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_H1DkN7ZCZ | Deep learning mutation prediction enables early stage lung cancer detection in liquid biopsy | Somatic cancer mutation detection at ultra-low variant allele frequencies (VAFs) is an unmet challenge that is intractable with current state-of-the-art mutation calling methods. Specifically, the limit of VAF detection is closely related to the depth of coverage, due to the requirement of multiple supporting reads in ... | workshop-papers | Authors present a method for representing DNA sequence reads as one-hot encoded vectors, with genomic context (expected original human sequence), read sequence, and CIGAR string (match operation encoding) concatenated as a single input into the framework. Method is developed on 5 lung cancer patients and 4 melanoma pat... | train | [
"rk6_mqulz",
"H18G3z5gM",
"rJKKR8qxM",
"B11xHxxVf",
"SJFOVOpXf",
"ryLJedamM",
"B1GXE_aQG",
"Hk7yNdamz",
"Byf2CVy-f",
"B1avSDnlM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"In this paper the author propose a CNN based solution for somatic mutation calling at ultra low allele frequencies.\nThe tackled problem is a hard task in computational biology, and the proposed solution Kittyhawk, although designed with very standard ingredients (several layers of CNN inspired to the VGG structur... | [
8,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1DkN7ZCZ",
"iclr_2018_H1DkN7ZCZ",
"iclr_2018_H1DkN7ZCZ",
"Byf2CVy-f",
"rk6_mqulz",
"rJKKR8qxM",
"Hk7yNdamz",
"H18G3z5gM",
"B1avSDnlM",
"iclr_2018_H1DkN7ZCZ"
] |
iclr_2018_ByaQIGg0- | AUTOMATED DESIGN USING NEURAL NETWORKS AND GRADIENT DESCENT | We propose a novel method that makes use of deep neural networks and gradient decent to perform automated design on complex real world engineering tasks. Our approach works by training a neural network to mimic the fitness function of a design optimization task and then, using the differential nature of the neural netw... | workshop-papers | Differentiable neural networks used as a measure of design optimality in order to improve efficiency of automated design.
Pros:
- Genetic algorithms, which are the dominant optimization routine for automated design systems, can be computationally expensive. This approach alleviates this bottleneck under certain circu... | val | [
"HJ_m58weG",
"HkWsjrOlz",
"rJ_jlsdlf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper proposes to use neural network and gradient descent to automatically design for engineering tasks. It uses two networks, parameterization network and prediction network to model the mapping from design parameters to fitness. It uses back propagation (gradient descent) to improve the design. The method i... | [
5,
7,
4
] | [
4,
4,
5
] | [
"iclr_2018_ByaQIGg0-",
"iclr_2018_ByaQIGg0-",
"iclr_2018_ByaQIGg0-"
] |
iclr_2018_HyDMX0l0Z | Towards Effective GANs for Data Distributions with Diverse Modes | Generative Adversarial Networks (GANs), when trained on large datasets with diverse modes, are known to produce conflated images which do not distinctly belong to any of the modes. We hypothesize that this problem occurs due to the interaction between two facts: (1) For datasets with large variety, it is likely that th... | workshop-papers | The paper presents a really interesting take on the mode collapse problem and argue that the issue arises because of the current GAN models try to model distributions with disconnected support using continuous noise and generators. The authors try to fix this issue by training multiple generators with shared parameters... | val | [
"HJjiZ-qef",
"H1BVVg9ez",
"HknROGcxG",
"rkoUufImz",
"S1sXdf8mz",
"ryaerzImM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary:\n\nThe paper studies the problem of learning distributions with disconnected support. The paper is very well written, and the analysis is mostly correct, with some important exceptions. However, there are a number of claims that are unverified, and very important baselines are missing. I suggest improving... | [
6,
4,
4,
-1,
-1,
-1
] | [
5,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_HyDMX0l0Z",
"iclr_2018_HyDMX0l0Z",
"iclr_2018_HyDMX0l0Z",
"H1BVVg9ez",
"HJjiZ-qef",
"HknROGcxG"
] |
iclr_2018_rkEtzzWAb | Parametric Adversarial Divergences are Good Task Losses for Generative Modeling | Generative modeling of high dimensional data like images is a notoriously difficult and ill-defined problem. In particular, how to evaluate a learned generative model is unclear.
In this paper, we argue that *adversarial learning*, pioneered with generative adversarial networks (GANs), provides an interesting fra... | workshop-papers | Pros:
- The paper proposes interesting new ideas on evaluating generative models.
- Paper provides hints at interesting links between structural prediction and adversarial learning.
- Authors propose a new dataset called Thin-8 to demonstrate the new ideas and argue that it is useful in general to study generative m... | train | [
"H1EMeWfgz",
"S1owSiOeM",
"SJvLO1WZf",
"ByMGLaMNf",
"BJYVF7TXG",
"SJlaT3nQf",
"H1pl71jQz",
"SJq0z1omM",
"rJeC1JiXM",
"rJRiJyoQf",
"HysGOaZQM",
"ryTCJFdfG",
"HJd9FudGf",
"ByBOYudfM",
"H1QeAUOfz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"public",
"author",
"author",
"author",
"author",
"public",
"author",
"author",
"author",
"public"
] | [
"This paper is in some sense a \"position paper,\" giving a framework for thinking about the loss functions implicitly used by the generator of GAN-type models. It advocates thinking about the loss in a way similar to how it is considered in structured prediction. It also proposes that approximating the dual formul... | [
6,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rkEtzzWAb",
"iclr_2018_rkEtzzWAb",
"iclr_2018_rkEtzzWAb",
"SJlaT3nQf",
"iclr_2018_rkEtzzWAb",
"iclr_2018_rkEtzzWAb",
"SJq0z1omM",
"SJvLO1WZf",
"rJRiJyoQf",
"H1EMeWfgz",
"iclr_2018_rkEtzzWAb",
"iclr_2018_rkEtzzWAb",
"ByBOYudfM",
"S1owSiOeM",
"iclr_2018_rkEtzzWAb"
] |
iclr_2018_r1YUtYx0- | Ensemble Robustness and Generalization of Stochastic Deep Learning Algorithms | The question why deep learning algorithms generalize so well has attracted increasing
research interest. However, most of the well-established approaches,
such as hypothesis capacity, stability or sparseness, have not provided complete
explanations (Zhang et al., 2016; Kawaguchi et al., 2017). In this... | workshop-papers | The paper proposes a new way to understand why neural networks generalize well. They introduce the concept of ensemble robustness and try to explain DNN generalization based on this concept. The reviewers feel the paper is a bit premature for publication in a top conference although this new way of explaining generaliz... | train | [
"rJhcwfLgf",
"SkLRjndlG",
"r1BvYvAeG",
"r1oRSIn7M",
"Hyon4U3Qz",
"Hyb7ir3Xf",
"Hk6PtE2XM",
"ry3UJraMG",
"BJ4tCEaGM",
"rJ1R2E6Mz",
"B1yPYVpMf",
"H1T7gLalM",
"SJx-TCsxz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Summary:\nThis paper presents an adaptation of the algorithmic robustness of Xu&Mannor'12 to a notion robustness of ensemble of hypothesis allowing the authors to study generalization ability of stochastic learning algorithms for Deep Learning Networks. \nGeneralization can be established as long as the sensitiven... | [
4,
4,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1YUtYx0-",
"iclr_2018_r1YUtYx0-",
"iclr_2018_r1YUtYx0-",
"ry3UJraMG",
"Hk6PtE2XM",
"H1T7gLalM",
"SkLRjndlG",
"SkLRjndlG",
"rJhcwfLgf",
"r1BvYvAeG",
"iclr_2018_r1YUtYx0-",
"SJx-TCsxz",
"iclr_2018_r1YUtYx0-"
] |
iclr_2018_SyKoKWbC- | Distributional Adversarial Networks | In most current formulations of adversarial training, the discriminators can be expressed as single-input operators, that is, the mapping they define is separable over observations. In this work, we argue that this property might help explain the infamous mode collapse phenomenon in adversarially-trained generative mod... | workshop-papers | All the reviewers and myself have concerns about the potentially incremental nature of this work. While I do understand that the proposed method goes beyond crafting minibatch losses, and instead parametrizes things via a neural network, ultimately it's roughly very similar to simply combining MMD and minibatch discrim... | train | [
"Hy1iAsugf",
"S1raRaKlG",
"H1u8M0Fgf",
"S13Nk4a7M",
"SJN1xV67z",
"HkbykEpmf",
"rkYvA7TQf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper proposes to replace single-sample discriminators in adversarial training with discriminators that explicitly operate on distributions of examples, so as to incentivize the generator to cover the full distribution of the training data and not collapse to isolated modes. \n\nThe idea of avoiding mode colla... | [
6,
6,
6,
-1,
-1,
-1,
-1
] | [
3,
4,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SyKoKWbC-",
"iclr_2018_SyKoKWbC-",
"iclr_2018_SyKoKWbC-",
"Hy1iAsugf",
"iclr_2018_SyKoKWbC-",
"S1raRaKlG",
"H1u8M0Fgf"
] |
iclr_2018_BkM3ibZRW | Adversarially Regularized Autoencoders | While autoencoders are a key technique in representation learning for continuous structures, such as images or wave forms, developing general-purpose autoencoders for discrete structures, such as text sequence or discretized images, has proven to be more challenging. In particular, discrete inputs make it more difficul... | workshop-papers | In general, the reviewers and myself find this work of some interest, though potentially somewhat incremental in terms of technical novelty compared to the work for Makhzani et al. Another bothersome aspect is the question of evaluation and understanding how well the model actually does; I am not convinced that the int... | val | [
"rkhahIeBz",
"By5yPxlBz",
"HyqcNqaEG",
"Bk0IN5TEz",
"BJntMDTEf",
"S1Q6dxjlz",
"rkzhgMpgM",
"rkevbtAgf",
"B1C3Tr4ZM",
"rkS56HVWG",
"SJoSaBN-G"
] | [
"author",
"official_reviewer",
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Indeed, the discussion around parametrized prior versus the classical prior is very interesting. In this work we only explore this in the universe of autoencoders, i.e., ARAE/AAE. The study of a generic form of parametrized prior may require research on numerous other machine learning framework/schemes, and that i... | [
-1,
-1,
-1,
-1,
5,
6,
3,
9,
-1,
-1,
-1
] | [
-1,
-1,
-1,
-1,
4,
3,
4,
3,
-1,
-1,
-1
] | [
"By5yPxlBz",
"rkevbtAgf",
"Bk0IN5TEz",
"BJntMDTEf",
"iclr_2018_BkM3ibZRW",
"iclr_2018_BkM3ibZRW",
"iclr_2018_BkM3ibZRW",
"iclr_2018_BkM3ibZRW",
"S1Q6dxjlz",
"rkzhgMpgM",
"rkevbtAgf"
] |
iclr_2018_Skk3Jm96W | Some Considerations on Learning to Explore via Meta-Reinforcement Learning | We consider the problem of exploration in meta reinforcement learning. Two new meta reinforcement learning algorithms are suggested: E-MAML and ERL2. Results are presented on a novel environment we call 'Krazy World' and a set of maze environments. We show E-MAML and ERL2 deliver better performance on tasks where expl... | workshop-papers | Overall, the paper is missing a couple of ingredients that would put it over the bar for acceptance:
- I am mystified by statements such as "RL2 no longer gets the best final performance." from one revision to another, as I have lower confidence in the results now.
- More importantly, the paper is missing comparisons... | train | [
"SJ0Q_6Hlf",
"SkE07mveG",
"ryse_yclM",
"H1NWlrYmM",
"BkQIdmTMG",
"S1Sbum6ff",
"BJCjPQpMM",
"rJfePXpMG",
"ByDI8XpMz",
"rJyE8QpzG",
"rkYuBQTfG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"This is an interesting paper about correcting some of the myopic bias in meta RL. For two existing algorithms (MAML, RL2) it proposes a modification of the metaloss that encourages more exploration in the first (couple of) test episodes. The approach is a reasonable one, the proposed methods seem to work, the (toy... | [
7,
6,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Skk3Jm96W",
"iclr_2018_Skk3Jm96W",
"iclr_2018_Skk3Jm96W",
"rJfePXpMG",
"SkE07mveG",
"SkE07mveG",
"SkE07mveG",
"SJ0Q_6Hlf",
"ryse_yclM",
"ryse_yclM",
"iclr_2018_Skk3Jm96W"
] |
iclr_2018_BkA7gfZAb | Stable Distribution Alignment Using the Dual of the Adversarial Distance | Methods that align distributions by minimizing an adversarial distance between them have recently achieved impressive results. However, these approaches are difficult to optimize with gradient descent and they often do not converge well without careful hyperparameter tuning and proper initialization. We investigate whe... | workshop-papers | All the reviewers noted that the dual formulation, as presented, only applies to the logistic family of classifiers. The kernelization is of course something that *can* be done, as argued by the authors, but is not in fact approached in the submission, only in the rebuttal. The toy-ish nature of the problems tackled in... | train | [
"r1u3MhYxG",
"S1eNChYxG",
"S1sYODRlG",
"SyfSHuTmG",
"HyKfsBd7f",
"B10bZyYMM",
"H1iDp4dfG",
"Sy6J64Ozz",
"rJRn3EOMG",
"SkMB24dMM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper deals with “fixing GANs at the computational level”, in a similar sprit to f-GANs and WGANs. The fix is very specific and restricted. It relies on the logistic regression model as the discriminator, and the dual formulation of logistic regression by Jaakkola and Haussler. \n\nComments: \n1) Experiments a... | [
5,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkA7gfZAb",
"iclr_2018_BkA7gfZAb",
"iclr_2018_BkA7gfZAb",
"S1eNChYxG",
"B10bZyYMM",
"Sy6J64Ozz",
"r1u3MhYxG",
"S1eNChYxG",
"S1sYODRlG",
"S1sYODRlG"
] |
iclr_2018_H18WqugAb | Still not systematic after all these years: On the compositional skills of sequence-to-sequence recurrent networks | Humans can understand and produce new utterances effortlessly, thanks to their systematic compositional skills. Once a person learns the meaning of a new verb "dax," he or she can immediately understand the meaning of "dax twice" or "sing and dax." In this paper, we introduce the SCAN domain, consisting of a set of sim... | workshop-papers | Reviewers were somewhat lukewarm about this paper, which seeks to present an analysis of the limitations of sequence models when it comes to understanding compositionality. Somewhat synthetic experiments show that such models generalise poorly on patterns not attested during training, even if the information required t... | test | [
"S1m7bXulz",
"By_6glcgf",
"BkpqEH5gf",
"ByhFlFIZM",
"BkQM5uI-f",
"rJVyquUZG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper focuses on the zero-shot learning compositional capabilities of modern sequence-to-sequence RNNs. Through a series of experiments and a newly defined dataset, it exposes the short-comings of current seq2seq RNN architectures. The proposed dataset, called the SCAN dataset, is a selected subset of the C... | [
6,
7,
6,
-1,
-1,
-1
] | [
3,
4,
5,
-1,
-1,
-1
] | [
"iclr_2018_H18WqugAb",
"iclr_2018_H18WqugAb",
"iclr_2018_H18WqugAb",
"S1m7bXulz",
"By_6glcgf",
"BkpqEH5gf"
] |
iclr_2018_HJ3d2Ax0- | Benefits of Depth for Long-Term Memory of Recurrent Networks | The key attribute that drives the unprecedented success of modern Recurrent Neural Networks (RNNs) on learning tasks which involve sequential data, is their ever-improving ability to model intricate long-term temporal dependencies. However, a well established measure of RNNs' long-term memory capacity is lacking, and t... | workshop-papers | This paper attempts a theoretical treatment of the influence of depth in RNNs on their ability to capture dependencies in the data. All reviewers found the theoretical contribution of the paper interesting, and while there were problems raised regarding formalisation, they appear to have been adequately addressed in th... | train | [
"B1L-7zDgz",
"SJl6_ov-G",
"BJpAdoezz",
"Hkb2oX6XG",
"BJCcFQTXG",
"S1XOyeaQf",
"rJvf61TXM",
"BJQjt8d-G",
"BysD9IxbG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"This paper investigates an effect of time dependencies in a specific type of RNN.\n\nThe idea is important and this paper seems sound. However, I am not sure that the main result (Theorem 1) explains an effect of depth sufficiently.\n\n--Main comment\nAbout the deep network case in Theorem 1, how $L$ affects the b... | [
5,
7,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
3,
3,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HJ3d2Ax0-",
"iclr_2018_HJ3d2Ax0-",
"iclr_2018_HJ3d2Ax0-",
"iclr_2018_HJ3d2Ax0-",
"B1L-7zDgz",
"BJpAdoezz",
"SJl6_ov-G",
"SJl6_ov-G",
"B1L-7zDgz"
] |
iclr_2018_ryG6xZ-RZ | DLVM: A modern compiler infrastructure for deep learning systems | Deep learning software demands reliability and performance. However, many of the existing deep learning frameworks are software libraries that act as an unsafe DSL in Python and a computation graph interpreter. We present DLVM, a design and implementation of a compiler infrastructure with a linear algebra intermediate ... | workshop-papers | This is a fascinating paper, and representative of the sort of work which is welcome in our field and in our community. It presents a compiler framework for the development of DSLs (and models) for Deep Learning and related methods. Overall, reviewers were supportive of and excited by this line of work, but questioned ... | train | [
"SyW2zW3VM",
"Syj85_q4M",
"H1YMhD9Vf",
"BJYvCt8Ef",
"ByG1QoMxz",
"rJs8O8Bgz",
"SJ_WummXM"
] | [
"official_reviewer",
"author",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"I appreciate the author's response (I'm a little confused why everything is promised in a future update, e.g. clarifications and improvements in the paper, is there no opportunity to update the pdf on openreview? This would make it easier to appreciate these changes).\n\nHowever, I still feel the lack of compariso... | [
-1,
-1,
-1,
-1,
5,
7,
5
] | [
-1,
-1,
-1,
-1,
4,
4,
3
] | [
"SJ_WummXM",
"H1YMhD9Vf",
"BJYvCt8Ef",
"iclr_2018_ryG6xZ-RZ",
"iclr_2018_ryG6xZ-RZ",
"iclr_2018_ryG6xZ-RZ",
"iclr_2018_ryG6xZ-RZ"
] |
iclr_2018_HkGJUXb0- | Learning Efficient Tensor Representations with Ring Structure Networks | \emph{Tensor train (TT) decomposition} is a powerful representation for high-order tensors, which has been successfully applied to various machine learning tasks in recent years. In this paper, we propose a more generalized tensor decomposition with ring structure network by employing circular multilinear products ov... | workshop-papers | This paper proposes a new way of learning tensors representation with ring-structured decompositions rather than through Tensor Train methods. The paper investigates the mathematical properties of this decomposition and provides synthetic experiments. There was some debate, with the reviewers, about the novelty and imp... | train | [
"BkQlqDiBG",
"BJSfkUNzf",
"SkhraDqVM",
"ry_11ijeM",
"SJvcooagM",
"r1hXHTCzG",
"BJSzKFRGz",
"BJE1YYAGz"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"We appreciate the reviewer’s positive feedbacks on our revised paper. \n\nFor experiments on one image, the compression rate achieved by our method is almost 2 times over TT method. For tensorizing neural networks, although the classification performance of our method is only slightly better than TT, given the ... | [
-1,
5,
-1,
5,
6,
-1,
-1,
-1
] | [
-1,
4,
-1,
4,
3,
-1,
-1,
-1
] | [
"SkhraDqVM",
"iclr_2018_HkGJUXb0-",
"r1hXHTCzG",
"iclr_2018_HkGJUXb0-",
"iclr_2018_HkGJUXb0-",
"BJSfkUNzf",
"ry_11ijeM",
"SJvcooagM"
] |
iclr_2018_BJypUGZ0Z | Accelerating Neural Architecture Search using Performance Prediction | Methods for neural network hyperparameter optimization and meta-modeling are computationally expensive due to the need to train a large number of model configurations. In this paper, we show that standard frequentist regression models can predict the final performance of partially trained model configurations using fea... | workshop-papers | The paper proposes to use simple regression models for predicting the accuracy of a neural network based on its initial training curve, architecture, and hyper-parameters; this can be used for speeding up architecture search. While this is an interesting direction and the presented experiments look quite encouraging, t... | test | [
"SJwuTOPNz",
"ryMvgqdgM",
"B12MLEclM",
"SJzxff6xM",
"HJeLUNhmz",
"rJEPHE3mG",
"BkaVm42Xz",
"Sy4RWE37f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Thanks for the careful response; it helped clarify several of my questions and also fixed the previously inflated speedup result of 7x at the end of the paper. Extending Figure 6 was also helpful.\n\nIt is nice to see the new results for BLR. I am confused, though: which features does this use? Why are the results... | [
-1,
6,
6,
4,
-1,
-1,
-1,
-1
] | [
-1,
4,
5,
3,
-1,
-1,
-1,
-1
] | [
"BkaVm42Xz",
"iclr_2018_BJypUGZ0Z",
"iclr_2018_BJypUGZ0Z",
"iclr_2018_BJypUGZ0Z",
"iclr_2018_BJypUGZ0Z",
"ryMvgqdgM",
"B12MLEclM",
"SJzxff6xM"
] |
iclr_2018_SkOb1Fl0Z | A Flexible Approach to Automated RNN Architecture Generation | The process of designing neural architectures requires expert knowledge and extensive trial and error.
