paper_id stringlengths 19 21 | paper_title stringlengths 8 170 | paper_abstract stringlengths 8 5.01k | paper_acceptance stringclasses 18
values | meta_review stringlengths 29 10k | label stringclasses 3
values | review_ids list | review_writers list | review_contents list | review_ratings list | review_confidences list | review_reply_tos list |
|---|---|---|---|---|---|---|---|---|---|---|---|
iclr_2018_rk6qdGgCZ | Fixing Weight Decay Regularization in Adam | We note that common implementations of adaptive gradient algorithms, such as Adam, limit the potential benefit of weight decay regularization, because the weights do not decay multiplicatively (as would be expected for standard weight decay) but by an additive constant factor.
We propose a simple way to resolve ... | rejected-papers | This paper generated quite a bit of controversy among reviewers. The main claim of the paper is that Adam and related optimizers are broken because their "weight decay" regularization is not actually weight decay. It proposes to modify Adam to decay all weights the same regardless of the gradient variances.
Calling Ad... | train | [
"SJs7uYYeM",
"rJvLpvdez",
"HJX7HvOez",
"HyXvNUTQG",
"rJ3e48aXM",
"BJ5T7IpXM",
"BJJu7LaXM",
"ryLEQIp7f",
"S1cWEZj7M",
"H1zlFEcxM",
"BJNHyGYlG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public",
"author",
"public"
] | [
"At the heart of the paper, there is a single idea: to decouple the weight decay from the number of steps taken by the optimization process (the paragraph at the end of page 2 is the key to the paper). This is an important and largely overlooked area of implementation and most off-the-shelf optimization algorithms,... | [
7,
4,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rk6qdGgCZ",
"iclr_2018_rk6qdGgCZ",
"iclr_2018_rk6qdGgCZ",
"iclr_2018_rk6qdGgCZ",
"S1cWEZj7M",
"SJs7uYYeM",
"H1zlFEcxM",
"rJvLpvdez",
"iclr_2018_rk6qdGgCZ",
"BJNHyGYlG",
"iclr_2018_rk6qdGgCZ"
] |
iclr_2018_BJaU__eCZ | Hallucinating brains with artificial brains | Human brain function as measured by functional magnetic resonance imaging
(fMRI), exhibits a rich diversity. In response, understanding the individual variability
of brain function and its association with behavior has become one of the
major concerns in modern cognitive neuroscience. Our work is moti... | rejected-papers | The submission proposes to use GANs to learn a generative model of fMRI scans that can then be used for downstream classification tasks. Although there was some appreciation from the reviewers of the approach, there were several important remaining concerns:
1) From Reviewer 1: "Generating high resolution images with... | train | [
"rJirqFBgz",
"B1LfYs_gf",
"rJw8PuhxM",
"SJx6w-67M",
"S1cT_-6XG",
"S1FBv-6QG",
"By-5yr6mf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Quality\n\nThis is a very clear contribution which elegantly demonstrates the use of extensions of GAN variants in the context of neuroimaging.\n\nClarity\n\nThe paper is well-written. Methods and results are clearly described. The authors state significant improvements in classification using generated data. Thes... | [
8,
6,
5,
-1,
-1,
-1,
-1
] | [
5,
4,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJaU__eCZ",
"iclr_2018_BJaU__eCZ",
"iclr_2018_BJaU__eCZ",
"B1LfYs_gf",
"rJirqFBgz",
"rJw8PuhxM",
"iclr_2018_BJaU__eCZ"
] |
iclr_2018_Sy1f0e-R- | An empirical study on evaluation metrics of generative adversarial networks | Despite the widespread interest in generative adversarial networks (GANs), few works have studied the metrics that quantitatively evaluate GANs' performance. In this paper, we revisit several representative sample-based evaluation metrics for GANs, and address the important problem of \emph{how to evaluate the evaluati... | rejected-papers | The problem addressed here is an important one: What is a good evaluation metric for generative models? A good selection of popular metrics are analyzed for their appropriateness for model selection of GANs. Two popular approaches are recommended: the kernel Maximum Mean Discrepancy (MMD) and the 1-Nearest-Neighbour ... | train | [
"HJ_M-jdBG",
"rkyzL1NHz",
"S1zLl6mlz",
"BJquSu9eM",
"HyK44e1bM",
"BJc3SHHZG",
"BJtIwUofM",
"S18QPUjGf",
"r1LDL8oMM",
"BkEGIIjMG",
"BkoCeoOef",
"HkLsEjDlf",
"BJhrEjDgf",
"H1bj91Ixz",
"HJ27tABef",
"r1E16Vmxz",
"Sy2r0j_yG"
] | [
"public",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"author",
"author",
"public",
"public",
"public",
"public"
] | [
"Can we apply these metrics to evaluate conditional versions of GANs or it needs some specific adaptations?\nWhat are the most relevant metrics for evaluating conditional GANs?",
"We have updated our paper with FID results included. The FID score does appear to be an appealing metric according to our criterions."... | [
-1,
-1,
8,
7,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
3,
4,
5,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Sy1f0e-R-",
"Sy2r0j_yG",
"iclr_2018_Sy1f0e-R-",
"iclr_2018_Sy1f0e-R-",
"iclr_2018_Sy1f0e-R-",
"iclr_2018_Sy1f0e-R-",
"S1zLl6mlz",
"BJquSu9eM",
"HyK44e1bM",
"BJc3SHHZG",
"H1bj91Ixz",
"HJ27tABef",
"Sy2r0j_yG",
"r1E16Vmxz",
"iclr_2018_Sy1f0e-R-",
"Sy2r0j_yG",
"iclr_2018_Sy1f0... |
iclr_2018_Syjha0gAZ | Loss Functions for Multiset Prediction | We study the problem of multiset prediction. The goal of multiset prediction is to train a predictor that maps an input to a multiset consisting of multiple items. Unlike existing problems in supervised learning, such as classification, ranking and sequence generation, there is no known order among items in a target mu... | rejected-papers | The submission addresses the problem of multiset prediction, which combines predicting which labels are present, and counting the number of each object. Experiments are shown on a somewhat artificial MNIST setting, and a more realistic problem of the COCO dataset.
There were several concerns raised by the reviewers, ... | train | [
"rktAPxrlG",
"SJVGWfceM",
"SJYge3sxG",
"r1k356QXG",
"rJnxOp7mf",
"SJcPDpX7G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper proposes a type of loss functions for the problem of multiset prediction. A detailed discussion on the intuition is provided and extensive experiments are conducted to show that this loss function indeed provides some performance gain in terms of Exact Matching and F1-score.\n\nThe idea of this paper is... | [
5,
7,
4,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_Syjha0gAZ",
"iclr_2018_Syjha0gAZ",
"iclr_2018_Syjha0gAZ",
"rktAPxrlG",
"SJVGWfceM",
"SJYge3sxG"
] |
iclr_2018_SJme6-ZR- | A Deep Learning Approach for Survival Clustering without End-of-life Signals | The goal of survival clustering is to map subjects (e.g., users in a social network, patients in a medical study) to K clusters ranging from low-risk to high-risk. Existing survival methods assume the presence of clear \textit{end-of-life} signals or introduce them artificially using a pre-defined timeout. In this pape... | rejected-papers | The submission proposes a Kuiper statistic based loss function for survival clustering. This loss function is applied to train a deep network. Results are presented on a Friendster dataset.
This submission received borderline/mixed reviews. The primary concerns were: justification of the Kuiper loss, lack of detail... | train | [
"ry3ETTtxG",
"HyM5JWhgG",
"S1QsSa1-M",
"BJqJHI67z",
"BJ4iN8p7M",
"S1hGEIa7f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Pros:\nThe paper is a nice read, clearly written, and its originality is well stated by the authors, “addressing the lifetime clustering problem without end-of-life signals for the first time”. I do not feel experienced enough in the field to evaluate the significance of this work.\n\nThe approach proposed in the ... | [
6,
4,
6,
-1,
-1,
-1
] | [
1,
4,
5,
-1,
-1,
-1
] | [
"iclr_2018_SJme6-ZR-",
"iclr_2018_SJme6-ZR-",
"iclr_2018_SJme6-ZR-",
"ry3ETTtxG",
"HyM5JWhgG",
"S1QsSa1-M"
] |
iclr_2018_Bys_NzbC- | Achieving Strong Regularization for Deep Neural Networks | L1 and L2 regularizers are critical tools in machine learning due to their ability to simplify solutions. However, imposing strong L1 or L2 regularization with gradient descent method easily fails, and this limits the generalization ability of the underlying neural networks. To understand this phenomenon, we investigat... | rejected-papers | The submission is motivated by an empirical observation of a phase transition when a sufficiently high L1 or L2 penalty on the weights is applied. The proposed solution is to optimize for several epochs without the penalty followed by introduction of the penalty. Although empirical results seem to moderately support ... | train | [
"rk43lpulz",
"BJk4jatgM",
"H1U0Hpcef",
"BJFUdLTQM",
"SJp9GptXM",
"HJNPfatQf",
"B1ej93KXz",
"rJLJ_3KXz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"The authors studied the behavior that a strong regularization parameter may lead to poor performance in training of deep neural networks. Experimental results on CIFAR-10 and CIFAR-100 were reported using AlexNet and VGG-16. The results seem to show that a delayed application of the regularization parameter leads ... | [
6,
4,
5,
-1,
-1,
-1,
-1,
-1
] | [
2,
5,
5,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Bys_NzbC-",
"iclr_2018_Bys_NzbC-",
"iclr_2018_Bys_NzbC-",
"iclr_2018_Bys_NzbC-",
"BJk4jatgM",
"BJk4jatgM",
"rk43lpulz",
"H1U0Hpcef"
] |
iclr_2018_HkOhuyA6- | Graph Classification with 2D Convolutional Neural Networks | Graph classification is currently dominated by graph kernels, which, while powerful, suffer some significant limitations. Convolutional Neural Networks (CNNs) offer a very appealing alternative. However, processing graphs with CNNs is not trivial. To address this challenge, many sophisticated extensions of CNNs have re... | rejected-papers | The submission proposes a strategy for creating vector representations of graphs, upon which a CNN can be applied. Although this is a useful problem to solve, there are multiple works in the existing literature for doing so. Given that the choice between these is essentially empirical, a through comparison is necessa... | train | [
"BJL5gdKgG",
"H1W1OsYxG",
"HkyP9hteG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The authors propose to use 2D CNNs for graph classification by transforming graphs to an image-like representation from its node embedding. The approach uses node2vec to obtain a node embedding, which is then compacted using PCA and turned into a stack of discretized histograms. Essentially the authors propose an ... | [
3,
4,
7
] | [
5,
3,
3
] | [
"iclr_2018_HkOhuyA6-",
"iclr_2018_HkOhuyA6-",
"iclr_2018_HkOhuyA6-"
] |
iclr_2018_SyW4Gjg0W | Kernel Graph Convolutional Neural Nets | Graph kernels have been successfully applied to many graph classification problems. Typically, a kernel is first designed, and then an SVM classifier is trained based on the features defined implicitly by this kernel. This two-stage approach decouples data representation from learning, which is suboptimal. On the other... | rejected-papers | The reviewers were unanimous in their assessment that the paper was not ready for publication in ICLR. Their concerns included:
- lack of novelty over Niepert, Ahmed, Kutzkov, ICML 2016
- The approach learns combinations of graph kernels and its expressive capacity is thus limited
- The results are close to the sta... | train | [
"rywsxcOxG",
"BJS9j2YlM",
"rJstP6Fef"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper proposes a graph classification method by integrating three techniques, community detection, graph kernels, and CNNs.\n\n* This paper is clearly written and easy to follow. Thus the clarity is high.\n\n* The originality is not high as the application of neural networks for graph classification has alrea... | [
5,
5,
4
] | [
5,
4,
5
] | [
"iclr_2018_SyW4Gjg0W",
"iclr_2018_SyW4Gjg0W",
"iclr_2018_SyW4Gjg0W"
] |
iclr_2018_BkVsWbbAW | Deep Generative Dual Memory Network for Continual Learning | Despite advances in deep learning, artificial neural networks do not learn the same way as humans do. Today, neural networks can learn multiple tasks when trained on them jointly, but cannot maintain performance on learnt tasks when tasks are presented one at a time -- this phenomenon called catastrophic forgetting is ... | rejected-papers | Thank you for submitting you paper to ICLR. The big-picture idea is fairly simple, although the implementation is certainly challenging requiring a deep generative model to be trained as part of the final system. The experimental validation is not sufficient to warrant publication. A comparison to a larger number of co... | train | [
"rkVGk_UEG",
"SydEmAKgM",
"rk8n0esxf",
"S1t718Tef",
"B1uEjlZXM",
"BJf-jeW7z",
"Hk_3rebmG",
"SJTSBxZmf",
"r1gLq2olz",
"rkb_ZMilM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Using inspirations from brain architecture and mechanism is certainly great direction and I applause authors for good referencing of some of the relevant literature . But I get a sense from the authors both in the paper and in their comment that they are over emphasizing the link of their work and how brain functi... | [
-1,
5,
6,
7,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
4,
2,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"BJf-jeW7z",
"iclr_2018_BkVsWbbAW",
"iclr_2018_BkVsWbbAW",
"iclr_2018_BkVsWbbAW",
"SydEmAKgM",
"SydEmAKgM",
"rk8n0esxf",
"S1t718Tef",
"rkb_ZMilM",
"iclr_2018_BkVsWbbAW"
] |
iclr_2018_HJ5AUm-CZ | The Variational Homoencoder: Learning to Infer High-Capacity Generative Models from Few Examples | Hierarchical Bayesian methods have the potential to unify many related tasks (e.g. k-shot classification, conditional, and unconditional generation) by framing each as inference within a single generative model. We show that existing approaches for learning such models can fail on expressive generative networks such as... | rejected-papers | Thank you for submitting your paper to ICLR. The reviewers agree that the idea of sharing the approximating distribution across sets of variables is an interesting one and that the Omniglot experiments are thorough. However, although the authors make the nice addition of some simple examples during the revision period ... | train | [
"rJX5uPLNG",
"BkviGptxG",
"rkJkoJogf",
"H1V4eZ3lG",
"HJyH8DTQM",
"SyleUDaQG",
"B1e9BP67M",
"H1ufSPa7z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Thank you for making edits to the technical portion of the paper. I believe the changes improve the paper's readability.",
"This paper presents an alternative approach to constructing variational lower bounds on data log likelihood in deep, directed generative models with latent variables. Specifically, the auth... | [
-1,
7,
5,
6,
-1,
-1,
-1,
-1
] | [
-1,
4,
5,
3,
-1,
-1,
-1,
-1
] | [
"HJyH8DTQM",
"iclr_2018_HJ5AUm-CZ",
"iclr_2018_HJ5AUm-CZ",
"iclr_2018_HJ5AUm-CZ",
"BkviGptxG",
"rkJkoJogf",
"H1V4eZ3lG",
"iclr_2018_HJ5AUm-CZ"
] |
iclr_2018_Hkp3uhxCW | Revisiting Bayes by Backprop | In this work we explore a straightforward variational Bayes scheme for Recurrent Neural Networks.
