venue stringclasses 2
values | paper_content stringlengths 7.54k 83.7k | prompt stringlengths 161 2.5k | format stringclasses 5
values | review stringlengths 293 9.84k |
|---|---|---|---|---|
ICLR | Title
On Representation Learning in the First Layer of Deep CNNs and the Dynamics of Gradient Descent
Abstract
It has previously been reported that the representation that is learned in the first layer of deep CNNs is very different from the initial representation and highly consistent across initialization and archite... | 1. What is the focus of the paper regarding the first layer representations learned by CNNs?
2. What are the strengths and weaknesses of the proposed approach in analyzing the filters?
3. Do you have any concerns or questions regarding the research scope and its limitations?
4. How does the reviewer assess the clarity,... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper looks more closely at the first layer representations learned by CNNs in more detail than has been done in previous studies, where it has been observed that irrespective of the architecture the first layer filters all seem qualitatively similar. The authors propose to compare the similar... |
ICLR | Title
Bridging Nonlinearities and Stochastic Regularizers with Gaussian Error Linear Units
Abstract
We propose the Gaussian Error Linear Unit (GELU), a high-performing neural network activation function. The GELU nonlinearity is the expected transformation of a stochastic regularizer which randomly applies the identity... | 1. What is the novelty and significance of the proposed regularizer in deep learning?
2. How does the reviewer assess the performance of the proposed method compared to other related works?
3. Is there an interesting or new insight provided by the paper regarding nonlinearities and stochastic regularizers? | Review | Review
The proposed regularizer seems to be a particular combination of existing methods. Though the implied connection between nonlinearities and stochastic regularizers is intriguing, in my opinion the empirical performance does not exceed the performance achieved by similar methods by a large enough margin to arrive... |
ICLR | Title
Bridging Nonlinearities and Stochastic Regularizers with Gaussian Error Linear Units
Abstract
We propose the Gaussian Error Linear Unit (GELU), a high-performing neural network activation function. The GELU nonlinearity is the expected transformation of a stochastic regularizer which randomly applies the identity... | 1. What is the main contribution of the paper, and how does it differ from other related works?
2. What are the strengths and weaknesses of the proposed method, particularly in terms of its ability to train neural networks without traditional nonlinearity?
3. How does the reviewer assess the experimental results presen... | Review | Review
The method proposed essential trains neural networks without a traditional nonlinearity, using multiplicative gating by the CDF of a Gaussian evaluated at the preactivation; this is motivated as a relaxation of a probit-Bernoulli stochastic gate. Experiments are performed with both.
The work is somewhat novel a... |
ICLR | Title
Bridging Nonlinearities and Stochastic Regularizers with Gaussian Error Linear Units
Abstract
We propose the Gaussian Error Linear Unit (GELU), a high-performing neural network activation function. The GELU nonlinearity is the expected transformation of a stochastic regularizer which randomly applies the identity... | 1. How does the proposed approach differ from existing methods like adaptive dropout?
2. What are the strengths and weaknesses of the proposed approach compared to other methods in terms of functionality and experimental validation?
3. Is the proposed approach a novel solution or an improvement over existing methods?
4... | Review | Review
Approaches like adaptive dropout also have the binary mask as a function of input to a neuron very similar to the proposed approach. It is not clear, even from the new draft, how the proposed approach differs to Adaptive dropout in terms of functionality. The experimental validation is also not extensive since c... |
ICLR | Title
Scenario-based Question Answering with Interacting Contextual Properties
Abstract
In the scenario-based Question Answering (QA) task, models are asked to find answers that are appropriate to the user scenarios associated with the question and identify information that is missing from the scenarios but is necessar... | 1. What is the focus and contribution of the paper on question answering?
2. What are the strengths of the proposed approach, particularly in terms of its ability to reason and infer conditions?
3. What are the weaknesses of the paper, especially regarding the experimental section?
4. How does the reviewer assess the c... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper looks at QA when questions are asked in a given scenario, and they can be answered only if providing the model information about the scenario. In addition, such questions require a high level of reasoning, and thus the model should also be able to infer how conditions interact with each o... |
ICLR | Title
Scenario-based Question Answering with Interacting Contextual Properties
Abstract
In the scenario-based Question Answering (QA) task, models are asked to find answers that are appropriate to the user scenarios associated with the question and identify information that is missing from the scenarios but is necessar... | 1. What is the main contribution of the paper in question-answering?
2. What are the strengths of the proposed model, particularly in its components?
3. What are the weaknesses of the paper regarding its experiments and error analysis?
4. How does the reviewer assess the clarity, quality, novelty, and reproducibility o... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper proposes a model to tackle scenario-based question-answering to predict the answer to a question along with unsatisfied conditions for the given user scenario. The proposed model comprises 3 components, an entailment module (to identify the condition), a reasoning module (that decides whe... |
ICLR | Title
Scenario-based Question Answering with Interacting Contextual Properties
Abstract
In the scenario-based Question Answering (QA) task, models are asked to find answers that are appropriate to the user scenarios associated with the question and identify information that is missing from the scenarios but is necessar... | 1. What is the focus of the paper regarding question-answering scenarios?
2. What are the strengths of the proposed approach, particularly in its components and empirical advances?
3. What are the weaknesses of the paper, such as the need for clearer articulation of challenges or design choice justifications?
4. How do... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper presents a method for answering questions about scenarios -- questions for which there isn’t a fixed answer but is varied depending on additional conditions that are unstated in the text.
Method: Given a scenario, a question, and a set of conditions (extracted from an input text) that are... |
ICLR | Title
Prediction, Consistency, Curvature: Representation Learning for Locally-Linear Control
Abstract
Many real-world sequential decision-making problems can be formulated as optimal control with high-dimensional observations and unknown dynamics. A promising approach is to embed the high-dimensional observations into ... | 1. What is the main contribution of the paper regarding low-dimensional representations for control purposes?
2. What are the strengths of the proposed PCC-Loss function and variational PCC method?
3. What are the limitations of the paper, particularly in terms of practical applications and comparisons with other metho... | Review | Review
This paper considers learning low-dimensional representations from high-dimensional observations for control purposes. The authors extend the E2C framework by introducing the new PCC-Loss function. This new loss function aims to reflect the prediction in the observation space, the consistency between latent and ... |
ICLR | Title
Prediction, Consistency, Curvature: Representation Learning for Locally-Linear Control
Abstract
Many real-world sequential decision-making problems can be formulated as optimal control with high-dimensional observations and unknown dynamics. A promising approach is to embed the high-dimensional observations into ... | 1. What is the focus of the paper regarding policy learning for dynamic control problems?
2. What are the strengths of the proposed regularization strategy, particularly in its design principles?
3. Do you have any concerns or questions regarding the approach, such as its notation, proof details, or practical implement... | Review | Review
This work proposes a regularization strategy for learning optimal policy for a dynamic control problem in a latent low-dimensional domain. The work is based on LCE approach, but with in-depth analysis on how to choose/design the regularization for the \hat{P} operator, which consists of an encoder, a decoder, an... |
ICLR | Title
Prediction, Consistency, Curvature: Representation Learning for Locally-Linear Control
Abstract
Many real-world sequential decision-making problems can be formulated as optimal control with high-dimensional observations and unknown dynamics. A promising approach is to embed the high-dimensional observations into ... | 1. What is the focus of the paper in terms of the problem it addresses?
2. What are the desired properties of the latent representation for LLC algorithms according to the authors?
3. How does the proposed learning framework satisfy these desired properties?
4. Are there any concerns or suggestions regarding the clarit... | Review | Review
This paper considers from a high level the problem of learning a latent representation of high dimensional observations with underlying dynamics for control. The authors specifically describe some desiredata for latent representations for LLC algorithms. The authors rigorously construct a learning framework tha... |
ICLR | Title
Hadamard Product for Low-rank Bilinear Pooling
Abstract
Bilinear models provide rich representations compared with linear models. They have been applied in various visual tasks, such as object recognition, segmentation, and visual question-answering, to get state-of-the-art performances taking advantage of the ex... | 1. What are the main contributions of the paper regarding low-rank bilinear pooling?
2. What are the strengths of the paper, particularly in providing new insights into element-wise multiplication?
3. What are the weaknesses of the paper, especially regarding the comparison with compact bilinear pooling and statistical... | Review | Review
Summary: The paper presents low-rank bilinear pooling that uses Hadamard product (commonly known as element-wise multiplication). The paper implements low-rank bilinear pooling on an existing model (Kim et al., 2016b) and builds a model for Visual Question Answering (VQA) that outperforms the current state-of-ar... |
ICLR | Title
Hadamard Product for Low-rank Bilinear Pooling
Abstract
Bilinear models provide rich representations compared with linear models. They have been applied in various visual tasks, such as object recognition, segmentation, and visual question-answering, to get state-of-the-art performances taking advantage of the ex... | 1. What is the focus of the paper regarding bilinear pooling and its approximation?
2. What are the strengths of the proposed approach in terms of experimental evaluation and novelty?
3. What are the weaknesses of the paper, particularly in terms of comparison with other works and theoretical analysis?
4. How does the ... | Review | Review
This work proposes to approximate the bilinear pooling (outer product) with a formulation which uses the Hadamard Product (element-wise product).
This formulation is evaluated on the visual question answering (VQA) task together with several other model variants.
Strength:
1. The paper discusses how the Hadama... |
ICLR | Title
Hadamard Product for Low-rank Bilinear Pooling
Abstract
Bilinear models provide rich representations compared with linear models. They have been applied in various visual tasks, such as object recognition, segmentation, and visual question-answering, to get state-of-the-art performances taking advantage of the ex... | 1. What are the strengths and contributions of the paper's approach to the VQA task?