While automated architecture search may simplify these requirements, the recurrent neural network (RNN) architectures generated by existing methods are limited in both flexibility and components.
We propos... | workshop-papers | The paper presents a domain-specific language for RNN architecture search, which can be used in combination with learned ranking function or RL-based search. While the approach is interesting and novel, the paper would benefit from an improved evaluation, as pointed out by reviewers. For example, the paper currently ev... | test | [
"SkQLSjmSG",
"H11b-JnVf",
"rkLDKb5Nf",
"SJgAX9tVf",
"r1wiKz5ef",
"SJ4ObrLEG",
"S1exhDQJf",
"BkuT3b9ef",
"rJOkNuzVf",
"Ska-dZ6mG",
"SkOgOWpmz",
"HyOUwZTQM"
] | [
"official_reviewer",
"author",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Agree that is not straight forward, but would be in my view important so that the community understands what do we gain with a new method. One could for example run the different methods (for different initial conditions) and quantify how often they end up in better solutions (or 'radically' new solutions).",
"D... | [
-1,
-1,
-1,
-1,
6,
-1,
4,
5,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
-1,
4,
-1,
4,
4,
-1,
-1,
-1,
-1
] | [
"H11b-JnVf",
"rkLDKb5Nf",
"SJgAX9tVf",
"SJ4ObrLEG",
"iclr_2018_SkOb1Fl0Z",
"HyOUwZTQM",
"iclr_2018_SkOb1Fl0Z",
"iclr_2018_SkOb1Fl0Z",
"Ska-dZ6mG",
"S1exhDQJf",
"BkuT3b9ef",
"r1wiKz5ef"
] |
iclr_2018_SySaJ0xCZ | Simple and efficient architecture search for Convolutional Neural Networks | Neural networks have recently had a lot of success for many tasks. However, neural
network architectures that perform well are still typically designed manually
by experts in a cumbersome trial-and-error process. We propose a new method
to automatically search for well-performing CNN architectures bas... | workshop-papers | The paper proposes a method for architecture search using network morphisms, which allows for faster search without retraining candidate models. The results on CIFAR are worse than the state of the art, but reasonably competitive, and achieved using limited computation resources. It would have been interesting to see h... | val | [
"BkGYrcIlG",
"HJ8nXbKeM",
"Hk4ciAteG",
"ByYaDP9XG",
"ByYtFtLQf",
"SJJLYtL7f",
"HkFZKt8mM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper proposes a neural architecture search method that achieves close to state-of-the-art accuracy on CIFAR10 and takes much less computational resources. The high-level idea is similar to the evolution method of [Real et al. 2017], but the mutation preserves net2net properties, which means the mutated netwo... | [
6,
5,
4,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SySaJ0xCZ",
"iclr_2018_SySaJ0xCZ",
"iclr_2018_SySaJ0xCZ",
"iclr_2018_SySaJ0xCZ",
"BkGYrcIlG",
"HJ8nXbKeM",
"Hk4ciAteG"
] |
iclr_2018_BkCV_W-AZ | Regret Minimization for Partially Observable Deep Reinforcement Learning | Deep reinforcement learning algorithms that estimate state and state-action value functions have been shown to be effective in a variety of challenging domains, including learning control strategies from raw image pixels. However, algorithms that estimate state and state-action value functions typically assume a fully ... | workshop-papers | The reviewers agree this is a really interesting paper, with an interesting idea (in particular
the use of regret clipping might provide a benefit over typical policy gradient methods). However,
there are two major concerns: 1) clarity / exposition and more importantly 2) lack of a strong
empirical motivation for the n... | train | [
"ByyLE_tgG",
"S17ehb1WM",
"SkYyvPyWf",
"SkNPS_pXz",
"rJIfV_TXz",
"H1a1BOTQM",
"r1fM8U3Xz",
"Hk29NGimM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper presents Advantage-based Regret Minimization, somewhat similar to advantage actor-critic with REINFORCE.\nThe main focus of the paper seems to be the motivation/justification of this algorithm with connection to the regret minimization literature (and without Markov assumptions).\nThe claim that ARM is ... | [
4,
7,
5,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkCV_W-AZ",
"iclr_2018_BkCV_W-AZ",
"iclr_2018_BkCV_W-AZ",
"S17ehb1WM",
"SkYyvPyWf",
"ByyLE_tgG",
"Hk29NGimM",
"iclr_2018_BkCV_W-AZ"
] |
iclr_2018_rkEfPeZRb | Variance-based Gradient Compression for Efficient Distributed Deep Learning | Due to the substantial computational cost, training state-of-the-art deep neural networks for large-scale datasets often requires distributed training using multiple computation workers. However, by nature, workers need to frequently communicate gradients, causing severe bottlenecks, especially on lower bandwidth conne... | workshop-papers | The reviewers find the gradient compression approach novel and interesting, but they find the empirical evaluation not fully satisfactory. Some aspects of the paper have improved with the feedback from the reviewers, but because of the domain of the paper, experimental evaluation is very important. I recommend improvin... | train | [
"B1O_32YeM",
"rkZd9y9xz",
"ByqfOWqlM",
"SkGeDz6Xf",
"B1aOTEKXG",
"Byiwed0fG",
"B1rp1uAGz",
"H1mSy_RMM",
"Hk1W1dRzf",
"BJkmTPAzM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper proposes a variance-based gradient compression method to reduce the communication overhead of distributed deep learning. Experiments on real datasets are used for evaluation. \n\nThe idea to adopt approximated variances of gradients to reduce communication cost seems to be interesting. However, there al... | [
6,
4,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rkEfPeZRb",
"iclr_2018_rkEfPeZRb",
"iclr_2018_rkEfPeZRb",
"B1aOTEKXG",
"H1mSy_RMM",
"iclr_2018_rkEfPeZRb",
"B1O_32YeM",
"Hk1W1dRzf",
"rkZd9y9xz",
"ByqfOWqlM"
] |
iclr_2018_SyVOjfbRb | LSH-SAMPLING BREAKS THE COMPUTATIONAL CHICKEN-AND-EGG LOOP IN ADAPTIVE STOCHASTIC GRADIENT ESTIMATION | Stochastic Gradient Descent or SGD is the most popular optimization algorithm for large-scale problems. SGD estimates the gradient by uniform sampling with sample size one. There have been several other works that suggest faster epoch wise convergence by using weighted non-uniform sampling for better gradient estimates... | workshop-papers | The reviewers think that the theoretical contribution is not significant on its own. The reviewers find the empirical aspect of the paper interesting, but more analysis of the empirical behavior is required, especially for large datasets. Even for small datasets with input augmentation (e.g. random crops in CIFAR-10) t... | train | [
"SJpaRgDNf",
"HyGpJF44z",
"HkmgURdlf",
"rkcg14qlz",
"SJl0YdfWM",
"ryd-Vb4Vf",
"SJBH-bE4z",
"SyEJNkVVf",
"BJBcbfAXG",
"HyQqaeRmM",
"HJZNvP2fG",
"S1Kj_sofM",
"ByTTs9ifM"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"Thanks for the discussions! \n\nWe will restress the subtleties and differenced of indexing, sub-linear similarity search, and the new line of sub-linear adaptive sampling and unbiased estimation in any future versions of the paper. \n\nLet us know if you think anything else will be helpful. ",
"Response to the... | [
-1,
-1,
8,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"HyGpJF44z",
"ryd-Vb4Vf",
"iclr_2018_SyVOjfbRb",
"iclr_2018_SyVOjfbRb",
"iclr_2018_SyVOjfbRb",
"SJBH-bE4z",
"SyEJNkVVf",
"HyQqaeRmM",
"HyQqaeRmM",
"ByTTs9ifM",
"HkmgURdlf",
"SJl0YdfWM",
"rkcg14qlz"
] |
iclr_2018_BJjquybCW | The loss surface and expressivity of deep convolutional neural networks | We analyze the expressiveness and loss surface of practical deep convolutional
neural networks (CNNs) with shared weights and max pooling layers. We show
that such CNNs produce linearly independent features at a “wide” layer which
has more neurons than the number of training samples. This condition ho... | workshop-papers | Dear authors,
While I appreciate the result that a convolutional layer can have full rank output, this allowing a dataset to be classified perfectly under mild conditions, the fact that all reviewers expressed concern about the statement is an indication that the presentation sill needs quite a bit of work.
Thus, I r... | train | [
"BkIW6fYxz",
"rkvS6-9gG",
"S136E0hZf",
"HJ0nIZkfM",
"rJoC5PTQM",
"ByYsQbYGz",
"SyA2UxD-z",
"SJsVLlv-f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper presents several theoretical results on the loss functions of CNNs and fully-connected neural networks. I summarize the results as follows:\n\n(1) Under certain assumptions, if the network contains a \"wide“ hidden layer, such that the layer width is larger than the number of training examples, then (wi... | [
4,
7,
5,
6,
-1,
-1,
-1,
-1
] | [
4,
2,
2,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJjquybCW",
"iclr_2018_BJjquybCW",
"iclr_2018_BJjquybCW",
"iclr_2018_BJjquybCW",
"HJ0nIZkfM",
"S136E0hZf",
"rkvS6-9gG",
"BkIW6fYxz"
] |
iclr_2018_SyfiiMZA- | Jointly Learning to Construct and Control Agents using Deep Reinforcement Learning | The physical design of a robot and the policy that controls its motion are inherently coupled. However, existing approaches largely ignore this coupling, instead choosing to alternate between separate design and control phases, which requires expert intuition throughout and risks convergence to suboptimal designs. In t... | workshop-papers | The chief contribution of this paper is to show that a single set of policy parameters can be optimized in an alternating fashion while the design parameters of the body are also optimized with policy gradients and sampled. The fact that this simple approach seems to work is interesting and worthy of note. However, the... | train | [
"SksyD3Dgz",
"ByrfSMcgz",
"rJweW2Sbf",
"Sydj1V_Mz",
"Hk8rkN_Mz",
"HkyWy4ufz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This is a well written paper, very nice work.\nIt makes progress on the problem of co-optimization of the physical parameters of a design\nand its control system. While it is not the first to explore this kind of direction,\nthe method is efficient for what it does; it shows that at least for some systems, \nthe ... | [
9,
4,
5,
-1,
-1,
-1
] | [
5,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_SyfiiMZA-",
"iclr_2018_SyfiiMZA-",
"iclr_2018_SyfiiMZA-",
"rJweW2Sbf",
"SksyD3Dgz",
"ByrfSMcgz"
] |
iclr_2018_SJZ2Mf-0- | Adaptive Memory Networks | Real-world Question Answering (QA) tasks consist of thousands of words that often represent many facts and entities. Existing models based on LSTMs require a large number of parameters to support external memory and do not generalize well for long sequence inputs. Memory networks attempt to address these limitations by... | workshop-papers | This paper presents an interesting model which at the time of submission was still quite confusingly described to the reviewers.
A lot of improvements have been made for which I applaud the authors.
However, at this point, the original 20 babi tasks are not quite that exciting and several other models are able to fully... | train | [
"SJIPkC2Nf",
"ByQPccw4G",
"HJ_BtypgM",
"rJBY85UEM",
"rJGuMGYef",
"By5ZMrqxG",
"S1Lg0TTfM",
"B1amYyGXG",
"S18rEFImM",
"SJzUv1GXG",
"Skkp0DGgM",
"SkYhYmGlz"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Thanks for taking time to revisit the paper. We are working on additional experiments and will add these results in the coming revision.",
"Dear reviewer, \n\nThe revised paper is available. We can see 03 Nov 2017 (modified: 05 Jan 2018) as the latest revision. Furthermore, if you click on revisions, you can see... | [
-1,
-1,
5,
-1,
7,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
-1,
5,
3,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"rJBY85UEM",
"HJ_BtypgM",
"iclr_2018_SJZ2Mf-0-",
"SJzUv1GXG",
"iclr_2018_SJZ2Mf-0-",
"iclr_2018_SJZ2Mf-0-",
"By5ZMrqxG",
"HJ_BtypgM",
"iclr_2018_SJZ2Mf-0-",
"rJGuMGYef",
"SkYhYmGlz",
"iclr_2018_SJZ2Mf-0-"
] |
iclr_2018_SJyfrl-0b | Fast Node Embeddings: Learning Ego-Centric Representations | Representation learning is one of the foundations of Deep Learning and allowed important improvements on several Machine Learning tasks, such as Neural Machine Translation, Question Answering and Speech Recognition. Recent works have proposed new methods for learning representations for nodes and edges in graphs. Sever... | workshop-papers | The authors addressed the reviewers concerns but the scores remain somewhat low.