Firstly, we show that a simple adaptation of truncated backpropagation through time can yield good quality uncertainty estimates and superior regularisation at only a small extra computational cost during training, a... | rejected-papers | Thank you for submitting you paper to ICLR. The revision improved the paper e.g. moving Appendix A3 to the main text has improved clarity, but, like reviewer 3, I still found section 4 hard to follow. As the authors suggest, shifting the terminology to "posterior shifting” rather than “sharpening" would help at a high ... | train | [
"BJZRkfFgG",
"Hy2OnpKeG",
"rk156h2gf",
"Sk5DnNTXz",
"rynmh4amz",
"ry8ynV67G",
"H1d6T1y7z",
"HkueR0FlG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public",
"public"
] | [
"*Summary*\n\nThe paper applies variational inference (VI) with the 'reparameterisation' trick for Bayesian recurrent neural networks (BRNNs). The paper first considers the \"Bayes by Backprop\" approach of Blundell et al. (2015) and then modifies the BRNN model with a hierarchical prior over the network parameters... | [
5,
6,
6,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Hkp3uhxCW",
"iclr_2018_Hkp3uhxCW",
"iclr_2018_Hkp3uhxCW",
"BJZRkfFgG",
"Hy2OnpKeG",
"rk156h2gf",
"Hy2OnpKeG",
"iclr_2018_Hkp3uhxCW"
] |
iclr_2018_S1fduCl0b | Lifelong Generative Modeling | Lifelong learning is the problem of learning multiple consecutive tasks in a sequential manner where knowledge gained from previous tasks is retained and used for future learning. It is essential towards the development of intelligent machines that can adapt to their surroundings. In this work we focus on a lifelong le... | rejected-papers | Thank you for submitting you paper to ICLR. The paper studies an interesting problem and the solution, which fuses student-teacher approaches to continual learning and variational auto-encoders, is interesting. The revision of the paper has improved readability. However, although the framework is flexible, it is comple... | train | [
"rJuN3apyf",
"SJuF9Eqez",
"rJokGmjgG",
"rJlVE72ZG",
"SJHQwXnWM",
"H1Y7BmhZz",
"r1Y8VQ3bz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"We have seen numerous variants of variational autoencoders, most of them introducing delta changes to the original architecture to address the same sort of modeling problems. This paper attacks a different kind of problem, namely lifelong learning. This key aspect of the paper, besides the fact that it constitutes... | [
9,
4,
4,
-1,
-1,
-1,
-1
] | [
5,
5,
2,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1fduCl0b",
"iclr_2018_S1fduCl0b",
"iclr_2018_S1fduCl0b",
"rJokGmjgG",
"rJuN3apyf",
"SJuF9Eqez",
"rJokGmjgG"
] |
iclr_2018_BkDB51WR- | Learning temporal evolution of probability distribution with Recurrent Neural Network | We propose to tackle a time series regression problem by computing temporal evolution of a probability density function to provide a probabilistic forecast. A Recurrent Neural Network (RNN) based model is employed to learn a nonlinear operator for temporal evolution of a probability density function. We use a softmax l... | rejected-papers | Thank you for submitting you paper to ICLR. Two of the reviewers are concerned that the paper’s contributions are not significant enough —either in terms of the theoretical or experimental contribution -- to warrant publication. The authors have improved the experimental aspect to include a more comprehensive compariso... | train | [
"rJwjY_1xG",
"rJrHPz9lG",
"B1fvgpNbz",
"HyNuJjMmz",
"ryBQ1sGQM",
"Hkvi0qMmM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Interesting ideas that extend LSTM to produce probabilistic forecasts for univariate time series, experiments are okay. Unclear if this would work at all in higher-dimensional time series. It is also unclear to me what are the sources of the uncertainties captured.\n\n\nThe author proposed to incorporate 2 differe... | [
6,
5,
6,
-1,
-1,
-1
] | [
2,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_BkDB51WR-",
"iclr_2018_BkDB51WR-",
"iclr_2018_BkDB51WR-",
"rJwjY_1xG",
"rJrHPz9lG",
"B1fvgpNbz"
] |
iclr_2018_B1nLkl-0Z | Learning Gaussian Policies from Smoothed Action Value Functions | State-action value functions (i.e., Q-values) are ubiquitous in reinforcement learning (RL), giving rise to popular algorithms such as SARSA and Q-learning. We propose a new notion of action value defined by a Gaussian smoothed version of the expected Q-value used in SARSA. We show that such smoothed Q-values still sat... | rejected-papers | Thank you for submitting you paper to ICLR. Two of the reviewers are concerned that the paper’s contributions are not significant enough —either in terms of the theoretical or experimental contribution -- to warrant publication. The authors have improved the experimental aspect to include a more comprehensive compariso... | train | [
"S1qQ8ZFlf",
"ByJ1CsYgG",
"HkzHPhmWf",
"rywix_TQG",
"BkBxlO6Qz",
"rk6c1_6Qf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"I think I should understand the gist of the paper, which is very interesting, where the action of \\tilde Q(s,a) is drawn from a distribution. The author also explains in detail the relation with PGQ/Soft Q learning, and the recent paper \"expected policy gradient\" by Ciosek & Whiteson. All these seems very sound... | [
6,
6,
5,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_B1nLkl-0Z",
"iclr_2018_B1nLkl-0Z",
"iclr_2018_B1nLkl-0Z",
"HkzHPhmWf",
"ByJ1CsYgG",
"S1qQ8ZFlf"
] |
iclr_2018_SySisz-CW | On the difference between building and extracting patterns: a causal analysis of deep generative models. | Generative models are important tools to capture and investigate the properties of complex empirical data. Recent developments such as Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs) use two very similar, but \textit{reverse}, deep convolutional architectures, one to generate and one to extr... | rejected-papers | Thank you for submitting you paper to ICLR. The paper presents an interesting analysis, but the utility of this analysis is questionable e.g. it is not clear how this might lead to improved VAEs/GANs. The authors did add an additional experimental result in their revised paper, but questions still remain. In light of t... | val | [
"H1CZob5Jz",
"rkA0vi8gz",
"BJT2Cecgz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper examines the nature of convolutional filters in the encoder and a decoder of a VAE, and a generator and a discriminator of a GAN. The authors treat the inputs (X) and outputs (Y) of each filter throughout each step of the convolving process as a time series, which allows them to do a Discrete Time Fouri... | [
2,
7,
7
] | [
4,
3,
2
] | [
"iclr_2018_SySisz-CW",
"iclr_2018_SySisz-CW",
"iclr_2018_SySisz-CW"
] |
iclr_2018_rkcya1ZAW | Continuous-Time Flows for Efficient Inference and Density Estimation | Two fundamental problems in unsupervised learning are efficient inference for latent-variable models and robust density estimation based on large amounts of unlabeled data. For efficient inference, normalizing flows have been recently developed to approximate a target distribution arbitrarily well. In practice, however... | rejected-papers | Thank you for submitting you paper to ICLR. The consensus from the reviewers is that there are some interesting theoretical contributions and some promising experimental support. However, although the paper is moving in the right direction, they believe that it is not quite ready for publication. | train | [
"HkCNqISxM",
"r1_fQy9lz",
"Hy8TV-qgf",
"ryZFenTGf",
"SyAtlT_zG",
"ByCUxp_GG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The authors try to use continuous time generalizations of normalizing flows for improving upon VAE-like models or for standard density estimation problems.\n\nClarity: the text is mathematically very sloppy / hand-wavy.\n\n1. I do not understand proposition (1). I do not think that the proof is correct (e.g. the g... | [
3,
6,
6,
-1,
-1,
-1
] | [
3,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_rkcya1ZAW",
"iclr_2018_rkcya1ZAW",
"iclr_2018_rkcya1ZAW",
"r1_fQy9lz",
"HkCNqISxM",
"Hy8TV-qgf"
] |
iclr_2018_rJLTTe-0W | Bayesian Time Series Forecasting with Change Point and Anomaly Detection | Time series forecasting plays a crucial role in marketing, finance and many other quantitative fields. A large amount of methodologies has been developed on this topic, including ARIMA, Holt–Winters, etc. However, their performance is easily undermined by the existence of change points and anomaly points, two structure... | rejected-papers | Thank you for submitting you paper to ICLR. The consensus from the reviewers is that this is not quite ready for publication. There is also concern about whether ICLR, with its focus on representational learning, is the right venue for this work.
One of the reviewers initially submitted an incorrect review, but this ... | train | [
"HJ0Hc82gM",
"HJhn9OtxG",
"HJL1pxqeG",
"rkAJXOnQz",
"BJljaSkzf",
"rkqRmuhQz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"\n\nSummary:\n\nThis paper develops a state space time series forecasting model in the Bayesian framework, jointly detects anomaly and change points. Integrated with an iterative MCMC method, the authors develop an efficient algorithm and use both synthetic and real data set to demonstrate that their algorithms ou... | [
5,
6,
4,
-1,
-1,
-1
] | [
5,
3,
5,
-1,
-1,
-1
] | [
"iclr_2018_rJLTTe-0W",
"iclr_2018_rJLTTe-0W",
"iclr_2018_rJLTTe-0W",
"HJL1pxqeG",
"HJ0Hc82gM",
"HJhn9OtxG"
] |
iclr_2018_r1drp-WCZ | State Space LSTM Models with Particle MCMC Inference | Long Short-Term Memory (LSTM) is one of the most powerful sequence models. Despite the strong performance, however, it lacks the nice interpretability as in state space models. In this paper, we present a way to combine the best of both worlds by introducing State Space LSTM (SSL), which generalizes the earlier work \c... | rejected-papers | Thank you for submitting you paper to ICLR. The consensus from the reviewers is that this is not quite ready for publication. The work is related to (although different from) Gu et al Neural Sequential Monte Carlo NIPS2015 and it would be useful to point this out in the related work section. | train | [
"SJLHg2OxG",
"SyI_srKgz",
"r1qsxyTlG",
"B134ZGMEM",
"Bk3ogjt7G",
"S1YA4adQz",
"rkFg-CTMM",
"S1-q0p6Mz",
"B1qlAppGG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"This article presents an approach for learning and inference in nonlinear state-space models (SSM) based on LSTMs. Learning is done using a stochastic EM where Particle PMCM is used to sample state trajectories.\n\nThe model is presented assuming that SSMs are linear. This is not necessarily the case since nonline... | [
3,
5,
7,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1drp-WCZ",
"iclr_2018_r1drp-WCZ",
"iclr_2018_r1drp-WCZ",
"Bk3ogjt7G",
"S1YA4adQz",
"S1-q0p6Mz",
"SJLHg2OxG",
"SyI_srKgz",
"r1qsxyTlG"
] |
iclr_2018_S1tWRJ-R- | Joint autoencoders: a flexible meta-learning framework | The incorporation of prior knowledge into learning is essential in achieving good performance based on small noisy samples. Such knowledge is often incorporated through the availability of related data arising from domains and tasks similar to the one of current interest. Ideally one would like to allow both the data f... | rejected-papers | Thank you for submitting you paper to ICLR. ICLR. The consensus from the reviewers is that this is not quite ready for publication. In particular, the experimental results are promising, but further work is required to fully demonstrate the efficacy of the approach. | train | [
"H1cgp9qxf",
"rkseIXolf",
"HJzFmcNZM",
"ByQN_opGz",
"By6WviTfz",
"ByOi8oaGf",
"Byv5BsTzM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The work proposed a generic framework for end-to-end transfer learning / domain adaptation with deep neural networks. The idea is to learn a joint autoencoders, containing private branch with task/domain-specific weights, as well as common branch consisting of shared weights used across tasks/domains, as well as t... | [
4,
5,
5,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1tWRJ-R-",
"iclr_2018_S1tWRJ-R-",
"iclr_2018_S1tWRJ-R-",
"H1cgp9qxf",
"rkseIXolf",
"HJzFmcNZM",
"iclr_2018_S1tWRJ-R-"
] |
iclr_2018_HkbJTYyAb | Convolutional Normalizing Flows | Bayesian posterior inference is prevalent in various machine learning problems. Variational inference provides one way to approximate the posterior distribution, however its expressive power is limited and so is the accuracy of resulting approximation. Recently, there has a trend of using neural networks to approximate... | rejected-papers | Thank you for submitting you paper to ICLR. ICLR. Although there revision has improved the paper, the consensus from the reviewers is that this is not quite ready for publication. | train | [
"HJNuic4gz",
"By8otd_ef",
"S1IcSyqxM",
"SkQLkdTXM",
"SJx9SdpXG",
"ry0g4ua7M",
"Bk5dG_pmf",
"BylwwsmlG",
"HyQ4crGlG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"official_reviewer"
] | [
"In this paper, the authors propose a type of Normalizing Flows (Rezende and Mohamed, 2015) for Variational Autoencoders (Kingma and Welling, 2014; Rezende et al., 2014) they call Convolutional Normalizing Flows.\nMore particularly, it aims at extending on the Planar Flow scheme proposed in Rezende and Mohamed (201... | [
3,
5,
3,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HkbJTYyAb",
"iclr_2018_HkbJTYyAb",
"iclr_2018_HkbJTYyAb",
"iclr_2018_HkbJTYyAb",
"By8otd_ef",
"HJNuic4gz",
"S1IcSyqxM",
"HyQ4crGlG",
"iclr_2018_HkbJTYyAb"
] |
iclr_2018_rkhCSO4T- | Distributed non-parametric deep and wide networks | In recent work, it was shown that combining multi-kernel based support vector machines (SVMs) can lead to near state-of-the-art performance on an action recognition dataset (HMDB-51 dataset). In the present work, we show that combining distributed Gaussian Processes with multi-stream deep convolutional neural networks ... | rejected-papers | Thank you for submitting you paper to ICLR. ICLR. The consensus from the reviewers is that this is not quite ready for publication. | test | [
"rkG37k9xM",
"SJ2q6Rqgf",
"HJ_YkBz-G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"- The paper is fairly written and it is clear what is being done\n- There is not much novelty in the paper; it combines known techniques and is a systems paper, so I \n would judge the contributions mainly in terms of the empirical results and messsage conveyed (see\n third point)\n- The paper builds on a previ... | [
3,
3,
3
] | [
4,
5,
4
] | [
"iclr_2018_rkhCSO4T-",
"iclr_2018_rkhCSO4T-",
"iclr_2018_rkhCSO4T-"
] |
iclr_2018_HJjePwx0- | Better Generalization by Efficient Trust Region Method | In this paper, we develop a trust region method for training deep neural networks. At each iteration, trust region method computes the search direction by solving a non-convex subproblem. Solving this subproblem is non-trivial---existing methods have only sub-linear convergence rate. In the first part, we show that a s... | rejected-papers | There are two parts to this paper (1) an efficient procedure for solving trust-region subproblems in second-order optimization of neural nets, and (2) evidence that the proposed trust region method leads to better generalization performance than SGD in the large-batch setting. In both cases, there are some promising le... | test | [
"B1BvO3Bez",
"HybcslTyf",
"HJK45bIez",
"H1S9sm8MG",
"S1CpgVUMz",
"rJKPRmIfM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"**I am happy to see some good responses from the authors to my questions. I am raising my score a bit higher. \n\nSummary: \nA new stochastic method based on trust region (TR) is proposed. Experiments show improved generalization over mini-batch SGD, which is the main positive aspect of this paper. The main algori... | [
6,
5,
6,
-1,
-1,
-1
] | [
5,
2,
3,
-1,
-1,
-1
] | [
"iclr_2018_HJjePwx0-",
"iclr_2018_HJjePwx0-",
"iclr_2018_HJjePwx0-",
"HJK45bIez",
"HybcslTyf",
"B1BvO3Bez"
] |
iclr_2018_Byk4My-RZ | Flexible Prior Distributions for Deep Generative Models | We consider the problem of training generative models with deep neural networks as generators, i.e. to map latent codes to data points. Whereas the dominant paradigm combines simple priors over codes with complex deterministic models,
we argue that it might be advantageous to use more flexible code distributions.... | rejected-papers | This paper presents a method for learning more flexible prior distributions for GANs by learning another distribution on top of the latent codes for training examples. It's reminiscent of layerwise training of deep generative models. This seems like a reasonable thing to do, but it's probably not a substantial enough c... | train | [
"H1k_ZpFlf",
"SyoujCYgG",
"S1TJm_gZG",
"SJ31YST7f",
"ByCPOSp7M",
"BJwavHp7z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary:\n\nThe paper proposes to learn new priors for latent codes z for GAN training. for this the paper shows that there is a mismatch between the gaussian prior and an estimated of the latent codes of real data by reversal of the generator . To fix this the paper proposes to learn a second GAN to learn the p... | [
6,
6,
5,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1
] | [
"iclr_2018_Byk4My-RZ",
"iclr_2018_Byk4My-RZ",
"iclr_2018_Byk4My-RZ",
"SyoujCYgG",
"H1k_ZpFlf",
"S1TJm_gZG"
] |
iclr_2018_H1rRWl-Cb | An information-theoretic analysis of deep latent-variable models | We present an information-theoretic framework for understanding trade-offs in unsupervised learning of deep latent-variables models using variational inference. This framework emphasizes the need to consider latent-variable models along two dimensions: the ability to reconstruct inputs (distortion) and the communicatio... | rejected-papers | This paper gives a coding theory interpretation of VAEs and uses it to motivate an additional knob for tuning and evaluating VAEs: namely, the tradeoff between the rate and the distortion. This is a useful set of dimensions to investigate, and past work on variational models has often found it advantageous to penalize ... | train | [
"SkPsD5dxz",
"HJoM4eclG",
"H1Je5milG",
"Skkvvu3mM",
"rJbQPO37f",
"Hk83yunQf",
"B1Gpxdn7M",
"BJL5eunmz",
"S10iNmsJM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Summary:\n\nThis paper optimizes the beta-VAE objective and analyzes the resulting models in terms of the two components of the VAE loss: the reconstruction error (which the authors refer to as distortion, “D”) and the KL divergence term (which the authors refer to as rate, “R”). Various VAEs using either PixelCNN... | [
5,
7,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1rRWl-Cb",
"iclr_2018_H1rRWl-Cb",
"iclr_2018_H1rRWl-Cb",
"S10iNmsJM",
"SkPsD5dxz",
"iclr_2018_H1rRWl-Cb",
"HJoM4eclG",
"H1Je5milG",
"iclr_2018_H1rRWl-Cb"
] |
iclr_2018_rkMt1bWAZ | Bias-Variance Decomposition for Boltzmann Machines | We achieve bias-variance decomposition for Boltzmann machines using an information geometric formulation. Our decomposition leads to an interesting phenomenon that the variance does not necessarily increase when more parameters are included in Boltzmann machines, while the bias always decreases. Our result gives a theo... | rejected-papers | This paper presents a bias/variance decomposition for Boltzmann machines using the generalized Pythagorean Theorem from information geometry. The main conclusion is that counterintuitively, the variance may decrease as the model is made larger. There are probably some interesting ideas here, but there isn't a clear tak... | train | [