2. What are the limitations and areas for improvement in the proposed method?
3. Can the authors provide more explanation or justification for their choices of hyperparameters, such as the embedding dimension and output dimension?
4. H... | Review | Review
Results on the VQA task are good for this simple model, the ablation study of table 1 gives some insights as to what is important.
Missing are some explanations about the language embedding and the importance in deciding embedding dimension and final output dimension, equivalent to deciding the projected dimen... |
ICLR | Title
NeuralStagger: accelerating physics constrained neural PDE solver with spatial-temporal decomposition
Abstract
Neural networks have shown great potential in accelerating the solution of partial differential equations (PDEs). Recently, there has been a growing interest in introducing physics constraints into train... | 1. What is the focus and contribution of the paper regarding fluid system dynamics prediction?
2. What are the strengths and weaknesses of the proposed method, particularly in terms of spatial and temporal staggering?
3. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the paper's cont... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes to address the challenge of learning models to predict the dynamics of fluid systems with high resolutions by interleaving discretization points spatially and temporally. I.e., under the assumption that solution is sufficiently continuous, multiple evaluations of a network can b... |
ICLR | Title
NeuralStagger: accelerating physics constrained neural PDE solver with spatial-temporal decomposition
Abstract
Neural networks have shown great potential in accelerating the solution of partial differential equations (PDEs). Recently, there has been a growing interest in introducing physics constraints into train... | 1. What is the focus and contribution of the paper regarding neural PDE solvers?
2. What are the strengths and weaknesses of the proposed approach, particularly in comparison to existing ideas in numerical methods for solving PDEs?
3. Do you have any concerns or questions about the paper's presentation, such as the cho... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper aims to speed up neural PDE solvers. The strategy is to decompose the spatial and temporal domains into smaller problems and use multiple neural networks to learn solutions to smaller problems in parallel. The paper presented results on Navier-Stokes equations.
Strengths And Weaknesses
... |
ICLR | Title
NeuralStagger: accelerating physics constrained neural PDE solver with spatial-temporal decomposition
Abstract
Neural networks have shown great potential in accelerating the solution of partial differential equations (PDEs). Recently, there has been a growing interest in introducing physics constraints into train... | 1. What is the focus and contribution of the paper on NeuralStagger?
2. What are the strengths of the proposed approach, particularly in terms of its simplicity and applicability to general grid-based simulations?
3. What are the weaknesses of the paper, especially regarding experiment evaluation and scalability?
4. Ho... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper introduces NeuralStagger, a method to divide the spatial and temporal resolution into staggered coarser grids, and use a neural model to evolve the coarse-grained grids (which also takes the relative location of the coarser grid as input). In inference, the different grids are combined t... |
ICLR | Title
NeuralStagger: accelerating physics constrained neural PDE solver with spatial-temporal decomposition
Abstract
Neural networks have shown great potential in accelerating the solution of partial differential equations (PDEs). Recently, there has been a growing interest in introducing physics constraints into train... | 1. What is the focus and contribution of the paper regarding partial differential equations?
2. What are the strengths of the proposed method, particularly in terms of efficiency and parallelization?
3. What are the weaknesses of the paper, especially regarding experimental results and comparisons with traditional meth... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The authors propose a method to accelerate the solution of partial differential equation (PDE). Numeric methods to solve PDEs require discretizing space and time into a fine mesh, and the higher the resolution of this discretization, the higher the computation cost. The authors propose to divide th... |
ICLR | Title
Word2net: Deep Representations of Language
Abstract
Word embeddings extract semantic features of words from large datasets of text. Most embedding methods rely on a log-bilinear model to predict the occurrence of a word in a context of other words. Here we propose word2net, a method that replaces their linear par... | 1. What is the main contribution of the paper, and how does it extend the previous work on SGNS?
2. What is the purpose of incorporating POS tags in the proposed method, and how does it improve the performance?
3. Why does the reviewer have reservations about the execution of the work, particularly regarding the choice... | Review | Review
The paper extends SGNS as follows. In SGNS, each word x is associated with vectors a_x and r_x. Given a set of context words C, the model calculates the probability that the target word is x by a dot product between a_x and the average of {r_c: c in C}. The paper generalizes this computation to an arbitrary net... |
ICLR | Title
Word2net: Deep Representations of Language
Abstract
Word embeddings extract semantic features of words from large datasets of text. Most embedding methods rely on a log-bilinear model to predict the occurrence of a word in a context of other words. Here we propose word2net, a method that replaces their linear par... | 1. What is the novel approach presented in the paper for learning vector representations of words?
2. What is the advantage of using non-linear combinations of context vectors in the proposed method?
3. How does the proposed method incorporate additional context information, such as POS tags, into word vector learning?... | Review | Review
The paper presents a method to use non-linear combination of context vectors for learning vector representation of words. The main idea is to replace each word embedding by a neural network, which scores how likely is the current word given the context words. This also allowed them to use other context informati... |
ICLR | Title
Word2net: Deep Representations of Language
Abstract
Word embeddings extract semantic features of words from large datasets of text. Most embedding methods rely on a log-bilinear model to predict the occurrence of a word in a context of other words. Here we propose word2net, a method that replaces their linear par... | 1. What is the focus of the paper regarding neural language models?
2. What are the strengths of the proposed approach, particularly in utilizing side information?
3. What are the weaknesses of the paper, especially in its evaluation methodology?
4. How does the reviewer assess the significance of the proposed method?
... | Review | Review
This paper presents another variant on neural language models used to learn word embeddings. In keeping with the formulation of Mikolov et al, the model learned is a set of independent binary classifiers, one per word. As opposed to other work, each classifier is not based on the dot product between an embedding... |
ICLR | Title
Laplacian Networks: Bounding Indicator Function Smoothness for Neural Networks Robustness
Abstract
For the past few years, Deep Neural Network (DNN) robustness has become a question of paramount importance. As a matter of fact, in sensitive settings misclassification can lead to dramatic consequences. Such miscla... | 1. What are the strengths and weaknesses of the paper regarding its contributions to improving neural network robustness?
2. How does the reviewer assess the clarity and quality of the paper's content, particularly in figure presentation and experimental design?
3. Are there any concerns or suggestions regarding the ch... | Review | Review
This paper proposes the interesting addition of a graph-based regularisers, in NNs architectures, for improving their robustness to different perturbations or noise. The regularisation enforces smoothness on a graph built on the different features at different layers of the NN system. The proposed ideas are quit... |
ICLR | Title
Laplacian Networks: Bounding Indicator Function Smoothness for Neural Networks Robustness
Abstract
For the past few years, Deep Neural Network (DNN) robustness has become a question of paramount importance. As a matter of fact, in sensitive settings misclassification can lead to dramatic consequences. Such miscla... | 1. What is the main contribution of the paper regarding the use of graph regularization for neural networks?
2. What are the strengths of the proposed approach, particularly in terms of its practicality and algorithmic perspective?
3. What are the weaknesses of the paper, especially regarding the experimental results a... | Review | Review
The paper proposes to use a regularization which preserves nearest-neighbor smoothness from layer to layer. The approach is based on controlling the extent to which examples from different classes are separated from one layer to the next, in deep neural networks. The criterion computes the smoothness of the labe... |
ICLR | Title
Laplacian Networks: Bounding Indicator Function Smoothness for Neural Networks Robustness
Abstract
For the past few years, Deep Neural Network (DNN) robustness has become a question of paramount importance. As a matter of fact, in sensitive settings misclassification can lead to dramatic consequences. Such miscla... | 1. What is the focus and contribution of the paper regarding neural network robustness?
2. What are the strengths of the proposed approach, particularly in its experimental results?
3. What are the weaknesses of the paper, specifically regarding its significance and empirical proofs?
4. How does the reviewer assess the... | Review | Review
To improve the robustness of neural networks under various conditions, this paper proposes a new regularizer defined on the graph of the training examples, which penalizes the large similarities between representations belonging to different classes, thus increase the stability of the transformations defined by ... |
ICLR | Title
Quasi-Taylor Samplers for Diffusion Generative Models based on Ideal Derivatives
Abstract
Diffusion generative models have emerged as a new challenger to popular deep neural generative models such as GANs, but have the drawback that they often require a huge number of neural function evaluations (NFEs) during syn... | 1. What is the focus of the paper regarding diffusion models?
2. What are the strengths and weaknesses of the proposed approach in improving sample efficiency?
3. Do you have any concerns or questions about the experimental results and comparisons with other works?
4. How does the reviewer assess the clarity, quality, ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes an approach to improving the sample efficiency of diffusion models by incorporating high-order derivatives. As high-order derivatives are often expensive to compute, the authors propose an approximation. Empirically, the proposed approach is able to generate images using small n... |
ICLR | Title
Quasi-Taylor Samplers for Diffusion Generative Models based on Ideal Derivatives
Abstract
Diffusion generative models have emerged as a new challenger to popular deep neural generative models such as GANs, but have the drawback that they often require a huge number of neural function evaluations (NFEs) during syn... | 1. What is the focus of the paper regarding Probability Flow ODE for diffusion models?
2. What are the strengths and weaknesses of the proposed novel solver?
3. Do you have any concerns about the "ideal derivatives" replacement?
4. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the p... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The authors propose a novel solver of the Probability Flow ODE for diffusion models introduced in [1]. By taking a Taylor expansion of the ODE and including higher-order terms, the solver can take larger steps. This speeds up sampling, which is a well-known computational bottleneck in diffusion mod... |
ICLR | Title
Quasi-Taylor Samplers for Diffusion Generative Models based on Ideal Derivatives
Abstract
Diffusion generative models have emerged as a new challenger to popular deep neural generative models such as GANs, but have the drawback that they often require a huge number of neural function evaluations (NFEs) during syn... | 1. What is the main contribution of the paper regarding acceleration of denoising diffusion models?
