authors describe their study in determining two distinctive styles of labelling behaviour and compared the accuracy performance from training the same structured classifiers.
The topic of study is interesting,... | 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 Empirical Study on the Effect of Quick and Careful Labeling Styles in Image Annotation
### Paper Abstract
Assigning a label to difficult data requires a long time, pa... |
r1lfga4KvS | ICLR.cc/2020/Conference | 2020 | Extreme Value k-means Clustering | ["Sixiao Zheng", "Yanxi Hou", "Yanwei Fu", "Jianfeng Feng"] | Clustering is the central task in unsupervised learning and data mining. k-means is one of the most widely used clustering algorithms. Unfortunately, it is generally non-trivial to extend k-means to cluster data points beyond Gaussian distribution, particularly, the clusters with non-convex shapes (Beliakov & King, 200... | ["unsupervised learning", "clustering", "k-means", "Extreme Value Theory"] | ABSTRACTClustering is the central task in unsupervised learning and data mining. k-meansis one of the most widely used clustering algorithms. Unfortunately, it is generallynon-trivial to extend k-means to cluster data samples beyond Gaussian distribu-tion, particularly, the clusters with non-convex shape (Beliakov & Ki... | SJglh4GrcB | Official Blind Review #3 | 6: Weak Accept | This paper utilizes EVT to improve k-means, which aims to address the problem of clustering data of nonconvex shape. GEV k-means and GPD k-means are proposed as two kind of Extreme Value k-means. A method for accelerating Euclidean distance computation has also been proposed to solve the bottleneck of k-means.
The pro... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Extreme Value k-means Clustering
### Paper Abstract
Clustering is the central task in unsupervised learning and data mining. k-means is one of the most widely used clust... | |
_O9YLet0wvN | ICLR.cc/2021/Conference | 2021 | Closing the Generalization Gap in One-Shot Object Detection | ["Claudio Michaelis", "Matthias Bethge", "Alexander S Ecker"] | Despite substantial progress in object detection and few-shot learning, detecting objects based on a single example - one-shot object detection - remains a challenge. A central problem is the generalization gap: Object categories used during training are detected much more reliably than novel ones. We here show that th... | ["One-Shot Learning", "Few-Shot Learning", "Object Detection", "One-Shot Object Detection", "Generalization"] | ABSTRACTDespite substantial progress in object detection and few-shot learning, detectingobjects based on a single example – one-shot object detection – remains a chal-lenge. A central problem is the generalization gap: Object categories used duringtraining are detected much more reliably than novel ones. We here show ... | ppwj8hWq7ZV | I appreciate the proposed hypothesis, but my major concerns are the experiments. Please refer to my comments. | 6: Marginally above acceptance threshold | Pros.:
1. The hypothesis is interesting and novel
2. The paper is clearly presented and well-organized.
3. The authors conduct many ablation studies to validate the proposed hypothesis.
Cons:
1. Claimed observation:
In your claimed observation, increasing the number of classes could improve performance. However, incr... | 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
Closing the Generalization Gap in One-Shot Object Detection
### Paper Abstract
Despite substantial progress in object detection and few-shot learning, detecting objects ... |
HJgkj0NFwr | ICLR.cc/2020/Conference | 2020 | Differentiable Architecture Compression | ["Shashank Singh", "Ashish Khetan", "Zohar Karnin"] | In many learning situations, resources at inference time are significantly more constrained than resources at training time. This paper studies a general paradigm, called Differentiable ARchitecture Compression (DARC), that combines model compression and architecture search to learn models that are resource-efficient a... | ["resources", "inference time", "darc", "accuracy", "memory footprint", "accuracy loss", "many learning situations", "constrained", "training time"] | ABSTRACTIn many learning situations, resources at inference time are much more con-strained than resources at training time. This paper studies a general paradigm,called Differentiable ARchitecture Compression (DARC), that combines modelcompression and architecture search to learn models that are resource-efficient ati... | Hyl64KDMqH | Official Blind Review #3 | 6: Weak Accept | This paper addresses the problem of limited resources that may generally be available at inference time (as compared to the lack of constraints during training time). The paper addresses architecture search as well as model compression by simultaneously optimizing all layers/weights/submodules of the network. The appro... | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Differentiable Architecture Compression
### Paper Abstract
In many learning situations, resources at inference time are significantly more constrained than resources at ... | |
-u4j4dHeWQi | ICLR.cc/2021/Conference | 2021 | Explore with Dynamic Map: Graph Structured Reinforcement Learning | ["Jiarui Jin", "Sijin Zhou", "Weinan Zhang", "Rasool Fakoor", "David Wipf", "Tong He", "Yong Yu", "Zheng Zhang", "Alex Smola"] | In reinforcement learning, a map with states and transitions built based on historical trajectories is often helpful in exploration and exploitation. Even so, learning and planning on such a map within a sparse environment remains a challenge. As a step towards this goal, we propose Graph Structured Reinforcement Learn... | ["Deep Reinforcement Learning", "Graph Structured Reinforcement Learning", "Exploration"] | ABSTRACTIn reinforcement learning, a map with states and transitions built using historicaltrajectories is often helpful in exploration and exploitation. Even so, learning andplanning on such a map within a sparse environment remains a challenge. Asa step towards this goal, we propose Graph Structured Reinforcement Lea... | eP4ecXXLZVs | Official Blind Review #1 | 6: Marginally above acceptance threshold | This paper introduces Graph Structured Reinforcement Learning (GSRL) framework, able to balance exploration and exploitation in RL. Actually, GSRL builds a dynamic graph based on historical trajectories. Then in order to learn from sparse or delayed rewards and be able to reach a distant goal, it decomposes the main t... | 3: The reviewer is fairly confident that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Explore with Dynamic Map: Graph Structured Reinforcement Learning
### Paper Abstract
In reinforcement learning, a map with states and transitions built based on historic... |
ryF7rTqgl | ICLR.cc/2017/conference | 2017 | Understanding intermediate layers using linear classifier probes | ["Guillaume Alain", "Yoshua Bengio"] | Neural network models have a reputation for being black boxes. We propose a new method to better understand the roles and dynamics of the intermediate layers. This has direct consequences on the design of such models and it enables the expert to be able to justify certain heuristics (such as adding auxiliary losses in ... | ["intermediate layers", "models", "probes", "model", "linear classifier probes", "reputation", "black boxes", "new", "better"] | ABSTRACTNeural network models have a reputation for being black boxes. We proposea new method to better understand the roles and dynamics of the intermediatelayers. This has direct consequences on the design of such models and it enablesthe expert to be able to justify certain heuristics (such as adding auxiliary losse... | HJ6BKc6Qg | Interesting problem, but technically and experimentally not solid enough | 4: Ok but not good enough - rejection | This paper proposes to use a linear classifier as the probe for the informativeness of the hidden activations from different neural network layers. The training of the linear classifier does not affect the training of the neural network.