The method is not super novel, but it is an incremental improvement over existing approaches. | train | [
"HkuS2uuef",
"ByVfm9uxz",
"Sy5ZqP9gM",
"B1qFOGOzf",
"HyWyDMdzG",
"Sk2-OTwGM",
"B1_ipj-GG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper demonstrates good experiment results on several tasks. There are some pros and comes as below:\n\nPros\nThe proposed model provides a new view to generate training examples for random-walk-based embedding models.\nExperiments are conducted on several datasets (6 datasets for link prediction and 3 datset... | [
5,
6,
4,
-1,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJyfrl-0b",
"iclr_2018_SJyfrl-0b",
"iclr_2018_SJyfrl-0b",
"B1_ipj-GG",
"HkuS2uuef",
"ByVfm9uxz",
"Sy5ZqP9gM"
] |
iclr_2018_S1LXVnxRb | Cross-Corpus Training with TreeLSTM for the Extraction of Biomedical Relationships from Text | A bottleneck problem in machine learning-based relationship extraction (RE) algorithms, and particularly of deep learning-based ones, is the availability of training data in the form of annotated corpora. For specific domains, such as biomedicine, the long time and high expertise required for the development of manuall... | workshop-papers | We encourage the authors to improve the mentioned aspects of their work in the reviews.
| train | [
"HydZAdHgG",
"rJFaZDqgz",
"Sk6i-Szbz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"SUMMARY.\n\nThe paper presents a cross-corpus approach for relation extraction from text.\nThe main idea is complementing small training data for relation extraction with training data with different relation types.\nThe model is also connected with multitask learning approaches where the encoder for the input is ... | [
4,
5,
3
] | [
4,
4,
5
] | [
"iclr_2018_S1LXVnxRb",
"iclr_2018_S1LXVnxRb",
"iclr_2018_S1LXVnxRb"
] |
iclr_2018_rkaT3zWCZ | Building Generalizable Agents with a Realistic and Rich 3D Environment | Teaching an agent to navigate in an unseen 3D environment is a challenging task, even in the event of simulated environments. To generalize to unseen environments, an agent needs to be robust to low-level variations (e.g. color, texture, object changes), and also high-level variations (e.g. layout changes of the enviro... | workshop-papers | The authors present an environment for semantic navigation that is based on an existing dataset, SUNCG. Datasets/environments are important for deep RL research, and the contribution of this paper is welcome. However, this paper does not offer enough novelty in terms of approach/method and its claims are somewhat misle... | val | [
"BySAhfGgM",
"ByK1GkteM",
"BJlAby9gM",
"ByREAIo7z",
"SywxCIj7f",
"SyJ9TUiXM",
"r1jXpIi7z",
"SJKZ7DPxM",
"ryfUh0Xlf",
"HJWt2VcAZ",
"HJAzR1nR-",
"Hk38Cy2Cb",
"Hkv6GtMRW"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public",
"public",
"author",
"author",
"public"
] | [
"Paper Summary: The paper proposes a simulator for the SUNCG dataset to perform rendering and collision detection. The paper also extends A3C and DDPG (reinforcement learning methods) by augmenting them with gated attention. These methods are applied for the task of navigation.\n\nPaper Strengths:\n- It is interest... | [
4,
5,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rkaT3zWCZ",
"iclr_2018_rkaT3zWCZ",
"iclr_2018_rkaT3zWCZ",
"BySAhfGgM",
"ByK1GkteM",
"BJlAby9gM",
"iclr_2018_rkaT3zWCZ",
"iclr_2018_rkaT3zWCZ",
"HJWt2VcAZ",
"iclr_2018_rkaT3zWCZ",
"HJWt2VcAZ",
"Hkv6GtMRW",
"iclr_2018_rkaT3zWCZ"
] |
iclr_2018_HyXNCZbCZ | Hierarchical Adversarially Learned Inference | We propose a novel hierarchical generative model with a simple Markovian structure and a corresponding inference model. Both the generative and inference model are trained using the adversarial learning paradigm. We demonstrate that the hierarchical structure supports the learning of progressively more abstract represe... | rejected-papers | Pros:
- The paper proposes to use a hierarchical structure to address reconstruction issues with ALI model.
- Obtaining multiple latent representations that individually achieve a different level of reconstructions is interesting.
- Paper is well written and the authors made a reasonable attempt to improve the paper ... | train | [
"B135Vq1rz",
"Sy-zrYT4f",
"BkAnkYaNz",
"rk7QytpNz",
"SkRgK-YxM",
"r1KKXM6Nf",
"SJjWbz6VM",
"r1s1wghEf",
"HJMQW6P4M",
"SJbtlZ5gf",
"SkI8jt8EG",
"SJbY1_5xG",
"HJ8Hbw6Xf",
"rkOvxDpQG",
"rJixxPpQG",
"B1tuJwTQG"
] | [
"author",
"official_reviewer",
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"We thank the reviewer for his answer. \n\nWhile we agree that HALI's objective is effectively unchanged from ALI, we feel that HALI's novelty lies in illustrating how the hierarchy can be leveraged to:\n\n* Improve reconstructions in adversarially trained generative models.\n* Learn a hierarchy of latent represent... | [
-1,
-1,
-1,
-1,
5,
-1,
-1,
-1,
-1,
5,
-1,
7,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
-1,
5,
-1,
-1,
-1,
-1,
5,
-1,
3,
-1,
-1,
-1,
-1
] | [
"Sy-zrYT4f",
"r1s1wghEf",
"r1KKXM6Nf",
"SJjWbz6VM",
"iclr_2018_HyXNCZbCZ",
"HJMQW6P4M",
"r1s1wghEf",
"SJbtlZ5gf",
"SkI8jt8EG",
"iclr_2018_HyXNCZbCZ",
"HJ8Hbw6Xf",
"iclr_2018_HyXNCZbCZ",
"SkRgK-YxM",
"SJbtlZ5gf",
"SJbY1_5xG",
"iclr_2018_HyXNCZbCZ"
] |
iclr_2018_B1CNpYg0- | Learning to Compute Word Embeddings On the Fly | Words in natural language follow a Zipfian distribution whereby some words are frequent but most are rare. Learning representations for words in the ``long tail'' of this distribution requires enormous amounts of data.
Representations of rare words trained directly on end tasks are usually poor, requiring us to ... | rejected-papers | The pros and cons of the paper can be summarized as follows:
Pros:
* The method of combining together multiple information sources is effective
* Experimental evaluation is thorough
Cons:
* The method is a relatively minor contribution, combining together multiple existing methods to improve word embeddings. This als... | train | [
"SJTAcW5xf",
"ryDrjZqxM",
"SJBZut5gM",
"H1MCKqz7f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author"
] | [
"This paper describes a method for computing representations for out-of-vocabulary words, e.g. based on their spelling or dictionary definitions. The main difference from previous approaches is that the model is that the embeddings are trained end-to-end for a specific task, rather than trying to produce genericall... | [
5,
7,
5,
-1
] | [
4,
3,
4,
-1
] | [
"iclr_2018_B1CNpYg0-",
"iclr_2018_B1CNpYg0-",
"iclr_2018_B1CNpYg0-",
"iclr_2018_B1CNpYg0-"
] |
iclr_2018_SJvu-GW0b | Graph2Seq: Scalable Learning Dynamics for Graphs | Neural networks are increasingly used as a general purpose approach to learning algorithms over graph structured data. However, techniques for representing graphs as real-valued vectors are still in their infancy. Recent works have proposed several approaches (e.g., graph convolutional networks), but as we show in this... | rejected-papers | The reviewers agree that the problem being studied is important and relevant but express serious concerns. I recommend the authors to carefully go through the reviews and significantly scale up their experiments. | test | [
"H1n6BG_HG",
"SyhZ-YerG",
"HyNr86Ylz",
"Hy9gZ2CxM",
"SkD9M_NZf",
"HklBwVKmG",
"SydlD4FQz",
"SJ67IVFmG",
"SJl9hMKXG"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"(1) Reg. length of sequence: \n\nSequence length depends on the graph, and in the worst case is exponential in # of edges. In any case, this is irrelevant to our (empirical) results. In fact, the theoretical fact of exponential dependence in # of edges only makes our empirical results more impressive; we only need... | [
-1,
-1,
4,
4,
4,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
3,
-1,
-1,
-1,
-1
] | [
"SyhZ-YerG",
"SydlD4FQz",
"iclr_2018_SJvu-GW0b",
"iclr_2018_SJvu-GW0b",
"iclr_2018_SJvu-GW0b",
"HyNr86Ylz",
"Hy9gZ2CxM",
"SkD9M_NZf",
"iclr_2018_SJvu-GW0b"
] |
iclr_2018_rJv4XWZA- | Generating Differentially Private Datasets Using GANs | In this paper, we present a technique for generating artificial datasets that retain statistical properties of the real data while providing differential privacy guarantees with respect to this data. We include a Gaussian noise layer in the discriminator of a generative adversarial network to make the output and the gr... | rejected-papers | This paper presents an interesting idea: employ GANs in a manner that guarantees the generation of differentially private data.
The reviewers liked the motivation but identified various issues. Also, the authors themselves discovered some problems in their formulation; on behalf of the community, thanks for letting th... | test | [
"B1p11ROxz",
"H1Ae8Z5eM",
"ByaqVKoxz",
"BJgtYeuXM",
"BktqSZGXf",
"SyaRpcAlG",
"SJ15NnjeG",
"rkZnVH9yz",
"HJBcVfDkM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"author",
"public",
"author",
"public"
] | [
"Summary: The paper addresses the problem of non-interactive differentially private mechanism via adversarial networks. Non-interactive mechanisms have been one of the most sought-after approaches in differentially private algorithm design. The reason is that once a differentially private data set is released, it c... | [
6,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rJv4XWZA-",
"iclr_2018_rJv4XWZA-",
"iclr_2018_rJv4XWZA-",
"iclr_2018_rJv4XWZA-",
"iclr_2018_rJv4XWZA-",
"SJ15NnjeG",
"iclr_2018_rJv4XWZA-",
"HJBcVfDkM",
"iclr_2018_rJv4XWZA-"
] |
iclr_2018_Sk7cHb-C- | Representing dynamically: An active process for describing sequential data | We propose an unsupervised method for building dynamic representations of sequential data, particularly of observed interactions. The method simultaneously acquires representations of input data and its dynamics. It is based on a hierarchical generative model composed of two levels. In the first level, a model learns r... | rejected-papers | This paper proposes a model which learns simultaneously the dynamics of sequential data, together with a static latent representation. The idea and motivation is interesting and the results are promising.
However, all reviewers agree that the presentation needs much more work to convey the messages correctly and convi... | train | [
"r1OraDNgf",
"By8SeCYez",
"BypLdzcxf",
"S1pvRc7bM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The authors propose an architecture and generative model for static images and video sequences, with the purpose of generating an image that looks as similar as possible to the one that is supplied. This is useful for for example frame prediction in video and detection of changes in video as a consequence to chang... | [
6,
3,
4,
4
] | [
3,
3,
4,
4
] | [
"iclr_2018_Sk7cHb-C-",
"iclr_2018_Sk7cHb-C-",
"iclr_2018_Sk7cHb-C-",
"iclr_2018_Sk7cHb-C-"
] |
iclr_2018_BkS3fnl0W | Semi-supervised Outlier Detection using Generative And Adversary Framework | In a conventional binary/multi-class classification task, the decision boundary is supported by data from two or more classes. However, in one-class classification task, only data from one class are available. To build an robust outlier detector using only data from a positive class, we propose a corrupted GAN(CorGAN),... | rejected-papers | This paper presents a framework where GANs are used to improve detection of outliers (in this context, instances of the “background class”). This is a very interesting and, as demonstrated, promising idea. However, the general feeling of the reviewers is that more work is needed to make the technical and evaluations pa... | train | [
"HJDhuQtlM",
"Sks502_lG",
"Hy2FsQKef",
"B1ufhqoQz",
"By7-yAXGG",
"HJSWKmQGf",
"HJT5g4XzM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The idea of using GANs for outlier detection is interesting and the problem is relevant. However, I have the following concerns about the quality and the significance:\n- The proposed formulation in Equation (2) is questionable. The authors say that this is used to generate outliers, and since it will generate inl... | [
4,
4,
3,
-1,
-1,
-1,
-1
] | [
3,
4,
5,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkS3fnl0W",
"iclr_2018_BkS3fnl0W",
"iclr_2018_BkS3fnl0W",
"iclr_2018_BkS3fnl0W",
"Sks502_lG",
"Hy2FsQKef",
"HJDhuQtlM"
] |
iclr_2018_HkinqfbAb | Automatic Parameter Tying in Neural Networks | Recently, there has been growing interest in methods that perform neural network compression, namely techniques that attempt to substantially reduce the size of a neural network without significant reduction in performance. However, most existing methods are post-processing approaches in that they take a learned neural... | rejected-papers | This paper presents yet another scheme for weight tying for compressing neural networks, which looks a lot like a less Bayesian version of recent related work, and gets good empirical results on realistic problems.
This paper is well-executed and is a good contribution, but falls below the bar on
1) Discovering someth... | train | [
"S13TkvpQz",
"S14u6LpQG",
"H1J8TLamG",
"HyrzaIp7G",
"Bky9cL_eG",
"Byk0Q2_xz",
"Hy-t_ztgG"
] | [
"author",
"author",
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"We have updated the draft to address the reviewers concerns (plus rearranging to still keep it within the 8 page limit). Most notably, we have added additional experiments on VGG-16 and improved the clarity of the presentation by adding additional details in both the algorithm description and experimental section... | [
-1,
-1,
-1,
-1,
6,
6,
6
] | [
-1,
-1,
-1,
-1,
5,
4,
4
] | [
"iclr_2018_HkinqfbAb",
"Bky9cL_eG",
"Byk0Q2_xz",
"Hy-t_ztgG",
"iclr_2018_HkinqfbAb",
"iclr_2018_HkinqfbAb",
"iclr_2018_HkinqfbAb"
] |
iclr_2018_H15RufWAW | GraphGAN: Generating Graphs via Random Walks | We propose GraphGAN - the first implicit generative model for graphs that enables to mimic real-world networks.
We pose the problem of graph generation as learning the distribution of biased random walks over a single input graph.
Our model is based on a stochastic neural network that generates discrete out... | rejected-papers | This paper proposes an implicit model of graphs, trained adversarially using the Gumbel-softmax trick. The main idea of feeding random walks to the discriminator is interesting and novel. However,
1) The task of generating 'sibling graphs', for some sort of bootstrap analysis, isn't well-motivated.
2) The method is c... | val | [
"BkCkJetef",
"SJhXxLYgz",
"rkKo_YeWG",
"HynfzR27M",
"SktZrEd-z",
"SkKn4NO-G",
"H1oI4EuZf",
"HyECmNOZz",
"B1CHXVu-M",
"rkG7qbQZz",
"SkRp15ZWM",
"rJa0pv5ef",
"B1M0aoceM",
"B1xzITYxf",
"SknlQLFef",
"B1wkj2BgM",
"SyDRyMBlf",
"SycFao-kf",
"Bk-RVrRAW"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"author",
"public",
"author",
"public",
"author",
"public"
] | [
"I am overall positive about the work but I would like to see some questions addressed. \n\nQuality: The paper is good but does not address some important issues. The paper proposes a GAN model to generate graphs with non-trivial properties. This is possibly one of the best papers on graph generation using GANs cur... | [
6,
7,
4,
-1,
-1,
-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,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H15RufWAW",
"iclr_2018_H15RufWAW",
"iclr_2018_H15RufWAW",
"iclr_2018_H15RufWAW",
"SJhXxLYgz",
"H1oI4EuZf",
"BkCkJetef",
"B1CHXVu-M",
"rkKo_YeWG",
"SkRp15ZWM",
"iclr_2018_H15RufWAW",
"B1xzITYxf",
"rJa0pv5ef",
"SknlQLFef",
"iclr_2018_H15RufWAW",
"SyDRyMBlf",
"iclr_2018_H15Ru... |
iclr_2018_rJiaRbk0- | Towards Binary-Valued Gates for Robust LSTM Training | Long Short-Term Memory (LSTM) is one of the most widely used recurrent structures in sequence modeling. Its goal is to use gates to control the information flow (e.g., whether to skip some information/transformation or not) in the recurrent computations, although its practical implementation based on soft gates only pa... | rejected-papers | This paper proposes training binary-values LSTMs for NLP using the Gumbel-softmax reparameterization. The motivation is that this will generalize better, and this is demonstrated in a couple of instances.
However, it's not clear how cherry-picked the examples are, since the training loss wasn't reported for most expe... | train | [
"HyA3jBqgG",
"S15OPlugz",
"Syo-smqgf",
"rJBQl4j7z",
"BJj7JzW7G",
"BJvcAWbQG",
"rJ4UxzbQz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper propose a new \"gate\" function for LSTM to enable the values of the gates towards 0 or 1. The motivation behind is a flat region of the loss surface is likely to generalize well. It shows the experimental results are comparable or better than vanilla LSTM and much more robust to low-precision approxim... | [
6,
4,
6,
-1,
-1,
-1,
-1
] | [
3,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rJiaRbk0-",
"iclr_2018_rJiaRbk0-",
"iclr_2018_rJiaRbk0-",
"iclr_2018_rJiaRbk0-",
"Syo-smqgf",
"HyA3jBqgG",
"S15OPlugz"
] |
iclr_2018_r1h2DllAW | Discrete-Valued Neural Networks Using Variational Inference | The increasing demand for neural networks (NNs) being employed on embedded devices has led to plenty of research investigating methods for training low precision NNs. While most methods involve a quantization step, we propose a principled Bayesian approach where we first infer a distribution over a discrete weight spac... | rejected-papers | This paper presents a somewhat new approach to training neural nets with ternary or low-precision weights. However the Bayesian motivation doesn't translate into an elegant and self-tuning method, and ends up seeming kind of complicated and ad-hoc. The results also seem somewhat toy. The paper is fairly clearly writ... | test | [
"Sy_eXHHfz",
"BkZJDo8-f",
"ry637oIWz",
"SJJ5dvJgM",
"HkshYX9xz",
"H1S_cEcxM"
] | [
"author",
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"We added a revision of our paper where we changed the following aspects.\n\n(1) We left the structure of the paper largely unchanged. We removed some details about our method from the introduction but we kept the related work there as it is needed to motivate our work and to highlight the gap in the literature we ... | [
-1,
-1,
-1,
6,
5,
5
] | [
-1,
-1,
-1,
1,
4,
4
] | [
"iclr_2018_r1h2DllAW",
"HkshYX9xz",
"H1S_cEcxM",
"iclr_2018_r1h2DllAW",
"iclr_2018_r1h2DllAW",
"iclr_2018_r1h2DllAW"
] |
iclr_2018_S1Y7OOlRZ | Massively Parallel Hyperparameter Tuning | Modern machine learning models are characterized by large hyperparameter search spaces and prohibitively expensive training costs. For such models, we cannot afford to train candidate models sequentially and wait months before finding a suitable hyperparameter configuration. Hence, we introduce the large-scale regime ... | rejected-papers | This paper presents a simple tweak to hyperband to allow it to be run asynchonously on a large cluster, and contains reasonably large-scale experiments.