2. What are the strengths and weaknesses of the proposed approach, particularly in its theoretical foundation and experimental results?
3. Do you have any concerns or questions about the paper's discussion of the Backwar... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
In this paper, the authors deal with the acceleration of denoising diffusion models. In particular, they propose improved samplers with higher order in order to reduce the number of steps requires at sampling times. The acceleration proposed in that paper is described for both Ordinary Differential... |
ICLR | Title
Quasi-Taylor Samplers for Diffusion Generative Models based on Ideal Derivatives
Abstract
Diffusion generative models have emerged as a new challenger to popular deep neural generative models such as GANs, but have the drawback that they often require a huge number of neural function evaluations (NFEs) during syn... | 1. What is the focus and contribution of the paper regarding solving SDE and ODE?
2. What are the strengths and weaknesses of the proposed approach, particularly in its approximation and flexibility?
3. Do you have any concerns or suggestions regarding the experiments and related works?
4. How does the reviewer assess ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This work proposes high order methods for solving both SDE and ODE, by approximating high order gradients in Taylar expansion. It seems that this method can be extended to arbitrary order by computing some constants. Experiments are conducted on CIFAR10 and CelebA 64.
Strengths And Weaknesses
Stre... |
ICLR | Title
Overcoming Catastrophic Forgetting via Hessian-free Curvature Estimates
Abstract
Learning neural networks with gradient descent over a long sequence of tasks is problematic as their fine-tuning to new tasks overwrites the network weights that are important for previous tasks. This leads to a poor performance on o... | 1. What is the main problem addressed by the paper regarding neural network learning?
2. How do existing works, such as Elastic Weight Consolidation (EWC), attempt to alleviate the issue of "catastrophic forgetting"?
3. What is the proposed approach in the paper, and how does it differ from previous methods like EWC an... | Review | Review
The paper focuses on alleviating the problem of "catastrophic forgetting", exhibited by neural networks learned with gradient-based algorithms over long sequence of tasks. In such learning scenarios, tuning of parameters over the new tasks lead to degradation of performance over the old tasks as the parameters i... |
ICLR | Title
Overcoming Catastrophic Forgetting via Hessian-free Curvature Estimates
Abstract
Learning neural networks with gradient descent over a long sequence of tasks is problematic as their fine-tuning to new tasks overwrites the network weights that are important for previous tasks. This leads to a poor performance on o... | 1. What is the main contribution of the paper regarding neural network training in continual learning?
2. What are the strengths and weaknesses of the proposed approach compared to existing methods such as EWC and Kronecker-factored online Laplace?
3. How does the paper address the issue of calculating the Hessian in c... | Review | Review
1. Summary:
The paper considers neural network training in the continual learning setting -- data arrive sequentially and we can not revisit past data. The paper proposes an approximate Laplace’s method, in which the Hessian the log likelihood of the data is approximated by some form of Hessian-vector project (?... |
ICLR | Title
Overcoming Catastrophic Forgetting via Hessian-free Curvature Estimates
Abstract
Learning neural networks with gradient descent over a long sequence of tasks is problematic as their fine-tuning to new tasks overwrites the network weights that are important for previous tasks. This leads to a poor performance on o... | 1. What is the main contribution of the paper in tackling catastrophic forgetting?
2. What are the strengths and weaknesses of the proposed method compared to previous approaches like EWC?
3. Do you have any concerns regarding the low-rank approximation to the Hessian and its impact on performance?
4. How does the revi... | Review | Review
This paper proposes a method for tackling catastrophic forgetting. Similar to previous methods such as EWC (Kirkpatrick et al., 2017), they penalize parameter updates that align with the Fisher information matrix of the previous tasks. This will prevent the model from changing the previously useful parameters. T... |
ICLR | Title
Learning to Represent Edits
Abstract
We introduce the problem of learning distributed representations of edits. By combining a “neural editor” with an “edit encoder”, our models learn to represent the salient information of an edit and can be used to apply edits to new inputs. We experiment on natural language an... | 1. What are the strengths and weaknesses of the proposed approach regarding its novelty, importance, and ability to capture the structure of edits?
2. How effective is the created source code edit dataset in supporting the study of the new task, and what are the limitations of the current implementation?
3. To what ext... | Review | Review
The authors state nicely and clearly the main contributions they see in their work (Intro, last paragraph). Specifically the state the paper: 1) present a new and important machine learning task, 2) present a family of models that capture the structure of edits and compute efficient representations, 3) create a ... |
ICLR | Title
Learning to Represent Edits
Abstract
We introduce the problem of learning distributed representations of edits. By combining a “neural editor” with an “edit encoder”, our models learn to represent the salient information of an edit and can be used to apply edits to new inputs. We experiment on natural language an... | 1. What is the main contribution of the paper regarding text revisions and code changes?
2. What are the strengths of the proposed approach, particularly in the utilization of the bidirectional LSTM and gated graph neural network?
3. What are the weaknesses of the paper, especially in the evaluation process?
4. How doe... | Review | Review
This paper looks at learning to represent edits for text revisions and code changes. The main contributions are as follows:
* They define a new task of representing and predicting textual and code changes
* They make available a new dataset of code changes (text edit dataset was already available) with labels o... |
ICLR | Title
Learning to Represent Edits
Abstract
We introduce the problem of learning distributed representations of edits. By combining a “neural editor” with an “edit encoder”, our models learn to represent the salient information of an edit and can be used to apply edits to new inputs. We experiment on natural language an... | 1. What are the main contributions and strengths of the paper regarding the edit encoder model and dataset?
2. How does the proposed approach compare to prior works such as Guu et al. 2017, specifically in terms of robustness and applicability to other tasks?
3. What are some potential limitations or areas for improvem... | Review | Review
The main contributions of the paper are an edit encoder model similar to (Guu et al. 2017 http://aclweb.org/anthology/Q18-1031), a new dataset of tree-structured source code edits, and thorough and well thought-out analysis of the edit encodings. The paper is clearly written, and provides clear support for each ... |
ICLR | Title
Adversarially robust transfer learning
Abstract
Transfer learning, in which a network is trained on one task and re-purposed on another, is often used to produce neural network classifiers when data is scarce or full-scale training is too costly. When the goal is to produce a model that is not only accurate but a... | 1. What is the main contribution of the paper regarding transfer learning using standard training?
2. What are the strengths of the paper, particularly in its exploration of robust models and representations?
3. Do you have any questions or concerns about the paper's experiments and results?
4. How does the reviewer as... | Review | Review
Paper summary: This paper explores the problem of robustly transfer learning using only standard training (as opposed to adversarial training (AT)) on the target domain. The authors start by highlighting that intermediate representations learned by adversarially trained networks are themselves fairly robust. The... |
ICLR | Title
Adversarially robust transfer learning
Abstract
Transfer learning, in which a network is trained on one task and re-purposed on another, is often used to produce neural network classifiers when data is scarce or full-scale training is too costly. When the goal is to produce a model that is not only accurate but a... | 1. What is the main contribution of the paper regarding transfer learning and adversarial robustness?
2. What are the strengths of the proposed approach, particularly in preserving robustness and improving validation accuracy?
3. How does the reviewer assess the significance of the individual contributions, such as ret... | Review | Review
The paper studies transfer learning from the point of view of adversarial robustness. The goal is, given a robust deep neural network classifier for a source domain, learn a robust classifier for a target domain as efficiently and with as few samples as possible. The authors empirically evaluate different strate... |
ICLR | Title
Adversarially robust transfer learning
Abstract
Transfer learning, in which a network is trained on one task and re-purposed on another, is often used to produce neural network classifiers when data is scarce or full-scale training is too costly. When the goal is to produce a model that is not only accurate but a... | 1. What is the main contribution of the paper regarding robust transfer learning?
2. What are the strengths and weaknesses of the paper's approach to robust transfer learning?
3. How does the reviewer assess the significance and novelty of the proposed approach compared to prior works?
4. What are the limitations of th... | Review | Review
Summary
-------
This paper addresses the problem of performing robust transfer learning. A first contribution of the paper is to robust and classic training with respect to usual validation accuracy and robustness to adversarial attacks on the CIFAR task. Then, the same comparison is made on a transfer learning ... |
ICLR | Title
Towards Learning Implicit Symbolic Representation for Visual Reasoning
Abstract
Visual reasoning tasks are designed to test a learning algorithm’s capability to infer causal relationships, discover object interactions, and understand temporal dynamics, all from visual cues. It is commonly believed that to achieve... | 1. What is the focus and contribution of the paper regarding self-supervised learning for compositional scene representation?
2. What are the strengths and weaknesses of the proposed approach, particularly in terms of its ability to capture temporal dynamics and encourage object discovery?
3. How does the reviewer asse... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
A self-supervised training over video is proposed for learning compositional scene representations. The paper explores the approach on the CATER and ACRE datasets (video versions of CLEVR), and show its benefits in the domain of visual reasoning and question answering over these videos.
Strengths ... |
ICLR | Title
Towards Learning Implicit Symbolic Representation for Visual Reasoning
Abstract
Visual reasoning tasks are designed to test a learning algorithm’s capability to infer causal relationships, discover object interactions, and understand temporal dynamics, all from visual cues. It is commonly believed that to achieve... | 1. What is the focus and contribution of the paper on visual reasoning tasks?