"HknVXoHez",
"rkv5HU5gG",
"Sy_JLe5lz",
"rJFkOigXM",
"rkp5vjgQM",
"ryDvPolQG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary of the paper:\nThe paper derives a lower bound on the expected squared KL-divergence between a true distribution and the sample based maximum likelihood estimate (MLE) of that distribution modelled by an Boltzmann machine (BM) based on methods from information geometry. This KL-divergence is first split ... | [
5,
7,
5,
-1,
-1,
-1
] | [
2,
5,
5,
-1,
-1,
-1
] | [
"iclr_2018_rkMt1bWAZ",
"iclr_2018_rkMt1bWAZ",
"iclr_2018_rkMt1bWAZ",
"HknVXoHez",
"Sy_JLe5lz",
"rkv5HU5gG"
] |
iclr_2018_SJOl4DlCZ | Classifier-to-Generator Attack: Estimation of Training Data Distribution from Classifier | Suppose a deep classification model is trained with samples that need to be kept private for privacy or confidentiality reasons. In this setting, can an adversary obtain the private samples if the classification model is given to the adversary? We call this reverse engineering against the classification model the Class... | rejected-papers | This paper addresses the very important problem of ensuring that sensitive training data remain private. It proposes an attack whereby the attacker can reconstruct information about the training data given only the trained classifier and an auxiliary dataset. If done well, such an attack would be a useful contribution ... | train | [
"r1-fz-MJG",
"r1IcQO8lz",
"BkQD60b-f",
"HyaMD51Qf",
"rJHgvqJ7z",
"SJz5U517G",
"HkmZ89kXz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper considers a new problem : given a classifier f trained from D_tr and a set of auxillary samples from D_aux, find D_tr conditioned on label t*. Its solution is based on a new GAN: preImageGAN. Three settings of the similarity between auxillary distribution and training distribution is considered: exact s... | [
7,
4,
4,
-1,
-1,
-1,
-1
] | [
3,
3,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJOl4DlCZ",
"iclr_2018_SJOl4DlCZ",
"iclr_2018_SJOl4DlCZ",
"r1IcQO8lz",
"r1-fz-MJG",
"BkQD60b-f",
"iclr_2018_SJOl4DlCZ"
] |
iclr_2018_ByuI-mW0W | Towards a Testable Notion of Generalization for Generative Adversarial Networks | We consider the question of how to assess generative adversarial networks, in particular with respect to whether or not they generalise beyond memorising the training data. We propose a simple procedure for assessing generative adversarial network performance based on a principled consideration of what the actual goal ... | rejected-papers | This paper proposes a method for quantitatively evaluating GANs. Better quantitative metrics for GANs are badly needed, as the field is being held back by excessive focus on generated samples. This paper proposes to estimate the Wasserstein distance to the data distribution. A paper which does this well would be a sign... | train | [
"rJk_SwYxz",
"B1P-gBclf",
"ry88vO5ez",
"rk1McFaQf",
"r1qCuET7G",
"HkY5m4aQM",
"HyY42ZLgf",
"BJKBqG7xf",
"BJJWvcexG",
"HkExdq2JG",
"rJOyommkf",
"Hka0Iobyf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"official_reviewer",
"author",
"author",
"public",
"public"
] | [
"`The papers aims to provide a quality measure/test for GANs. The objective is ambitious an deserve attention. As GANs are minimizing some f-divergence measure, the papers remarks that computing a Wasserstein distance between two distributions made of a sum of Diracs is not a degenerate case and is tractable. So... | [
5,
6,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByuI-mW0W",
"iclr_2018_ByuI-mW0W",
"iclr_2018_ByuI-mW0W",
"B1P-gBclf",
"rJk_SwYxz",
"ry88vO5ez",
"BJKBqG7xf",
"iclr_2018_ByuI-mW0W",
"Hka0Iobyf",
"rJOyommkf",
"iclr_2018_ByuI-mW0W",
"iclr_2018_ByuI-mW0W"
] |
iclr_2018_S1EwLkW0W | Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients | The ADAM optimizer is exceedingly popular in the deep learning community. Often it works very well, sometimes it doesn’t. Why? We interpret ADAM as a combination of two aspects: for each weight, the update direction is determined by the sign of the stochastic gradient, whereas the update magnitude is solely determined ... | rejected-papers | This paper presents a theoretical justification for the Adam optimizer in terms of decoupling the signs and magnitudes of the gradients. The overall analysis seems reasonable, though there's been much back-and-forth with the reviewers about particular claims and assumptions. Overall, the contributions don't feel quite ... | val | [
"r1mT8HDgf",
"S1urbvOgf",
"By-KPs9eM",
"SJtm7zOZM",
"HJbFyfdWz",
"HkqnfeuZf",
"HJOTq8P-M",
"HJOpoC8Zf",
"H1c9cRIbM",
"rkV11bcAW"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary: \nThe paper is trying to improve Adam based on variance adaption with momentum. Two algorithms are proposed, M-SSD (Stochastic Sign Descent with Momentum) and M-SVAG (Stochastic Variance-Adapted Gradient with Momentum) to solve finite sum minimization problem. The convergence analysis is provided for SVAG... | [
4,
4,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1EwLkW0W",
"iclr_2018_S1EwLkW0W",
"iclr_2018_S1EwLkW0W",
"HkqnfeuZf",
"By-KPs9eM",
"HJOTq8P-M",
"HJOpoC8Zf",
"r1mT8HDgf",
"S1urbvOgf",
"iclr_2018_S1EwLkW0W"
] |
iclr_2018_HJYoqzbC- | A comparison of second-order methods for deep convolutional neural networks | Despite many second-order methods have been proposed to train neural networks, most of the results were done on smaller single layer fully connected networks, so we still cannot conclude whether it's useful in training deep convolutional networks. In this study, we conduct extensive experiments to answer the question "... | rejected-papers | This paper investigates the performance of various second-order optimization methods for training neural networks. Comparing different optimizers is worthwhile, but as this is an empirical paper which doesn't present novel techniques, the bar is very high for the experimental methodology. Unfortunately, I don't think t... | train | [
"rk_w_Rrlz",
"Bkc6tWIgf",
"rkOSPgqef",
"SyLhb3IXz",
"SJMkznLQf",
"BJbwWnIQz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"A good experimentation of second order methods for training large DNNs in comparison with the popular SGD method has been lacking in the literature. This paper tries to fill that gap. Though there are some good experiments, I feel it could have been much better and more complete.\n\nSeveral candidates for second o... | [
5,
6,
3,
-1,
-1,
-1
] | [
5,
3,
4,
-1,
-1,
-1
] | [
"iclr_2018_HJYoqzbC-",
"iclr_2018_HJYoqzbC-",
"iclr_2018_HJYoqzbC-",
"rkOSPgqef",
"Bkc6tWIgf",
"iclr_2018_HJYoqzbC-"
] |
iclr_2018_ByJDAIe0b | Integrating Episodic Memory into a Reinforcement Learning Agent Using Reservoir Sampling | Episodic memory is a psychology term which refers to the ability to recall specific events from the past. We suggest one advantage of this particular type of memory is the ability to easily assign credit to a specific state when remembered information is found to be useful. Inspired by this idea, and the increasing pop... | rejected-papers | This paper presents a memory architecture for RL based on reservoir sampling, and is meant to be an alternative to RNNs. The reviewers consider the idea to be potentially interesting and useful, but have concerns about the mathematical justification. They also point out limitations in the experiments: in particular, us... | train | [
"HkoFNQpeG",
"SkCS_v6ez",
"SyZdbUQZz",
"SkOS7RHWG",
"Byo6WRS-z",
"SJl_lCrZz",
"BkAxeRHWG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper considers a new way to incorporate episodic memory with shallow-neural-nets RL using reservoir sampling. The authors propose a reservoir sampling algorithm for drawing samples from the memory. Some theoretical guarantees for the efficiency of reservoir sampling are provided. The whole algorithm is teste... | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByJDAIe0b",
"iclr_2018_ByJDAIe0b",
"iclr_2018_ByJDAIe0b",
"HkoFNQpeG",
"SyZdbUQZz",
"SkCS_v6ez",
"iclr_2018_ByJDAIe0b"
] |
iclr_2018_H13WofbAb | Faster Distributed Synchronous SGD with Weak Synchronization | Distributed training of deep learning is widely conducted with large neural networks and large datasets. Besides asynchronous stochastic gradient descent~(SGD), synchronous SGD is a reasonable alternative with better convergence guarantees. However, synchronous SGD suffers from stragglers. To make things worse, althoug... | rejected-papers | This paper introduces a method for making synchronous SGD more resistant to failed or slow workers. The idea seems plausible, but as the reviewers point out, the novelty and the experimental validation are somewhat limited. For a contribution such as this, it would be good to see some experiments on a wider range of ta... | train | [
"H1SAHAdlf",
"Sy8xDgYxG",
"HkVrKHYlG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper introduces a parameter server architecture to improve distributed training of CNNs in the presence of stragglers. Specifically, the paper proposes partial pulling where a worker only waits for first b blocks rather than all the blocks of the parameters. This technique is combined with existing methods s... | [
4,
3,
4
] | [
5,
4,
5
] | [
"iclr_2018_H13WofbAb",
"iclr_2018_H13WofbAb",
"iclr_2018_H13WofbAb"
] |
iclr_2018_BJLmN8xRW | Character Level Based Detection of DGA Domain Names | Recently several different deep learning architectures have been proposed that take a string of characters as the raw input signal and automatically derive features for text classification. Little studies are available that compare the effectiveness of these approaches for character based text classification with each ... | rejected-papers | meta score: 4
This is basically an application in which some different deep learning approaches are compared on the task of automatically identifying domain names automatically generated by malware. The experiments are well-constructed and reported. However, the work does not have novelty beyond the application doma... | train | [
"H18dy2IEM",
"HJukYvYxf",
"SJ7ulqYxz",
"S1SG_l5gz",
"S1ZaxLkfz",
"H1RFyLkMf",
"r1xf1LJGz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"I appreciate the effort to include the additional experiments.\n\nThe positive points of this paper remain the correct technical evaluation and the multiple models being evaluated.\nTechnically the work appears to be solid and improved in the revision thanks to the additional experiments.\n\nUnfortunately, given t... | [
-1,
7,
5,
4,
-1,
-1,
-1
] | [
-1,
4,
3,
4,
-1,
-1,
-1
] | [
"H1RFyLkMf",
"iclr_2018_BJLmN8xRW",
"iclr_2018_BJLmN8xRW",
"iclr_2018_BJLmN8xRW",
"HJukYvYxf",
"SJ7ulqYxz",
"S1SG_l5gz"
] |
iclr_2018_rkrWCJWAW | Unbiasing Truncated Backpropagation Through Time | \emph{Truncated Backpropagation Through Time} (truncated BPTT, \cite{jaeger2002tutorial}) is a widespread method for learning recurrent computational graphs. Truncated BPTT keeps the computational benefits of \emph{Backpropagation Through Time} (BPTT \cite{werbos:bptt}) while relieving the need for a complete backtrack... | rejected-papers | Meta score: 5
The paper explores an interesting idea, addressing a known bias in truncated BPTT by sampling across different truncated history lengths. Limited theoretical analysis is presented along with PTB language modelling experimentation. The experimental part could be stronger (e.g. trying to improve over th... | train | [
"rkvD7QulM",
"rJRZM0txz",
"r11tZl5xG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper proposes stochastic determination methods for truncation points in backpropagation through time. The previous truncation methods naively determine truncation points with fixed intervals, however, these methods cannot ensure the unbiasedness of gradients. The proposed methods stochastically determine tru... | [
6,
5,
5
] | [
3,
4,
4
] | [
"iclr_2018_rkrWCJWAW",
"iclr_2018_rkrWCJWAW",
"iclr_2018_rkrWCJWAW"
] |
iclr_2018_rJJzTyWCZ | Large-scale Cloze Test Dataset Designed by Teachers | Cloze test is widely adopted in language exams to evaluate students' language proficiency. In this paper, we propose the first large-scale human-designed cloze test dataset CLOTH in which the questions were used in middle-school and high-school language exams. With the missing blanks carefully created by teachers and c... | rejected-papers | Meta score: 4
The paper presents a manually-constructed cloze-style fill-in-the-missing-word dataset, with baseline language modelling experiments that aim to show that this dataset is difficult for machines relative to human performance. The dataset is interesting but the fact that the experiments are confined to b... | train | [
"BynNGX9eG",
"SJU2A_Yyf",
"BJAUOGclz",
"Sye2r5h7z",
"HJMLr93Xz",
"SJ3RG937f",
"H1zBE92QM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper collects a cloze-style fill-in-the-missing-word dataset constructed manually by English teachers to test English proficiency. Experiments are given which are claimed to show that this dataset is difficult for machines relative to human performance. The dataset seems interesting but I find the empiric... | [
4,
7,
4,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rJJzTyWCZ",
"iclr_2018_rJJzTyWCZ",
"iclr_2018_rJJzTyWCZ",
"SJU2A_Yyf",
"BJAUOGclz",
"iclr_2018_rJJzTyWCZ",
"BynNGX9eG"
] |
iclr_2018_Bk346Ok0W | Sensor Transformation Attention Networks | Recent work on encoder-decoder models for sequence-to-sequence mapping has shown that integrating both temporal and spatial attentional mechanisms into neural networks increases the performance of the system substantially. We report on a new modular network architecture that applies an attentional mechanism not on temp... | rejected-papers | Meta-score: 4
This paper presents an approach which uses attention across multiple speech or video channels. After some synthetic experiments, presents experiments on chime-3, but has a rather weak baseline system
Pros:
- addresses an interesting task
Cons:
- does not take account of other recent papers in the ar... | train | [
"BJWWXpKeM",
"SyBQzAteM",
"BJKqHlqgM",
"B1o7pzaQM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author"
] | [
"This paper proposes sensor transformation attention network (STAN), which dynamically select appropriate sequential sensor inputs based on an attention mechanism. \n\nPros:\nOne of the main focuses of this paper is to apply this method to a real task, multichannel speech recognition based on CHiME-3, by providing ... | [
7,
4,
3,
-1
] | [
4,
4,
4,
-1
] | [
"iclr_2018_Bk346Ok0W",
"iclr_2018_Bk346Ok0W",
"iclr_2018_Bk346Ok0W",
"iclr_2018_Bk346Ok0W"
] |
iclr_2018_SkNQeiRpb | Training Deep AutoEncoders for Recommender Systems | This paper proposes a new model for the rating prediction task in recommender systems which significantly outperforms previous state-of-the art models on a time-split Netflix data set. Our model is based on deep autoencoder with 6 layers and is trained end-to-end without any layer-wise pre-training. We empirically demo... | rejected-papers | meta score: 4
The paper uses a deep autoencoder to rating prediction, with experiments on netflix.
Pros
- Proposed dense refeeding approach appears novel
- Good experimental results
Cons
- limited experimentation
- main novelty (dense refeeding) is not well linked to existing data imputation approaches
- novel ... | train | [
"S17oyqIgM",
"r1eVIIqgM",
"rk5q9Vixz",
"BJoaxnnzz",
"Sk3ikh3fG",
"BkfXAonGG",
"HkcDh0A1M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public"
] | [
"This paper presents a deep autoencoder model for rating prediction. The autoencoder takes the user’s rating over all the items as input and tries to predict the observed ratings in the output with mean squared error. A few techniques are applied to make the training feasible without layer-wise pre-training: 1) SEL... | [
4,
3,
6,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SkNQeiRpb",
"iclr_2018_SkNQeiRpb",
"iclr_2018_SkNQeiRpb",
"S17oyqIgM",
"r1eVIIqgM",
"rk5q9Vixz",
"iclr_2018_SkNQeiRpb"
] |
iclr_2018_H1DGha1CZ | Enhancing Batch Normalized Convolutional Networks using Displaced Rectifier Linear Units: A Systematic Comparative Study | In this paper, we turn our attention to the interworking between the activation functions and the batch normalization, which is a virtually mandatory technique to train deep networks currently. We propose the activation function Displaced Rectifier Linear Unit (DReLU) by conjecturing that extending the identity functio... | rejected-papers | meta score: 4
This paper proposes an activation function, called displaced ReLU (DReLU), to improve the performance of CNNs that use batch normalization.
Pros
- good set of experiments using CIFAR, with good results
- attempt to explain the approach using expectations
Cons
- theoretical explanations are not so con... | train | [
"ByyhLzKgM",
"S1WbYz5gM",
"SJwceNsxG",
"ryV4TxjzM",
"HJJL5Lz-z",
"BkOQHGmWf",
"B1juMp1WM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The key argument authors present against ReLU+BN is the fact that using ReLU after BN skews the values resulting in non-normalized activations. Although the BN paper suggests using BN before non-linearity many articles have been using BN after non-linearity which then gives normalized activations (https://github.c... | [
3,
4,
5,
-1,
-1,
-1,
-1
] | [
5,
4,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1DGha1CZ",
"iclr_2018_H1DGha1CZ",
"iclr_2018_H1DGha1CZ",
"BkOQHGmWf",
"ByyhLzKgM",
"S1WbYz5gM",
"SJwceNsxG"
] |
iclr_2018_ryY4RhkCZ | DEEP DENSITY NETWORKS AND UNCERTAINTY IN RECOMMENDER SYSTEMS | Building robust online content recommendation systems requires learning com- plex interactions between user preferences and content features. The field has evolved rapidly in recent years from traditional multi-arm bandit and collabora- tive filtering techniques, with new methods integrating Deep Learning models that e... | rejected-papers | Meta score: 4
The paper concerns the development of a density network for estimating uncertainty in recommender systems. The submitted paper is not very clear and it is hard to completely understand the proposed method from the way it is presented. This makes assessing the contribution of the paper difficult.
Pros... | train | [
"SyCyVT-ez",
"rJtu_bqlf",
"HyJutQ6gz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper adresses a very interesting question about the handling of the dynamics of a recommender systems at scale (here for linking to some articles).\nThe defended idea is to use the context to fit a mixture of Gaussian with a NN and to assume that the noise could be additively split into two terms. One depend ... | [
4,
3,
4
] | [
5,
4,
3
] | [
"iclr_2018_ryY4RhkCZ",
"iclr_2018_ryY4RhkCZ",
"iclr_2018_ryY4RhkCZ"
] |
iclr_2018_rkfbLilAb | Improving Search Through A3C Reinforcement Learning Based Conversational Agent | We develop a reinforcement learning based search assistant which can assist users through a set of actions and sequence of interactions to enable them realize their intent. Our approach caters to subjective search where the user is seeking digital assets such as images which is fundamentally different from the tasks wh... | rejected-papers | meta score: 4
This paper is primarily an application paper applying known RL techniques to dialogue. Very little reference to the extensive literature in this area.