The paper is written clearly enough, and will be of interest to anyone running large-scale ML experiments. However, it falls below the bar by:
1) Not exploring the s... | test | [
"ry6owJ9lM",
"ByZovF3eM",
"r1GP9AFeM",
"SJgMG1tmM",
"H1W9naEXG",
"SkRUhpVmz",
"Bk5no64Xf",
"Bk0Qc6EmG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper introduces a simple extension to parallelize Hyperband. \n\nPoints in favor of the paper:\n* Addresses an important problem\n\nPoints against:\n* Only 5-fold speedup by parallelization with 5 x 25 workers, and worse performance in the same budget than Google Vizier (even though that treats the problem a... | [
5,
6,
5,
-1,
-1,
-1,
-1,
-1
] | [
5,
3,
5,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1Y7OOlRZ",
"iclr_2018_S1Y7OOlRZ",
"iclr_2018_S1Y7OOlRZ",
"iclr_2018_S1Y7OOlRZ",
"Bk0Qc6EmG",
"Bk0Qc6EmG",
"Bk0Qc6EmG",
"iclr_2018_S1Y7OOlRZ"
] |
iclr_2018_SyAbZb-0Z | Transfer Learning to Learn with Multitask Neural Model Search | Deep learning models require extensive architecture design exploration and hyperparameter optimization to perform well on a given task. The exploration of the model design space is often made by a human expert, and optimized using a combination of grid search and search heuristics over a large space of possible choices... | rejected-papers | This paper presents a sensible, but somewhat incremental, generalization of neural architecture search. However, the experiments are only done in a single artificial setting (albeit composed of real, large-scale subtasks). It's also not clear that such an expensive meta-learning based approach is even necessary, comp... | train | [
"HkKJIALgM",
"HyU-Egtef",
"S1dRXMqxG",
"BkrXu907M",
"B1_TYtaQM",
"rycuQ6nmz",
"S1J62Ih7f",
"Bkls2InmM",
"HJDO3Un7z",
"SkeU2L3Qz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper proposes an extension of the Neural Architecture Search approach, in which a single RNN controller is trained with RL to select hyperparameters for child networks that must perform different tasks. The architecture includes the notion of a \"task embedding\", that helps the controller keeping track of si... | [
5,
7,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SyAbZb-0Z",
"iclr_2018_SyAbZb-0Z",
"iclr_2018_SyAbZb-0Z",
"B1_TYtaQM",
"rycuQ6nmz",
"S1J62Ih7f",
"HkKJIALgM",
"HyU-Egtef",
"SkeU2L3Qz",
"S1dRXMqxG"
] |
iclr_2018_SyrGJYlRZ | YellowFin and the Art of Momentum Tuning | Hyperparameter tuning is one of the most time-consuming workloads in deep learning. State-of-the-art optimizers, such as AdaGrad, RMSProp and Adam, reduce this labor by adaptively tuning an individual learning rate for each variable. Recently researchers have shown renewed interest in simpler methods like momentum SGD ... | rejected-papers | This paper asks when SGD+M can beat adaptive methods such as Adam, and then suggests a variant of SGD+M with an adaptive controller for a single learning rate and momentum parameter. There is are comparisons with some popular alternatives. However, the bulk of the paper is concerned with a motivation that didn't conv... | train | [
"ryUK51INM",
"HyuhIWYez",
"B1RZJ1cxG",
"SJ0CHgbbM",
"SkdSe3zNz",
"BJsMnO6mG",
"B1jF7rpQz",
"SJ6oqITQG",
"Bkf6euOGM",
"ryz7Z_OzM",
"rJ9XCPOMG",
"BkE6RP_fG",
"HJWNy__GM",
"Hy4hvwuff"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"public",
"public",
"public",
"public",
"public",
"public",
"public",
"public"
] | [
"Dear Carlos,\n\nThanks for the clarification. Rescaling the learning rate as per your suggestion, on multiple experiments, has the following consequences:\n\n-- As expected, training is sped-up compared to using YF’s LR un-adjusted and just setting momentum to 0. In some cases, like the one you conduct experiment ... | [
-1,
4,
4,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
3,
5,
1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"BJsMnO6mG",
"iclr_2018_SyrGJYlRZ",
"iclr_2018_SyrGJYlRZ",
"iclr_2018_SyrGJYlRZ",
"iclr_2018_SyrGJYlRZ",
"SJ6oqITQG",
"iclr_2018_SyrGJYlRZ",
"B1jF7rpQz",
"HyuhIWYez",
"HyuhIWYez",
"B1RZJ1cxG",
"B1RZJ1cxG",
"B1RZJ1cxG",
"SJ0CHgbbM"
] |
iclr_2018_H1bM1fZCW | GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks | Deep multitask networks, in which one neural network produces multiple predictive outputs, are more scalable and often better regularized than their single-task counterparts. Such advantages can potentially lead to gains in both speed and performance, but multitask networks are also difficult to train without finding t... | rejected-papers | This paper proposes a way to automatically weight different tasks in a multi-task setting. The problem is a bit niche, and the paper had a lot of problems with clarity, as well as the motivation for the experimental setup and evaluation. | train | [
"H10ZQaugz",
"Bycjn6tef",
"Sy3BPIMbM",
"ryLKvyUzM",
"rJQmwyUzM",
"r1NpL1UGz",
"SkRYF0DZG",
"SJkNF0PbM",
"S1Ia_CP-G",
"S1Sqw0P-f",
"H1kyHRslM",
"rkHMM4KeM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"The paper proposes a method to train deep multi-task networks using gradient normalization. The key idea is to enforce the gradients from multi tasks balanced so that no tasks are ignored in the training. The authors also demonstrated that the technique can improve test errors over single task learning and uncerta... | [
6,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1bM1fZCW",
"iclr_2018_H1bM1fZCW",
"iclr_2018_H1bM1fZCW",
"H10ZQaugz",
"Bycjn6tef",
"Sy3BPIMbM",
"H10ZQaugz",
"Sy3BPIMbM",
"Bycjn6tef",
"iclr_2018_H1bM1fZCW",
"rkHMM4KeM",
"iclr_2018_H1bM1fZCW"
] |
iclr_2018_H1OQukZ0- | Online Hyper-Parameter Optimization | We propose an efficient online hyperparameter optimization method which uses a joint dynamical system to evaluate the gradient with respect to the hyperparameters. While similar methods are usually limited to hyperparameters with a smooth impact on the model, we show how to apply it to the probability of dropout in neu... | rejected-papers | This paper presents an update to the method of Franceschi 2017 to optimize regularization hyperparameters, to improve stability. However, the theoretical story isn't so clear, and the results aren't much of an improvement. Overall, the presentation and development of the idea needs work. | train | [
"S1Qv42tlz",
"HkNOs3g-z",
"SkFNKkUZf",
"rJmsNRjXz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author"
] | [
"Summary of the paper\n---------------------------\nThe paper addresses the issue of online optimization of hyper-parameters customary involved in deep architectures learning. The covered framework is limited to regularization parameters. These hyper-parameters, noted $\\lambda$, are updated along the training of ... | [
4,
5,
4,
-1
] | [
3,
3,
3,
-1
] | [
"iclr_2018_H1OQukZ0-",
"iclr_2018_H1OQukZ0-",
"iclr_2018_H1OQukZ0-",
"iclr_2018_H1OQukZ0-"
] |
iclr_2018_SyjjD1WRb | Evolutionary Expectation Maximization for Generative Models with Binary Latents | We establish a theoretical link between evolutionary algorithms and variational parameter optimization of probabilistic generative models with binary hidden variables.
While the novel approach is independent of the actual generative model, here we use two such models to investigate its applicability and scalabili... | rejected-papers | This method makes a connection between evolutionary and variational methods in a particular model. This is a good contribution, but there has been little effort to position it in comparison to standard methods that do the same thing, showing relative strengths and weaknesses.
Also, please shorten the abstract. | train | [
"BykoTdLNG",
"B1HSi_8Vf",
"ryWjGsdef",
"SyHYDG5lf",
"S15xOyjgf",
"BJaIxMSGz",
"SkqzzmSzG",
"ByuaozrzM",
"ByFAWfSGz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Just to clarify my position, I would suggest that any \"proof of concept\" paper DOES need to position itself clearly with respect to other approaches. You want your reader to walk away with clear understand of when you use your approach and why.\n\nEven a \"conceptual\" positioning would be a step in the right di... | [
-1,
-1,
4,
4,
4,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"BJaIxMSGz",
"ByuaozrzM",
"iclr_2018_SyjjD1WRb",
"iclr_2018_SyjjD1WRb",
"iclr_2018_SyjjD1WRb",
"iclr_2018_SyjjD1WRb",
"ryWjGsdef",
"SyHYDG5lf",
"S15xOyjgf"
] |
iclr_2018_r1kP7vlRb | Toward learning better metrics for sequence generation training with policy gradient | Designing a metric manually for unsupervised sequence generation tasks, such as text generation, is essentially difficult. In a such situation, learning a metric of a sequence from data is one possible solution. The previous study, SeqGAN, proposed the framework for unsupervised sequence generation, in which a metric i... | rejected-papers | The pros and cons of this paper can be summarized as follows:
Pros:
* It seems that the method has very good intuitions: consideration of partial rewards, estimation of rewards from modified sequences, etc.
Cons:
* The writing of the paper is scattered and not very well structured, which makes it difficult to follow ... | train | [
"H104OpbgM",
"SkY1f6Hlf",
"HJ2pirpxG",
"SJBLEn27z",
"SJRHEvMfG",
"Sku0lvfzz",
"ByWVfK6bf",
"r13qwSzbf",
"ByDj_rG-z",
"S1clBHfZG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"This article is a follow-up from recent publications (especially the one on \"seqGAN\" by Yu et al. @ AAAI 2017) which tends to assimilate Generative Adversarial Networks as an Inverse Reinforcement Learning task in order to obtain a better stability.\nThe adversarial learning is replaced here by a combination of ... | [
7,
4,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
1,
3,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1kP7vlRb",
"iclr_2018_r1kP7vlRb",
"iclr_2018_r1kP7vlRb",
"SJRHEvMfG",
"iclr_2018_r1kP7vlRb",
"ByWVfK6bf",
"r13qwSzbf",
"SkY1f6Hlf",
"HJ2pirpxG",
"H104OpbgM"
] |
iclr_2018_rkdU7tCaZ | Dynamic Evaluation of Neural Sequence Models | We present methodology for using dynamic evaluation to improve neural sequence models. Models are adapted to recent history via a gradient descent based mechanism, causing them to assign higher probabilities to re-occurring sequential patterns. Dynamic evaluation outperforms existing adaptation approaches in our compar... | rejected-papers | The pros and cons of the paper are summarized below:
Pros:
* The proposed tweaks to the dynamic evaluation of Mikolov et al. 2010 are somewhat effective, and when added on top of already-strong baseline models improve them substantially
Cons:
* Novelty is limited. This is essentially a slightly better training scheme... | train | [
"Sk-EjzwxG",
"B1Z3O3HeG",
"BkGlZbceG",
"SywY2Zvmz",
"HJ-esWv7G",
"HyTDubw7z",
"S19dzZ4bM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"official_reviewer"
] | [
"The authors provide an improved implementation of the idea of dynamic evaluation, where the update of the parameters used in the last time step proposed in (Mikolov et al. 2010) is replaced with a back-propagation through the last few time steps, and uses RMSprop rather than vanilla SGD. The method is applied to ... | [
7,
7,
3,
-1,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rkdU7tCaZ",
"iclr_2018_rkdU7tCaZ",
"iclr_2018_rkdU7tCaZ",
"B1Z3O3HeG",
"Sk-EjzwxG",
"BkGlZbceG",
"BkGlZbceG"
] |
iclr_2018_SkYibHlRb | SQLNet: Generating Structured Queries From Natural Language Without Reinforcement Learning | Synthesizing SQL queries from natural language is a long-standing open problem and has been attracting considerable interest recently. Toward solving the problem, the de facto approach is to employ a sequence-to-sequence-style model. Such an approach will necessarily require the SQL queries to be serialized. Since the ... | rejected-papers | The pros and cons of the paper cited by the reviewers can be summarized as follows:
Pros:
- good problem, NL2SQL is an important task given how dominant SQL is
- incorporating a grammar ("sketch") is a sensible improvement.
Cons:
- The dataset used makes very strong simplification assumptions (that every token is an ... | train | [
"BJnEgv8Nf",
"B1y7_3YgM",
"HksQE4cez",
"HkTzAHqxf",
"S1LubAnQz",
"HykdZyNmz",
"HyrNZJ4mM",
"ryjOuKfQG",
"B1SI_FfQz",
"rJP3AS0fM",
"SkkIarCfz",
"BkOfprAzG",
"ByNjir0Mf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"I have looked at the latest revision and it has not addressed my main concern adequately, i.e. the positioning of the paper wrt the literature. It is argued that previous work on question answering over tables is not relevant to the paper, which is clearly not the case: even though SQL can express more complex que... | [
-1,
4,
7,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
5,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"ByNjir0Mf",
"iclr_2018_SkYibHlRb",
"iclr_2018_SkYibHlRb",
"iclr_2018_SkYibHlRb",
"ryjOuKfQG",
"HyrNZJ4mM",
"B1SI_FfQz",
"B1SI_FfQz",
"BkOfprAzG",
"B1y7_3YgM",
"HksQE4cez",
"HkTzAHqxf",
"iclr_2018_SkYibHlRb"
] |
iclr_2018_r1nmx5l0W | SIC-GAN: A Self-Improving Collaborative GAN for Decoding Sketch RNNs | Variational RNNs are proposed to output “creative” sequences. Ideally, a collection of sequences produced by a variational RNN should be of both high quality and high variety. However, existing decoders for variational RNNs suffer from a trade-off between quality and variety. In this paper, we seek to learn a variation... | rejected-papers | Pros and cons of the paper can be summarized as follows:
Pros:
* The underlying idea may be interesting
* Results are reasonably strong on the test set used
Cons:
* Testing on the single dataset indicates that the model may be of limited applicability
* As noted by reviewer 2, core parts of the paper are extremely di... | train | [
"SJK75ZFef",
"Skie1qFxM",
"ByfDSjtlf",
"H1dfBd6Xz",
"Sk1SSO6Xz",
"rJtkrOaQf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper proposed a method that tries to generate both accurate and diverse samples from RNNs. \nI like the basic intuition of this paper, i.e., using mistakes for creativity and refining on top of it. I also think the evaluation is done properly. I think my biggest concern is that the method was only tested on a... | [
5,
4,
7,
-1,
-1,
-1
] | [
3,
5,
3,
-1,
-1,
-1
] | [
"iclr_2018_r1nmx5l0W",
"iclr_2018_r1nmx5l0W",
"iclr_2018_r1nmx5l0W",
"Skie1qFxM",
"SJK75ZFef",
"ByfDSjtlf"
] |
iclr_2018_BkUDW_lCb | Pointing Out SQL Queries From Text | The digitization of data has resulted in making datasets available to millions of users in the form of relational databases and spreadsheet tables. However, a majority of these users come from diverse backgrounds and lack the programming expertise to query and analyze such tables. We present a system that allows for qu... | rejected-papers | The pros and cons of the paper can be summarized below:
Pro:
* The improvements afforded by the method are significant over baselines, although these baselines are very preliminary baselines on a new dataset.