2. What are the strengths of the proposed approach, particularly in terms of its ability to handle various visual reasoning tasks without requiring specific object abstraction?
3. Do you have any concerns regarding the explanations provided i... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposed an end-to-end implicit symbolic representation learning framework for visual reasoning tasks. It wisely adopts slot tokens for its bottleneck information properties, masked autoencoding objective, and transformers to learn implicit representations in a self-supervised way. The l... |
ICLR | Title
Towards Learning Implicit Symbolic Representation for Visual Reasoning
Abstract
Visual reasoning tasks are designed to test a learning algorithm’s capability to infer causal relationships, discover object interactions, and understand temporal dynamics, all from visual cues. It is commonly believed that to achieve... | 1. What is the main contribution of the paper regarding compositional visual reasoning in videos?
2. What are the strengths of the proposed approach, particularly in terms of self-supervised pretraining and transfer learning?
3. What are the weaknesses of the paper, especially regarding the claims of implicit symbolic ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes a method for compositional visual reasoning in videos, based on recent advances in self-supervised pretraining. The method first trains a spacial-temporal transformer that reconstructs the video frames under the masked autoencoder paradigm, then performs reasoning via transfer l... |
ICLR | Title
Towards Learning Implicit Symbolic Representation for Visual Reasoning
Abstract
Visual reasoning tasks are designed to test a learning algorithm’s capability to infer causal relationships, discover object interactions, and understand temporal dynamics, all from visual cues. It is commonly believed that to achieve... | 1. What is the main contribution of the paper, and how does it build upon previous works in the field?
2. What are the strengths and weaknesses of the proposed framework, particularly in its ability to reason about videos?
3. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the paper's... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes the framework "Implicit Symbolic Concept Learner" (IS-CL), a transformer-based architecture that is capable of reasoning about videos. The framework follows a pretraining-transfer-learning pipeline. That is, the model is first pretrained on a collection of unlabelled videos (by ... |
ICLR | Title
Identifying nonlinear dynamical systems with multiple time scales and long-range dependencies
Abstract
A main theoretical interest in biology and physics is to identify the nonlinear dynamical system (DS) that generated observed time series. Recurrent Neural Networks (RNNs) are, in principle, powerful enough to a... | 1. What is the focus of the paper regarding dynamical system identification?
2. What are the strengths of the proposed approach, particularly in addressing the issue of gradient vanishing or exploding?
3. Do you have any concerns about the choice of regularization technique or its application to RNN models?
4. How does... | Review | Review
The paper explores a very important question in dynamical system identification of how to make recurrent neural networks (RNNs) learn both long-term and short-term dependencies without the gradient vanishing or exploding limitation. They suggest using piece-wise linear RNNs (PLRNNs) with a novel regularization t... |
ICLR | Title
Identifying nonlinear dynamical systems with multiple time scales and long-range dependencies
Abstract
A main theoretical interest in biology and physics is to identify the nonlinear dynamical system (DS) that generated observed time series. Recurrent Neural Networks (RNNs) are, in principle, powerful enough to a... | 1. What is the main contribution of the paper, and how does it address the vanishing and exploding gradient problem in piecewise linear RNNs (PLRNNs)?
2. What is the proposed regularization method, and how does it encourage plane or line attractors in the network dynamics?
3. What are the results of the numerical exper... | Review | Review
This paper proposes a type of regularization for piecewise linear RNNs (PLRNNs) that encourages the network to learn line or plane attractors. The paper argues, through mathematical analysis of the regularized network as well as numerical experiments, that this regularization alleviates the vanishing and explodi... |
ICLR | Title
Identifying nonlinear dynamical systems with multiple time scales and long-range dependencies
Abstract
A main theoretical interest in biology and physics is to identify the nonlinear dynamical system (DS) that generated observed time series. Recurrent Neural Networks (RNNs) are, in principle, powerful enough to a... | 1. What is the focus and contribution of the paper on PLRNN?
2. What are the strengths of the proposed approach, particularly its novelty and theoretical connections?
3. Do you have any concerns regarding the paper, such as jumpy explanations or lack of definitions?
4. How does the reviewer assess the clarity, quality,... | Review | Review
The paper proposes a novel regularization term to PLRNN. PLRNN has nice numerical properties given its simple mathematical structure, but is able to capture complicate dynamics. It's also easy to establish a theoretical connection between PLRNN dynamics and the behavior of its gradients, which is nice. Given suc... |
ICLR | Title
Identifying nonlinear dynamical systems with multiple time scales and long-range dependencies
Abstract
A main theoretical interest in biology and physics is to identify the nonlinear dynamical system (DS) that generated observed time series. Recurrent Neural Networks (RNNs) are, in principle, powerful enough to a... | 1. What is the focus of the paper regarding regularization schemes for training vanilla Relu RNN?
2. What are the strengths of the proposed approach, particularly in connecting RNN dynamics and gradient theoretically?
3. What are the weaknesses of the paper, especially regarding the setting of RNN and working memory?
4... | Review | Review
This paper proposes a regularization scheme for training vanilla Relu RNN to tackle the exploding and vanishing gradients issue. The work eases the analysis of RNN in the dynamical system point of view and connects the RNN dynamics and gradient theoretically. The experiments show the competitive performance comp... |
ICLR | Title
CaPC Learning: Confidential and Private Collaborative Learning
Abstract
Machine learning benefits from large training datasets, which may not always be possible to collect by any single entity, especially when using privacy-sensitive data. In many contexts, such as healthcare and finance, separate parties may wis... | 1. What are the strengths and weaknesses of the proposed federated system for classification?
2. How does the system protect the privacy of the training data and the sample to be classified?
3. What are the limitations of the statistical security used in the system?
4. Can the system be improved by using secret sharing... | Review | Review
Summary: The authors combine several cryptographic techniques to create a federated systems that allows several entities to run classification against all the model held be the participants without revealing information in the process. In particular, the sample to be classified is not revealed to any other party... |
ICLR | Title
CaPC Learning: Confidential and Private Collaborative Learning
Abstract
Machine learning benefits from large training datasets, which may not always be possible to collect by any single entity, especially when using privacy-sensitive data. In many contexts, such as healthcare and finance, separate parties may wis... | 1. What is the focus and contribution of the paper regarding confidentiality and privacy in collaborative learning?
2. What are the strengths and weaknesses of the proposed method compared to other works like InstaHide and TextHide?
3. Do you have any concerns or suggestions regarding the presentation of the paper, suc... | Review | Review
This work motivated by healthcare and finance where separate parties may wish to collaborate and learn from each other's data but are prevented from doing so due to privacy regulations. This paper propose Confidential and Private Collaborative (CaPC) learning, the first method provably achieving both confidentia... |
ICLR | Title
CaPC Learning: Confidential and Private Collaborative Learning
Abstract
Machine learning benefits from large training datasets, which may not always be possible to collect by any single entity, especially when using privacy-sensitive data. In many contexts, such as healthcare and finance, separate parties may wis... | 1. What is the focus of the paper regarding collaborative learning?
2. What are the strengths of the proposed approach, particularly in terms of privacy and security?
3. What are the weaknesses of the paper, especially regarding the evaluation and experiment section?
4. How does the reviewer assess the novelty and sign... | Review | Review
This paper works on the problem of collaborative learning while preserving both confidentiality and privacy of the data points. It combines techniques from secure multi-party computation and differential privacy for the same, and improves on confidential inference and PATE in the process. The new technique is ca... |
ICLR | Title
DiffusER: Diffusion via Edit-based Reconstruction
Abstract
In text generation, models that generate text from scratch one token at a time are currently the dominant paradigm. Despite being performant, these models lack the ability to revise existing text, which limits their usability in many practical scenarios. ... | 1. What is the focus of the paper regarding text generation?
2. What are the strengths of the proposed approach, particularly in its design and training objective?
3. What are the weaknesses of the paper, especially in terms of the training objective and evaluation methods?
4. How does the reviewer assess the clarity, ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper studies an edit-based generative text model that starts with a complete noise distribution (random gibberish) as input and then produces a series of edits to reach high-quality output. Inspired by diffusion models in CV, this Diffuser model rivals or outperforms standard autoregressive m... |
ICLR | Title
DiffusER: Diffusion via Edit-based Reconstruction
Abstract
In text generation, models that generate text from scratch one token at a time are currently the dominant paradigm. Despite being performant, these models lack the ability to revise existing text, which limits their usability in many practical scenarios. ... | 1. What is the main contribution of the paper, and how does it build upon prior work in iterative text generation models and diffusion models?
2. What are the strengths and weaknesses of the proposed Diffuser model, particularly regarding its ability to refine text and be bootstrapped by autoregressive translation or s... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
Motivated by how humans revise content and the success of diffusion models with continuous inputs, the authors propose a generative model of text based on multiple editing steps, with each step based on one or more text span editing operations (insert, delete, replace, keep). The generative model i... |
ICLR | Title
DiffusER: Diffusion via Edit-based Reconstruction
Abstract
In text generation, models that generate text from scratch one token at a time are currently the dominant paradigm. Despite being performant, these models lack the ability to revise existing text, which limits their usability in many practical scenarios. ... | 1. What is the focus and contribution of the paper on text generative tasks?
2. What are the strengths and weaknesses of the proposed DIFFUSER model?
3. Do you have any concerns or questions regarding the model's approach, experimental settings, or explanations?
4. How does the reviewer assess the clarity, quality, nov... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper proposes DIFFUSER, a denoising diffusion model for text generative tasks. It treats text generation as a Markov chain of Levenshtein edit steps to denoise from the initial text. An editing step is modeled as an editing process in existing work (Reid & Neubig).