The paper is well motivated for investigating how much useful information (or ho... | 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
Understanding intermediate layers using linear classifier probes
### Paper Abstract
Neural network models have a reputation for being black boxes. We propose a new metho... |
WWxviNxYwoc | graphicsinterface.org/Graphics_Interface/2020/Conference | 2020 | Presenting Information Closer to Mobile Crane Operators’ Line of Sight: Designing and Evaluating Visualization Concepts Based on Transparent Displays | ["Taufik Akbar Sitompul", "Rikard Lindell", "Markus Wallmyr", "Antti Siren"] | We have investigated the visualization of safety information for mobile crane operations utilizing transparent displays, where the information can be presented closer to operators’ line of sight with minimum obstruction on their view. The intention of the design is to help operators in acquiring supportive information ... | ["human-machine interface", "transparent display", "mobile cranes", "heavy machinery", "design process"] | ABSTRACTWe have investigated the visualization of safety information formobile crane operations utilizing transparent displays, where theinformation can be presented closer to operators’ line of sight withminimum obstruction on their view. The intention of the designis to help operators in acquiring supportive informat... | pf3zhSYY-Gl | Interesting first step towards addressing information dispersion in mobile cranes | 6: Marginally above acceptance threshold | The authors have prototyped potential transparent display based solution to bringing critical information closer to the person operating a mobile crane. The authors identified a problem of information disconnect wherein the operators are looking out through the glass to move the crane and all the supplementary informat... | 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
Presenting Information Closer to Mobile Crane Operators’ Line of Sight: Designing and Evaluating Visualization Concepts Based on Transparent Displays
### Paper Abstract
... |
ByD6xlrFe | ICLR.cc/2017/workshop | 2017 | Hybrid Neural Networks over Time Series for Trend Forecasting | ["Tao Lin", "Tian Guo", "Karl Aberer"] | The trend of time series characterize the intermediate upward and downward patterns of time series. Learning and forecasting the trend in time series data play an important role in many real applications, ranging from resource allocation in data centers, load schedule in smart grid and so on. Inspired by the recent suc... | ["time series", "trend", "trenet", "lstm", "hybrid neural networks", "cnn", "intermediate upward", "downward patterns", "time series data", "important role"] | ABSTRACTThe trend of time series characterize the intermediate upward and downward pat-terns of time series. Learning and forecasting the trend in time series data playan important role in many real applications, ranging from resource allocation indata centers and load schedule in smart grid. Inspired by the recent suc... | S13x6Smil | Paper was not invited for submission to workshop tract | 4: Ok but not good enough - rejection | This paper was not invited for submission to the workshop track of ICLR and suffers from the same limitations as the main paper submission: ad-hoc local trend computation. The authors merely shifted parts of the paper to an appendix to fit into 3 pages.
| 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
Hybrid Neural Networks over Time Series for Trend Forecasting
### Paper Abstract
The trend of time series characterize the intermediate upward and downward patterns of t... |
pwwVuSICBgt | ICLR.cc/2021/Conference | 2021 | Enabling Binary Neural Network Training on the Edge | ["Erwei Wang", "James J. Davis", "Daniele Moro", "Piotr Zielinski", "Claudionor Coelho", "Satrajit Chatterjee", "Peter Y. K. Cheung", "George Anthony Constantinides"] | The ever-growing computational demands of increasingly complex machine learning models frequently necessitate the use of powerful cloud-based infrastructure for their training. Binary neural networks are known to be promising candidates for on-device inference due to their extreme compute and memory savings over higher... | ["Binary neural network", "edge computing", "neural network training"] | ABSTRACTThe ever-growing computational demands of increasingly complex machine learn-ing models frequently necessitate the use of powerful cloud-based infrastructurefor their training. Binary neural networks are known to be promising candidatesfor on-device inference due to their extreme compute and memory savings over... | -n9jdfNsfqF | I think this paper is very interesting and if some minor comments are addressed, it should be accepted for presentation at ICLR21 | 8: Top 50% of accepted papers, clear accept | I think this a very good contribution to ICLR given the topic and the quality of the submission (originality, contribution to the state of the art, experimental evidence, e
Some of the strong points of the submission are summarized as follows, along with some points for clarification
1. Enabling training on any embe... | 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
Enabling Binary Neural Network Training on the Edge
### Paper Abstract
The ever-growing computational demands of increasingly complex machine learning models frequently ... |
rsf1z-JSj87 | ICLR.cc/2021/Conference | 2021 | End-to-end Adversarial Text-to-Speech | ["Jeff Donahue", "Sander Dieleman", "Mikolaj Binkowski", "Erich Elsen", "Karen Simonyan"] | Modern text-to-speech synthesis pipelines typically involve multiple processing stages, each of which is designed or learnt independently from the rest. In this work, we take on the challenging task of learning to synthesise speech from normalised text or phonemes in an end-to-end manner, resulting in models which oper... | ["text-to-speech", "speech synthesis", "adversarial", "GAN", "end-to-end", "feed-forward", "generative model"] | ABSTRACTModern text-to-speech synthesis pipelines typically involve multiple processingstages, each of which is designed or learnt independently from the rest. In thiswork, we take on the challenging task of learning to synthesise speech fromnormalised text or phonemes in an end-to-end manner, resulting in models which... | KAiTv-9pk4 | Review of End-to-end adversarial text-to-speech | 8: Top 50% of accepted papers, clear accept |
## Summary
The authors propose EATS, a method for TTS from unaligned audio and text data, directly to the waveform. Previous work either use aligned phonetic features, or output spectrograms that are later converted to a waveform by a deep vocoder model.