Pros:
- interesting application (digital search)
- revised version contains subjective evaluation of experiments
Cons:
- limited technical novelty... | train | [
"Hy4tIW5xf",
"H1f_jh_ef",
"BkL816Ygf",
"SkPmHMpXz",
"ryJv-MaXM",
"Hy3jUZaXf",
"SkDTDVamf",
"Byxcyzp7M",
"H1QVkGpmM",
"HkAuVWpmz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"The paper \"IMPROVING SEARCH THROUGH A3C REINFORCEMENT LEARNING BASED CONVERSATIONAL AGENT\" proposes to define an agent to guide users in information retrieval tasks. By proposing refinements of the query, categorizations of the results or some other bookmarking actions, the agent is supposed to help the user in ... | [
5,
2,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rkfbLilAb",
"iclr_2018_rkfbLilAb",
"iclr_2018_rkfbLilAb",
"BkL816Ygf",
"BkL816Ygf",
"Hy4tIW5xf",
"H1f_jh_ef",
"Hy4tIW5xf",
"Hy4tIW5xf",
"iclr_2018_rkfbLilAb"
] |
iclr_2018_rJVruWZRW | Dense Recurrent Neural Network with Attention Gate | We propose the dense RNN, which has the fully connections from each hidden state to multiple preceding hidden states of all layers directly. As the density of the connection increases, the number of paths through which the gradient flows can be increased. It increases the magnitude of gradients, which help to prevent t... | rejected-papers | meta score: 4
This paper concerns a variant to previous RNN architectures using temporal skip connections, with experimentation on the PTB language modelling task
The reviewers all recommend that the paper is not ready for publication and thus should be rejected from ICLR. The novelty of the paper and its relation to ... | train | [
"SkLFMZ9gG",
"Hk37PGqlz",
"SJLNIX9lG",
"BJ_ph527G",
"BylCs9n7z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author"
] | [
" This paper proposes a new type of RNN architectures called Dense RNNs. The authors combine several different RNN architectures and claim that their RNN can model long-term dependencies better, can learn multiscale representation of the sequential data, and can sidestep the exploding or vanishing gradients problem... | [
2,
4,
4,
-1,
-1
] | [
4,
4,
4,
-1,
-1
] | [
"iclr_2018_rJVruWZRW",
"iclr_2018_rJVruWZRW",
"iclr_2018_rJVruWZRW",
"Hk37PGqlz",
"SJLNIX9lG"
] |
iclr_2018_SkYXvCR6W | Compact Encoding of Words for Efficient Character-level Convolutional Neural Networks Text Classification | This paper puts forward a new text to tensor representation that relies on information compression techniques to assign shorter codes to the most frequently used characters. This representation is language-independent with no need of pretraining and produces an encoding with no information loss. It provides an adequate... | rejected-papers | meta score: 4
The paper has been extensively edited during the review process - the edits are so extensive that I think the paper requires a re-review, which is not possible for ICLR 2018
Pros:
- potentially interesting and novel approach to prefix encoding for character level CNN text classification
- some experim... | train | [
"HJPT3oIVM",
"rkst5E4Jz",
"BJ1fw5Flf",
"rkB5-TcgG",
"SyLE3TzXz",
"Hk2cEo9QG",
"HkRp2TG7z",
"BkP93TM7M",
"SyLL2pfQM",
"rygb3pfQf",
"BJWqQmMGG",
"BJ9KbpZMz",
"ryxk9FaZM",
"HJFJeMf-G",
"rkHO_abWG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"public",
"author",
"public"
] | [
"After looking at the revision, the manuscript looks in a much better shape at this point.\nHowever, due to the amount of changes, \nI believe it has to go trough a full review process again as I mentioned in the original review.\n\nTherefore I stand by my original opinion the paper cannot be accepted now. \n\nIf ... | [
-1,
4,
3,
2,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
5,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"SyLE3TzXz",
"iclr_2018_SkYXvCR6W",
"iclr_2018_SkYXvCR6W",
"iclr_2018_SkYXvCR6W",
"BJ1fw5Flf",
"ryxk9FaZM",
"BJ9KbpZMz",
"BJWqQmMGG",
"rkst5E4Jz",
"rkB5-TcgG",
"iclr_2018_SkYXvCR6W",
"iclr_2018_SkYXvCR6W",
"iclr_2018_SkYXvCR6W",
"rkHO_abWG",
"iclr_2018_SkYXvCR6W"
] |
iclr_2018_Sy5OAyZC- | On the Use of Word Embeddings Alone to Represent Natural Language Sequences | To construct representations for natural language sequences, information from two main sources needs to be captured: (i) semantic meaning of individual words, and (ii) their compositionality. These two types of information are usually represented in the form of word embeddings and compositional functions, respectively.... | rejected-papers | This work presents a strong baseline model for several NLP-ish tasks such as document classification, sentence classification, representation learning based NLI, and text matching. In terms of originality, reviewers found that "there is not much contribution in terms of technical novelty" but that "one might also concl... | train | [
"BJxQqeTEf",
"B1wBlYKNz",
"rJ54aZ9gG",
"r1HVB4DxG",
"rkEpbo5xz",
"H1ls7qmVf",
"ryDqLvfNz",
"SJC634fVG",
"SJjhTLgEG",
"BJkidEgNz",
"S1V6EaJ4M",
"r1QGIf37M",
"S1Hg83yVz",
"rJpcO1gQG",
"r1UNBe6Mz",
"SyDybe6zz",
"BkZ7kepfG"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"More experiments have been conducted for the sequence tagging tasks: we shuffled all the words within each input sentence (along with the corresponding labels) for the training set and trained a BI-LSTM-CRF model on both datasets. For NER, the F1 score drops from 90.10 to 85.79; while for chunking, the F1 score dr... | [
-1,
-1,
7,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"r1HVB4DxG",
"r1HVB4DxG",
"iclr_2018_Sy5OAyZC-",
"iclr_2018_Sy5OAyZC-",
"iclr_2018_Sy5OAyZC-",
"ryDqLvfNz",
"SJC634fVG",
"SJjhTLgEG",
"BJkidEgNz",
"S1V6EaJ4M",
"S1Hg83yVz",
"rJpcO1gQG",
"r1QGIf37M",
"iclr_2018_Sy5OAyZC-",
"r1HVB4DxG",
"rJ54aZ9gG",
"rkEpbo5xz"
] |
iclr_2018_Byht0GbRZ | STRUCTURED ALIGNMENT NETWORKS | Many tasks in natural language processing involve comparing two sentences to compute some notion of relevance, entailment, or similarity. Typically this comparison is done either at the word level or at the sentence level, with no attempt to leverage the inherent structure of the sentence. When sentence structure is u... | rejected-papers | This work introduces a new type of structured attention network that learn latent structured alignments between sentences in a fully differentiable manner, which allows the network to learn not only the target task, but also the latent relationships. Reviewers seem partial to the idea of the work, and it's originality,... | train | [
"HJem9rYlf",
"Sybe_7qlG",
"S1zkYGm-G",
"SJ2-1ibmz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"Summary:\nThis paper introduces a structured attention mechanisms to compute alignment scores among all possible spans in two given sentences. The span representations are weighted by the spans marginal scores given by the inside-outside algorithm. Experiments on TREC-QA and SNLI show modest improvement over the w... | [
6,
5,
5,
-1
] | [
4,
4,
4,
-1
] | [
"iclr_2018_Byht0GbRZ",
"iclr_2018_Byht0GbRZ",
"iclr_2018_Byht0GbRZ",
"iclr_2018_Byht0GbRZ"
] |
iclr_2018_SJZsR7kCZ | Iterative Deep Compression : Compressing Deep Networks for Classification and Semantic Segmentation | Machine learning and in particular deep learning approaches have outperformed many traditional techniques in accomplishing complex tasks such as
image classfication, natural language processing or speech recognition. Most of the state-of-the art deep networks have complex architecture and use a vast number of par... | rejected-papers | This paper presents a new pipeline for nn compression that extends that of Han et. al, but show that it reduces parameters further, maintains higher accuracy and can be applied to methods behind classification (semantic segmentation). While the authors found the paper clearly written, excepting for some typos, and pote... | train | [
"Syn5-0_lM",
"BkYtPqKez",
"rJrltU5gf",
"H1a6sypQf",
"SkdxEmgmM",
"rkhukfWfG",
"H1X-IeeGz",
"HyqT8n1fM",
"ryU_V00bf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"quality: this paper is of good quality\nclarity: this paper is very clear\noriginality: this paper combines original ideas with existing approaches for pruning to obtain dramatic space reduction in NN parameters.\nsignificance: this paper seems significant.\n\nPROS\n- a new approach to sparsifying that considers d... | [
6,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJZsR7kCZ",
"iclr_2018_SJZsR7kCZ",
"iclr_2018_SJZsR7kCZ",
"iclr_2018_SJZsR7kCZ",
"Syn5-0_lM",
"Syn5-0_lM",
"BkYtPqKez",
"rJrltU5gf",
"rJrltU5gf"
] |
iclr_2018_BkQCGzZ0- | Discrete Autoencoders for Sequence Models | Recurrent models for sequences have been recently successful at many tasks, especially for language modeling
and machine translation. Nevertheless, it remains challenging to extract good representations from
these models. For instance, even though language has a clear hierarchical structure going from chara... | rejected-papers | This paper presents a different method for learning autoencoders with discrete hidden states (compared to recent discrete-like VAE type models). The reviewers in general like the method being proposed and are convinced that there is worth to the underlying proposal. However there are several shared complaints about the... | train | [
"Hk_BLd_ef",
"HJGJNQcez",
"ryeycmTlz",
"HJrln8hmG",
"H1T2PNnzG",
"Bk_yWAofM",
"Sk0B0pifG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author"
] | [
"The topic is interesting however the description in the paper is lacking clarity. The paper is written in a procedural fashion - I first did that, then I did that and after that I did third. Having proper mathematical description and good diagrams of what you doing would have immensely helped. Another big issue is... | [
5,
4,
6,
-1,
-1,
-1,
-1
] | [
5,
4,
1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkQCGzZ0-",
"iclr_2018_BkQCGzZ0-",
"iclr_2018_BkQCGzZ0-",
"H1T2PNnzG",
"Bk_yWAofM",
"Hk_BLd_ef",
"HJGJNQcez"
] |
iclr_2018_Bkl1uWb0Z | Inducing Grammars with and for Neural Machine Translation | Previous work has demonstrated the benefits of incorporating additional linguistic annotations such as syntactic trees into neural machine translation. However the cost of obtaining those syntactic annotations is expensive for many languages and the quality of unsupervised learning linguistic structures is too poor to ... | rejected-papers | In this work reviewers use structured attention as a way to induce grammatical structure in NMT models. Reviewers liked th motivation of the work and found experiments mostly well done. However reviewers found the paper a bit difficult to follow, with several commenting that distinctions made between the different sub ... | val | [
"SyOq7nXxf",
"Hyg5qAtgf",
"rylV679xz",
"B1KCT7WNz",
"rJtGoXWzz",
"r1Jhq7-Gz",
"SkrGscezf",
"Sk8occlMG",
"Bkl159lfG",
"HJwRq_ilz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper adds source side dependency syntax trees to an NMT model without explicit supervision. Exploring the use of syntax in neural translation is interesting but I am not convinced that this approach actually works based on the experimental results.\n\nThe paper distinguishes between syntactic and semantic ob... | [
3,
6,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Bkl1uWb0Z",
"iclr_2018_Bkl1uWb0Z",
"iclr_2018_Bkl1uWb0Z",
"iclr_2018_Bkl1uWb0Z",
"iclr_2018_Bkl1uWb0Z",
"HJwRq_ilz",
"SyOq7nXxf",
"Hyg5qAtgf",
"rylV679xz",
"iclr_2018_Bkl1uWb0Z"
] |
iclr_2018_Syx6bz-Ab | Seq2SQL: Generating Structured Queries From Natural Language Using Reinforcement Learning | Relational databases store a significant amount of the worlds data. However, accessing this data currently requires users to understand a query language such as SQL. We propose Seq2SQL, a deep neural network for translating natural language questions to corresponding SQL queries. Our model uses rewards from in the loop... | rejected-papers | This paper introduces a new dataset and method for a "semantic parsing" problem of generating logical sql queries from text. Reviews generally seemed to be very impressed by the dataset portion of the work saying "the creation of a large scale semantic parsing dataset is fantastic," but were less compelled by the mode... | train | [
"SkGbiIKxz",
"ByL2SX9ez",
"HJ75M8ogM",
"r1E6tYEmz",
"ByyiYYEXz",
"ByeStYNQM",
"S1cptoZXG",
"H1PC1EX1M",
"Hk2vAMlJM",
"SJmanaxJz",
"HJg8aSa0-",
"S184E1gyz",
"r1qqBPk1z",
"By6SuETCZ"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"official_reviewer",
"public",
"author",
"public",
"author",
"public",
"author",
"public"
] | [
"This paper presents a new approach to support the conversion from natural language to database queries. \n\nOne of the major contributions of the work is the introduction of a new real-world benchmark dataset based on questions over Wikipedia. The scale of the data set is significantly larger than any existing one... | [
5,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Syx6bz-Ab",
"iclr_2018_Syx6bz-Ab",
"iclr_2018_Syx6bz-Ab",
"SkGbiIKxz",
"ByL2SX9ez",
"HJ75M8ogM",
"SJmanaxJz",
"iclr_2018_Syx6bz-Ab",
"S184E1gyz",
"HJg8aSa0-",
"By6SuETCZ",
"HJg8aSa0-",
"iclr_2018_Syx6bz-Ab",
"iclr_2018_Syx6bz-Ab"
] |
iclr_2018_BJInMmWC- | Generative Entity Networks: Disentangling Entitites and Attributes in Visual Scenes using Partial Natural Language Descriptions | Generative image models have made significant progress in the last few years, and are now able to generate low-resolution images which sometimes look realistic. However the state-of-the-art models utilize fully entangled latent representations where small changes to a single neuron can effect every output pixel in rela... | rejected-papers | This paper presents a novel model for generating images and natural language descriptions simultaneously. The aim is to distangle representations learned for image generation by connecting them to the paired text. The reviews praise the problem setup and the mathematical formulation. However they point out significant ... | test | [
"ryU5DROgM",
"r1MR39Kgz",
"BJbx0qYlz",
"HyK9VuT7M",
"H1WUNO6Xz",
"H1zZE_6mf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper presented a Generative entity networks (GEN). It is a multi-view extension of variational autoencoder (VAE) for disentangled representation. It uses the image and its attributes. The paper is very well motivated and tackles an important problem. However, the presentation of the method is not clear, the... | [
4,
5,
5,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1
] | [
"iclr_2018_BJInMmWC-",
"iclr_2018_BJInMmWC-",
"iclr_2018_BJInMmWC-",
"H1zZE_6mf",
"H1zZE_6mf",
"iclr_2018_BJInMmWC-"
] |
iclr_2018_r1Zi2Mb0- | EXPLORING NEURAL ARCHITECTURE SEARCH FOR LANGUAGE TASKS | Neural architecture search (NAS), the task of finding neural architectures automatically, has recently emerged as a promising approach for unveiling better models over human-designed ones. However, most success stories are for vision tasks and have been quite limited for text, except for a small language modeling setup... | rejected-papers | This paper extends work on neural architecture search by introducing a new framework for searching and experiments on new domains of NMT and QA. The results of the work are beneficial and show improvements using this approach. However the reviewers point out significant issues with the approach itself:
- There is ske... | train | [
"Bkyu46Xlz",
"r1E8UmKgz",
"SJVkJ0Axf",
"Hkv5oLTQf",
"BywoFAUkM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public"
] | [
"The paper explores neural architecture search for translation and reading comprehension tasks. It is fairly clearly written and required a lot of large-scale experimentation. However, the paper introduces few new ideas and seems very much like applying an existing framework to new problems. It is probably better s... | [
3,
4,
3,
-1,
-1
] | [
4,
4,
4,
-1,
-1
] | [
"iclr_2018_r1Zi2Mb0-",
"iclr_2018_r1Zi2Mb0-",
"iclr_2018_r1Zi2Mb0-",
"iclr_2018_r1Zi2Mb0-",
"iclr_2018_r1Zi2Mb0-"
] |
iclr_2018_S1347ot3b | Exploring Sentence Vectors Through Automatic Summarization | Vector semantics, especially sentence vectors, have recently been used successfully in many areas of natural language processing. However, relatively little work has explored the internal structure and properties of spaces of sentence vectors. In this paper, we will explore the properties of sentence vectors by studyin... | rejected-papers | This work is interested in using sentence vector representations as a method for both doing extractive summarization and as a way to better understand the structure of vector representations. While the methodological aspects utilize representation learning, the reviewers felt that the main thrust of the work would be b... | train | [
"BkSq8vBxG",
"Sk3CQYOgG",
"rk-MJ-ceM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The authors report a number of experiments using off-the-shelf sentence embedding methods for performing extractive summarisation, using a number of simple methods for choosing the extracted sentences. Unfortunately the contribution is too minor, and the work too incremental, to be worthy of a place at a top-tier ... | [
2,
2,
3
] | [
5,
5,
5
] | [
"iclr_2018_S1347ot3b",
"iclr_2018_S1347ot3b",
"iclr_2018_S1347ot3b"
] |
iclr_2018_BJDEbngCZ | Global Convergence of Policy Gradient Methods for Linearized Control Problems | Direct policy gradient methods for reinforcement learning and continuous control problems are a popular
approach for a variety of reasons:
1) they are easy to implement without explicit knowledge of the underlying model;
2) they are an "end-to-end" approach, directly optimizing the performance metric... | rejected-papers | The paper studies the global convergence for policy gradient methods for linear control problems. Multiple reviewers point out strong concerns about the novelty of the results. | train | [
"ry1QoRKlf",
"Hy0XRb5lG",
"HJaedAnWz",
"S1SMngjMf",
"r1dg3lsMz",
"SyFAixjzf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The work investigates convergence guarantees of gradient-type policies for reinforcement learning and continuous control\nproblems, both in deterministic and randomized case, whiling coping with non-convexity of the objective. I found that the paper suffers many shortcomings that must be addressed:\n\n1) The writi... | [
6,
5,
5,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_BJDEbngCZ",
"iclr_2018_BJDEbngCZ",
"iclr_2018_BJDEbngCZ",
"ry1QoRKlf",
"Hy0XRb5lG",
"HJaedAnWz"
] |
iclr_2018_SJICXeWAb | Depth separation and weight-width trade-offs for sigmoidal neural networks | Some recent work has shown separation between the expressive power of depth-2 and depth-3 neural networks. These separation results are shown by constructing functions and input distributions, so that the function is well-approximable by a depth-3 neural network of polynomial size but it cannot be well-approximated und... | rejected-papers | The reviewers point out that most of the results are already known and are not novel. There are also issues with the presentation. Studying only depth 2 and depth 3 networks is very limiting. | train | [
"rkb3kqxxf",
"ry5vY1teG",
"rJ7xdwpeM",
"SJ63ZL2Xz",
"Bk84l8nQM",
"BkcJxU2XG",
"H1F_0H3XG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper contributes to the growing literature on depth separations in neural network, showing cases where depth is provably needed to express certain functions. Specifically, the paper shows that there are functions on R^d that can be approximated well by a depth-3 sigmoidal network with poly(d) weights, that c... | [
6,
5,
3,
-1,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJICXeWAb",
"iclr_2018_SJICXeWAb",
"iclr_2018_SJICXeWAb",
"iclr_2018_SJICXeWAb",
"rkb3kqxxf",
"ry5vY1teG",
"rJ7xdwpeM"
] |
iclr_2018_rJR2ylbRb | Spectral Graph Wavelets for Structural Role Similarity in Networks | Nodes residing in different parts of a graph can have similar structural roles within their local network topology. The identification of such roles provides key insight into the organization of networks and can also be used to inform machine learning on graphs. However, learning structural representations of nodes is ... | rejected-papers | The reviewers present strong concerns about the lack of novelty in the paper. Further there are strong concerns about how the experiments are conducted. I recommend the authors to carefully go through the reviews. | train | [