The contribution of the pape... |
ICLR | Title
Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples
Abstract
Exploration in reinforcement learning is, in general, a challenging problem. In this work, we study a more tractable class of reinforcement learning problems defined by data that provides examples of successf... | 1. What is the focus of the paper regarding learning a policy for an MDP with unspecified reward?
2. What are the strengths of the proposed approach, particularly in using CNML and meta-learning?
3. Do you have any concerns or questions regarding the effectiveness and feasibility of the algorithm?
4. How does the revie... | Review | Review
This paper considers the problem of learning a policy for an MDP with unspecified reward, given user-provided goal states. To this end, a reward model and a policy are jointly learned: the reward model is the conditional normalized maximum likelihood (CNML) learned from a training set consisting of the example g... |
ICLR | Title
Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples
Abstract
Exploration in reinforcement learning is, in general, a challenging problem. In this work, we study a more tractable class of reinforcement learning problems defined by data that provides examples of successf... | 1. What is the main contribution of the paper regarding reinforcement learning problems?
2. How does the proposed approach differ from previous works, specifically Fu et al.'s (2018b) event framework?
3. What are the strengths and weaknesses of the proposed CNML classifier compared to standard neural network classifier... | Review | Review
This manuscript aims to solve reinforcement learning problems where the reward is unknown but a set of successful states are available. Iteratively, it trains a classifier using provided successful states as positive and on-policy samples as negative and use its predictions as the reward function to learn RL pol... |
ICLR | Title
Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples
Abstract
Exploration in reinforcement learning is, in general, a challenging problem. In this work, we study a more tractable class of reinforcement learning problems defined by data that provides examples of successf... | 1. What is the focus of the paper regarding reinforcement learning and inverse RL?
2. What is the novel approach proposed by the authors in addressing the problem?
3. What are the strengths and weaknesses of the paper, particularly in the motivation, methodology, and experimental results?
4. How does the reviewer asses... | Review | Review
Summary
This paper addresses a reinforcement learning problem where the reward function is learned through a classifier that decides whether states are successful or not based on previous examples (i.e. RL after inverse RL). The authors show that this requires uncertainty-aware predictions, which are difficult w... |
ICLR | Title
Reinforcement Learning with Bayesian Classifiers: Efficient Skill Learning from Outcome Examples
Abstract
Exploration in reinforcement learning is, in general, a challenging problem. In this work, we study a more tractable class of reinforcement learning problems defined by data that provides examples of successf... | 1. What is the main contribution of the paper regarding solving RL problems?
2. What are the strengths and weaknesses of the proposed approach, particularly in its use of CNML and meta-learning?
3. Do you have any questions or concerns regarding the experimental results and comparisons with other methods?
4. How does t... | Review | Review
This paper studies how to solve RL problems with a set of success states instead of a standard reward function. The central idea is to firstly train a Bayesian classifier from both the input success examples and the on-policy sampling using the conditional normalized maximum likelihood (CNML) and then use the le... |
ICLR | Title
On Feature Diversity in Energy-based Models
Abstract
Energy-based learning is a powerful learning paradigm that encapsulates various discriminative and generative approaches. An energy-based model (EBM) is typically formed of inner-model(s) that learn a combination of the different features to generate an energy ... | 1. What is the focus and contribution of the paper regarding energy-based models?
2. What are the strengths and weaknesses of the proposed method for feature diversity measurement?
3. Do you have any concerns or suggestions regarding the experimental setup or results?
4. How does the reviewer assess the clarity, qualit... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper has provided an analysis method to evaluate the feature representation of energy-based models based on feature diversity. The authors extend the probably approximately correct (PAC) theory in the view of redundancy reduction on the performance of energy-based models.
Strengths And Weakn... |
ICLR | Title
On Feature Diversity in Energy-based Models
Abstract
Energy-based learning is a powerful learning paradigm that encapsulates various discriminative and generative approaches. An energy-based model (EBM) is typically formed of inner-model(s) that learn a combination of the different features to generate an energy ... | 1. What is the main contribution of the paper regarding energy-based models?
2. What are the strengths and weaknesses of the proposed regularizer term?
3. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the paper's content?
4. Are there any questions or concerns regarding the paper's ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper provides a new regulariser term for training energy-based models, which promotes feature diversity. A theoretical analysis on the generalisation performance of energy-based models using the PAC-learning framework gives a solid motivating evidence on the need of the regularizer. Specifical... |
ICLR | Title
On Feature Diversity in Energy-based Models
Abstract
Energy-based learning is a powerful learning paradigm that encapsulates various discriminative and generative approaches. An energy-based model (EBM) is typically formed of inner-model(s) that learn a combination of the different features to generate an energy ... | 1. What is the focus and contribution of the paper regarding feature diversity in PAC theory of EBMs?
2. What are the strengths and weaknesses of the proposed approach, particularly in its theoretical analysis and experimental results?
3. Do you have any concerns regarding the similarity between the paper and a previou... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper extends the PAC theory of EBMs to analyze the impact of feature diversity on the performance of EBMs. The generalization bounds for regression, binary classification and implicit regression w.r.t. the feature set redundancy are derived. Experimental results on MNIST for image generation a... |
ICLR | Title
On Feature Diversity in Energy-based Models
Abstract
Energy-based learning is a powerful learning paradigm that encapsulates various discriminative and generative approaches. An energy-based model (EBM) is typically formed of inner-model(s) that learn a combination of the different features to generate an energy ... | 1. What is the main contribution of the paper regarding energy-based approaches?
2. What are the strengths and weaknesses of the proposed method, particularly in terms of its simplicity and experimental results?
3. Do you have any concerns about the definition of v-diversity and its representation of feature set divers... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes a unifying PAC theory proof, showing the potential of improving the generalization error of energy-based approaches by reducing the redundancy of the feature set. It combines the Rademacher complexity and the definition of v-diversity to compute the upper-bound of the gap betwee... |
ICLR | Title
On Feature Diversity in Energy-based Models
Abstract
Energy-based learning is a powerful learning paradigm that encapsulates various discriminative and generative approaches. An energy-based model (EBM) is typically formed of inner-model(s) that learn a combination of the different features to generate an energy ... | 1. What is the main contribution of the paper regarding energy-based models' generalization ability?
2. How does the paper enhance feature diversity in inner models, and what are the theoretical upper bounds provided?
3. Are there any concerns or suggestions regarding the tightness of the upper bounds or the sufficienc... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper studies the influence of feature diversity on the generalization of energy-based models (EBM), which refers to the gap between the estimated energy function and the true energy distribution. The authors propose to improve the performance of Energy-Based Models (EBMs) by enhancing the fea... |
ICLR | Title
Dependency Structure Discovery from Interventions
Abstract
Promising results have driven a recent surge of interest in continuous optimization methods for Bayesian network structure learning from observational data. However, there are theoretical limitations on the identifiability of underlying structures obtaine... | 1. What is the focus of the paper regarding causal graph learning?
2. What are the strengths and weaknesses of the proposed algorithm compared to other methods like JCI?
3. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the paper's content?
4. What are some questions or concerns rega... | Review | Review
The authors propose a 3-phase heuristic algorithm to learn a causal graph from interventional data using continuous optimization. Unfortunately, the paper is hard to follow. Specifically, the exact procedure should be clarified by the authors. If I understand correctly, first they fit to observational data by se... |
ICLR | Title
Dependency Structure Discovery from Interventions
Abstract
Promising results have driven a recent surge of interest in continuous optimization methods for Bayesian network structure learning from observational data. However, there are theoretical limitations on the identifiability of underlying structures obtaine... | 1. What is the focus and contribution of the paper on causal discovery?
2. What are the strengths and weaknesses of the proposed method, particularly in terms of its empirical performance and lack of theoretical guarantees?
3. Do you have any concerns regarding the heuristic used for predicting an unknown intervention ... | Review | Review
This paper aims to extend the continuous optimization approach to causal discovery to handle interventional data as well as observational data. It describes a method for learning the causal structure over a set of categorical variables and reports strong empirical performance. However, no theoretical guarantee o... |
ICLR | Title
Dependency Structure Discovery from Interventions
Abstract
Promising results have driven a recent surge of interest in continuous optimization methods for Bayesian network structure learning from observational data. However, there are theoretical limitations on the identifiability of underlying structures obtaine... | 1. What is the focus and contribution of the paper on structure learning for causal Bayesian networks?
2. What are the strengths of the proposed approach, particularly in its iterative method and use of the do-formalism?
3. What are the weaknesses of the paper, especially regarding the brief definition of interventions... | Review | Review
Recommendation to Accept
########################################################################## Summary: The paper provides a novel approach in the area of structure learning for causal bayesian networks. The authors suggest an iterative method, that builds on the widely accepted do-formalism. The approach s... |
ICLR | Title
Dependency Structure Discovery from Interventions
Abstract
Promising results have driven a recent surge of interest in continuous optimization methods for Bayesian network structure learning from observational data. However, there are theoretical limitations on the identifiability of underlying structures obtaine... | 1. What is the focus of the paper regarding structure learning from observational and interventional data?
2. What are the concerns regarding the method's ability to handle limited interventional datasets and samples?
3. How does the reviewer assess the effectiveness and efficiency of the proposed three-phase score-bas... | Review | Review
The authors propose a method for structure learning from observational and interventional data that uses a continuous optimization method. Data is discrete-valued, there are no hidden confounders, each intervention affects only one variable, but the location of it may be unknown. A three-phase score-based, itera... |
ICLR | Title
A Fine-Grained Spectral Perspective on Neural Networks
Abstract
Are neural networks biased toward simple functions? Does depth always help learn more complex features? Is training the last layer of a network as good as training all layers? How to set the range for learning rate tuning? These questions seem unrela... | 1. What are the main contributions and findings of the paper regarding conjugate kernel and neural tangent kernels?