In order to achieve this, the authors had to use several trick... | 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
End-to-end Adversarial Text-to-Speech
### Paper Abstract
Modern text-to-speech synthesis pipelines typically involve multiple processing stages, each of which is designe... |
BJlhEs09YQ | ICLR.cc/2019/Conference | 2019 | End-to-end Learning of a Convolutional Neural Network via Deep Tensor Decomposition | ["Samet Oymak", "Mahdi Soltanolkotabi"] | In this paper we study the problem of learning the weights of a deep convolutional neural network. We consider a network where convolutions are carried out over non-overlapping patches with a single kernel in each layer. We develop an algorithm for simultaneously learning all the kernels from the training data. Our app... | ["convolutional neural network", "tensor decomposition", "sample complexity", "approximation"] | ABSTRACTIn this paper we study the problem of learning the weights of a deep convolutionalneural network. We consider a network where convolutions are carried out over non-overlapping patches with a single kernel in each layer. We develop an algorithm forsimultaneously learning all the kernels from the training data. O... | SygjJH2hh7 | Recovery of true weights for a special class of CNNs by reduction to tensor decomposition | 5: Marginally below acceptance threshold | Summary:
This paper analyzes the problem of learning a very special class of CNNs: each layer consists of a single filter, applied to non-overlapping patches of the input. Specifically, the task is: given data generated from a planted CNN of this kind with gaussian inputs, recover the ground truth weights. In the main... | 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
End-to-end Learning of a Convolutional Neural Network via Deep Tensor Decomposition
### Paper Abstract
In this paper we study the problem of learning the weights of a de... |
cfqJY583gim | ICLR.cc/2021/Conference | 2021 | On the Benefits of Early Fusion in Multimodal Representation Learning | ["Sabera J Talukder", "George Barnum", "Yisong Yue"] | Intelligently reasoning about the world often requires integrating data from multiple modalities, as any individual modality may contain unreliable or incomplete information. Prior work in multimodal learning fuses input modalities only after significant independent processing. On the other hand, the brain performs mul... | ["Multimodal Learning", "Representations", "Noise", "Audio", "Visual", "Fusion", "Biologically Inspired"] | ABSTRACTIntelligently reasoning about the world often requires integrating data from mul-tiple modalities, as any individual modality may contain unreliable or incompleteinformation. Prior work in multimodal learning fuses input modalities only af-ter significant independent processing. On the other hand, the brain per... | clpTWecycMy | Study of early fusion of different modalities, with questionable experimental settings | 4: Ok but not good enough - rejection | This paper studies early fusion with the introduction of a convolutional LSTM network architecture and compares it to intermediate and fully connected layer fusion. Experiments were conducted on a data set of combining two separate image and audio data sets, in addition to adding white noise to the data. Results show t... | 4: The reviewer is confident but not absolutely certain that the evaluation is correct | <|im_start|>system
You are a helpful assistant that reviews papers and provides feedback on their quality.<|im_end|>
<|im_start|>user
### Paper Title
On the Benefits of Early Fusion in Multimodal Representation Learning
### Paper Abstract
Intelligently reasoning about the world often requires integrating data from mul... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.