Con
* There is already a significant amount of work in using grammars to guide semantic parsing or code gener... | train | [
"S1IbWw_gM",
"rk5LF4OeM",
"BkYlT4ieG",
"S1VLPVn7f",
"rJhlPN27G",
"SJb6IEnQf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper claims to develop a novel method to map natural language queries to SQL. They claim to have the following contributions: \n\n1. Using a grammar to guide decoding \n2. Using a new loss function for pointer / copy mechanism. For each output token, they aggregate scores for all positions that the output tok... | [
4,
3,
7,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_BkUDW_lCb",
"iclr_2018_BkUDW_lCb",
"iclr_2018_BkUDW_lCb",
"rk5LF4OeM",
"S1IbWw_gM",
"BkYlT4ieG"
] |
iclr_2018_HyTrSegCb | Achieving morphological agreement with Concorde | Neural conversational models are widely used in applications like personal assistants and chat bots. These models seem to give better performance when operating on word level. However, for fusion languages like French, Russian and Polish vocabulary size sometimes become infeasible since most of the words have lots of w... | rejected-papers | The pros and cons of this paper cited by the reviewers can be summarized below:
Pros:
* Empirical results demonstrate decent improvements over other reasonable models
* The method is well engineered to the task
Cons:
* The paper is difficult to read due to grammar and formatting issues
* Experiments are also lacking ... | train | [
"r1dQH78gM",
"By3d5LqlM",
"H18dJfpxM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper is a pain to read. Most of the citation styles are off (i.e., without parentheses). Most of the sentences are not grammatically correct. Most, if not all, of the determiners are missing. It is ironic that the paper is proposing a model to generate grammatically correct sentences, while most of the senten... | [
2,
5,
6
] | [
5,
4,
5
] | [
"iclr_2018_HyTrSegCb",
"iclr_2018_HyTrSegCb",
"iclr_2018_HyTrSegCb"
] |
iclr_2018_rJ7RBNe0- | Generative Models for Alignment and Data Efficiency in Language | We examine how learning from unaligned data can improve both the data efficiency of supervised tasks as well as enable alignments without any supervision. For example, consider unsupervised machine translation: the input is two corpora of English and French, and the task is to translate from one language to the other b... | rejected-papers | The pros and cons of this paper cited by the reviewers (with a small amount of my personal opinion) can be summarized below:
Pros:
* The method itself seems to be tackling an interesting problem, which is feature matching between encoders within a generative model
Cons:
* The paper is sloppily written and symbols are... | train | [
"H1fmtttez",
"r1dDwZqxM",
"HyZEwgCgM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This work propose a generative model for unsupervised learning of translation model using a variant of auto-encoder which reconstruct internal layer representation in two directions. Basic idea is to treat the intermediate layers as feature representation which is reconstructed from the other direction. Experiment... | [
5,
4,
2
] | [
3,
3,
3
] | [
"iclr_2018_rJ7RBNe0-",
"iclr_2018_rJ7RBNe0-",
"iclr_2018_rJ7RBNe0-"
] |
iclr_2018_r1HNP0eCW | Estimation of cross-lingual news similarities using text-mining methods | Every second, innumerable text data, including all kinds news, reports, messages, reviews, comments, and twits have been generated on the Internet, which is written not only in English but also in other languages such as Chinese, Japanese, French and so on. Not only SNS sites but also worldwide news agency such as Tho... | rejected-papers | The pros and cons of this paper cited by the reviewers can be summarized below:
Pros:
* The motivation of the problem is presented well
* The architecture is simple and potentially applicable to real-world applications
Cons:
* The novel methodological contribution is limited to non-existant
* Comparison against other... | test | [
"HylRsRmgz",
"ByIyxIKef",
"r1fg60Kgz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"* PAPER SUMMARY *\n\nThis paper proposes a siamese net architecture to compare text in different languages. The proposed architecture builds upon siamese RNN by Mueller and Thyagarajan. The proposed approach is evaluated on cross lingual bitext retrieval.\n\n* REVIEW SUMMARY * \n\nThis paper is hard to read and ne... | [
2,
6,
2
] | [
5,
4,
4
] | [
"iclr_2018_r1HNP0eCW",
"iclr_2018_r1HNP0eCW",
"iclr_2018_r1HNP0eCW"
] |
iclr_2018_ryacTMZRZ | Jiffy: A Convolutional Approach to Learning Time Series Similarity | Computing distances between examples is at the core of many learning algorithms for time series. Consequently, a great deal of work has gone into designing effective time series distance measures. We present Jiffy, a simple and scalable distance metric for multivariate time series. Our approach is to reframe the task a... | rejected-papers | R1 was neutral on the paper: they liked the problem, simplicity of the approach, and thought the custom pooling layer was novel, but raised issues with the motivation and design of experiments. R1 makes a reasonable point that training a CNN to classify time series, then throw away the output layer and use the internal... | train | [
"Sk4LBfY4f",
"r1VHCbtNM",
"r1aYq6ieM",
"Hkns5PSlM",
"r1cIB5Fxf",
"SJUPi53zG",
"Hy94i92fz",
"SyDfYeqMz",
"r1rhVX1GM",
"SJ2qtgpWz",
"SJ6szzaWG",
"r1izTZpZf",
"Bk-UjRn-z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"public",
"public",
"author",
"author",
"public",
"public"
] | [
"I did in fact read the reproducibility report, but I want to clarify that I did NOT take it into account in my review (and for the record, I think its impact would have been negligible in this case). Reproduction during review represents an additional level of scrutiny for a submission, and while I feel that is Go... | [
-1,
-1,
6,
4,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"r1aYq6ieM",
"r1aYq6ieM",
"iclr_2018_ryacTMZRZ",
"iclr_2018_ryacTMZRZ",
"iclr_2018_ryacTMZRZ",
"Hy94i92fz",
"iclr_2018_ryacTMZRZ",
"iclr_2018_ryacTMZRZ",
"SJ6szzaWG",
"Bk-UjRn-z",
"r1izTZpZf",
"SJ2qtgpWz",
"iclr_2018_ryacTMZRZ"
] |
iclr_2018_ryj38zWRb | Optimizing the Latent Space of Generative Networks | Generative Adversarial Networks (GANs) have achieved remarkable results in the task of generating realistic natural images. In most applications, GAN models share two aspects in common. On the one hand, GANs training involves solving a challenging saddle point optimization problem, interpreted as an adversarial game be... | rejected-papers | This paper attempts to decouple two factors underlying the success of GANs: the inductive bias of deep CNNs and adversarial training. It shows that, surprisingly, the second factor is not essential. R1 thought that comparisons to Generative Moment Matching Networks and Variational Autoencoders should be provided (note:... | val | [
"BkILtntlz",
"SynXdTKeM",
"HyE2oHixz",
"ryNFIO6mf",
"SkvIrda7G",
"HJ8VH_pmz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary: The authors observe that the success of GANs can be attributed to two factors; leveraging the inductive bias of deep CNNs and the adversarial training protocol. In order to disentangle the factors of success, and they propose to eliminate the adversarial training protocol while maintaining the first facto... | [
4,
6,
6,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_ryj38zWRb",
"iclr_2018_ryj38zWRb",
"iclr_2018_ryj38zWRb",
"iclr_2018_ryj38zWRb",
"iclr_2018_ryj38zWRb",
"iclr_2018_ryj38zWRb"
] |
iclr_2018_B1ZZTfZAW | Real-valued (Medical) Time Series Generation with Recurrent Conditional GANs | Generative Adversarial Networks (GANs) have shown remarkable success as a framework for training models to produce realistic-looking data. In this work, we propose a Recurrent GAN (RGAN) and Recurrent Conditional GAN (RCGAN) to produce realistic real-valued multi-dimensional time series, with an emphasis on their appli... | rejected-papers | Overall I agree with the assessment of R1 that the paper touches on many interesting issues (deep learning for time series, privacy-respecting ML, simulated-to-real-world adaptation) but does not make a strong contribution to any of these. Especially with respect to the privacy-respecting aspect, there needs to be more... | train | [
"H1q2baOxM",
"SyhqIGYxG",
"Hk6aJkmWM",
"HJwRYDdXG",
"B1nOSPu7z",
"BJ0gSwd7f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The authors propose to use synthetic data generated by GANs as a replacement for personally identifiable data in training ML models for privacy-sensitive applications such as medicine. In particular it demonstrates adversarial training of a recurrent generator for an ICU monitoring multidimensional time series, pr... | [
4,
6,
5,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_B1ZZTfZAW",
"iclr_2018_B1ZZTfZAW",
"iclr_2018_B1ZZTfZAW",
"H1q2baOxM",
"SyhqIGYxG",
"Hk6aJkmWM"
] |
iclr_2018_SkfNU2e0Z | Statestream: A toolbox to explore layerwise-parallel deep neural networks | Building deep neural networks to control autonomous agents which have to interact in real-time with the physical world, such as robots or automotive vehicles, requires a seamless integration of time into a network’s architecture. The central question of this work is, how the temporal nature of reality should be reflect... | rejected-papers | This paper presents a toolbox for the exploration of layerwise-parallel deep neural networks. The reviewers were consistent in their analysis of this paper: it provided an interesting class of models which warranted further investigation, and that the toolbox would be useful to those who are interested in exploring fur... | val | [
"SJ1tsSFgf",
"HkeBFwYgf",
"B1KY-MqgG",
"H1VGzc9zf",
"SJMlG95Mf",
"r1ka-99zM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Quality and clarity\n\nThe paper goes to some length to explain that update order in a neural network matters in the sense that different update orders give different results. While standard CNN like architectures are fine with the layer parallel updating process typically used in standard tools, for recurrent net... | [
3,
5,
5,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_SkfNU2e0Z",
"iclr_2018_SkfNU2e0Z",
"iclr_2018_SkfNU2e0Z",
"SJ1tsSFgf",
"HkeBFwYgf",
"B1KY-MqgG"
] |
iclr_2018_HyIFzx-0b | BinaryFlex: On-the-Fly Kernel Generation in Binary Convolutional Networks | In this work we present BinaryFlex, a neural network architecture that learns weighting coefficients of predefined orthogonal binary basis instead of the conventional approach of learning directly the convolutional filters. We have demonstrated the feasibility of our approach for complex computer vision datasets such a... | rejected-papers | The paper proposes using a set of orthogonal bases that combine to form convolution kernels for CNNs leading to a significant reduction of memory usage. The main concerns raised by the reviewers were 1) clarity; 2) issues with writing and presentation of results; 3) some missing experiments. The authors released a revi... | train | [
"BkA2-XteM",
"B1wZVecxM",
"B1hW9m5gG",
"HybCNdpQM",
"rk5aXua7M",
"ryiZXupXf",
"r1E4fOp7G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper presents a binary neural network architecture that operated on predefined orthogonal binary basis. The binary filters that are used as basis are generated using Orthogonal Variable Spreading Factor. \nBecause the filters are weighted combinations of predefined basis, only the weights need to be trained a... | [
5,
5,
3,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HyIFzx-0b",
"iclr_2018_HyIFzx-0b",
"iclr_2018_HyIFzx-0b",
"BkA2-XteM",
"B1wZVecxM",
"B1hW9m5gG",
"iclr_2018_HyIFzx-0b"
] |
iclr_2018_SJtfOEn6- | ResBinNet: Residual Binary Neural Network | Recent efforts on training light-weight binary neural networks offer promising execution/memory efficiency. This paper introduces ResBinNet, which is a composition of two interlinked methodologies aiming to address the slow convergence speed and limited accuracy of binary convolutional neural networks. The first method... | rejected-papers | R1 and R3’s main concern was that the work was not actually outperforming existing work and therefore its advantages were unclear. R2 brought up several questions on the experiments and asked for clarification with respect to previous work. R3 had several other detailed questions for the authors. The authors did not p... | train | [
"S1X0siBxz",
"HkG6r4Kgf",
"Byi5CZcxz",
"Hy4pBX-kM",
"HyUlYClJM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public"
] | [
"This paper proposes a method to quantize weights and activations in neural network during propagations.\n\nThe residual binarization idea is interesting. However, the experimental results are not sufficiently convincing that this method is meaningfully improving over previous methods. Specifically:\n\n1) In table ... | [
4,
4,
4,
-1,
-1
] | [
4,
4,
4,
-1,
-1
] | [
"iclr_2018_SJtfOEn6-",
"iclr_2018_SJtfOEn6-",
"iclr_2018_SJtfOEn6-",
"HyUlYClJM",
"iclr_2018_SJtfOEn6-"
] |
iclr_2018_HyFaiGbCW | Generalization of Learning using Reservoir Computing | We investigate the methods by which a Reservoir Computing Network (RCN) learns concepts such as 'similar' and 'different' between pairs of images using a small training dataset and generalizes these concepts to previously unseen types of data. Specifically, we show that an RCN trained to identify relationships between ... | rejected-papers | Both R1 and R2 suggested that Conceptors (Jaeger, 2014) had previously explored learning transformations in the context of reservoir computing. The authors acknowledged this in their response and added a reference. The main concern raised by the reviewers was lack of novelty and weak experiments (both the MNIST and dep... | train | [
"r1IUXROxz",
"SkMZxYKgz",
"rJNd7A1bz",
"rkisUzUQf",
"r1EuVzUXz",
"BkKkRW8mM",
"BkVgNWIQM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The claimed results of \"combining transformations\" in the context of RC was done in the works of Herbert Jaeger on conceptors [1], which also should be cited here.\n\nThe argument of biological plausibility is not justified. The authors use an echo-state neural network with standard tanh activations, which is ... | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
3,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HyFaiGbCW",
"iclr_2018_HyFaiGbCW",
"iclr_2018_HyFaiGbCW",
"iclr_2018_HyFaiGbCW",
"SkMZxYKgz",
"r1IUXROxz",
"rJNd7A1bz"
] |
iclr_2018_H1vCXOe0b | Interpreting Deep Classification Models With Bayesian Inference | In this paper, we propose a novel approach to interpret a well-trained classification model through systematically investigating effects of its hidden units on prediction making. We search for the core hidden units responsible for predicting inputs as the class of interest under the generative Bayesian inference framew... | rejected-papers | The paper proposes a new method for interpreting the hidden units of neural networks by employing an Indian Buffet Process. The reviewers felt that the approach was interesting, but at times hard to follow and more analysis was needed. In particular, it was difficult to glean any advantage of this method over others. T... | train | [
"rye1W_5lf",
"HJmy1rjlM",
"HyESiU7WG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper intends to interpret a well-trained multi-class classification deep neural network by discovering the core units of one or multiple hidden layers for prediction making. However, these discovered core units are specific to a particular class, which are retained to maintain the deep neural network’s abilit... | [
3,
5,
3
] | [
3,
3,
4
] | [
"iclr_2018_H1vCXOe0b",
"iclr_2018_H1vCXOe0b",
"iclr_2018_H1vCXOe0b"
] |
iclr_2018_HJ1HFlZAb | Evaluation of generative networks through their data augmentation capacity | Generative networks are known to be difficult to assess. Recent works on generative models, especially on generative adversarial networks, produce nice samples of varied categories of images. But the validation of their quality is highly dependent on the method used. A good generator should generate data which contain ... | rejected-papers | Given that the paper proposes a new evaluation scheme for generative models, I agree with the reviewers that it is essential that the paper compare with existing metrics (even if they are imperfect). The choice of datasets was very limited as well, given the nature of the paper. I acknowledge that the authors took care... | val | [
"HyjUd0Kgf",
"ryYjvicxM",
"B1u5na9lG",
"ryaVoRcfz",
"B1EXiA5Mf",
"By4esCczz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The main idea is to use the accuracy of a classifier trained on synthetic training examples produced by a generative model to define an evaluation metric for the generative model. Specifically, compare the accuracy of a classifier trained on a noise-perturbed version of the real dataset to that of a classifier tra... | [
3,
3,
5,
-1,
-1,
-1
] | [
5,
5,
3,
-1,
-1,
-1
] | [
"iclr_2018_HJ1HFlZAb",
"iclr_2018_HJ1HFlZAb",
"iclr_2018_HJ1HFlZAb",
"HyjUd0Kgf",
"ryYjvicxM",
"B1u5na9lG"
] |
iclr_2018_r1RQdCg0W | MACH: Embarrassingly parallel K-class classification in O(dlogK) memory and O(KlogK+dlogK) time, instead of O(Kd) | We present Merged-Averaged Classifiers via Hashing (MACH) for K-classification with large K. Compared to traditional one-vs-all classifiers that require O(Kd) memory and inference cost, MACH only need O(dlogK) memory while only requiring O(KlogK+dlogK) operation for inference. MACH is the first generic K-classificat... | rejected-papers | There is a very nice discussion with one of the reviewers on the experiments, that I think would need to be battened down in an ideal setting. I'm also a bit surprised at the lack of discussion or comparison to two seemingly highly related papers:
1. T. G. Dietterich and G. Bakiri (1995) Solving Multiclass via Error C... | train | [
"SJB-0Mtlz",
"H1tJH9FxM",
"H1VwD15lG",
"Sk9lYDj7M",
"r1GeQQxmG",
"S1YDsMlXf",
"SJfRHbeQz",
"Skch5gafG",
"HkC5WzafM",
"HJVtSZaMz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"The manuscript proposes an efficient hashing method, namely MACH, for softmax approximation in the context of large output space, which saves both memory and computation. In particular, the proposed MACH uses 2-universal hashing to randomly group classes, and trains a classifier to predict the group membership. It... | [
6,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1RQdCg0W",
"iclr_2018_r1RQdCg0W",
"iclr_2018_r1RQdCg0W",
"r1GeQQxmG",
"S1YDsMlXf",
"SJfRHbeQz",
"Skch5gafG",
"H1tJH9FxM",
"SJB-0Mtlz",
"H1VwD15lG"
] |
iclr_2018_r1AoGNlC- | Code Synthesis with Priority Queue Training | We consider the task of program synthesis in the presence of a reward function over the output of programs, where the goal is to find programs with maximal rewards. We introduce a novel iterative optimization scheme, where we train an RNN on a dataset of K best programs from a priority queue of the generated programs s... | rejected-papers | This paper introduces a possibly useful new RL idea (though it's a incremental on Liang et al), but the evaluations don't say much about why it works (when it does), and we didn't find the target application convincing.