"BJnpN5cgf",
"rk9wKvM-z",
"SJqU9kAWf",
"SkYmLUkGz",
"HygUoe6Wz",
"SJq3YgabG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The term \"structural equivalence\" is used incorrectly in the paper. From sociology, two nodes with the same position are in an equivalence relation. An equivalence, Q, is any relation that satisfies these three conditions:\n - Transitivity: (a,b), (b,c) ∈ Q ⇒ (a,c) ∈Q\n - Symmetry: (a, b) ∈ Q if and only if ... | [
5,
5,
3,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_rJR2ylbRb",
"iclr_2018_rJR2ylbRb",
"iclr_2018_rJR2ylbRb",
"SJqU9kAWf",
"BJnpN5cgf",
"rk9wKvM-z"
] |
iclr_2018_r1CE9GWR- | Understanding GANs: the LQG Setting | Generative Adversarial Networks (GANs) have become a popular method to learn a probability model from data. Many GAN architectures with different optimization metrics have been introduced recently. Instead of proposing yet another architecture, this paper aims to provide an understanding of some of the basic issues su... | rejected-papers | While the reviewers agree that this is an important topic, there are numerous concerns novelty, correctness and limitations. | test | [
"H1PuapUef",
"HyW8p6vxM",
"r1WP0c2bz",
"B1cmO6LQM",
"B1UKOpIQG",
"BJG4hf8lz",
"SJFExL7eM",
"BJ7s1MTyz",
"Sks3SeNJf",
"BJxqtFwAW",
"B1zBD53A-",
"H1oRM3cCZ",
"HJkAme4AZ"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public",
"author",
"public",
"author",
"author",
"public",
"public"
] | [
"*Paper summary*\n\nThe paper considers GANs from a theoretical point of view. The authors approach GANs from the 2-Wasserstein point of view and provide several insights for a very specific setting. In my point of view, the main novel contribution of the paper is to notice the following fact:\n\n(*) It is well kno... | [
4,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1CE9GWR-",
"iclr_2018_r1CE9GWR-",
"iclr_2018_r1CE9GWR-",
"iclr_2018_r1CE9GWR-",
"iclr_2018_r1CE9GWR-",
"SJFExL7eM",
"BJ7s1MTyz",
"Sks3SeNJf",
"iclr_2018_r1CE9GWR-",
"HJkAme4AZ",
"H1oRM3cCZ",
"BJxqtFwAW",
"iclr_2018_r1CE9GWR-"
] |
iclr_2018_HyY0Ff-AZ | Representing Entropy : A short proof of the equivalence between soft Q-learning and policy gradients | Two main families of reinforcement learning algorithms, Q-learning and policy gradients, have recently been proven to be equivalent when using a softmax relaxation on one part, and an entropic regularization on the other. We relate this result to the well-known convex duality of Shannon entropy and the softmax function... | rejected-papers | The reviewers point out that this is a well known result and is not novel. | train | [
"r1FjCQKxG",
"rJ9BTHFez",
"B1_rmNyZz",
"S1-FOw67M",
"Sk1y8waXf",
"SyDO4PpXf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper uses a well-known variational representation of the relative entropy (the so-called Donsker-Varadhan formula) to derive an expression for the Bellman error with entropy regularization in terms of a certain log-partition function. This is stated in Equation (13) in the paper. However, this precise repres... | [
5,
2,
5,
-1,
-1,
-1
] | [
5,
5,
4,
-1,
-1,
-1
] | [
"iclr_2018_HyY0Ff-AZ",
"iclr_2018_HyY0Ff-AZ",
"iclr_2018_HyY0Ff-AZ",
"rJ9BTHFez",
"r1FjCQKxG",
"B1_rmNyZz"
] |
iclr_2018_BJcAWaeCW | Graph Topological Features via GAN | Inspired by the success of generative adversarial networks (GANs) in image domains, we introduce a novel hierarchical architecture for learning characteristic topological features from a single arbitrary input graph via GANs. The hierarchical architecture consisting of multiple GANs preserves both local and global topo... | rejected-papers | The reviewers present strong concerns regarding presentation of the paper. The approach appears overly complex, some design choices are not clear and the experiments are not conducted properly. I recommend the authors to carefully go through the reviews. | train | [
"rycdDEtxz",
"HJukFfcxf",
"BypRYU5xM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The authors try to combine the power of GANs with hierarchical community structure detections. While the idea is sound, many design choices of the system is questionable. The problem is particularly aggravated by the poor presentation of the paper, creating countless confusions for readers. I do not recommend the ... | [
3,
4,
4
] | [
4,
4,
5
] | [
"iclr_2018_BJcAWaeCW",
"iclr_2018_BJcAWaeCW",
"iclr_2018_BJcAWaeCW"
] |
iclr_2018_HyEi7bWR- | Orthogonal Recurrent Neural Networks with Scaled Cayley Transform | Recurrent Neural Networks (RNNs) are designed to handle sequential data but suffer from vanishing or exploding gradients. Recent work on Unitary Recurrent Neural Networks (uRNNs) have been used to address this issue and in some cases, exceed the capabilities of Long Short-Term Memory networks (LSTMs). We propose a si... | rejected-papers | The authors use the Cayley transform representation of an orthogonal matrix to provide a parameterization of an RNN with orthogonal weights. The paper is clearly written and the formulation is simple and elegant. However, I share the concerns of reviewer 3 about the significance of another method for parameterizing... | train | [
"SkLk8W9lM",
"rkQbrzqxM",
"HyGpBPslM",
"HkNx0LpXM",
"SyFq28pQM",
"HkhweWhGG",
"HywrlZhzz",
"HJGWlb2zf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"This manuscript introduce a scheme for learning the recurrent parameter matrix in a neural network that uses the Cayley transform and a scaling weight matrix. This scheme leads to good performance on sequential data tasks and requires fewer parameters than other techniques\n\nComments:\n-- It’s not clear to me how... | [
7,
6,
5,
-1,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HyEi7bWR-",
"iclr_2018_HyEi7bWR-",
"iclr_2018_HyEi7bWR-",
"HkhweWhGG",
"HywrlZhzz",
"SkLk8W9lM",
"rkQbrzqxM",
"HyGpBPslM"
] |
iclr_2018_B1EPYJ-C- | Federated Learning: Strategies for Improving Communication Efficiency | Federated Learning is a machine learning setting where the goal is to train a high-quality centralized model while training data remains distributed over a large number of clients each with unreliable and relatively slow network connections. We consider learning algorithms for this setting where on each round, each cl... | rejected-papers | The authors study the problem of reducing uplink communication costs in training a ML model where the training data is distributed over many clients. The reviewers consider the problem interesting, but have concerns about the extent of the novelty of the approach. As the reviewers and authors agree that the paper is... | train | [
"Hk_LRZ5gG",
"BJhrvHcgf",
"BkWhzt0ez",
"ByrcIc-GM",
"rkZHL9ZMz",
"HkJcS5WGz",
"H15iEq-MM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper proposes several client-server neural network gradient update strategies aimed at reducing uplink usage while maintaining prediction performance. The main approaches fall into two categories: structured, where low-rank/sparse updates are learned, and sketched, where full updates are either sub-sampled ... | [
5,
7,
5,
-1,
-1,
-1,
-1
] | [
3,
5,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_B1EPYJ-C-",
"iclr_2018_B1EPYJ-C-",
"iclr_2018_B1EPYJ-C-",
"Hk_LRZ5gG",
"BJhrvHcgf",
"BkWhzt0ez",
"iclr_2018_B1EPYJ-C-"
] |
iclr_2018_r1SuFjkRW | Discrete Sequential Prediction of Continuous Actions for Deep RL | It has long been assumed that high dimensional continuous control problems cannot be solved effectively by discretizing individual dimensions of the action space due to the exponentially large number of bins over which policies would have to be learned. In this paper, we draw inspiration from the recent success of sequ... | rejected-papers | The reviewers consider the paper to promising, but raise issues with the increase in the complexity of the MDP caused by the authors' parameterization of the action space, and comparisons with earlier work (Pazis and Lagoudakis). While the authors cite this work, and say that they that they needed to make changes to ... | train | [
"H13MV5tkM",
"rkaH5MsgM",
"S1m6NXjlM",
"SJ0-Es9XG",
"rkI8QoqXz",
"S1fQmo5mz",
"S15a-iqQG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper presents Sequential Deep Q-Networks (SDQNs), which select actions from discretized high-dimensional action spaces. This is done by introducing another, undiscounted MDP in which each action dimension is chosen sequentially by an agent. By training a Q network to best choose these action dimensions, and... | [
7,
5,
4,
-1,
-1,
-1,
-1
] | [
5,
1,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1SuFjkRW",
"iclr_2018_r1SuFjkRW",
"iclr_2018_r1SuFjkRW",
"iclr_2018_r1SuFjkRW",
"H13MV5tkM",
"rkaH5MsgM",
"S1m6NXjlM"
] |
iclr_2018_HyXBcYg0b | Residual Gated Graph ConvNets | Graph-structured data such as social networks, functional brain networks, gene regulatory networks, communications networks have brought the interest in generalizing deep learning techniques to graph domains. In this paper, we are interested to design neural networks for graphs with variable length in order to solve le... | rejected-papers | The authors make an experimental study of the relative merits of RNN-type approaches and graph-neural-network approaches to solving node-labeling problems on graphs. They discuss various improvements in gnn constructions, such as residual connections.
This is a borderline paper. On one hand, the reviewers feel that... | train | [
"BJWd98x4z",
"ryFFX8Yxf",
"Sy7QPPYxM",
"r1-tSfqeG",
"HycHDznMM",
"rJ_yvM3fz",
"SyffDMnff",
"HyshLz3fz"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"We would like to thank the referee for her/his time reviewing the revised paper and for improving her/his evaluation score. ",
"The authors revised the paper according to all reviewers suggestions, I am satisfied with the current version.\n\nSummary: this works proposes to employ recurrent gated convnets to solv... | [
-1,
7,
3,
6,
-1,
-1,
-1,
-1
] | [
-1,
4,
4,
3,
-1,
-1,
-1,
-1
] | [
"HycHDznMM",
"iclr_2018_HyXBcYg0b",
"iclr_2018_HyXBcYg0b",
"iclr_2018_HyXBcYg0b",
"ryFFX8Yxf",
"Sy7QPPYxM",
"rJ_yvM3fz",
"r1-tSfqeG"
] |
iclr_2018_HyI5ro0pW | Neural Networks with Block Diagonal Inner Product Layers | Artificial neural networks have opened up a world of possibilities in data science and artificial intelligence, but neural networks are cumbersome tools that grow with the complexity of the learning problem. We make contributions to this issue by considering a modified version of the fully connected layer we call a blo... | rejected-papers | The authors propose a technique for weight pruning that leaves block diagonal weights, instead of unstructured sparse weights, leading to faster inference. However, the experiments demonstrating the quality of the pruned models are insufficient. The authors also discuss connections to random matrix theory; but these... | train | [
"rygiBN8Vz",
"Bk9pmiuxM",
"S1IUpXKgG",
"H1hqfaKeM",
"BkRhT_OMG",
"HJabtddzG",
"r1FWLOuzG",
"HJWme931M",
"r15TXzn1z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public"
] | [
"Thanks to the authors for responding (we have read it and taken it into account).",
"This is a mostly experimental paper which evaluates the capabilities of neural networks with weight matrices that are block diagonal. The authors describe two methods to obtain this structure: (1) enforced during training, (2) e... | [
-1,
5,
6,
4,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
4,
3,
-1,
-1,
-1,
-1,
-1
] | [
"HJabtddzG",
"iclr_2018_HyI5ro0pW",
"iclr_2018_HyI5ro0pW",
"iclr_2018_HyI5ro0pW",
"Bk9pmiuxM",
"S1IUpXKgG",
"H1hqfaKeM",
"r15TXzn1z",
"iclr_2018_HyI5ro0pW"
] |
iclr_2018_SJ1fQYlCZ | Training with Growing Sets: A Simple Alternative to Curriculum Learning and Self Paced Learning | Curriculum learning and Self paced learning are popular topics in the machine learning that suggest to put the training samples in order by considering their difficulty levels. Studies in these topics show that starting with a small training set and adding new samples according to difficulty levels improves the learnin... | rejected-papers | The authors give evidence that is certain cases, the ordering of sample inclusion in a curriculum is not important. However, the reviewers believe the experiments are inconclusive, both in the sense that as reported, they do not demonstrate the authors' hypothesis, and that they may leave out many relevant factors of ... | test | [
"BJL_WOKgz",
"Hy4VEO9gM",
"rkh23GClz",
"r1UuG5AWz",
"Sy-ml5RbG",
"BJCdg50bG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper addresses an interesting problem of curriculum/self-paced versus random order of samples for faster learning. Specifically, the authors argue that adding samples in random order is as beneficial as adding them with some curriculum strategy, i.e. from easiest to hardest, or reverse. \nThe main learning s... | [
4,
6,
4,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1
] | [
"iclr_2018_SJ1fQYlCZ",
"iclr_2018_SJ1fQYlCZ",
"iclr_2018_SJ1fQYlCZ",
"BJL_WOKgz",
"rkh23GClz",
"Hy4VEO9gM"
] |
iclr_2018_rJaE2alRW | Autoregressive Convolutional Neural Networks for Asynchronous Time Series | We propose Significance-Offset Convolutional Neural Network, a deep convolutional network architecture for regression of multivariate asynchronous time series. The model is inspired by standard autoregressive (AR) models and gating mechanisms used in recurrent neural networks. It involves an AR-like weighting system,... | rejected-papers | The reviewers feel that the novelties in the model are not significant. Furthermore, they suggest that empirical results could be improved by
1: analyses showing how the significance network functions and directly measuring its impact
2: More reproducible experiments. In particular, this is really an applications p... | test | [
"Hkl6sRTyf",
"H1JQgkAJG",
"BkaINb9xz",
"r1UnLEpmz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author"
] | [
"To begin with, the authors seem to be missing some recent developments in the field of deep learning which are closely related to the proposed approach; e.g.:\n\nSotirios P. Chatzis, “Recurrent Latent Variable Conditional Heteroscedasticity,” Proc. 42nd IEEE International Conference on Acoustics, Speech and Signal... | [
4,
5,
5,
-1
] | [
5,
3,
4,
-1
] | [
"iclr_2018_rJaE2alRW",
"iclr_2018_rJaE2alRW",
"iclr_2018_rJaE2alRW",
"iclr_2018_rJaE2alRW"
] |
iclr_2018_B1KFAGWAZ | Revisiting The Master-Slave Architecture In Multi-Agent Deep Reinforcement Learning | Many tasks in artificial intelligence require the collaboration of multiple agents. We exam deep reinforcement learning for multi-agent domains. Recent research efforts often take the form of two seemingly conflicting perspectives, the decentralized perspective, where each agent is supposed to have its own controller; ... | rejected-papers | The authors present a centralized neural controller for multi-agent reinforcement learning. The reviewers are are not convinced that there is sufficient novelty, considering the authors setup as essentially a special case of other recent works, with added adjustments to the neural-networks that are standard in the li... | train | [
"BkbZ3Ip7f",
"ByhCSC3EM",
"S141qTNVG",
"HyuecsNxf",
"HynlSGDxM",
"B1qTwXcgz",
"H1aoqUTmz",
"S1vBOLaQG",
"BkSjwU6Xz"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"We summarize the updates made to the paper as follow:\n\n1. Add one more StarCraft micromanagement task (both task description and new evaluation results etc.) \"Dragoons vs. Zealots\" where heterogeneous agents are involved.\n\n2. A new experiment is included in section 4.4 to compare models of different centrali... | [
-1,
-1,
-1,
5,
5,
4,
-1,
-1,
-1
] | [
-1,
-1,
-1,
5,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_B1KFAGWAZ",
"S141qTNVG",
"H1aoqUTmz",
"iclr_2018_B1KFAGWAZ",
"iclr_2018_B1KFAGWAZ",
"iclr_2018_B1KFAGWAZ",
"HyuecsNxf",
"HynlSGDxM",
"B1qTwXcgz"
] |
iclr_2018_S1q_Cz-Cb | Training Neural Machines with Partial Traces | We present a novel approach for training neural abstract architectures which in- corporates (partial) supervision over the machine’s interpretable components. To cleanly capture the set of neural architectures to which our method applies, we introduce the concept of a differential neural computational machine (∂NCM) an... | rejected-papers | While the reviewers considered the basic idea of adding supervision intermediate to differentiable programming style architectures to be interesting and worthy of effort, they were unsure if
1: the proposed abstractions for discussing ntm and nram are well motivated/more generally applicable
2: the methods used in this... | train | [
"SkhI_TPez",
"rkRHKcFgG",
"SJdsxW5xG",
"r1f5rvp7z",
"HJaUrD6Qz",
"Sy9-HwpQf",
"SJ8n4P6Qf",
"Skp04D6QG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"The authors introduce the general concept of a differential neural computational machine, dNCM. It can apply to any fully differentiable neural programming machine, such as the Neural Turing Machine (NTM) or NRAM or the Neural GPU, but not to non-fully-differentiable architecture such as NPI. The author show how p... | [
4,
5,
4,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
3,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1q_Cz-Cb",
"iclr_2018_S1q_Cz-Cb",
"iclr_2018_S1q_Cz-Cb",
"SkhI_TPez",
"rkRHKcFgG",
"SJdsxW5xG",
"iclr_2018_S1q_Cz-Cb",
"iclr_2018_S1q_Cz-Cb"
] |
iclr_2018_HkMCybx0- | Improving Deep Learning by Inverse Square Root Linear Units (ISRLUs) | We introduce the “inverse square root linear unit” (ISRLU) to speed up learning in deep neural networks. ISRLU has better performance than ELU but has many of the same benefits. ISRLU and ELU have similar curves and characteristics. Both have negative values, allowing them to push mean unit activation closer to zero, a... | rejected-papers | The authors introduce a new activation function which is similar in shape to ELU, but is faster to compute. The reviewers consider this to not be a significant innovation because the amount of time spent in computing the activation function is small compared to other neural network operations. | test | [
"rk3rjfYgG",
"rkDaCCtlz",
"SJNibmcgz",
"rk5N6Gm-M",
"HJ4fWYDC-",
"rJ8gr8kyM",
"B1zM2bHAb"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public"
] | [
"This paper introduces a new nonlinear activation function for neural networks, i.e., Inverse Square Root Linear Units (ISRLU). Experiments show that ISRLU is promising compared to competitors like ReLU and ELU.\n\nPros:\n(1) The paper is clearly written.\n\n(2) The proposed ISRLU function has similar curves with ... | [
4,
5,
3,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HkMCybx0-",
"iclr_2018_HkMCybx0-",
"iclr_2018_HkMCybx0-",
"rkDaCCtlz",
"B1zM2bHAb",
"HJ4fWYDC-",
"iclr_2018_HkMCybx0-"
] |
iclr_2018_SJxE3jlA- | Now I Remember! Episodic Memory For Reinforcement Learning | Humans rely on episodic memory constantly, in remembering the name of someone they met 10 minutes ago, the plot of a movie as it unfolds, or where they parked the car. Endowing reinforcement learning agents with episodic memory is a key step on the path toward replicating human-like general intelligence. We analyze why... | rejected-papers | The authors show evidence that an RL agent with a new neural architecture with an external memory is superior on a version of the concentration game to a baseline. However, other works have proposed neural architectures with episodic memories, and the reviewers feel that the proposed model was not adequately compared... | train | [
"r1gpVv_gM",
"SyG_vjueM",
"r1o2Xfcgf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"There are a number of attempts to add episodic memory to RL agents. A common approach is to use some sort of recurrent model with a model-free agent. This work follows this approach using what could be considered a memory network with a identity embedding function and tests on 'Concentration', a game which require... | [
4,
4,
4
] | [
5,
4,
5
] | [
"iclr_2018_SJxE3jlA-",
"iclr_2018_SJxE3jlA-",
"iclr_2018_SJxE3jlA-"
] |
iclr_2018_Sk1NTfZAb | Key Protected Classification for GAN Attack Resilient Collaborative Learning | Large-scale publicly available datasets play a fundamental role in training deep learning models. However, large-scale
datasets are difficult to collect in problems that involve processing of sensitive information.