2. How do the authors develop spectral theory for CK and NTK on boolean cube, spheres, and Gaussian distributions?
3. What are some interesting empirical observations clarified by the authors using their ... | Review | Review
Updates:
Thanks for the updates.
I find the new theoretical results interesting and potentially useful, which shows, in the large $d$ setting, spectrums of CKs/NTKs for boolean cube, sphere and isotropic Gaussian are closed to each other in some sense. Thus, I raise my score to weakly accepted but lower dow... |
ICLR | Title
A Fine-Grained Spectral Perspective on Neural Networks
Abstract
Are neural networks biased toward simple functions? Does depth always help learn more complex features? Is training the last layer of a network as good as training all layers? How to set the range for learning rate tuning? These questions seem unrela... | 1. What is the main contribution of the paper regarding neural networks' conjugate kernel and neural tangent kernel?
2. What are the strengths and weaknesses of the paper's analysis on boolean cube?
3. Do you have any concerns about the paper's claims and conclusions, particularly regarding the simplicity bias theorem?... | Review | Review
Aiming to resolve the question whether and why deep networks are biased towards simple functions, this paper gives a spectral analysis on neural networks' conjugate kernel(CK) and neural tangent kernel(NTK) on boolean cube. The eigenfunctions are identified and the eigenvalues are shown computable in polynomial ... |
ICLR | Title
A Fine-Grained Spectral Perspective on Neural Networks
Abstract
Are neural networks biased toward simple functions? Does depth always help learn more complex features? Is training the last layer of a network as good as training all layers? How to set the range for learning rate tuning? These questions seem unrela... | 1. How do randomly initialized and trained deep networks perform on simple functions?
2. How does the performance change with depth, activation function, and initialization?
3. How does the input distribution affect the results of spectral analysis?
4. What are the limitations of restricting input distributions to bool... | Review | Review
This paper examined the spectrum of NNGP and NTK kernels and answer several questions about deep networks using both analytical results and experimental evidence:
* Are randomly initialized and trained deep networks biased to simple functions?
* How does this change with depth, activation function, and initializ... |
ICLR | Title
In Your Pace: Learning the Right Example at the Right Time
Abstract
Training neural networks is traditionally done by sequentially providing random mini-batches sampled uniformly from the entire dataset. In our work, we show that sampling mini-batches non-uniformly can both enhance the speed of learning and impro... | 1. What is the reviewer's overall assessment of the paper's quality and significance?
2. What are the strengths and weaknesses of the paper, according to the reviewer?
3. Does the reviewer have any concerns regarding the novelty of the proposed method?
4. How does the reviewer evaluate the practical usefulness of the p... | Review | Review
In my opinion this paper is generally of good quality and clarity, modest originality and significance.
Strengths:
- The experiments are very thorough. Hyperparameters were honestly optimized. The method does show some modest improvements in the experiments provided by the authors.
- The analysis of the results... |
ICLR | Title
In Your Pace: Learning the Right Example at the Right Time
Abstract
Training neural networks is traditionally done by sequentially providing random mini-batches sampled uniformly from the entire dataset. In our work, we show that sampling mini-batches non-uniformly can both enhance the speed of learning and impro... | 1. What are the strengths and weaknesses of the paper regarding its contribution to understanding curriculum learning in deep neural networks?
2. Are there any important related works missing from the paper's references? If so, what are they, and how do they contribute to the topic of curriculum learning?
3. How convin... | Review | Review
This paper studies an interesting and meaningful topic that what is the potential of curriculum learning (CL) in training dnn. The authors decompose CL into two main parts: scoring function and pacing function. Towards both parts, several candidate functions are proposed and verified. The paper is presented qu... |
ICLR | Title
In Your Pace: Learning the Right Example at the Right Time
Abstract
Training neural networks is traditionally done by sequentially providing random mini-batches sampled uniformly from the entire dataset. In our work, we show that sampling mini-batches non-uniformly can both enhance the speed of learning and impro... | 1. What is the main contribution of the paper regarding Curriculum Learning (CL)?
2. How does the proposed approach differ from previous works, specifically Weinshall et al?
3. Can you elaborate on the bootstrapping approach for estimating the scoring function?
4. What are the advantages and disadvantages of using easy... | Review | Review
This problem of interest in this paper is Curriculum Learning (CL), in the context of deep learning in particular. CL refers to learning a non-random order of presenting the training examples to the learner, typically with easier examples presented before difficult ones, to guide learning more effectively. This ... |
ICLR | Title
For self-supervised learning, Rationality implies generalization, provably
Abstract
We prove a new upper bound on the generalization gap of classifiers that are obtained by first using self-supervision to learn a representation r of the training data, and then fitting a simple (e.g., linear) classifier g to the l... | 1. What is the primary contribution of the paper regarding generalization error?
2. What are the three independent components of generalization error discussed in the paper, and how do they relate to self-supervised learning?
3. How does the paper bound the memorization error of simple algorithms in terms of their info... | Review | Review
This paper gives a new perspective on generalization, motivated by the success of self-supervised learning, especially on noisy data. They view the generalization error as consisting of 3 independent components: robustness, rationality and memorization. Informally, robustness measures the degradation in training... |
ICLR | Title
For self-supervised learning, Rationality implies generalization, provably
Abstract
We prove a new upper bound on the generalization gap of classifiers that are obtained by first using self-supervision to learn a representation r of the training data, and then fitting a simple (e.g., linear) classifier g to the l... | 1. What is the focus of the paper regarding generalization capability in self-supervised learning?
2. What are the strengths of the proposed approach in understanding generalization error?
3. How does the paper address the data re-use problem in analyzing generalization?
4. Can the authors provide more explanations reg... | Review | Review
The present paper aims to understand the generalization capability of self-supervised learning algorithms that fine-tune a simple linear classifier to the labels. Analyzing generalization in this case is challenging due to a data re-use problem: the same training data that is used for self-supervised learning is... |
ICLR | Title
For self-supervised learning, Rationality implies generalization, provably
Abstract
We prove a new upper bound on the generalization gap of classifiers that are obtained by first using self-supervision to learn a representation r of the training data, and then fitting a simple (e.g., linear) classifier g to the l... | 1. What is the main contribution of the paper regarding generalization gaps in self-supervised learning?
2. What are the strengths and weaknesses of the proposed RRM decomposition?
3. How does the reviewer assess the novelty and significance of the paper's contributions in isolation to self-supervised learning?
4. What... | Review | Review
The paper analyzes the generalization gap for self-supervised learning. This paper's contribution includes the proposal of decomposing the generalization bound into three terms: robustness, rationality, and memorization (RRM). The three terms explain the generalization gap with some different perspectives. With ... |
ICLR | Title
For self-supervised learning, Rationality implies generalization, provably
Abstract
We prove a new upper bound on the generalization gap of classifiers that are obtained by first using self-supervision to learn a representation r of the training data, and then fitting a simple (e.g., linear) classifier g to the l... | 1. What are the concerns regarding the paper's claims about bounding the generalization gap?
2. How does the reviewer assess the robustness gap and rationality gap discussed in the paper?
3. Is the proposed method effective in indicating the generalization of the algorithm?
4. What are the issues with the bound propose... | Review | Review
The authors propose to upper bound the generalization gap via three quantities, namely robustness gap, rationality gap and memorization gap, shows that the memorization gap can be bounded via standard learning theory arguments, and empirically show that all of the three terms are small. The authors also argue th... |
ICLR | Title
Spatio-temporal point processes with deep non-stationary kernels
Abstract
Point process data are becoming ubiquitous in modern applications, such as social networks, health care, and finance. Despite the powerful expressiveness of the popular recurrent neural network (RNN) models for point process data, they may ... | 1. What is the focus and contribution of the paper on temporal/spatio-temporal point processes?
2. What are the strengths of the proposed approach, particularly in terms of kernel functions and positivity ensurance?
3. What are the weaknesses of the paper regarding additional empirical studies and comparisons with othe... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
the paper proposes a more general form of kernel function that is typically used in temporal/spatio-temporal point process, by considering an absolute time-dependent component in addition to the relative spatial-time inputs. In addition, authors made another contribution by proposing a more efficie... |
ICLR | Title
Spatio-temporal point processes with deep non-stationary kernels
Abstract
Point process data are becoming ubiquitous in modern applications, such as social networks, health care, and finance. Despite the powerful expressiveness of the popular recurrent neural network (RNN) models for point process data, they may ... | 1. What is the focus and contribution of the paper regarding spatio-temporal point processes?
2. What are the strengths of the proposed approach, particularly in terms of reducing model complexity and computational efficiency?
3. What are the weaknesses of the paper, such as the need for more explanations, clarificatio... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes a deep non-stationary kernel for spatio-temporal point processes using a different parameterization scheme, which reduces the model complexity. The non-negativity of the solution is guaranteed by a log-barrier method which maintains the linearity of the conditional intensity fun... |
ICLR | Title
Spatio-temporal point processes with deep non-stationary kernels
Abstract
Point process data are becoming ubiquitous in modern applications, such as social networks, health care, and finance. Despite the powerful expressiveness of the popular recurrent neural network (RNN) models for point process data, they may ... | 1. What is the focus of the paper regarding modeling non-stationary spatio-temporal events?