| train | [
"B1xdjLPef",
"Sk8MrZ9gM",
"HJK2Mt3lG",
"rJ7ZnliGM",
"SyF8pGsmM",
"Bk0b7t57f",
"HJDzCgsMz",
"Bk9r6xsfM",
"rJXb6gjGG",
"BJWFneiGM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper presents an algorithm called Priority Queue Training (PQT) for\nprogram synthesis using an RNN where the RNN is trained in presence of a \nreward signal over the desired program outputs. The RNN learns a policy \nthat generates a sequence of characters in BF conditioned on a prefix of characters.\nThe k... | [
6,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1AoGNlC-",
"iclr_2018_r1AoGNlC-",
"iclr_2018_r1AoGNlC-",
"iclr_2018_r1AoGNlC-",
"Bk0b7t57f",
"HJDzCgsMz",
"rJXb6gjGG",
"B1xdjLPef",
"Sk8MrZ9gM",
"HJK2Mt3lG"
] |
iclr_2018_r154_g-Rb | Composable Planning with Attributes | The tasks that an agent will need to solve often aren’t known during training. However, if the agent knows which properties of the environment we consider im- portant, then after learning how its actions affect those properties the agent may be able to use this knowledge to solve complex tasks without training specifi-... | rejected-papers | Overall the reviewers appear to like the ideas in this paper, though this is some disagreement about novelty (I agree with the reviewer who believes that the top-level search can very easily be interpreted as an MDP, making this very similar to SMDPs). The reviewers generally felt that the experimental results need to ... | train | [
"S1bTbEyrG",
"H1G5f2A4G",
"rkxi5qA4G",
"HJhTjeREz",
"Hk10FQqef",
"r1cviMjgf",
"B1CBt8ilG",
"r1i-SlcQG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author"
] | [
"I'm not sure that's quite fair to the authors, as the paper you linked to was only published on arXiv about three weeks before the ICLR deadline; I would consider that concurrent work.",
"- The paper needs to be better contextualized with prior work. As other reviewers agree, the connections to MDPs and semi-MDP... | [
-1,
-1,
-1,
-1,
5,
4,
7,
-1
] | [
-1,
-1,
-1,
-1,
4,
5,
3,
-1
] | [
"rkxi5qA4G",
"r1cviMjgf",
"Hk10FQqef",
"B1CBt8ilG",
"iclr_2018_r154_g-Rb",
"iclr_2018_r154_g-Rb",
"iclr_2018_r154_g-Rb",
"iclr_2018_r154_g-Rb"
] |
iclr_2018_B1NOXfWR- | Neural Task Graph Execution | In order to develop a scalable multi-task reinforcement learning (RL) agent that is able to execute many complex tasks, this paper introduces a new RL problem where the agent is required to execute a given task graph which describes a set of subtasks and dependencies among them. Unlike existing approaches which explici... | rejected-papers | Paper presents and interesting new direction, but the evaluation leaves many questions open, and situation with respect to state of the art is lacking | train | [
"Skjfeg5gG",
"B1d-B8jgz",
"r14EMwbzf",
"r13JqCBMz",
"Hy8rL0HMG",
"Hy3nSRHGM",
"By167CrfM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"In the context of multitask reinforcement learning, this paper considers the problem of learning behaviours when given specifications of subtasks and the relationship between them, in the form of a task graph. The paper presents a neural task graph solver (NTS), which encodes this as a recursive-reverse-recursive ... | [
6,
6,
4,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_B1NOXfWR-",
"iclr_2018_B1NOXfWR-",
"iclr_2018_B1NOXfWR-",
"r14EMwbzf",
"B1d-B8jgz",
"Skjfeg5gG",
"iclr_2018_B1NOXfWR-"
] |
iclr_2018_By3VrbbAb | Realtime query completion via deep language models | Search engine users nowadays heavily depend on query completion and correction to shape their queries. Typically, the completion is done by database lookup which does not understand the context and cannot generalize to prefixes not in the database. In the paper, we propose to use unsupervised deep language models to c... | rejected-papers | This paper has some interesting ideas that have been implemented in a rather ad hoc way; the presentation focuses perhaps too much on engineering aspects. | train | [
"r1BkYcJeM",
"HJdsrHOxf",
"B1kHjjhlM",
"rkM1XV3mG",
"BksB6zhXf",
"rk8SBP6Jz",
"B16Bg6ARW",
"HJkXhY9Rb"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"public",
"author",
"public"
] | [
"This paper presents methods for query completion that includes prefix correction, and some engineering details to meet particular latency requirements on a CPU. Regarding the latter methods: what is described in the paper sounds like competent engineering details that those performing such a task for launch in a ... | [
4,
6,
5,
-1,
-1,
-1,
-1,
-1
] | [
5,
3,
3,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_By3VrbbAb",
"iclr_2018_By3VrbbAb",
"iclr_2018_By3VrbbAb",
"r1BkYcJeM",
"rk8SBP6Jz",
"iclr_2018_By3VrbbAb",
"HJkXhY9Rb",
"iclr_2018_By3VrbbAb"
] |
iclr_2018_BJ4prNx0W | Learning what to learn in a neural program | Learning programs with neural networks is a challenging task, addressed by a long line of existing work. It is difficult to learn neural networks which will generalize to problem instances that are much larger than those used during training. Furthermore, even when the learned neural program empirically works on all te... | rejected-papers | This paper is novel, but relatively incremental and relatively niche; the reviewers (despite discussion) are still unsure why this approach is needed. | train | [
"B10flwLgz",
"Sk-AwdKlf",
"HJ0ww5VbG",
"HkQ4xH9zf",
"Hke1lSqzz",
"ByOs1BcMM",
"H1ZeDgDJG",
"ByvPTmHkM",
"ryp3LaFA-"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public",
"author",
"public"
] | [
"Quality\nThe paper is well-written and clear, and includes relevant comparisons to previous work (NPI and recursive NPI).\n\nClarity\nThe paper is clearly written.\n\nOriginality\nTo my knowledge the method proposed in this work is novel. It is the first to study constructing minimal training sets for NPI given a ... | [
5,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
2,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJ4prNx0W",
"iclr_2018_BJ4prNx0W",
"iclr_2018_BJ4prNx0W",
"B10flwLgz",
"Sk-AwdKlf",
"HJ0ww5VbG",
"ByvPTmHkM",
"ryp3LaFA-",
"iclr_2018_BJ4prNx0W"
] |
iclr_2018_ryZElGZ0Z | Discovery of Predictive Representations With a Network of General Value Functions | The ability of an agent to {\em discover} its own learning objectives has long been considered a key ingredient for artificial general intelligence. Breakthroughs in autonomous decision making and reinforcement learning have primarily been in domains where the agent's goal is outlined and clear: such as playing a game ... | rejected-papers | There was substantial disagreement between reviewers on how this paper contributes to the literature; it seems (having read the paper) that the problem tackled here is clearly quite interesting, but it is hard to tease out in the current version exactly what the contribution does to extend beyond current art. | train | [
"ByqXgftlz",
"HJf2G-cgf",
"Bk0FsiXbM",
"BJt9fysmz",
"BytVjsWGG",
"HJh6poWzM",
"S15W3ibfG",
"HJ4g2obGG",
"SyU8oj-zG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"I have to say that I do not have all the background of this paper, and the paper is not written very clearly. I think the major contribution of the paper is represented in a very vague way.",
"I really enjoyed reading this paper and stopped a few time to write down new ideas it brought up. Well written and very ... | [
4,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
1,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ryZElGZ0Z",
"iclr_2018_ryZElGZ0Z",
"iclr_2018_ryZElGZ0Z",
"SyU8oj-zG",
"iclr_2018_ryZElGZ0Z",
"ByqXgftlz",
"HJf2G-cgf",
"Bk0FsiXbM",
"BytVjsWGG"
] |
iclr_2018_SyvCD-b0W | Autostacker: an Automatic Evolutionary Hierarchical Machine Learning System | This work provides an automatic machine learning (AutoML) modelling architecture called Autostacker. Autostacker improves the prediction accuracy of machine learning baselines by utilizing an innovative hierarchical stacking architecture and an efficient parameter search algorithm. Neither prior domain knowledge about ... | rejected-papers | The reviewers have pointed out that there is a substantial amount of related work that this paper should be acknowledging and building on. | val | [
"HkIzq6Vez",
"SkYou4teG",
"HJePBM9lz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The author present Autostacker, a new algorithm for combining the strength of different learning algorithms during hyper parameter search. During the first step, the hyperparameter search is done in a conventional way. At the second step, the output of each primitives is added to the features of the original datas... | [
4,
3,
4
] | [
5,
4,
5
] | [
"iclr_2018_SyvCD-b0W",
"iclr_2018_SyvCD-b0W",
"iclr_2018_SyvCD-b0W"
] |
iclr_2018_H1Ww66x0- | Lifelong Learning with Output Kernels | Lifelong learning poses considerable challenges in terms of effectiveness (minimizing prediction errors for all tasks) and overall computational tractability for real-time performance. This paper addresses continuous lifelong multitask learning by jointly re-estimating the inter-task relations (\textit{output} kernel)... | rejected-papers | The output kernel idea for lifelong learning is interesting, but insufficiently developed in the current draft. | test | [
"SyZQxkmxG",
"SySnNRUxz",
"rklBKmcgG",
"SJBuJy0Xz",
"S1Ikdi67G",
"Skv3vop7M",
"SJ0_PiTXM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"CONTRIBUTION\nThe main contribution of the paper is not clearly stated. To the reviewer, It seems “life-long learning” is the same as “online learning”. However, the whole paper does not define what “life-long learning” is.\nThe limited budget scheme is well established in the literature. \n1. J. Hu, H. Yang, I.... | [
3,
2,
4,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1Ww66x0-",
"iclr_2018_H1Ww66x0-",
"iclr_2018_H1Ww66x0-",
"iclr_2018_H1Ww66x0-",
"SyZQxkmxG",
"SySnNRUxz",
"rklBKmcgG"
] |
iclr_2018_SkFvV0yC- | Network Iterative Learning for Dynamic Deep Neural Networks via Morphism | In this research, we present a novel learning scheme called network iterative learning for deep neural networks. Different from traditional optimization algorithms that usually optimize directly on a static objective function, we propose in this work to optimize a dynamic objective function in an iterative fashion capa... | rejected-papers | The paper presents a variant of network morphism (Wei et al., 2016) for dynamically growing deep neural networks. There are some novel contributions (such as OptGD for finding a morphism given the parent network layer). However, in the current form, the experiments mostly focus on comparisons against fixed network stru... | train | [
"S1YX1ptgz",
"BkdUwGqgz",
"By4qQIQ-f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This submission develops a learning scheme for training deep neural networks with adoption of network morphism (Wei et al., 2016), which optimizes a dynamic objective function in an iterative fashion capable of adapting its function form when being optimized, instead of directly optimizing a static objective funct... | [
7,
5,
5
] | [
4,
2,
3
] | [
"iclr_2018_SkFvV0yC-",
"iclr_2018_SkFvV0yC-",
"iclr_2018_SkFvV0yC-"
] |
iclr_2018_SJQO7UJCW | Adversarial Learning for Semi-Supervised Semantic Segmentation | We propose a method for semi-supervised semantic segmentation using the adversarial network. While most existing discriminators are trained to classify input images as real or fake on the image level, we design a discriminator in a fully convolutional manner to differentiate the predicted probability maps from the grou... | rejected-papers | The paper presents a reasonable idea, probably an improved version of method (combination of GAN and SSL for semantic segmentation) over the existing works. Novelty is not ground-breaking (e.g., discriminator network taking only pixel-labeling predictions, application of self-training for semantic segmentation---each o... | train | [
"r1RDwROeG",
"H1Op4eqlM",
"SJRdYLhgM",
"ByPmwO0bf",
"B1B1w_AWf",
"ByN58_AWf",
"HkbN6Aplz",
"B1exQRalz",
"BkXRaopxz",
"HydNvqQgf",
"BJ_zPCggz",
"BJJhDZakf",
"B1YQsg61M",
"H1wcIfh1z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"public",
"author",
"public",
"author",
"public",
"public",
"author",
"public"
] | [
"This paper describes techniques for training semantic segmentation networks. There are two key ideas:\n\n- Attach a pixel-level GAN loss to the output semantic segmentation map. That is, add a discriminator network that decides whether each pixel in the label map belongs to a real label map or not. Of course, this... | [
5,
5,
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_SJQO7UJCW",
"iclr_2018_SJQO7UJCW",
"iclr_2018_SJQO7UJCW",
"r1RDwROeG",
"H1Op4eqlM",
"SJRdYLhgM",
"B1exQRalz",
"BkXRaopxz",
"HydNvqQgf",
"BJ_zPCggz",
"BJJhDZakf",
"B1YQsg61M",
"H1wcIfh1z",
"iclr_2018_SJQO7UJCW"
] |
iclr_2018_SyVVXngRW | Deep Asymmetric Multi-task Feature Learning | We propose Deep Asymmetric Multitask Feature Learning (Deep-AMTFL) which can learn deep representations shared across multiple tasks while effectively preventing negative transfer that may happen in the feature sharing process. Specifically, we introduce an asymmetric autoencoder term that allows reliable predictors fo... | rejected-papers | The paper proposes a multitask deep learning method (called Deep-AMFTL) for preventing negative transfer. Despite some positive experimental results, the contribution of the paper is not sufficient for publication at ICLR due to several issues: similarity between the proposed method and existing method (e.g., AMTL), un... | train | [
"ryAf2-ugz",
"H11NN0KgG",
"S1T4ik9ef"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper presents a deep asymmetric multi-task feature learning method (Deep-AMTFL).\n\nOne concern is that the high similarity between the proposed Deep-AMTFL and an existing AMTL method. Even though AMTL operates on task relations and Deep-AMTFL is on feature learning, the main ideas of both methods are very s... | [
6,
3,
5
] | [
4,
4,
4
] | [
"iclr_2018_SyVVXngRW",
"iclr_2018_SyVVXngRW",
"iclr_2018_SyVVXngRW"
] |
iclr_2018_Hy_o3x-0b | Feature Map Variational Auto-Encoders | There have been multiple attempts with variational auto-encoders (VAE) to learn powerful global representations of complex data using a combination of latent stochastic variables and an autoregressive model over the dimensions of the data. However, for the most challenging natural image tasks the purely autoregressive ... | rejected-papers | The paper proposes a VAE variant by embedding spatial information with multiple layers of latent variables. Although the paper reports state-of-the-art results on multiple datasets, some results may be due to a bug. This has been discussed, and the author acknowledges the bug. We hope the problem can be fixed, and the ... | train | [
"HkYC48PxG",
"SyjePb9gz",
"Hkk3C-5lM",
"ByZxuMamf",
"ryUdttWzG",
"BJZQRBZMf",
"BkUk6H-zz",
"SyaUOmUZf",
"rJNRG94lM",
"HkWqOPVgf",
"SkjKJ5Nlf",
"HyIvKY4ez",
"Hy0Pcd4gf",
"HkHWP_Vef",
"Syq-KRzlz",
"HksGNSfgG",
"H19apQuyM",
"ryo_Mw0Cb",
"SJbqRnO0b",
"S183EOuC-",
"S1RfdFwRb",
"... | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"public",
"public",
"public",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"official_reviewer",
"public",
"official_reviewer",
"author",
"public",
"public",
"public",
... | [
"The paper combines several recent advances on generative modelling including a ladder variational posterior and a PixelCNN decoder together with the proposed convolutional stochastic layers to boost the NLL results of the current VAEs. The numbers in the tables are good but I have several comments on the motivatio... | [
5,
3,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Hy_o3x-0b",
"iclr_2018_Hy_o3x-0b",
"iclr_2018_Hy_o3x-0b",
"iclr_2018_Hy_o3x-0b",
"BJZQRBZMf",
"SyaUOmUZf",
"iclr_2018_Hy_o3x-0b",
"SyjePb9gz",
"SkjKJ5Nlf",
"HksGNSfgG",
"HyIvKY4ez",
"Hy0Pcd4gf",
"HkHWP_Vef",
"HkWqOPVgf",
"H19apQuyM",
"H19apQuyM",
"SJbqRnO0b",
"S183EOuC-"... |
iclr_2018_HJIhGXWCZ | Prediction Under Uncertainty with Error Encoding Networks | In this work we introduce a new framework for performing temporal predictions
in the presence of uncertainty. It is based on a simple idea of disentangling com-
ponents of the future state which are predictable from those which are inherently
unpredictable, and encoding the unpredictable components in... | rejected-papers | The paper proposes a novel predictive model (e.g., from videos), called error encoding networks, by first learning a deterministic prediction model and then learning to minimize the residual error using latent variables. The latent variables given the sample are estimated by sampling from the prior then updating via gr... | train | [
"Hk1_7QS4z",
"SykHOLdxz",
"B1YP4d5lf",
"Byn8p7V-z",
"BJEMFNXEG",
"H1sUBupmf",
"BkNb8pXyM",
"SJb-YQmyG",
"Sko_8QFAb",
"H1MzsbDRW"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"public",
"author",
"public"
] | [
"Thank you for the revised version of the paper and adding new experiments. Even though the core idea is interesting I am not still convinced of the architecture and the experiment results. One reason could be the fact that video generation might have been wrong choice of task. As there has been many recent works i... | [
-1,
4,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
3,
2,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"H1sUBupmf",
"iclr_2018_HJIhGXWCZ",
"iclr_2018_HJIhGXWCZ",
"iclr_2018_HJIhGXWCZ",
"H1sUBupmf",
"iclr_2018_HJIhGXWCZ",
"SJb-YQmyG",
"iclr_2018_HJIhGXWCZ",
"H1MzsbDRW",
"iclr_2018_HJIhGXWCZ"
] |
iclr_2018_rJa90ceAb | Learning to Generate Filters for Convolutional Neural Networks | Conventionally, convolutional neural networks (CNNs) process different images with the same set of filters. However, the variations in images pose a challenge to this fashion. In this paper, we propose to generate sample-specific filters for convolutional layers in the forward pass. Since the filters are generated on-t... | rejected-papers | The paper proposes a method for learning convolutional networks with dynamic input-conditioned filters. There are several prior work along this idea, but there is no comparison agaist them. Overall, experimental results are not convincing enough. | test | [
"HygXOMDxf",
"Syy4M8qxf",
"BJFxOpcez",
"SJzHA_zyf",
"rknWFdWyf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public"
] | [
"The authors propose an approach to dynamically generating filters in a CNN based on the input image. The filters are generated as linear combinations of a basis set of filters, based on features extracted by an auto-encoder. The authors test the approach on recognition tasks on three datasets: MNIST, MTFL (facial ... | [
4,
5,
4,
-1,
-1
] | [
4,
4,
5,
-1,
-1
] | [
"iclr_2018_rJa90ceAb",
"iclr_2018_rJa90ceAb",
"iclr_2018_rJa90ceAb",
"rknWFdWyf",
"iclr_2018_rJa90ceAb"
] |
iclr_2018_HkCvZXbC- | 3C-GAN: AN CONDITION-CONTEXT-COMPOSITE GENERATIVE ADVERSARIAL NETWORKS FOR GENERATING IMAGES SEPARATELY | We present 3C-GAN: a novel multiple generators structures, that contains one conditional generator that generates a semantic part of an image conditional on its input label, and one context generator generates the rest of an image. Compared to original GAN model, this model has multiple generators and gives control ove... | rejected-papers | The paper presents a layered image generation model (e.g., foreground vs background) using GANs. The high-level idea is interesting, but novelty is somewhat limited. For example, layered generation with VAE/GAN has been explored in Yan et al. 2016 (VAEs) and Vondrick et al. 2016 (GANs). In addition, there are earlier ... | train | [
"By9jZQukf",
"ByfJW0vlf",
"SygpIWqlM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"Summary: This paper studied the conditional image generation with two-stream generative adversarial networks. More specifically, this paper proposed an unsupervised learning approach to generate (1) foreground region conditioned on class label and (2) background region without semantic meaning in the label. During... | [
5,
4,
4
] | [
5,
4,
5
] | [
"iclr_2018_HkCvZXbC-",
"iclr_2018_HkCvZXbC-",
"iclr_2018_HkCvZXbC-"
] |
iclr_2018_rkQsMCJCb | Generative Adversarial Networks using Adaptive Convolution | Most existing GANs architectures that generate images use transposed convolution or resize-convolution as their upsampling algorithm from lower to higher resolution feature maps in the generator. We argue that this kind of fixed operation is problematic for GANs to model objects that have very different visual appearan... | rejected-papers | The paper proposes a GAN model with adaptive convolution kernels. The proposed idea is reasonable, but the novelty is somewhat minor and the experimental results are limited. More comprehensive experiments (e.g., other evaluation metrics) will strengthen the future revision of paper. No rebuttal was submitted.