Collaborative learning techniques provide a privacy-preserving solution in such cases, by ena... | rejected-papers | While the reviewers feel there might be some merit to this work, they find enough ambiguities and inaccuracies that I think this paper would be better served by a resubmission. | train | [
"r1qWlNtlM",
"SJyQ2wqlf",
"SkS7qR3-M",
"SkMMwxszG",
"ry0o8eofG",
"BJMyLesGG",
"SyOgBeizf",
"SJ43M9ZMz",
"S1ZKptezz",
"S1xr_5abG",
"Hyku30n-G",
"SJ227bqbM",
"B197fWcbf",
"S1B5euOZz",
"B1xjZwmbz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public",
"public",
"official_reviewer",
"public",
"public",
"public",
"public"
] | [
"This paper is a follow-up work to the CCS'2017 paper on the GAN-based attack on collaborative learning system where multiple users contribute their private and sensitive data to joint learning tasks. In order to avoid the potential risk of adversary's mimic based on information flow among distributed users, the au... | [
4,
5,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
2,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Sk1NTfZAb",
"iclr_2018_Sk1NTfZAb",
"iclr_2018_Sk1NTfZAb",
"SkS7qR3-M",
"r1qWlNtlM",
"SJyQ2wqlf",
"iclr_2018_Sk1NTfZAb",
"S1ZKptezz",
"B197fWcbf",
"Hyku30n-G",
"B1xjZwmbz",
"B1xjZwmbz",
"S1B5euOZz",
"B1xjZwmbz",
"iclr_2018_Sk1NTfZAb"
] |
iclr_2018_Bk_fs6gA- | Long Term Memory Network for Combinatorial Optimization Problems | This paper introduces a framework for solving combinatorial optimization problems by learning from input-output examples of optimization problems. We introduce a new memory augmented neural model in which the memory is not resettable (i.e the information stored in the memory after processing an input example is kept fo... | rejected-papers | The authors use a memory-augmented neural architecture to learn solve combinatorial optimization problems. The reviewers consider the approach worth studying, but find the authors' experimental protocol and baselines flawed. | train | [
"rJpWE-cgM",
"ByEoHz5ez",
"rJT9cfqlz",
"Hy5pspL1f",
"rJjGPGcAZ",
"Skx-mCeAZ"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public",
"public"
] | [
"Learning to solve combinatorial optimization problems using recurrent networks is a very interesting research topic. However, I had a very hard time understanding the paper. It certainly doesn’t help that I’m not familiar with the architectures the model is based on, nor with state-of-the-art integer programming s... | [
4,
4,
3,
-1,
-1,
-1
] | [
1,
2,
4,
-1,
-1,
-1
] | [
"iclr_2018_Bk_fs6gA-",
"iclr_2018_Bk_fs6gA-",
"iclr_2018_Bk_fs6gA-",
"rJjGPGcAZ",
"iclr_2018_Bk_fs6gA-",
"iclr_2018_Bk_fs6gA-"
] |
iclr_2018_r17Q6WWA- | Multi-Task Learning by Deep Collaboration and Application in Facial Landmark Detection | Convolutional neural networks (CNN) have become the most successful and popular approach in many vision-related domains. While CNNs are particularly well-suited for capturing a proper hierarchy of concepts from real-world images, they are limited to domains where data is abundant. Recent attempts have looked into mitig... | rejected-papers | The experimental work in this paper leaves it just short of being suitable for acceptance.
The work needs more comparisons with prior work and other approaches.
The numerical ratings of the work by reviewers are just too low.
| train | [
"Bylfi7tez",
"ry0lbHclz",
"SyuPmP3lM",
"SytmUAi7M",
"Hy-UWWjff",
"HJDRBqufM",
"ryXDrc_fG",
"HyyB0FuGf",
"S1MZTYuzG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Pros:\n1. This paper proposed a new block which can aggregate features from different tasks. By doing this, it can take advantage of common information between related tasks and improve the generalization of target tasks.\n\n2. The achievement in this paper seems good, which is 24.31%.\n\nCons:\n1. The novelty of ... | [
5,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r17Q6WWA-",
"iclr_2018_r17Q6WWA-",
"iclr_2018_r17Q6WWA-",
"Hy-UWWjff",
"HJDRBqufM",
"ryXDrc_fG",
"ry0lbHclz",
"Bylfi7tez",
"SyuPmP3lM"
] |
iclr_2018_SJCPLLpaW | Exploring the Hidden Dimension in Accelerating Convolutional Neural Networks | DeePa is a deep learning framework that explores parallelism in all parallelizable dimensions to accelerate the training process of convolutional neural networks. DeePa optimizes parallelism at the granularity of each individual layer in the network. We present an elimination-based algorithm that finds an optimal paral... | rejected-papers | While this paper has some very interesting ideas the majority view of the reviewers and their aggregate numerical ratings are just too low to warrant acceptance. | val | [
"SkhvCRBVz",
"r1ggBqSVf",
"S18Z1_H4f",
"rkMXszYgM",
"BJ_nVijxM",
"rkCp66Tef",
"HkXoonNzf",
"HJ3rchNfG",
"BJuQ834Gf"
] | [
"author",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"We thank the reviewer for the constructive feedback.\n\nFor the TensorFlow experiments, we use synchronous training with a batch size of 32 and train the models on the ImageNet dataset. The performance numbers reported on https://www.tensorflow.org/performance/benchmarks are measured by using the asynchronous trai... | [
-1,
-1,
-1,
4,
5,
7,
-1,
-1,
-1
] | [
-1,
-1,
-1,
5,
4,
4,
-1,
-1,
-1
] | [
"r1ggBqSVf",
"S18Z1_H4f",
"iclr_2018_SJCPLLpaW",
"iclr_2018_SJCPLLpaW",
"iclr_2018_SJCPLLpaW",
"iclr_2018_SJCPLLpaW",
"rkMXszYgM",
"BJ_nVijxM",
"rkCp66Tef"
] |
iclr_2018_H1srNebAZ | Discovering the mechanics of hidden neurons | Neural networks trained through stochastic gradient descent (SGD) have been around for more than 30 years, but they still escape our understanding. This paper takes an experimental approach, with a divide-and-conquer strategy in mind: we start by studying what happens in single neurons. While being the core building bl... | rejected-papers | While one reviewer did upgrade their Rating from 6 to 7, the most negative reviewer maintains: "Overall, I find this work interesting and current results surprising. However, I find it to be a preliminary work and not yet ready for publication. The paper still lacks a conclusion / a leading hypothesis / an explanation ... | val | [
"rJ07mFogG",
"Hk5Z4hOlf",
"S1bAZxcxf",
"B1WYpDrff",
"Sk5XpPBfG",
"rJcxTPHGf",
"SkTChPSfz",
"SJT6iPSff",
"ByGR9PHzz",
"B1x99DrzM",
"rk3wqOUxz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"--------------------\nReview updates:\nRating 6 -> 7\nConfidence 2 -> 4\n\nThe rebuttal and update addressed a number of my concerns, cleared up confusing sections, and moved the paper materially closer to being publication-worthy, thus I’ve increased my score.\n--------------------\n\nI want to love this paper. T... | [
7,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1srNebAZ",
"iclr_2018_H1srNebAZ",
"iclr_2018_H1srNebAZ",
"rk3wqOUxz",
"Hk5Z4hOlf",
"Hk5Z4hOlf",
"Hk5Z4hOlf",
"S1bAZxcxf",
"rJ07mFogG",
"rJ07mFogG",
"iclr_2018_H1srNebAZ"
] |
iclr_2018_r16Vyf-0- | Image Transformer | Image generation has been successfully cast as an autoregressive sequence generation
or transformation problem. Recent work has shown that self-attention is
an effective way of modeling textual sequences. In this work, we generalize a
recently proposed model architecture based on self-attention, the T... | rejected-papers | This paper had some quality and clarity issues and the lack of motivation for the approach was pointed out by multiple reviewers. Just too far away from the acceptance threshold. | train | [
"SyA8u9OlG",
"BkYvzAixG",
"SJIFtpEbM",
"r1pOCmnMM",
"rkNH073Mf",
"H1vxCX3GM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary\n\nThis paper extends self-attention layers (Vaswani et al., 2017) from sequences to images and proposes to use the layers as part of PixelCNNs (van den Oord et al., 2016). The proposed model is evaluated in terms of visual appearance of samples and log-likelihoods. The authors find a small improvement in ... | [
6,
3,
5,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1
] | [
"iclr_2018_r16Vyf-0-",
"iclr_2018_r16Vyf-0-",
"iclr_2018_r16Vyf-0-",
"SyA8u9OlG",
"BkYvzAixG",
"SJIFtpEbM"
] |
iclr_2018_HklpCzC6- | Image Segmentation by Iterative Inference from Conditional Score Estimation | Inspired by the combination of feedforward and iterative computations in the visual cortex, and taking advantage of the ability of denoising autoencoders to estimate the score of a joint distribution, we propose a novel approach to iterative inference for capturing and exploiting the complex joint distribution of outpu... | rejected-papers | The experimental work was seen as one of the main weaknesses. | train | [
"B1hVeGtez",
"S1cNDW9eM",
"rynv0Uf-f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper proposes an image segmentation method which iteratively refines the semantic segmentation mask obtained from a deep net. To this end the authors investigate a denoising auto-encoder (DAE). Its purpose is to provide a semantic segmentation which improves upon its input in terms of the log-likelihood.\n\nM... | [
5,
4,
4
] | [
5,
4,
4
] | [
"iclr_2018_HklpCzC6-",
"iclr_2018_HklpCzC6-",
"iclr_2018_HklpCzC6-"
] |
iclr_2018_SkymMAxAb | AirNet: a machine learning dataset for air quality forecasting | In the past decade, many urban areas in China have suffered from serious air pollution problems, making air quality forecast a hot spot. Conventional approaches rely on numerical methods to estimate the pollutant concentration and require lots of computing power. To solve this problem, we applied the widely used deep l... | rejected-papers | This is an interesting application area, but the quality of the presentation and experimental work here is not sufficient for acceptance. The numerical ratings from reviewers are just not high enough to warrant acceptance. | train | [
"S1AztKDlf",
"SJl24wuxM",
"SJ7ACl5xM",
"ryR8G8j7M",
"rkuYI9UXG",
"rJBNjY8mM",
"Hk5P7LSQz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper is about open sourcing AirNet, a database that has interpolated air quality metrics in a spatial form along with matching meteorological data obtained elsewhere. In addition, the paper also develops a few baseline methods and evaluated using standard metrics such as detection rate, false alarms etc. The ... | [
5,
4,
4,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SkymMAxAb",
"iclr_2018_SkymMAxAb",
"iclr_2018_SkymMAxAb",
"iclr_2018_SkymMAxAb",
"S1AztKDlf",
"SJl24wuxM",
"SJ7ACl5xM"
] |
iclr_2018_r1nzLmWAb | Video Action Segmentation with Hybrid Temporal Networks | Action segmentation as a milestone towards building automatic systems to understand untrimmed videos has received considerable attention in the recent years. It is typically being modeled as a sequence labeling problem but contains intrinsic and sufficient differences than text parsing or speech processing. In this pap... | rejected-papers | All reviewers believed that the novelty of the contribution was limited. | train | [
"B1tWwoKxz",
"H1lTTDulG",
"HJ5VPLYxG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper discusses the problem of action segmentation in long videos, up to 10 minutes long. The basic idea is to use a temporal convolutional encoder-decoder architecture, where in the enconder 1-D temporal convolutions are used. In the decoder three variants are studied:\n\n(1) One that uses only several bidir... | [
3,
4,
3
] | [
5,
4,
5
] | [
"iclr_2018_r1nzLmWAb",
"iclr_2018_r1nzLmWAb",
"iclr_2018_r1nzLmWAb"
] |
iclr_2018_ByZmGjkA- | Understanding Grounded Language Learning Agents | Neural network-based systems can now learn to locate the referents of words and phrases in images, answer questions about visual scenes, and even execute symbolic instructions as first-person actors in partially-observable worlds. To achieve this so-called grounded language learning, models must overcome certain well-s... | rejected-papers | This paper resulted in significant discussion -- both between R2 and the authors, and between the AC, PCs, and other solicited experts.