2. What are the strengths of the proposed approach, particularly in terms of its novelty and computational complexity?
3. What are the weaknesses of the paper, especially regarding the design of the kernel function and its prope... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This manuscript proposes a method to model non-stationary spatio-temporal events in the framework of hawkes process. The specific method is to construct a more refined kernel function.
Strengths And Weaknesses
Strength:
The idea of using deep non-stationary kernel in the point process to model spa... |
ICLR | Title
Spatio-temporal point processes with deep non-stationary kernels
Abstract
Point process data are becoming ubiquitous in modern applications, such as social networks, health care, and finance. Despite the powerful expressiveness of the popular recurrent neural network (RNN) models for point process data, they may ... | 1. What is the main contribution of the paper regarding Spatio-Temporal Point Processes?
2. What are the strengths and weaknesses of the proposed DNSK model?
3. Do you have any questions or suggestions regarding the empirical results and comparisons with other methods?
4. How does the reviewer assess the clarity, quali... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The authors of this submission proposed a deep non-stationary kernel (DNSK) modeling Spatio-Temporal Point Processes (STPP) for potentially non-stationary events in continuous time and space.
The authors focused on Hawkes process assuming that the influences from the past events are linearly additi... |
ICLR | Title
New Perspective on the Global Convergence of Finite-Sum Optimization
Abstract
Deep neural networks (DNNs) have shown great success in many machine learning tasks. Their training is challenging since the loss surface of the network architecture is generally non-convex, or even non-smooth. How and under what assump... | 1. What is the focus of the paper regarding optimization problems?
2. What are the strengths of the proposed approach, particularly in its novelty and transformation?
3. What are the weaknesses of the paper regarding its significance and usefulness in solving real-world problems?
4. How does the reviewer suggest improv... | Summary Of The Paper
Review | Summary Of The Paper
The authors formulate a way to transform finite-sum optimization problems in a proxy strongly convex problem, and prove that it converges to a global minimum in a number of gradient steps that scales inverse quadratically with the tolerance.
Review
Strengths: the paper is fairly clear, and the pro... |
ICLR | Title
New Perspective on the Global Convergence of Finite-Sum Optimization
Abstract
Deep neural networks (DNNs) have shown great success in many machine learning tasks. Their training is challenging since the loss surface of the network architecture is generally non-convex, or even non-smooth. How and under what assump... | 1. What is the focus of the paper regarding machine learning loss functions?
2. What are the strengths of the proposed approach, particularly in terms of smoothness and convergence results?
3. Do you have any concerns about the novelty of the formulation and the existing works in the optimization community?
4. How does... | Summary Of The Paper
Review | Summary Of The Paper
The submitted paper considers a composite formulation of machine learning loss functions, where both the inner and outer functions are required to be smooth. Two algorithms are proposed for the considered formulation, with corresponding convergence results.
Review
The submitted work presents some ... |
ICLR | Title
New Perspective on the Global Convergence of Finite-Sum Optimization
Abstract
Deep neural networks (DNNs) have shown great success in many machine learning tasks. Their training is challenging since the loss surface of the network architecture is generally non-convex, or even non-smooth. How and under what assump... | 1. What is the focus and contribution of the paper regarding gradient-based algorithms?
2. What are the strengths of the proposed approach, particularly in terms of its design and analysis?
3. What are the weaknesses of the paper, especially regarding its claims of novelty and the absence of experimental studies?
4. Do... | Summary Of The Paper
Review | Summary Of The Paper
The paper provides a new gradient-based algorithm. The algorithm is based on the observation that a loss function for a single sample can be written as composition of two functions (the logits and the actual loss function). It computes the direction by means of solving a quadratic MSE problem. They... |
ICLR | Title
New Perspective on the Global Convergence of Finite-Sum Optimization
Abstract
Deep neural networks (DNNs) have shown great success in many machine learning tasks. Their training is challenging since the loss surface of the network architecture is generally non-convex, or even non-smooth. How and under what assump... | 1. What is the main contribution of the paper regarding optimization methods for nonconvex finite sum problems?
2. What are the strengths and weaknesses of the proposed method, particularly in terms of mathematical errors and convergence analysis?
3. Do you have any questions or concerns regarding the proof of Theorems... | Summary Of The Paper
Review | Summary Of The Paper
This paper presents a new optimization method for finding global minima of nonconvex finite sum problems. In particular, the summands are functions of the form
ϕ
i
∘
h
where
ϕ
i
is convex and Lipschitz smooth, while
h
is nonconvex. Each iteration of the method consists of solving an auxiliary regul... |
ICLR | Title
Effective Offline Reinforcement Learning via Conservative State Value Estimation
Abstract
Offline RL seeks to learn effective policies solely from historical data, which expects to perform well in the online environment. However, it faces a major challenge of value over-estimation introduced by the distributional... | 1. What is the focus and contribution of the paper regarding offline RL?
2. What are the strengths and weaknesses of the proposed method, particularly in its design decisions and comparisons with other works?
3. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the paper's content?
4. A... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper proposes a CQL-like method for penalizing out-of-distribution states in offline RL. The method first describes an approach for off-policy evaluation with OOD state penalization, and then incorporates into the training pipeline for Q-functions, and then uses it for offline RL.
Strengths A... |
ICLR | Title
Effective Offline Reinforcement Learning via Conservative State Value Estimation
Abstract
Offline RL seeks to learn effective policies solely from historical data, which expects to perform well in the online environment. However, it faces a major challenge of value over-estimation introduced by the distributional... | 1. What is the focus and contribution of the paper on offline reinforcement learning?
2. What are the strengths of the proposed approach, particularly in terms of its theoretical properties and empirical performance?
3. What are the weaknesses of the paper, especially regarding its similarity to other existing algorith... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The authors propose a new algorithm--CSVE--that learns conservative estimates of state value functions by penalizing values of OOD states. The authors prove that the estimated state value functions are lower-bounds of the true value in expectation over any state distribution. Finally, the authors e... |
ICLR | Title
Effective Offline Reinforcement Learning via Conservative State Value Estimation
Abstract
Offline RL seeks to learn effective policies solely from historical data, which expects to perform well in the online environment. However, it faces a major challenge of value over-estimation introduced by the distributional... | 1. What is the focus of the paper in terms of offline reinforcement learning?
2. What are the strengths and weaknesses of the proposed approach compared to prior works like CQL?
3. Do you have concerns regarding the theoretical analysis and algorithmic design?
4. How does the reviewer assess the clarity, quality, novel... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
A major challenge in offline RL is the distribution mismatch between the behavior policy generating the offline data and the target policy we want to learn. Such a mismatch can result in an overestimation error in the Bellman update, which can further result in divergence. The paper proposes to lea... |
ICLR | Title
Effective Offline Reinforcement Learning via Conservative State Value Estimation
Abstract
Offline RL seeks to learn effective policies solely from historical data, which expects to perform well in the online environment. However, it faces a major challenge of value over-estimation introduced by the distributional... | 1. What is the focus and contribution of the paper on reinforcement learning?
2. What are the strengths of the proposed approach, particularly in its theoretical analysis and originality?
3. What are the weaknesses of the paper, especially regarding its comparison with other methods and lack of interpretability?
4. How... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper proposes an algorithm called CSVE, which imposes penalization on states rather than state actions used in CQL. By doing such, we may benefit from learning a better conservative value since learning a conservative Q-value over joint state-action space is more challenging. The paper provid... |
ICLR | Title
DiscoBAX - Discovery of optimal intervention sets in genomic experiment design
Abstract
The discovery of therapeutics to treat genetically-driven pathologies relies on identifying genes involved in the underlying disease mechanism. With billions of potential hypotheses to test, an exhaustive exploration of the en... | 1. What is the focus and contribution of the paper regarding subset selection in genomic intervention?
2. What are the strengths and weaknesses of the proposed DiscoBAX algorithm?
3. How does the reviewer assess the clarity, quality, novelty, and reproducibility of the paper's content?
4. Are there any concerns or ques... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This work presents a new probabilistic algorithm ("DiscoBAX") for subset selection that aims to approximately optimize phenotype movement in genomic intervention and can be useful in drug discovery tasks according to the authors. The method identifies a set of interventions whose elements will trig... |
ICLR | Title
DiscoBAX - Discovery of optimal intervention sets in genomic experiment design
Abstract
The discovery of therapeutics to treat genetically-driven pathologies relies on identifying genes involved in the underlying disease mechanism. With billions of potential hypotheses to test, an exhaustive exploration of the en... | 1. What is the focus of the paper regarding experiment design?
2. What are the strengths and weaknesses of the proposed approach, particularly in terms of its formalization and exploration?
3. Do you have any concerns about the investigation of discoBAX's sample efficiency, sensitivity to hyperparameter settings, and c... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper considers the problem of designing an experiment where there are two stages: in the first (in vitro) stage our task is to (efficiently) design an experiment that will have good chance of success in the second (in vivo) stage. The authors formalize this elegantly in equation (3).
The propo... |
ICLR | Title
DiscoBAX - Discovery of optimal intervention sets in genomic experiment design
Abstract
The discovery of therapeutics to treat genetically-driven pathologies relies on identifying genes involved in the underlying disease mechanism. With billions of potential hypotheses to test, an exhaustive exploration of the en... | 1. What is the focus and contribution of the paper regarding iterative selection of optimal targets for genetic interventions?
2. What are the strengths and weaknesses of the proposed method, particularly in its application and technical aspects?
3. Do you have any concerns or questions about the paper's content, such ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
This paper is concerned with the iterative selection of an optimal set of targets for genetic interventions based on a scalar readout. The methods section described a particular instance of BAX, that models the singularity of the biological problem that is considered. A few aspects that makes it di... |
ICLR | Title
DiscoBAX - Discovery of optimal intervention sets in genomic experiment design
Abstract
The discovery of therapeutics to treat genetically-driven pathologies relies on identifying genes involved in the underlying disease mechanism. With billions of potential hypotheses to test, an exhaustive exploration of the en... | 1. What is the main contribution of the paper regarding the optimization of genomic interventions?