| train | [
"SkzQ2hFxf",
"H1JNBZTef",
"Sy686Qplz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper operates under the hypothesis that the rigidity of the convolution operator is responsible in part for the poor performance of GANs on diverse visual datasets. The authors propose to replace convolutions in the generator with an Adaptive Convolution Block, which learns to generate the convolution weights... | [
4,
4,
4
] | [
5,
4,
5
] | [
"iclr_2018_rkQsMCJCb",
"iclr_2018_rkQsMCJCb",
"iclr_2018_rkQsMCJCb"
] |
iclr_2018_HJrJpzZRZ | Self-Supervised Learning of Object Motion Through Adversarial Video Prediction | Can we build models that automatically learn about object motion from raw, unlabeled videos? In this paper, we study the problem of multi-step video prediction, where the goal is to predict a sequence of future frames conditioned on a short context. We focus specifically on two aspects of video prediction: accurately m... | rejected-papers | The paper proposes adversarial flow-based neural network architecture with adversarial training for video prediction. Although the reported experimental results are promising, the paper seems below ICLR threshold due to limited novelty and issues in evaluation (e.g., mechanical turk experiment). No rebuttal was submitt... | train | [
"BkGsQ4Ixz",
"ByW0MJqlM",
"BJAg3e7ZM",
"HJ99sdVbG",
"rJrjsrIkf",
"rk2BBWM1z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public"
] | [
"This is a fine paper that generally reads as a new episode in a series on motion-based video prediction with an eye towards robotic manipulation [Finn et al. 2016, Finn and Levine 2017, Ebert et al. 2017]. The work is rather incremental but is competently executed. It is in line with current trends in the research... | [
7,
3,
3,
3,
-1,
-1
] | [
5,
4,
5,
5,
-1,
-1
] | [
"iclr_2018_HJrJpzZRZ",
"iclr_2018_HJrJpzZRZ",
"iclr_2018_HJrJpzZRZ",
"iclr_2018_HJrJpzZRZ",
"rk2BBWM1z",
"iclr_2018_HJrJpzZRZ"
] |
iclr_2018_Hy8hkYeRb | A Deep Predictive Coding Network for Learning Latent Representations | It has been argued that the brain is a prediction machine that continuously learns how to make better predictions about the stimuli received from the external environment. For this purpose, it builds a model of the world around us and uses this model to infer the external stimulus. Predictive coding has been proposed a... | rejected-papers | The paper attempts to develop a method for learning latent representations using deep predictive coding and deconvolutional networks. However, the theoretical motivation for the proposed model in relation to existing methods (such as original predictive coding, deconvolutional networks, ladder networks, etc.), as well ... | train | [
"SJpl9FSlz",
"ryJ6sRYlf",
"HkrTEd9gf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"Quality\n\nThe authors introduce a deep network for predictive coding. It is unclear how the approach improves on the original predictive coding formulation of Rao and Ballard, who also use a hierarchy of transformations. The results seem to indicate that all layers are basically performing the same. No insight is... | [
4,
3,
3
] | [
4,
4,
5
] | [
"iclr_2018_Hy8hkYeRb",
"iclr_2018_Hy8hkYeRb",
"iclr_2018_Hy8hkYeRb"
] |
iclr_2018_ry4S90l0b | A Self-Training Method for Semi-Supervised GANs | Since the creation of Generative Adversarial Networks (GANs), much work has been done to improve their training stability, their generated image quality, their range of application but nearly none of them explored their self-training potential. Self-training has been used before the advent of deep learning in order to ... | rejected-papers | The paper presents self-training scheme for GANs. The proposed idea is simple but reasonable, and the experimental results show promise for MNIST and CIFAR10. However, the novelty of the proposed method seems relatively small and experimental results lack comparison against other stronger baselines (e.g., state-of-the-... | test | [
"SknKUAteG",
"S1e2kO9gG",
"r1UheZ6gG",
"B11QNrOfM",
"Sk7Azr_ff",
"rkoVGHOMf",
"r1e4YlmWG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper proposes to use self-training strategies for using unlabeled data in GAN. Experiments on only one data set, i.e., MNIST, are conducted \n\nPros:\n* Studying how to use unlabeled data to improve performance of GAN is of technical importance. The use of the self-training in GAN for exploiting unlabeled da... | [
3,
4,
3,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ry4S90l0b",
"iclr_2018_ry4S90l0b",
"iclr_2018_ry4S90l0b",
"SknKUAteG",
"S1e2kO9gG",
"r1UheZ6gG",
"iclr_2018_ry4S90l0b"
] |
iclr_2018_rJg4YGWRb | Attention-based Graph Neural Network for Semi-supervised Learning | Recently popularized graph neural networks achieve the state-of-the-art accuracy on a number of standard benchmark datasets for graph-based semi-supervised learning, improving significantly over existing approaches. These architectures alternate between a propagation layer that aggregates the hidden states of the local... | rejected-papers | A version of GCNs of Kipf and Welling is introduced with (1) no non-linearity; (2) a basic form of (softmax) attention over neighbors where the attention scores are computed as the cosine of endpoints' representations (scaled with a single learned scalar). There is a moderate improvement on Citeseer, Cora, Pubmed.
Sin... | train | [
"HJb-xDvHG",
"S1Z9bmyZf",
"rJmKbdIgM",
"HJvS2zhgz",
"S1rCoFiQz",
"Hk129YimG",
"H1w_FFiQz",
"HyJAutiQM",
"rk5cCabxf",
"H1buZ4xlz",
"rkWt6QPyG",
"rk8rhql1G"
] | [
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public",
"author",
"public"
] | [
"To add to the discussion, we would like to draw the attention to our recent paper accepted by AAAI-2018 as oral presentation (https://arxiv.org/abs/1801.07606). We have shown in our paper that the convolution layer of GCNs acts as \"Laplacian smoothing\" on the vertex features, which is the key reason why GCNs wor... | [
-1,
6,
6,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
3,
2,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"S1Z9bmyZf",
"iclr_2018_rJg4YGWRb",
"iclr_2018_rJg4YGWRb",
"iclr_2018_rJg4YGWRb",
"iclr_2018_rJg4YGWRb",
"S1Z9bmyZf",
"HJvS2zhgz",
"rJmKbdIgM",
"H1buZ4xlz",
"iclr_2018_rJg4YGWRb",
"rk8rhql1G",
"iclr_2018_rJg4YGWRb"
] |
iclr_2018_HJRV1ZZAW | FAST READING COMPREHENSION WITH CONVNETS | State-of-the-art deep reading comprehension models are dominated by recurrent
neural nets. Their sequential nature is a natural fit for language, but it also precludes
parallelization within an instances and often becomes the bottleneck for
deploying such models to latency critical scenarios. This is ... | rejected-papers | The key motivation for the work is producing both an efficient (parallelizable / fast) and accurate reading comprehension model. At least two reviewers are not convinced that this goal is really achieved (e.g., no comparison to hierarchical modeling, performance is not as strong). I also share concerns of R1 that, wi... | test | [
"SyyBD2Vxf",
"ry0IS1Kxz",
"H1xKadcgf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper borrows the idea from dilated CNN and proposes a dilated convolution based module for fast reading comprehension, in order to deal with the processing of very long documents in many reading comprehension tasks. The method part is clear and well-written. The results are fine when the idea is applied to t... | [
4,
7,
5
] | [
4,
3,
4
] | [
"iclr_2018_HJRV1ZZAW",
"iclr_2018_HJRV1ZZAW",
"iclr_2018_HJRV1ZZAW"
] |
iclr_2018_BJMuY-gRW | Jointly Learning Sentence Embeddings and Syntax with Unsupervised Tree-LSTMs | We introduce a neural network that represents sentences by composing their words according to induced binary parse trees. We use Tree-LSTM as our composition function, applied along a tree structure found by a fully differentiable natural language chart parser. Our model simultaneously optimises both the composition fu... | rejected-papers | Though the general direction is interesting and relevant to ICLR, the novelty is limited. As reviewers point out it is very similar to Le & Zuidema (2015), with few modifications (using LSTM word representations, a different type of pooling). However, it is not clear if they are necessary as there is no direct compari... | train | [
"HkMU7q5lf",
"HJwHbiqlG",
"SyTH-BCxz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper proposes to jointly learning a semantic objective and inducing a binary tree structure for word composition, which is similar to (Yogatama et al, 2017). Differently from (Yogatama et al, 2017), this paper doesn’t use reinforcement learning to induce a hard structure, but adopts a chart parser manner and... | [
5,
6,
4
] | [
4,
4,
4
] | [
"iclr_2018_BJMuY-gRW",
"iclr_2018_BJMuY-gRW",
"iclr_2018_BJMuY-gRW"
] |
iclr_2018_B1kIr-WRb | LEARNING SEMANTIC WORD RESPRESENTATIONS VIA TENSOR FACTORIZATION | Many state-of-the-art word embedding techniques involve factorization of a cooccurrence
based matrix. We aim to extend this approach by studying word embedding
techniques that involve factorization of co-occurrence based tensors (N-
way arrays). We present two new word embedding techniques based on te... | rejected-papers | The reviewers are concerned that the evaluation quality is not sufficient to convince readers that the proposed embedding method is indeed superior to alternatives. Though the authors attempted to address these comments in a subsequent revision but still, e.g., the evaluation is only intrinsic or on contrived problems.... | train | [
"HJR9ENn4G",
"r1aR0zMgf",
"Hy4sVp_lf",
"rkyEiZKef",
"HywV8xJNz",
"H1UafWrXG",
"SyX1tqVMG",
"r1wGD94GM",
"B1Qh854MM"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Hi Reviewer 3,\n\nWe see your point about needing to compare against another tensor method. While we are unaware of another embedding method that directly compares to ours (semantically-focused tensor factorization-based embedding), we see the utility in comparing against another embedding method that utilizes ten... | [
-1,
5,
5,
5,
-1,
-1,
-1,
-1,
-1
] | [
-1,
3,
5,
5,
-1,
-1,
-1,
-1,
-1
] | [
"HywV8xJNz",
"iclr_2018_B1kIr-WRb",
"iclr_2018_B1kIr-WRb",
"iclr_2018_B1kIr-WRb",
"SyX1tqVMG",
"iclr_2018_B1kIr-WRb",
"r1aR0zMgf",
"Hy4sVp_lf",
"rkyEiZKef"
] |
iclr_2018_H113pWZRb | Topology Adaptive Graph Convolutional Networks | Convolution acts as a local feature extractor in convolutional neural networks (CNNs). However, the convolution operation is not applicable when the input data is supported on an irregular graph such as with social networks, citation networks, or knowledge graphs. This paper proposes the topology adaptive graph convolu... | rejected-papers | The authors provide an extension to GCNs of Kipf and Welling in order to incorporate information about higher order neighborhoods. The extension is well motivated (and though I agree that it is not trivial modification of the K&W approach to the second order, thanks to the authors for the clarification). The improve... | train | [
"Hkqc-xL4z",
"ry6GiiKlz",
"H1kIb-Kef",
"r1XXuJcgM",
"rkSkNfj7G",
"ryXsYOBXz",
"Bk6GVtrmf",
"B1z3_FBXf",
"H1tKKKBXG",
"rJxc2drmG",
"B1kA7KrQz",
"S1-ItYrQM",
"SJvZYKHXG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"Thank you for pointing out that the Arxiv paper was updated recently. Please note that I did not mean to require that the two mentioned articles should have been referenced and discussed in your initial submission.\n\nI understand that there are technical differences between TAGCN and higher-order extensions of GC... | [
-1,
6,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
3,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"rJxc2drmG",
"iclr_2018_H113pWZRb",
"iclr_2018_H113pWZRb",
"iclr_2018_H113pWZRb",
"iclr_2018_H113pWZRb",
"r1XXuJcgM",
"iclr_2018_H113pWZRb",
"ryXsYOBXz",
"B1kA7KrQz",
"ry6GiiKlz",
"H1kIb-Kef",
"SJvZYKHXG",
"B1z3_FBXf"
] |
iclr_2018_rylejExC- | Stochastic Training of Graph Convolutional Networks | Graph convolutional networks (GCNs) are powerful deep neural networks for graph-structured data. However, GCN computes nodes' representation recursively from their neighbors, making the receptive field size grow exponentially with the number of layers. Previous attempts on reducing the receptive field size by subsampl... | rejected-papers | The paper studies subsampling techniques necessary to handle large graphs with graph convolutional networks. The paper introduces two ideas: (1) preprocessing for GCNs (basically replacing dropout followed by linear transformation with linear transformation followed by drop out); (2) adding control variates based on h... | train | [
"HkW14AGSz",
"B1jssYMHG",
"rJA4cxJlf",
"B1FrpdOeM",
"S1g5R5Ogz",
"rJ5i42a7M",
"BJz3D4TQf",
"rkC02767G",
"S1Cz67aQz",
"Hy-NnmTQz",
"SyLvq76Xf",
"HJu46VFxG"
] | [
"author",
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Thanks for you interest in our paper!\n\nFor CV (without +PP), the running time is 95.85 seconds, while CV+PP takes 56 seconds in Table 4. Our implementation does not support CVD without PP. The improvement of PP is not as large as CV, but is reasonable given its simplicity. Furthermore, some theoretical results o... | [
-1,
-1,
3,
4,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"B1jssYMHG",
"iclr_2018_rylejExC-",
"iclr_2018_rylejExC-",
"iclr_2018_rylejExC-",
"iclr_2018_rylejExC-",
"iclr_2018_rylejExC-",
"S1g5R5Ogz",
"B1FrpdOeM",
"rkC02767G",
"rJA4cxJlf",
"HJu46VFxG",
"iclr_2018_rylejExC-"
] |
iclr_2018_SkJKHMW0Z | Recurrent Relational Networks for complex relational reasoning | Humans possess an ability to abstractly reason about objects and their interactions, an ability not shared with state-of-the-art deep learning models. Relational networks, introduced by Santoro et al. (2017), add the capacity for relational reasoning to deep neural networks, but are limited in the complexity of the rea... | rejected-papers | The proposed relational reasoning algorithm is basically a fairly standard graph neural network, with a few modifications (e.g., the prediction loss at each layer - also not a new idea per se).
The claim that previously reasoning has not been considered in previous applications of graph neural networks (see discussio... | train | [
"r17v3MDxG",
"Syc551clG",
"rJFvPvqgz",
"HyLeDbvZf",
"r10uXbvbf",
"HymqZbDbz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper introduced recurrent relational network (RRNs), an enhanced version of the\nexisting relational network, that can be added to any neural networks to add\nrelational reasoning capacity. RRNs are illustrated on sudoku puzzles and textual QA.\n\nOverall the paper is well written and structured. It also addr... | [
3,
5,
5,
-1,
-1,
-1
] | [
5,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_SkJKHMW0Z",
"iclr_2018_SkJKHMW0Z",
"iclr_2018_SkJKHMW0Z",
"r17v3MDxG",
"Syc551clG",
"rJFvPvqgz"
] |
iclr_2018_ByquB-WC- | Finding ReMO (Related Memory Object): A Simple neural architecture for Text based Reasoning | Memory Network based models have shown a remarkable progress on the task of relational reasoning.
Recently, a simpler yet powerful neural network module called Relation Network (RN) has been introduced.