The problem of language grounding (and instruction following) in virtual environments is clearly important, this work was one of the first in the recent resurgence, and the goal of un... | train | [
"SyTAvN8yf",
"r1-nPRulz",
"HJULpw9gz",
"Sknip6OmM",
"Hy5uZa8Xz",
"HJzGTcLXf",
"S1bLjyVmM",
"BJ2ysCXQf",
"rJig5077M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper presents an analysis of an agent trained to follow linguistic commands in a 3D environment. The behaviour of the agent is analyzed by means of a set of \"psycholinguistic\" experiments probing what it learned, and by inspection of its visual component through an attentional mechanism.\n\nOn the positiv... | [
4,
5,
7,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByZmGjkA-",
"iclr_2018_ByZmGjkA-",
"iclr_2018_ByZmGjkA-",
"Hy5uZa8Xz",
"HJzGTcLXf",
"S1bLjyVmM",
"SyTAvN8yf",
"r1-nPRulz",
"HJULpw9gz"
] |
iclr_2018_SkBcLugC- | Fast and Accurate Inference with Adaptive Ensemble Prediction for Deep Networks | Ensembling multiple predictions is a widely-used technique to improve the accuracy of various machine learning tasks. In image classification tasks, for example, averaging the predictions for multiple patches extracted from the input image significantly improves accuracy. Using multiple networks trained independently t... | rejected-papers | The manuscript proposes a simple technique for adaptive ensemble prediction. Unfortunately, several significant concerns were raised (by R2 and R3) that this AC agrees with. Both R2 and R3 asked fairly specific questions and requested follow-up experiments, which have not been addressed. | train | [
"H1-ki5Bef",
"H1tMJeseM",
"ry6zwoief",
"ry3YAaQ4f",
"S1Rnr37Ez",
"HkW5mV67G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author"
] | [
"In this paper it is described a method that can be used to speed up the prediction process of ensembles of classifiers that output probabilistic predictions. The method proposed is very simple and it is based on the observation that in the case that the individual predictors are very sure about the potential class... | [
6,
5,
5,
-1,
-1,
-1
] | [
3,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_SkBcLugC-",
"iclr_2018_SkBcLugC-",
"iclr_2018_SkBcLugC-",
"S1Rnr37Ez",
"HkW5mV67G",
"iclr_2018_SkBcLugC-"
] |
iclr_2018_rJe7FW-Cb | A Painless Attention Mechanism for Convolutional Neural Networks | We propose a novel attention mechanism to enhance Convolutional Neural Networks for fine-grained recognition. The proposed mechanism reuses CNN feature activations to find the most informative parts of the image at different depths with the help of gating mechanisms and without part annotations. Thus, it can be used to... | rejected-papers | This paper received borderline reviews. Initially, all reviewers raised a number of concerns (clarity, small improvements, etc). Even after some back and forth discussion, concerns remain, and it's clear that while the idea has potential, another round of reviewing is needed before a decision can be reached. This would... | train | [
"SJypm6QBf",
"H1VpcYQSz",
"ByogdHI4M",
"BJZ7a4ING",
"rkzOQxcgM",
"Sky96rolf",
"ry2OdYCeM",
"H1H_vJ_QM",
"HkNSDkdQG",
"HkabPkOQM",
"SyTmfkumG",
"r1V-bJOXG",
"B10s11dmz"
] | [
"author",
"public",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"Thank you for the interest in our paper. We have included this explanation in the “Related Work” section, as a specialized solution for multilabel classification, where instead of learning universal modules, a ResNet is modified to improve its multilabel classification by enhancing the predictions with the learned... | [
-1,
-1,
-1,
-1,
5,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
-1,
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"H1VpcYQSz",
"iclr_2018_rJe7FW-Cb",
"BJZ7a4ING",
"H1H_vJ_QM",
"iclr_2018_rJe7FW-Cb",
"iclr_2018_rJe7FW-Cb",
"iclr_2018_rJe7FW-Cb",
"rkzOQxcgM",
"rkzOQxcgM",
"rkzOQxcgM",
"Sky96rolf",
"ry2OdYCeM",
"iclr_2018_rJe7FW-Cb"
] |
iclr_2018_BkiIkBJ0b | Do Deep Reinforcement Learning Algorithms really Learn to Navigate? | Deep reinforcement learning (DRL) algorithms have demonstrated progress in learning to find a goal in challenging environments. As the title of the paper by Mirowski et al. (2016) suggests, one might assume that DRL-based algorithms are able to “learn to navigate” and are thus ready to replace classical mapping and pat... | rejected-papers | This paper received divergent ratings (7, 3, 3). While there is value in thorough evaluation papers, this manuscript has significant presentation issues. As all three reviewers point out, the way it is currently written, it misrepresents the claims made by Mirowski et al 2016 and over-reaches in its findings. Unfortuna... | train | [
"rypaPgFxz",
"HyDUxCKlf",
"ryYlB69gz",
"rkXA-Wvzf",
"H1lxCJ-bM",
"BJ3G3JbZf",
"BkID51bWM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"author",
"author",
"author"
] | [
"The paper evaluates one proposed Deep RL-based model (Mirowski et al. 2016) on its ability to generally navigate. This evaluation includes training the agent on a set of training mazes and testing it's performance on a set of held-out test mazes. Evaluation metrics include repeated latency to the goal and comparis... | [
7,
3,
3,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkiIkBJ0b",
"iclr_2018_BkiIkBJ0b",
"iclr_2018_BkiIkBJ0b",
"BkID51bWM",
"rypaPgFxz",
"HyDUxCKlf",
"ryYlB69gz"
] |
iclr_2018_HJPSN3gRW | Learning to navigate by distilling visual information and natural language instructions | In this work, we focus on the problem of grounding language by training an agent
to follow a set of natural language instructions and navigate to a target object
in a 2D grid environment. The agent receives visual information through raw
pixels and a natural language instruction telling what task need... | rejected-papers | This paper was reviewed by 3 expert reviews and received largely negative reviews, with concerns about the toy-ish nature of the 2D environments and limited novelty.
Since ICLR18 received multiple papers on similar topics, we took additional measures to ensure that papers were similar papers were judged under the same... | train | [
"SyHpaZUrG",
"B1ji6W8Bz",
"BJLtabUHf",
"HJ57T-UBz",
"HyE1BPlrM",
"S1ZnEPxHM",
"S1UE88gSM",
"rku1fPgSM",
"rkf84UgSz",
"BJZpwtUNM",
"BJtd58mlM",
"BkLPOycez",
"SyXYbebMG",
"ryZgGvaQz",
"SJgErPTQz",
"S19J6I6mG",
"rJlcD8p7G",
"S1z0sST7G",
"S1C4UyGGf"
] | [
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"We have added more results showing significant improvement over baseline both in terms of accuracy and speed of convergence. \nAlso added preliminary results of hard scenario which looks promising ... ",
"We have added more results showing significant improvement over baseline both in terms of accuracy and speed... | [
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
4,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
4,
5,
3,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"BJtd58mlM",
"BkLPOycez",
"rkf84UgSz",
"SyXYbebMG",
"SyXYbebMG",
"BkLPOycez",
"BJtd58mlM",
"BJZpwtUNM",
"iclr_2018_HJPSN3gRW",
"ryZgGvaQz",
"iclr_2018_HJPSN3gRW",
"iclr_2018_HJPSN3gRW",
"iclr_2018_HJPSN3gRW",
"BJtd58mlM",
"iclr_2018_HJPSN3gRW",
"SyXYbebMG",
"BkLPOycez",
"S1C4UyGGf"... |
iclr_2018_HJNGGmZ0Z | What is image captioning made of? | We hypothesize that end-to-end neural image captioning systems work seemingly well because they exploit and learn ‘distributional similarity’ in a multimodal feature space, by mapping a test image to similar training images in this space and generating a caption from the same space. To validate our hypothesis, we focus... | rejected-papers | Paper reviewed by three experts who have provided detailed feedback. All three recommend rejection, and this AC sees no reason to overrule their recommendation. | train | [
"S1wHgNwHz",
"ryXXgEPHG",
"SyZwTXVrz",
"BybWlKXgz",
"r1fNl7Cgz",
"By5_q5y-z",
"H1xPSyofM",
"rkqoKB9fz",
"HybctS5zf",
"SkCgOrcGG",
"Hk60UB5GM",
"rJGFUBqfM",
"S1P_G7Mfz"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"We appreciate and thank the reviewer for going through our rebuttal and revised manuscript and for the comments. However, we disagree with many of the points raised in the above review.\n\n> - \"End-to-end IC models are remarkably capable of separating structure from noisy input representations, as demonstrated by... | [
-1,
-1,
-1,
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
5,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"SyZwTXVrz",
"SyZwTXVrz",
"iclr_2018_HJNGGmZ0Z",
"iclr_2018_HJNGGmZ0Z",
"iclr_2018_HJNGGmZ0Z",
"iclr_2018_HJNGGmZ0Z",
"iclr_2018_HJNGGmZ0Z",
"BybWlKXgz",
"BybWlKXgz",
"r1fNl7Cgz",
"By5_q5y-z",
"iclr_2018_HJNGGmZ0Z",
"iclr_2018_HJNGGmZ0Z"
] |
iclr_2018_rkWN3g-AZ | XGAN: Unsupervised Image-to-Image Translation for many-to-many Mappings | Style transfer usually refers to the task of applying color and texture information from a specific style image to a given content image while preserving the structure of the latter. Here we tackle the more generic problem of semantic style transfer: given two unpaired collections of images, we aim to learn a mapping b... | rejected-papers | This paper was reviewed by 3 expert reviewers. All three recommend rejection citing significant concerns (e.g. missing baselines). | train | [
"HJFBUEEBM",
"Skj8GNn4z",
"Sk2PtWDVz",
"rybCT6Olf",
"BJklMMFez",
"HJSBtHZZM",
"r1Tur03Qz",
"SkLHB03mG",
"rJWFEC27f",
"H102m03mG",
"By_ZiUJzf",
"SygfwiYgM",
"HkcKlK_1f",
"rJ8xTtx1f"
] | [
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"public"
] | [
"The semantic consistency loss is ||e2(d2(e1(x1))), e1(x1)||. There are two possible implementations:\n1. Treat e1(x1) as the label/target and do not back-propagate through it.\n2. Use the gradient w.r.t. the e1(x1) to update e1. So the e1 is updated with two gradient flows.\n\nI'm wondering which one the authors u... | [
-1,
-1,
-1,
3,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
5,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rkWN3g-AZ",
"BJklMMFez",
"SkLHB03mG",
"iclr_2018_rkWN3g-AZ",
"iclr_2018_rkWN3g-AZ",
"iclr_2018_rkWN3g-AZ",
"By_ZiUJzf",
"HJSBtHZZM",
"BJklMMFez",
"rybCT6Olf",
"iclr_2018_rkWN3g-AZ",
"HkcKlK_1f",
"rJ8xTtx1f",
"iclr_2018_rkWN3g-AZ"
] |
iclr_2018_BJDH5M-AW | Synthesizing Robust Adversarial Examples | Neural network-based classifiers parallel or exceed human-level accuracy on many common tasks and are used in practical systems. Yet, neural networks are susceptible to adversarial examples, carefully perturbed inputs that cause networks to misbehave in arbitrarily chosen ways. When generated with standard methods, the... | rejected-papers | This paper studies the problem of synthesizing adversarial examples that will succeed at fooling a classification system under unknown viewpoint, lighting, etc conditions. For that purpose, the authors propose a data-augmentation technique (called "EOT") that makes adversarial examples robust against a predetermined fa... | train | [
"HyIrp18rz",
"H1FaG8FgM",
"SyRDNVqxz",
"H1nifI9eG",
"SJnDHwnQf",
"SkwQmwnXM",
"Skul7w3mz",
"H1B9fD3Qf",
"BJ2bfP3QM",
"HymvZv37G",
"r1bnlcZMf",
"Byp5w7OAW",
"HJb4TgKAb",
"BkwU6xF0b",
"rJkwuN9RZ",
"Hk3iatqCW",
"Bk-AZjY0W",
"rk4nKr_0b",
"B1p3Q4_Rb",
"B1e1yJLCb"
] | [
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"author",
"author",
"author",
"author",
"public",
"public",
"public",
"public",
"public"
] | [
"Can you comment on what aspects can be explored further?",
"Summary: This work proposes a way to create 3D objects to fool the classification of their pictures from different view points by a neural network.\nRather than optimizing the log-likelihood of a single example, the optimization if performed over a the ... | [
-1,
5,
6,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"HymvZv37G",
"iclr_2018_BJDH5M-AW",
"iclr_2018_BJDH5M-AW",
"iclr_2018_BJDH5M-AW",
"iclr_2018_BJDH5M-AW",
"H1FaG8FgM",
"SyRDNVqxz",
"H1nifI9eG",
"rk4nKr_0b",
"r1bnlcZMf",
"iclr_2018_BJDH5M-AW",
"B1e1yJLCb",
"B1p3Q4_Rb",
"rk4nKr_0b",
"Bk-AZjY0W",
"HJb4TgKAb",
"iclr_2018_BJDH5M-AW",
"... |
iclr_2018_S1680_1Rb | CAYLEYNETS: SPECTRAL GRAPH CNNS WITH COMPLEX RATIONAL FILTERS | The rise of graph-structured data such as social networks, regulatory networks, citation graphs, and functional brain networks, in combination with resounding success of deep learning in various applications, has brought the interest in generalizing deep learning models to non-Euclidean domains.
In this paper, w... | rejected-papers | This paper considers graph neural representations that use Cayley polynomials of the graph Laplacian as generators. These polynomials offer better frequency localization than Chebyshev polynomials. The authors illustrate the advantages of Cayleynets on several benchmarks, producing modest improvements.
Reviewers were ... | train | [
"rJKsozOxM",
"BJWjA85xz",
"HyYek2cgM",
"ByM26osmf",
"H16j_wsQf",
"HyPqsdq7G",
"Skudq_cQf",
"rJYQYd97G",
"S1vcOrtmM",
"S18cwVtQG",
"Hks9aT_Qz",
"Sy_LzeO7M",
"BylmKBImM",
"r1fAlRBQM",
"BkO6CcXQM",
"HyqGHjDzz",
"rka7EjvGM",
"B1E7loPGG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"public",
"author",
"public",
"author",
"public",
"author",
"public",
"author",
"public",
"author",
"author",
"author",
"author"
] | [
"The paper proposes a new filter for spectral analysis on graphs for graph CNNs. The filter is a rational function based on the Cayley transform. Unlike other popular variants, it is not strictly supported on a small graph neighborhood, but the paper proves an exponential-decay property on the norm of a filtered ve... | [
6,
4,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
3,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1680_1Rb",
"iclr_2018_S1680_1Rb",
"iclr_2018_S1680_1Rb",
"iclr_2018_S1680_1Rb",
"HyPqsdq7G",
"BylmKBImM",
"rJYQYd97G",
"S1vcOrtmM",
"S18cwVtQG",
"Hks9aT_Qz",
"Sy_LzeO7M",
"iclr_2018_S1680_1Rb",
"r1fAlRBQM",
"iclr_2018_S1680_1Rb",
"rJKsozOxM",
"rka7EjvGM",
"BJWjA85xz",
"... |
iclr_2018_SkaPsfZ0W | Network of Graph Convolutional Networks Trained on Random Walks | Graph Convolutional Networks (GCNs) are a recently proposed architecture which has had success in semi-supervised learning on graph-structured data. At the same time, unsupervised learning of graph embeddings has benefited from the information contained in random walks. In this paper we propose a model, Network of GCNs... | rejected-papers | This paper proposes a multiscale variant of Graph Convolutional Networks (GCN) , obtained by combining separate GCN modules using powers of normalized adjacency as generators. The model is tested on several node classification semi-supervised tasks obtaining excellent numerical performance.
Reviewers acknowledged the ... | train | [
"BJququIlf",
"H1kTclOlf",
"H12U-wYgG",
"rJGeK9pXG",
"HkZAY5T7G",
"HJLHwTdQM",
"SkeGD2hff",
"HJeR42nGG",
"BkNk82nMz",
"BJkjr22MG",
"H1rQ5HmgG",
"BJfkzLgxf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"public",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"The paper proposes a novel graph convolutional network in which a variety of random walk steps are involved with multiple GCNs.\n\nThe basic idea, introducing long rage dependecy, would be interesting. Robustness for the feature remove is also interesting.\n\nThe validation set would be important for the proposed ... | [
5,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SkaPsfZ0W",
"iclr_2018_SkaPsfZ0W",
"iclr_2018_SkaPsfZ0W",
"H1kTclOlf",
"H12U-wYgG",
"SkeGD2hff",
"BJfkzLgxf",
"H12U-wYgG",
"BJququIlf",
"H1kTclOlf",
"BJfkzLgxf",
"iclr_2018_SkaPsfZ0W"
] |
iclr_2018_SyBBgXWAZ | Optimal transport maps for distribution preserving operations on latent spaces of Generative Models | Generative models such as Variational Auto Encoders (VAEs) and Generative Adversarial Networks (GANs) are typically trained for a fixed prior distribution in the latent space, such as uniform or Gaussian.
After a trained model is obtained, one can sample the Generator in various forms for exploration and understa... | rejected-papers | This paper exposes a simple recipe to manipulate the latent space of generative models in such a way to minimize the mismatch between the prior distribution and that of the manipulated latent space. Manipulations such as linear interpolation are commonplace in the literature, and this work will be helpful to improve as... | train | [
"HyBft3dgM",
"SJuG7tqxz",
"SJ13MSaxf",
"Bylsiz3QG",
"H1EeeKM7G",
"Bk6VytfXz",
"S1jQT_G7z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Authors note that models may be trained for a certain distribution (e.g. uniform or Gaussian) but then \"used\" by interpolating or jittering known examples, which has a different distribution. While the authors are clear about the fact that this is a mismatch, I did not find it well-motivated why it was \"the r... | [
4,
6,
6,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SyBBgXWAZ",
"iclr_2018_SyBBgXWAZ",
"iclr_2018_SyBBgXWAZ",
"iclr_2018_SyBBgXWAZ",
"HyBft3dgM",
"SJuG7tqxz",
"SJ13MSaxf"
] |
iclr_2018_H139Q_gAW | Learning Graph Convolution Filters from Data Manifold | Convolution Neural Network (CNN) has gained tremendous success in computer vision tasks with its outstanding ability to capture the local latent features. Recently, there has been an increasing interest in extending CNNs to the general spatial domain. Although various types of graph convolution and geometric convolutio... | rejected-papers | This paper proposes to combine Depthwise separable convolutions developed for 2d grids with recent graph convolutional architectures. The resulting architecture can be seen as learning both node and edge features, the latter encoding node similarities with learnt weights.