2. What are the strengths of the proposed approach, particularly in its application to real-world problems?
3. What are some weaknesses or limitations of the paper, such as the estimation of the noise distribution or the ... | Summary Of The Paper
Strengths And Weaknesses
Clarity, Quality, Novelty And Reproducibility | Summary Of The Paper
The paper considers the problem of finding a diverse set of genomic interventions that maximize a phenotype of interest. Formally, the problem can be modeled as optimizing expensive black-box functions over sets of inputs from a given design space. Bayesian Algorithm eXecution (BAX) is a recently p... |
ICLR | Title
Feature-Robust Optimal Transport for High-Dimensional Data
Abstract
Optimal transport is a machine learning problem with applications including distribution comparison, feature selection, and generative adversarial networks. In this paper, we propose feature-robust optimal transport (FROT) for highdimensional dat... | 1. What is the focus and contribution of the paper on robust OT/p-wasserstein-dist?
2. What are the strengths and weaknesses of the proposed formulation and solution method?
3. Do you have any concerns regarding the restriction on non-overlapping groups?
4. How does the reviewer assess the novelty and technical contrib... | Review | Review
This work proposes variants of robust OT/p-wasserstein-dist (3)/(4), where the ground cost is in some sense the maximum over costs with (prefixed) groups of features. The motivation is similar to that for feature selection: where perhaps only few of these groups of features are critical/sufficient for OT purpose... |
ICLR | Title
Feature-Robust Optimal Transport for High-Dimensional Data
Abstract
Optimal transport is a machine learning problem with applications including distribution comparison, feature selection, and generative adversarial networks. In this paper, we propose feature-robust optimal transport (FROT) for highdimensional dat... | 1. What is the focus and contribution of the paper on optimal transport?
2. What are the strengths and weaknesses of the proposed framework, FROT, particularly in terms of feature selection and robustness to noise?
3. Do you have any concerns regarding the novelty and extensiveness of the experiments in the paper?
4. H... | Review | Review
The proposed framework FROT - feature-robust optimal transport - seeks to select feature groups to both speed up OT computation for high-dimensional data and make it more robust to noise. The exposition is generally clear. My main concerns are limited novelty and lack of extensive experiments.
The paper draws th... |
ICLR | Title
Feature-Robust Optimal Transport for High-Dimensional Data
Abstract
Optimal transport is a machine learning problem with applications including distribution comparison, feature selection, and generative adversarial networks. In this paper, we propose feature-robust optimal transport (FROT) for highdimensional dat... | 1. What is the focus of the paper, and what are the proposed solutions to the featured robust optimal transport problem?
2. What are the strengths and weaknesses of the paper, particularly regarding the presentation and the convergence guarantee?
3. How does the reviewer assess the potential applications of FROT, and h... | Review | Review
Summary: The authors try to solve a special kind of high-dimensional optimal transport problem. Specifically, they consider the cases when features are grouped and the grouping is known a-priori. The authors formulate the problem into the feature-robust optimal transport (FROT) problem. The authors propose two s... |
ICLR | Title
Deep Convolution for Irregularly Sampled Temporal Point Clouds
Abstract
We consider the problem of modeling the dynamics of continuous spatial-temporal processes represented by irregular samples through both space and time. Such processes occur in sensor networks, citizen science, multi-robot systems, and many ot... | 1. What is the focus of the review, and what are the reviewer's main concerns regarding the paper?
2. What are the strengths and weaknesses of the proposed approach in the paper, according to the reviewer?
3. How does the reviewer assess the significance and novelty of the work compared to other works in the field?
4. ... | Review | Review
Summary:
This paper proposes a new spatial-temporal point cloud processing technique, which extends the prior work of PointConv for spatial point processing to the temporal domain. Through experiments on two datasets, this paper shows improved performance over a few baselines.
Paper Strengths:
The direction of s... |
ICLR | Title
Deep Convolution for Irregularly Sampled Temporal Point Clouds
Abstract
We consider the problem of modeling the dynamics of continuous spatial-temporal processes represented by irregular samples through both space and time. Such processes occur in sensor networks, citizen science, multi-robot systems, and many ot... | 1. What is the focus of the paper regarding spatial-temporal point clouds?
2. What are the strengths of the proposed Temporal PointConv model?
3. What are the weaknesses of the paper, particularly concerning its technical contribution?
4. Do you have any questions about the experimental results or the presentation of t... | Review | Review
This paper studies the problem of modeling spatial-temporal point clouds which are sampled at irregular space and time points. It proposes the Temporal PointConv model which is an extension of the PointConv model (Wu et al., 2019). In particular, PointConv computes a convolution by aggregating the features of ne... |
ICLR | Title
Deep Convolution for Irregularly Sampled Temporal Point Clouds
Abstract
We consider the problem of modeling the dynamics of continuous spatial-temporal processes represented by irregular samples through both space and time. Such processes occur in sensor networks, citizen science, multi-robot systems, and many ot... | 1. What is the focus of the paper, and what are the proposed approaches?
2. What are the strengths of the presented method, particularly in its effectiveness?
3. Do you have any concerns or questions regarding the method's components and their motivation?
4. How does the reviewer assess the significance of the paper's ... | Review | Review
In this paper spatio-temporal point convolution are proposed, which can be used for sequences of sparse and unordered data. For this purpose, a spatial convolution and a temporal convolution are applied separately, which are then combined in a further step.The presented method was evaluated with two different da... |
ICLR | Title
Deep Convolution for Irregularly Sampled Temporal Point Clouds
Abstract
We consider the problem of modeling the dynamics of continuous spatial-temporal processes represented by irregular samples through both space and time. Such processes occur in sensor networks, citizen science, multi-robot systems, and many ot... | 1. What is the focus of the paper, and how does it extend previous work in PointConv?
2. What are the limitations of the proposed approach, particularly in its application and evaluation?
3. How does the reviewer assess the novelty and significance of the paper's contribution?
4. Are there any concerns regarding the ch... | Review | Review
The paper proposes an extension of PointConv for spatial-temporal point cloud modeling. The model can be used for prediction or forecasting and is evaluated on Starcraft II and weather nowcasting.
The TemporalPointConv follows PointConv and the current work extends this by appending time. I think the paper is he... |
ICLR | Title
Connecting the Dots Between MLE and RL for Sequence Prediction
Abstract
Sequence prediction models can be learned from example sequences with a variety of training algorithms. Maximum likelihood learning is simple and efficient, yet can suffer from compounding error at test time. Reinforcement learning such as po... | 1. What is the main contribution of the paper on policy optimization?
2. What are the strengths and weaknesses of the proposed optimization framework?
3. How does the reviewer assess the novelty and effectiveness of the interpolation algorithm?
4. Are there any concerns regarding the annealing mechanism and its effecti... | Review | Review
This paper claims to propose a general entropy regularized policy optimization paradigm. MLE and RL are special cases of this training paradigm. Paper is well written, and the experimental results are convincing enough.
However, there are still some minor problems in the paper. For the optimization framework E... |
ICLR | Title
Connecting the Dots Between MLE and RL for Sequence Prediction
Abstract
Sequence prediction models can be learned from example sequences with a variety of training algorithms. Maximum likelihood learning is simple and efficient, yet can suffer from compounding error at test time. Reinforcement learning such as po... | 1. What is the main contribution of the paper in the field of sequence modeling?
2. What are the strengths and weaknesses of the proposed unified view on training algorithms?
3. How does the reviewer assess the presentation and clarity of the paper's content?
4. What are the limitations of the experimental validation p... | Review | Review
This submission belongs to the field of sequence modelling. In particular, this submission presents a unified view on a range of training algorithms including maximum likelihood (ML) and reinforcement learning (RL). The unified view presented I believe is interesting and could be of interest to a large community... |
ICLR | Title
Connecting the Dots Between MLE and RL for Sequence Prediction
Abstract
Sequence prediction models can be learned from example sequences with a variety of training algorithms. Maximum likelihood learning is simple and efficient, yet can suffer from compounding error at test time. Reinforcement learning such as po... | 1. What is the focus of the paper regarding policy optimization?
2. What are the strengths of the proposed formalism, particularly in its ability to encompass various policy gradient algorithms?
3. What are the concerns regarding the experimental results, specifically for text summarization?
4. How does the reviewer as... | Review | Review
This paper presents a formalism of entropy regularized policy optimization. They also show that various policy gradients algorithms can be reformulated as special instances of the presented novel formalism. The only difference between them being the reward function and two weight hyperparameters. Further, the pa... |
ICLR | Title
Kernel Deformed Exponential Families for Sparse Continuous Attention
Abstract
Attention mechanisms take an expectation of a data representation with respect to probability weights. This creates summary statistics that focus on important features. Recently, Martins et al. (2020; 2021) proposed continuous attention... | 1. What is the focus of the paper regarding attention mechanisms in neural networks?
2. What are the strengths of the proposed approach, particularly in its theoretical foundation?
3. What are the weaknesses or limitations of the method, especially regarding computational efficiency?
4. Do you have any concerns or ques... | Summary Of The Paper
Review | Summary Of The Paper
Many modern neural architectures, especially in natural language processing, rely heavily on the attention mechanism. Previous work in the literature proposed to extend the softmax-based attention mechanism by using different distribution families. In particular, the authors of this paper focus on ... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.