Despite its architectural simplicity, the time complexity of relation network grows quadratically with d... | rejected-papers | The contribution of this paper basically consists of using MLPs in the attention mechanism of end-2-end memory networks. Though it leads to some improvements on bAbI (which may not be so surprising - MLP attention has been shown preferable in certain scenarious), it does not seem to be a sufficient contribution. The mo... | train | [
"r1Z9q7Ygf",
"rk-hlXcez",
"SyuT1isxG",
"B1NJePCzM",
"ByV3YdaGf",
"HyCnDI6Gz",
"HyH5K8Tzf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper proposes to address the quadratic memory/time requirement of Relation Network (RN) by sequentially attending (via multiple layers) on objects and gating the object vectors with the attention weights of each layer. The proposed model obtains state of the art in bAbI story-based QA and bAbI dialog task.\n\... | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByquB-WC-",
"iclr_2018_ByquB-WC-",
"iclr_2018_ByquB-WC-",
"SyuT1isxG",
"rk-hlXcez",
"iclr_2018_ByquB-WC-",
"r1Z9q7Ygf"
] |
iclr_2018_rJBwoM-Cb | Neural Tree Transducers for Tree to Tree Learning | We introduce a novel approach to tree-to-tree learning, the neural tree transducer (NTT), a top-down depth first context-sensitive tree decoder, which is paired with recursive neural encoders. Our method works purely on tree-to-tree manipulations rather than sequence-to-tree or tree-to-sequence and is able to encode an... | rejected-papers | The proposed neural tree transduction framework is basically a combination of tree encoding and tree decoding. The tree encoding component is simply reused from previous work (TreeLSTM) whereas the decoding components is somewhat different from the previous work. They key problems (acknowledge also by at least 2 review... | train | [
"B1ueBCKeM",
"B1ISgaRez",
"B1BFRS7ZM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper introduces a neural tree decoder architecture for binary trees that conditions the next node prediction on \nrepresentations of its ascendants (encoded with an LSTM recurrent net) and left sibling subtree (encoded with a binary LSTM recursive net) for right sibling nodes. \nTo perform tree to tree tr... | [
3,
7,
2
] | [
4,
4,
5
] | [
"iclr_2018_rJBwoM-Cb",
"iclr_2018_rJBwoM-Cb",
"iclr_2018_rJBwoM-Cb"
] |
iclr_2018_S1sRrN-CW | Revisiting Knowledge Base Embedding as Tensor Decomposition | We study the problem of knowledge base (KB) embedding, which is usually addressed through two frameworks---neural KB embedding and tensor decomposition. In this work, we theoretically analyze the neural embedding framework and subsequently connect it with tensor based embedding. Specifically, we show that in neural KB ... | rejected-papers | The reviewers are not convinced by a number of aspects: including originality and clarity. Whereas the assessment of clarity and originality may be somewhat subjective (though the connections between margin-based loss and negative sampling is indeed well known), it is pretty clear that evaluation is very questionable. ... | train | [
"BytzlNjez",
"SyeeEtTef",
"BkwNvgRgf",
"r1S26N5lG",
"BkU9F4qlf",
"HJhXGX8AW",
"HkjsLOd0W",
"Hkw2SfERW"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public",
"author",
"public",
"public"
] | [
"The paper proposes a unified view of multiple methods for learning knowledge base embeddings.\n\nThe paper's motivations are interesting but the execution does fit standard for a publication at ICLR.\nMain reasons:\n* Section 3 does not bring much value. It is a rewriting trick that many knew but never thought of ... | [
3,
5,
3,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1sRrN-CW",
"iclr_2018_S1sRrN-CW",
"iclr_2018_S1sRrN-CW",
"HJhXGX8AW",
"HkjsLOd0W",
"Hkw2SfERW",
"Hkw2SfERW",
"iclr_2018_S1sRrN-CW"
] |
iclr_2018_SJ71VXZAZ | Learning To Generate Reviews and Discovering Sentiment | We explore the properties of byte-level recurrent language models. When given sufficient amounts of capacity, training data, and compute time, the representations learned by these models include disentangled features corresponding to high-level concepts. Specifically, we find a single unit which performs sentiment anal... | rejected-papers | The paper reports experiments where a LSTM language model is pretrained on a large corpus of reviews, and then the produced representation is used within a classifier on a number of sentiment classification datasets. The relative success of the method is not surprising. The novelty is very questionable, the writing qu... | val | [
"SJXeNaYlM",
"Sk7lrK5ez",
"ryrN2K9gf",
"HJABp-xyM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public"
] | [
"The authors propose to use a byte level RNN to classify reviews. In the meantime, they learn to generate reviews. The authors rely on the multiplicative LSTM proposed by Krause et al. 2016, a generative model predicting the next byte. They apply this architecture on the same task as the original article: document ... | [
4,
2,
4,
-1
] | [
3,
5,
5,
-1
] | [
"iclr_2018_SJ71VXZAZ",
"iclr_2018_SJ71VXZAZ",
"iclr_2018_SJ71VXZAZ",
"iclr_2018_SJ71VXZAZ"
] |
iclr_2018_S1XXq6lRW | Zero-shot Cross Language Text Classification | Labeled text classification datasets are typically only available in a few select languages. In order to train a model for e.g news categorization in a language Lt without a suitable text classification dataset there are two options. The first option is to create a new labeled dataset by hand, and the second option is ... | rejected-papers | Unfortunately, it falls short of ICLR standards -- from evaluation, novelty and clarity perspectives. The method is also not discussed in all details. | train | [
"HkGqBf5ef",
"r1-k8XqxG",
"Bk81W32lf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper proposes a language independent text encoding method for cross-language classification. The proposed approach demonstrates better performance than machine translation based classifier. \n\nThe proposed approach performs language independent common representation learning for cross-lingual text classifi... | [
4,
2,
3
] | [
3,
4,
4
] | [
"iclr_2018_S1XXq6lRW",
"iclr_2018_S1XXq6lRW",
"iclr_2018_S1XXq6lRW"
] |
iclr_2018_BkM27IxR- | Learning to Optimize Neural Nets | Learning to Optimize is a recently proposed framework for learning optimization algorithms using reinforcement learning. In this paper, we explore learning an optimization algorithm for training shallow neural nets. Such high-dimensional stochastic optimization problems present interesting challenges for existing reinf... | rejected-papers | The presented work is a good attempt to expand the work of Li and Malik to the high-dimensional, stochastic setting. Given the reviewer comments, I think the paper would benefit from highlighting the comparatively novel aspects, and in particular doing so earlier in the paper.
It is very important, given the nature of... | train | [
"Bynx0wHgM",
"BydHF89lz",
"Skd5kh5ef",
"HJt6cPaQz",
"HyZeqPTmM",
"HJ4qKvpmM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"[Main comments]\n\n* I would advice the authors to explain in more details in the intro\nwhat's new compared to Li & Malik (2016) and Andrychowicz et al. (2016).\nIt took me until section 3.5 to figure it out.\n\n* If I understand correctly, the only new part compared to Li & Malik (2016) is\nsection 3.5, where bl... | [
5,
6,
6,
-1,
-1,
-1
] | [
3,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_BkM27IxR-",
"iclr_2018_BkM27IxR-",
"iclr_2018_BkM27IxR-",
"Bynx0wHgM",
"BydHF89lz",
"Skd5kh5ef"
] |
iclr_2018_ByuP8yZRb | Censoring Representations with Multiple-Adversaries over Random Subspaces | Adversarial feature learning (AFL) is one of the promising ways for explicitly constrains neural networks to learn desired representations; for example, AFL could help to learn anonymized representations so as to avoid privacy issues. AFL learn such a representations by training the networks to deceive the adversary th... | rejected-papers | The reviewers tend to agree that the empirical results in this paper are good compared to the baselines. However, the paper in its current form is considered a bit too incremental. Some reviewers also suggested additional theory could help strengthen the paper. | train | [
"B143HDlWM",
"SJtlwgqlf",
"rJNERHjlf",
"By_NEbamM",
"BJHdYMX-z",
"B1S0ipGbz",
"rJ6ncpMWM",
"HyuIqafZG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"The below review addresses the first revision of the paper. The revised version does address my concerns. The fact that the paper does not come with substantial theoretical contributions/justification still stands out.\n\n---\n\nThe authors present a variant of the adversarial feature learning (AFL) approach by Ed... | [
6,
5,
6,
-1,
-1,
-1,
-1,
-1
] | [
3,
4,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByuP8yZRb",
"iclr_2018_ByuP8yZRb",
"iclr_2018_ByuP8yZRb",
"iclr_2018_ByuP8yZRb",
"B1S0ipGbz",
"B143HDlWM",
"rJNERHjlf",
"SJtlwgqlf"
] |
iclr_2018_SJzMATlAZ | Deep Continuous Clustering | Clustering high-dimensional datasets is hard because interpoint distances become less informative in high-dimensional spaces. We present a clustering algorithm that performs nonlinear dimensionality reduction and clustering jointly. The data is embedded into a lower-dimensional space by a deep autoencoder. The autoenco... | rejected-papers | After careful consideration, I think that this paper in its current form is just under the threshold for acceptance. Please note that I did take into account the comments, including the reviews and rebuttals, noting where arguments may be inconsistent or misleading.
The paper is a promising extension of RCC, albeit to... | train | [
"S1ruhov4G",
"SyKQPRLEM",
"HJe0S9VEG",
"SyqWgxzxf",
"H1ySNZVgf",
"HJ90m_PeG",
"ByAn4R6XG",
"SJKJMO67z",
"SkV2zAMmG",
"SkhYD5bGG",
"HyxSDcZfG",
"Hy53I9-Mz",
"H1gdIc-zz"
] | [
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"No, argument about canonical ACC definition is not my focus. Now it reads that the advantages claimed in the paper is conditional on the choice of clustering performance measure. Even if AMI is used, it is still hard to convince that the proposed method brings significant improvement because the authors refuse to ... | [
-1,
-1,
-1,
6,
3,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
3,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"SyKQPRLEM",
"HJe0S9VEG",
"Hy53I9-Mz",
"iclr_2018_SJzMATlAZ",
"iclr_2018_SJzMATlAZ",
"iclr_2018_SJzMATlAZ",
"SJKJMO67z",
"SkV2zAMmG",
"iclr_2018_SJzMATlAZ",
"iclr_2018_SJzMATlAZ",
"SyqWgxzxf",
"H1ySNZVgf",
"HJ90m_PeG"
] |
iclr_2018_ryjw_eAaZ | Unsupervised Deep Structure Learning by Recursive Dependency Analysis | We introduce an unsupervised structure learning algorithm for deep, feed-forward, neural networks. We propose a new interpretation for depth and inter-layer connectivity where a hierarchy of independencies in the input distribution is encoded in the network structure. This results in structures allowing neurons to conn... | rejected-papers | The updated draft has helped to address some of the issues that the reviewers had, however the reviewers believe there are still outstanding issues. With regard to the technical flaw, one reviewer has pointed out that the update changes the story of the paper by breaking the connection between the generative and discri... | train | [
"ryilanteG",
"HJZz1Wqef",
"SJGyhgwZz",
"S1DioOtMf",
"SykdYpPbz",
"rJqw9TDbz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper proposes an unsupervised structure learning method for deep neural networks. It first constructs a fully visible DAG by learning from data, and decomposes variables into autonomous sets. Then latent variables are introduced and stochastic inverse is generated. Later a deep neural network structure is con... | [
4,
5,
5,
-1,
-1,
-1
] | [
4,
2,
3,
-1,
-1,
-1
] | [
"iclr_2018_ryjw_eAaZ",
"iclr_2018_ryjw_eAaZ",
"iclr_2018_ryjw_eAaZ",
"SJGyhgwZz",
"HJZz1Wqef",
"ryilanteG"
] |
iclr_2018_rk9kKMZ0- | LEAP: Learning Embeddings for Adaptive Pace | Determining the optimal order in which data examples are presented to Deep Neural Networks during training is a non-trivial problem. However, choosing a non-trivial scheduling method may drastically improve convergence. In this paper, we propose a Self-Paced Learning (SPL)-fused Deep Metric Learning (DML) framework, wh... | rejected-papers | Although paper has been improved with new quantitative results and additional clarity, the reviewers agree though that larger-scale experiments would better highlight the utility of the method. There are some concerns with computational cost, despite the fact that the two networks are trained asynchronously. A baseline... | train | [
"ry9RWezWM",
"S1p86uteG",
"Byjs3NyZz",
"rJRhbuTXf",
"r1UD-d6mf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author"
] | [
"The authors purpose a method for creating mini batches for a student network by using a second learned representation space to dynamically selecting examples by their 'easiness and true diverseness'. The framework is detailed and results on MNIST, cifar10 and fashion-MNIST are presented. The work presented is nov... | [
6,
3,
4,
-1,
-1
] | [
3,
4,
4,
-1,
-1
] | [
"iclr_2018_rk9kKMZ0-",
"iclr_2018_rk9kKMZ0-",
"iclr_2018_rk9kKMZ0-",
"iclr_2018_rk9kKMZ0-",
"iclr_2018_rk9kKMZ0-"
] |
iclr_2018_SkFEGHx0Z | Nearest Neighbour Radial Basis Function Solvers for Deep Neural Networks | We present a radial basis function solver for convolutional neural networks that can be directly applied to both distance metric learning and classification problems. Our method treats all training features from a deep neural network as radial basis function centres and computes loss by summing the influence of a featu... | rejected-papers | This paper proposes a non-parametric method for metric learning and classification. One of the reviewers points out that it can be viewed as an extension of NCA. There is in fact a non-linear version of NCA that was subsequently published, see [1]. In this sense, the approach here appears to be a version of nonlinear N... | train | [
"r1jeC_Kgf",
"BJ1FJQ5lG",
"SkzzOIcez",
"SyHYEUjGz",
"ryz17UoMf",
"SJnwxLjzz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"(Summary)\nThis paper proposes weighted RBF distance based loss function where embeddings for cluster centroids and data are learned and used for class probabilities (eqn 3). The authors experiment on CUB200-2011, Cars106, Oxford 102 Flowers datasets.\n\n(Pros)\nThe citations and related works cover fairly compreh... | [
5,
3,
4,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_SkFEGHx0Z",
"iclr_2018_SkFEGHx0Z",
"iclr_2018_SkFEGHx0Z",
"r1jeC_Kgf",
"BJ1FJQ5lG",
"SkzzOIcez"
] |
iclr_2018_rJ695PxRW | Discovering Order in Unordered Datasets: Generative Markov Networks | The assumption that data samples are independently identically distributed is the backbone of many learning algorithms. Nevertheless, datasets often exhibit rich structures in practice, and we argue that there exist some unknown orders within the data instances. Aiming to find such orders, we introduce a novel Generati... | rejected-papers | The problem of discovering ordering in an unordered dataset is quite interesting, and the authors have outlined a few potential applications. However, the reviewer consensus is that this draft is too preliminary for acceptance. The main issues were clarity, lack of quantitative results for the order discovery experimen... | train | [
"B1ySxEolG",
"Hy97waqxM",
"r1bzglTgG",
"rksTsaYfM",
"ryXIsptMG",
"HJFOc6tfz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"\nThe authors deal with the problem of implicit ordering in a dataset and the challenge of recovering it, i.e. when given a random dataset with no explicit ordering in the samples, the model is able to recover an ordering. They propose to learn a distance-metric-free model that assumes a Markov chain as the genera... | [
4,
4,
4,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_rJ695PxRW",
"iclr_2018_rJ695PxRW",
"iclr_2018_rJ695PxRW",
"Hy97waqxM",
"B1ySxEolG",
"r1bzglTgG"
] |
iclr_2018_SyuWNMZ0W | Directing Generative Networks with Weighted Maximum Mean Discrepancy | The maximum mean discrepancy (MMD) between two probability measures P
and Q is a metric that is zero if and only if all moments of the two measures
are equal, making it an appealing statistic for two-sample tests. Given i.i.d. samples
from P and Q, Gretton et al. (2012) show that we can construct an u... | rejected-papers | The reviewers agree that the problem being addressed is interesting, however there are concerns with novelty and with the experimental results. An experiment beyond dealing with class imbalance would help strengthen this paper, as would experiments with other kinds of GANs. | train | [
"H15HuyMlz",
"SJ0Oxotlf",
"B1X0w52xz",
"H1NHCc3-f",
"HJckC9h-M",
"Bk-Npc2WG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper proposes an importance-weighted estimator of the MMD, in order to estimate the MMD between distributions based on samples biased according to a known scheme. It then discusses how to estimate the scheme when it is unknown, and further proposes using it in either the MMD-based generative models of Y. Li ... | [
4,
4,
4,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1
] | [
"iclr_2018_SyuWNMZ0W",
"iclr_2018_SyuWNMZ0W",
"iclr_2018_SyuWNMZ0W",
"H15HuyMlz",
"SJ0Oxotlf",
"B1X0w52xz"
] |
iclr_2018_HydnA1WCb | Gaussian Prototypical Networks for Few-Shot Learning on Omniglot | We propose a novel architecture for k-shot classification on the Omniglot dataset. Building on prototypical networks, we extend their architecture to what we call Gaussian prototypical networks. Prototypical networks learn a map between images and embedding vectors, and use their clustering for classification. In our m... | rejected-papers | The reviewers agree that the idea of utilizing covariance information in the few-shot setting is interesting. There are concerns with the novelty of the paper, as well as the correctness in terms of ensuring the covariance matrix is PSD in all cases. There are some concerns with the experimental evaluation as well. In ... | train | [
"rkHVojvez",
"r1LJyjOlM",
"BJ9tT6Fxz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"This paper presents an interesting extension to Snell et al.'s prototypical networks, by introducing uncertainty through a parameterised estimation of covariance along side the image embeddings (means). Uncertainty may be particularly important in the few-shot learning case this paper examines, when it is helpful ... | [
4,
3,
3
] | [
4,
4,
4
] | [
"iclr_2018_HydnA1WCb",
"iclr_2018_HydnA1WCb",
"iclr_2018_HydnA1WCb"
] |
iclr_2018_ryH_bShhW | DOUBLY STOCHASTIC ADVERSARIAL AUTOENCODER | Any autoencoder network can be turned into a generative model by imposing an arbitrary prior distribution on its hidden code vector. Variational Autoencoder uses a KL divergence penalty to impose the prior, whereas Adversarial Autoencoder uses generative adversarial networks. A straightforward modification of Adversar... | rejected-papers | The reviewers all outlined concerns regarding novelty and the maturity of this work. It would be helpful to clarify the relation to doubly stochastic kernel machines as opposed to random kitchen sinks, and to provide more insight into how this stochasticity helps. Finally, the approach should be tried on more difficult... | val | [
"By7B42BxM",
"B1BsWE9lM",
"BJQGTw5lM",
"Hk9p9pV-G",
"BJxDZRVZM",
"SkJ9YnNbz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public",
"public"
] | [
"Thank you for the feedback, and I have read it.\n\nThe authors claimed that they used techniques in [6] in which I am not an expert for this. However I cannot find the comparison that the authors mentioned in the feedback, so I am not sure if the claim is true.\n\nI still recommend rejection for the paper, and as ... | [
3,
3,
2,
-1,
-1,
-1
] | [
4,
5,
5,
-1,
-1,
-1
] | [
"iclr_2018_ryH_bShhW",
"iclr_2018_ryH_bShhW",
"iclr_2018_ryH_bShhW",
"B1BsWE9lM",
"By7B42BxM",
"BJQGTw5lM"
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.