Reviewers agreed that this is an interesting li... | test | [
"BkCxP2Fez",
"S1rTmy9xG",
"BJFoMD9eG",
"HJaNmI3ZM",
"B1e8aB3WM",
"Hkoc3H2bG",
"Hywv3ShbM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper presents a Depthwise Separable Graph Convolution network that aims\nat generalizing Depthwise convolutions, that exhibit a nice performance in image\nrelated tasks, to the graph domain. In particular it targets\nGraph Convolutional Networks.\n\nIn the abstract the authors mention that the Depthwise Separ... | [
4,
6,
5,
-1,
-1,
-1,
-1
] | [
5,
4,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H139Q_gAW",
"iclr_2018_H139Q_gAW",
"iclr_2018_H139Q_gAW",
"BkCxP2Fez",
"S1rTmy9xG",
"Hywv3ShbM",
"BJFoMD9eG"
] |
iclr_2018_SyjsLqxR- | Universality, Robustness, and Detectability of Adversarial Perturbations under Adversarial Training | Classifiers such as deep neural networks have been shown to be vulnerable against adversarial perturbations on problems with high-dimensional input space. While adversarial training improves the robustness of classifiers against such adversarial perturbations, it leaves classifiers sensitive to them on a non-negligible... | rejected-papers | This paper studies to what extent adversarial training affects the properties of adversarial examples in object classification.
Reviewers found the work going in the right direction, but agreed that it needs further evidence/focus in order to constitute a significant contribution to the ICLR community. In particular, ... | train | [
"rydWCNKxz",
"HkmB_HqlG",
"ryMchFseG",
"rkSVxjvGM",
"B10EqVUGf",
"B1h7pVUGf",
"r14Id_uyM",
"BJ29Juukz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public"
] | [
"Summary:\n\nThis paper empirically studies adversarial perturbations dx and what the effects are of adversarial training (AT) with respect to shared (dx fools for many x) and singular (only for a single x) perturbations. Experiments use a (previously published) iterative fast-gradient-sign-method and use a Resnet ... | [
3,
6,
6,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SyjsLqxR-",
"iclr_2018_SyjsLqxR-",
"iclr_2018_SyjsLqxR-",
"rydWCNKxz",
"ryMchFseG",
"HkmB_HqlG",
"BJ29Juukz",
"iclr_2018_SyjsLqxR-"
] |
iclr_2018_Hki-ZlbA- | Ground-Truth Adversarial Examples | The ability to deploy neural networks in real-world, safety-critical systems is severely limited by the presence of adversarial examples: slightly perturbed inputs that are misclassified by the network. In recent years, several techniques have been proposed for training networks that are robust to such examples; and ea... | rejected-papers | This paper describes a method to generate provably 'optimal' adversarial examples, leveraging the so-called 'Reluplex' technique, which can evaluate properties of piece-wise linear representations.
Reviewers agreed that incorporating optimality certificates into adversarial examples is a promising direction to follow, ... | train | [
"S1Q_cbqxf",
"H1TnZzcgz",
"Sy5sYncgM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"Summary: The paper proposes a method to compute adversarial examples with minimum distance to the original inputs, and to use the method to do two things: Show how well heuristic methods do in finding \"optimal/minimal\" adversarial examples (how close the come to the minimal change that flips the label) and to as... | [
5,
4,
6
] | [
4,
4,
3
] | [
"iclr_2018_Hki-ZlbA-",
"iclr_2018_Hki-ZlbA-",
"iclr_2018_Hki-ZlbA-"
] |
iclr_2018_SyqAPeWAZ | CNNs as Inverse Problem Solvers and Double Network Superresolution | In recent years Convolutional Neural Networks (CNN) have been used extensively for Superresolution (SR). In this paper, we use inverse problem and sparse representation solutions to form a mathematical basis for CNN operations. We show how a single neuron is able to provide the optimum solution for inverse problem, giv... | rejected-papers | This paper addresses the question of how to solve image super-resolution, building on a connection between sparse regularization and neural networks.
Reviewers agreed that this paper needs to be rewritten, taking into account recent work in the area and significantly improving the grammar. The AC thus recommends reject... | train | [
"rkDK2NwgG",
"rke8ggtxG",
"rkHX_Bjlf",
"SysQi-Fff",
"Hk7WiZFGG",
"HJincWtMf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper proposes an understanding of the relation between inverse problems, CNNs and sparse representations. Using the ground work for each proposes a new competitive super resolution technique using CNNs. Overall I liked authors' endeavors bringing together different fields of research addressing similar issues... | [
6,
3,
4,
-1,
-1,
-1
] | [
2,
5,
4,
-1,
-1,
-1
] | [
"iclr_2018_SyqAPeWAZ",
"iclr_2018_SyqAPeWAZ",
"iclr_2018_SyqAPeWAZ",
"rkDK2NwgG",
"rke8ggtxG",
"rkHX_Bjlf"
] |
iclr_2018_HklZOfW0W | UPS: optimizing Undirected Positive Sparse graph for neural graph filtering | In this work we propose a novel approach for learning graph representation of the data using gradients obtained via backpropagation. Next we build a neural network architecture compatible with our optimization approach and motivated by graph filtering in the vertex domain. We demonstrate that the learned graph has rich... | rejected-papers | This paper addresses the problem of learning neural graph representations, based on graph filtering techniques in the vertex domain.
Reviewers agreed on the fact that this paper has limited interest in its current form, and has serious grammatical issues. The AC thus recommends rejection at this time. | train | [
"HJb3ygfxf",
"rk9JKKwxM",
"BkuAX_h-M",
"rJgKwD67M",
"HkzgwPTmf",
"ByT1_Pa7G",
"HyeSDv6Qz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Learning adjacency matrix of a graph with sparsely connected undirected graph with nonnegative edge weights is the goal of this paper. A projected sub-gradient descent algorithm is used. The UPS optimizer by itself is not new.\n\nGraph Polynomial Signal (GPS) neural network is proposed to address two shortcomings ... | [
6,
3,
4,
-1,
-1,
-1,
-1
] | [
3,
3,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HklZOfW0W",
"iclr_2018_HklZOfW0W",
"iclr_2018_HklZOfW0W",
"rk9JKKwxM",
"iclr_2018_HklZOfW0W",
"HJb3ygfxf",
"BkuAX_h-M"
] |
iclr_2018_H11lAfbCW | On Characterizing the Capacity of Neural Networks Using Algebraic Topology | The learnability of different neural architectures can be characterized directly by computable measures of data complexity. In this paper, we reframe the problem of architecture selection as understanding how data determines the most expressive and generalizable architectures suited to that data, beyond inductive bias.... | rejected-papers | This paper attempts to connect the expressivity of neural networks with a measure of topological complexity. The authors present some empirical results on simplified datasets.
All reviewers agreed that this is an intriguing line of research, but that the current manuscript is still presenting preliminary results, and t... | train | [
"SJHp-7Klz",
"B19Fsy5gM",
"Sy4l2B9gG",
"Sk2OMTrMz",
"Sk4KcpHMf",
"BkmNYaBGz",
"rkA_SaHMz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Paper Summary:\n\nThis paper looks at empirically measuring neural network architecture expressivity by examining performance on a variety of complex datasets, measuring dataset complexity with algebraic topology. The paper first introduces the notion of topological equivalence for datasets -- a desirable measure ... | [
3,
4,
4,
-1,
-1,
-1,
-1
] | [
5,
5,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H11lAfbCW",
"iclr_2018_H11lAfbCW",
"iclr_2018_H11lAfbCW",
"SJHp-7Klz",
"Sy4l2B9gG",
"rkA_SaHMz",
"B19Fsy5gM"
] |
iclr_2018_Bk7wvW-C- | Exploring Asymmetric Encoder-Decoder Structure for Context-based Sentence Representation Learning | Context information plays an important role in human language understanding, and it is also useful for machines to learn vector representations of language. In this paper, we explore an asymmetric encoder-decoder structure for unsupervised context-based sentence representation learning. As a result, we build an encoder... | rejected-papers | here, yet another sentence representation method is proposed. i agree with R1 and R3 that this does not contribute significantly to be a full-length conference paper. | train | [
"SyU_UK2lf",
"Skrfq_Jlz",
"Hy4cMGVlf",
"r1OWJupQz",
"HJXUns5QG",
"HyeAFeUGM",
"BJBaulIMG",
"ryqHPgUMG",
"B1hPJJd0-",
"rJBQsOt0b",
"rknG_y_Cb",
"rJtgBTDCZ"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"\n-- updates to review: --\n\nThanks for trying to respond to my comments. I find the new results very interesting and fill in some empirical gaps that I think were worth investigating. I'm now more confident that this paper is worth publishing and I increased my rating from 6 to 7. \n\nI admit that this is a pret... | [
7,
6,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Bk7wvW-C-",
"iclr_2018_Bk7wvW-C-",
"iclr_2018_Bk7wvW-C-",
"iclr_2018_Bk7wvW-C-",
"ryqHPgUMG",
"Skrfq_Jlz",
"Hy4cMGVlf",
"SyU_UK2lf",
"rJtgBTDCZ",
"rknG_y_Cb",
"B1hPJJd0-",
"iclr_2018_Bk7wvW-C-"
] |
iclr_2018_ryHM_fbA- | Learning Document Embeddings With CNNs | This paper proposes a new model for document embedding. Existing approaches either require complex inference or use recurrent neural networks that are difficult to parallelize. We take a different route and use recent advances in language modeling to develop a convolutional neural network embedding model. This allows u... | rejected-papers | there are two separate ideas embedded in this submission; (1) language modelling (with the negative sampling objective by mikolov et al.) is a good objective to use for extracting document representation, and (2) CNN is a faster alternative to RNN's, both of which have been studied in similar contexts earlier (e.g., pa... | train | [
"B1qFEzr4M",
"HJmMNVDlz",
"BkBvQaFez",
"B1KZkIqxG",
"BJ1ZUHtGf",
"HyO-3YVGM",
"Hk5UoFVzM",
"S1uR9tEGG",
"r1tz5yoeM",
"SJOi664ez",
"B1siqc_yz",
"ry21MYLJG"
] | [
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public",
"public",
"author",
"public"
] | [
"The reported accuracies for doc2vec on IMDB are wrong, presumably a consequence of a suboptimal re-implementation. In the doc2vec paper, they report accuracy of 92.58%, significantly higher than your reported doc2vec accuracy, 88.73%, and the accuracy for the proposed method, 90.15%. Given this extremely poor impl... | [
-1,
6,
4,
2,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
3,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ryHM_fbA-",
"iclr_2018_ryHM_fbA-",
"iclr_2018_ryHM_fbA-",
"iclr_2018_ryHM_fbA-",
"Hk5UoFVzM",
"HJmMNVDlz",
"BkBvQaFez",
"B1KZkIqxG",
"B1siqc_yz",
"iclr_2018_ryHM_fbA-",
"ry21MYLJG",
"iclr_2018_ryHM_fbA-"
] |
iclr_2018_ByUEelW0- | Modifying memories in a Recurrent Neural Network Unit | Long Short-Term Memory (LSTM) units have the ability to memorise and use long-term dependencies between inputs to generate predictions on time series data. We introduce the concept of modifying the cell state (memory) of LSTMs using rotation matrices parametrised by a new set of trainable weights. This addition shows s... | rejected-papers | the idea is interesting, but as pointed out the reviewers (and also agreed by the authors), the current manuscript lacks clear motivations, reasons underlying specific design choices and convincing empirical evaluation. | train | [
"SkYqiWteM",
"HyoHt8YlG",
"rJeml9qlf",
"BJ0XIGzMz",
"r1Jn2Mq0Z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public"
] | [
"The paper proposes to add a rotation operation in long short-term memory (LSTM) cells. It performs experiments on bAbI tasks and showed that the results are better than the simple baselines with original LSTM cells. There are a few problems with the paper.\n\nFirstly, the title and abstract discuss \"modifying mem... | [
4,
4,
3,
-1,
-1
] | [
3,
3,
4,
-1,
-1
] | [
"iclr_2018_ByUEelW0-",
"iclr_2018_ByUEelW0-",
"iclr_2018_ByUEelW0-",
"iclr_2018_ByUEelW0-",
"iclr_2018_ByUEelW0-"
] |
iclr_2018_Syt0r4bRZ | Tree2Tree Learning with Memory Unit | Traditional recurrent neural network (RNN) or convolutional neural net- work (CNN) based sequence-to-sequence model can not handle tree structural data well. To alleviate this problem, in this paper, we propose a tree-to-tree model with specially designed encoder unit and decoder unit, which recursively encodes tree in... | rejected-papers | the problem is interesting, and the reviewers acknowledge it's worth an effort to tackle. unfortunately all the reviewers found the work to be too preliminary without a convincing evidence supporting the proposed approach against other alternatives (or on its own.) | train | [
"HJA-R_Fxf",
"rkP1vRFxz",
"rkAf3gcgM",
"Hy6misoxf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public"
] | [
"Summary: the paper proposes a tree2tree architecture for NLP tasks. Both the encoder and decoder of this architecture make use of memory cells: the encoder looks like a tree-lstm to encode a tree bottom-up, the decoder generates a tree top-down by predicting the number of children first. The objective function is ... | [
2,
5,
4,
-1
] | [
4,
4,
4,
-1
] | [
"iclr_2018_Syt0r4bRZ",
"iclr_2018_Syt0r4bRZ",
"iclr_2018_Syt0r4bRZ",
"iclr_2018_Syt0r4bRZ"
] |
iclr_2018_Hk2MHt-3- | Coupled Ensembles of Neural Networks | We investigate in this paper the architecture of deep convolutional networks. Building on existing state of the art models, we propose a reconfiguration of the model parameters into several parallel branches at the global network level, with each branch being a standalone CNN. We show that this arrangement is an effici... | rejected-papers | The paper studies end-to-end training of a multi-branch convolutional network. This appears to lead to strong accuracies on the CIFAR and SVHN datasets, but it remains unclear whether or not this results transfers to ImageNet. The proposed approach is hardly novel, and lacks a systematic comparison with "regular" ensem... | train | [
"SJBEkBpmz",
"SJXrqMPgf",
"Hk8Nwx9xf",
"rkbHBeSbM",
"B1jYCEamG",
"H1OW0NpmM",
"SJovT46Qz"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"We have updated the paper based on the reviewer suggestions and also added responses to their questions.\n\nMain updates:\n\n- added figure to demonstarte the model architecture and fusion scheme (Figure 1)\n- added Section G to compare between single-branch and multi-branch models for a fixed training time budget... | [
-1,
6,
6,
6,
-1,
-1,
-1
] | [
-1,
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_Hk2MHt-3-",
"iclr_2018_Hk2MHt-3-",
"iclr_2018_Hk2MHt-3-",
"iclr_2018_Hk2MHt-3-",
"rkbHBeSbM",
"SJXrqMPgf",
"Hk8Nwx9xf"
] |
iclr_2018_SJLy_SxC- | Log-DenseNet: How to Sparsify a DenseNet | Skip connections are increasingly utilized by deep neural networks to improve accuracy and cost-efficiency. In particular, the recent DenseNet is efficient in computation and parameters, and achieves state-of-the-art predictions by directly connecting each feature layer to all previous ones. However, DenseNet's extreme... | rejected-papers | The paper presents an empirical study into sparse connectivity patterns for DenseNets.
Whilst sparse connectivity is potentially interesting, the paper does not make a strong argument for such sparse connectivity patterns: in particular, the results on ImageNet suggest that sparse connectivity performs substantially w... | val | [
"BJaucU_gM",
"H12ujAYlG",
"Hkxey1cxM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper investigates how to impose layer-wise connections in DenseNets most efficiently. The authors propose a connection-pattern, which connects layer i to layer i-2^k, k=0,1,2... The authors also propose maximum backpropgation distance (MBD) for measuring the fluency of gradient flow in the network, and justi... | [
6,
6,
5
] | [
4,
4,
4
] | [
"iclr_2018_SJLy_SxC-",
"iclr_2018_SJLy_SxC-",
"iclr_2018_SJLy_SxC-"
] |
iclr_2018_HknbyQbC- | Generating Adversarial Examples with Adversarial Networks | Deep neural networks (DNNs) have been found to be vulnerable to adversarial examples resulting from adding small-magnitude perturbations to inputs. Such adversarial examples can mislead DNNs to produce adversary-selected results.
Different attack strategies have been proposed to generate adversarial examples, but... | rejected-papers | The paper presents AdvGAN: a GAN that is trained to generate adversarial examples against a convolutional network. The motivation for this method is unclear: the proposed attack does not outperform simpler attack methods such as Carlini-Wagner attack. In white-box settings, a clear downside for the attacker is that it ... | train | [
"BkkpVvWEM",
"rJFANDH4f",
"SJgWlg6yM",
"BJjc-tsef",
"S1gL2gTlf",
"Skv0tazVz",
"Hk6T-Bomz",
"B1-sJrjQz",
"B1OZgBoQG",
"Sy3y-GGbf",
"rkkYlehxz",
"HJZ_Lhjgf",
"BkGQbf9lz",
"BylMzPmxM",
"B1IfJszxM",
"BknKcXbxz",
"HJgQmzbgM",
"HJpmU2JeG",
"HkZgMFp1z",
"HJ2cJmgJf"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"public",
"public",
"author",
"public",
"author",
"public",
"author",
"public"
] | [
"Changes made in our revised version are listed as below：\n- Added perturbation plot in Figure 3(c)(d) and Fig 4(b) for CIFAR-10 and ImageNet, respectively.\n- Added a comparison between AdvGAN, FGSM, and optimization methods, comparing the relative changes from original images to adversarial examples (Table 7 in t... | [
-1,
-1,
4,
6,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HknbyQbC-",
"SJgWlg6yM",
"iclr_2018_HknbyQbC-",
"iclr_2018_HknbyQbC-",
"iclr_2018_HknbyQbC-",
"Sy3y-GGbf",
"SJgWlg6yM",
"S1gL2gTlf",
"BJjc-tsef",
"iclr_2018_HknbyQbC-",
"HJZ_Lhjgf",
"BkGQbf9lz",
"B1IfJszxM",
"BknKcXbxz",
"BknKcXbxz",
"HJgQmzbgM",
"HJpmU2JeG",
"HkZgMFp1z"... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.