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_S19dR9x0b | Alternating Multi-bit Quantization for Recurrent Neural Networks | Recurrent neural networks have achieved excellent performance in many applications. However, on portable devices with limited resources, the models are often too large to deploy. For applications on the server with large scale concurrent requests, the latency during inference can also be very critical for costly comput... | accepted-poster-papers | The reviewers unanimously agree that this paper is worth publication at ICLR. Please address the feedback of the reviewers and discuss exactly how the potential speed up rates are computed in the appendix. I speed up rates to be different for different devices. | train | [
"HyOWIZjeM",
"r1pQDxEZf",
"HJ-WByrVG",
"BJz5LyclM",
"rJr11K2mz",
"Bkbq0_2Qf",
"rkgh3_n7f",
"ryQ5eXRzM",
"rJFXe7Czf",
"rJUq0zAGz",
"SJh8yXRMG",
"SJ6WAf0MM",
"S1PH3zAMM",
"HyeXldDZG",
"rkaECW7bz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"I have read the comments and clarifications from the authors. They have added extra experiments, and clarified the speed-ups concern raised by others. I keep my original rating of the paper.\n\n---------------\nORIGINAL REVIEW:\n\nThis paper introduces a multi-bit quantization method for recurrent neural networks,... | [
8,
7,
-1,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
-1,
2,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S19dR9x0b",
"iclr_2018_S19dR9x0b",
"Bkbq0_2Qf",
"iclr_2018_S19dR9x0b",
"SJ6WAf0MM",
"SJh8yXRMG",
"iclr_2018_S19dR9x0b",
"HyeXldDZG",
"rkaECW7bz",
"HyOWIZjeM",
"BJz5LyclM",
"r1pQDxEZf",
"iclr_2018_S19dR9x0b",
"iclr_2018_S19dR9x0b",
"iclr_2018_S19dR9x0b"
] |
iclr_2018_HJNMYceCW | Residual Loss Prediction: Reinforcement Learning With No Incremental Feedback | We consider reinforcement learning and bandit structured prediction problems with very sparse loss feedback: only at the end of an episode. We introduce a novel algorithm, RESIDUAL LOSS PREDICTION (RESLOPE), that solves such problems by automatically learning an internal representation of a denser reward function. RESL... | accepted-poster-papers | The reviewers agree that the problem of learning learning credit assignment from terminal rewards is interesting, and that the presented approach is promising. There are some concerns regarding the rigor and correctness of the theoretical results, and I ask the authors to improve those aspects of the paper. I also ask ... | train | [
"r14M3-KxM",
"S1BCCpjrM",
"H159Cpirz",
"ryRaK9sHM",
"HJ3M8diHM",
"ByCeFUNgz",
"r1ajkbceM",
"rJeBU_a7M",
"HkZ05YT7z",
"Bk6bEOT7f",
"HkjWXd67G",
"Hy1aVOp7z",
"Bkcg5uT7z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"After reading the other reviews and the authors' responses, I am satisfied that this paper is above the accept threshold. I think there are many areas of further discussion that the authors can flesh out (as mentioned below and in other reviews), but overall the contribution seems solid. I also appreciate the r... | [
7,
-1,
-1,
-1,
-1,
7,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
-1,
-1,
-1,
-1,
2,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HJNMYceCW",
"Hy1aVOp7z",
"ryRaK9sHM",
"HJ3M8diHM",
"rJeBU_a7M",
"iclr_2018_HJNMYceCW",
"iclr_2018_HJNMYceCW",
"iclr_2018_HJNMYceCW",
"Bk6bEOT7f",
"r1ajkbceM",
"ByCeFUNgz",
"r14M3-KxM",
"r14M3-KxM"
] |
iclr_2018_SyOK1Sg0W | Adaptive Quantization of Neural Networks | Despite the state-of-the-art accuracy of Deep Neural Networks (DNN) in various classification problems, their deployment onto resource constrained edge computing devices remains challenging due to their large size and complexity. Several recent studies have reported remarkable results in reducing this complexity throug... | accepted-poster-papers | Given the changes to the paper, the reviewers agree that the paper meets the bar for publication at ICLR. There are some concerns regarding the practical impact on CPUs and GPUs. I ask the authors to clearly discuss the impact on different hardware. One can argue if adaptive quantization techniques are helpful, then th... | train | [
"rkIpwEslz",
"Hkcb6tG-M",
"HkrUKW5eM",
"B1ck33hXM",
"BkFMZah7f",
"SJxPAhhXG",
"BJSHpnnmM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"I have read the responses to the concerns raised by all reviewers. I find the clarifications and modifications satisfying, therefore I keep my rating of the paper to above acceptance threshold.\n\n-----------------\nORIGINAL REVIEW:\n\nThe paper proposes a method for quantizing neural networks that allows weights ... | [
6,
6,
6,
-1,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SyOK1Sg0W",
"iclr_2018_SyOK1Sg0W",
"iclr_2018_SyOK1Sg0W",
"Hkcb6tG-M",
"HkrUKW5eM",
"rkIpwEslz",
"B1ck33hXM"
] |
iclr_2018_BkUp6GZRW | Boosting the Actor with Dual Critic | This paper proposes a new actor-critic-style algorithm called Dual Actor-Critic or Dual-AC. It is derived in a principled way from the Lagrangian dual form of the Bellman optimality equation, which can be viewed as a two-player game between the actor and a critic-like function, which is named as dual critic. Compared... | accepted-poster-papers | All of the reviewers agree that the paper clearly presents promising ideas in developing a novel actor critic algorithm. The experiments do not show a significant gain against the baselines, but they support the presented ideas. I appreciated the ablation study on dual-AC.
Detailed comments:
My understanding is that t... | train | [
"HkJ6DWtgf",
"Bysjjx5lG",
"Hyu5lW5xf",
"B1Px5vamf",
"Byg6DR5QM",
"Byg5DCqQM",
"S1iWD09Qz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper studies a new architecture DualAC. The author give strong and convincing justifications based on the Lagrangian dual of the Bellman equation (although not new, introducing this as the justification for the architecture design is plausible).\n\nThere are several drawbacks of the current format of the pap... | [
7,
6,
5,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkUp6GZRW",
"iclr_2018_BkUp6GZRW",
"iclr_2018_BkUp6GZRW",
"iclr_2018_BkUp6GZRW",
"HkJ6DWtgf",
"Bysjjx5lG",
"Hyu5lW5xf"
] |
iclr_2018_BJk59JZ0b | Guide Actor-Critic for Continuous Control | Actor-critic methods solve reinforcement learning problems by updating a parameterized policy known as an actor in a direction that increases an estimate of the expected return known as a critic. However, existing actor-critic methods only use values or gradients of the critic to update the policy parameter. In this pa... | accepted-poster-papers | The reviewers agree that the formulation is novel and interesting, but they raised concerns regarding the motivation and the complexity of the approach. I find the authors' response mostly satisfying, and I ask them to improve the paper by incorporating the comments.
Detailed comments:
The maximum-entropy objective us... | train | [
"rJwNjgqef",
"Hk6bJG9gf",
"H1broNZ-G",
"r1_35T2Xz",
"B1pXz9jbz",
"Hy4vbqjWf",
"S1qXxqiWz",
"B11ykcs-M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"\n\nThe authors devise and explore use of the hessian of the\n(approximate/learned) value function (the critic) to update the policy\n(actor) in the actor-critic approach to RL. They connect their\ntechnique, 'guide actor-critic' or GAC, to existing actor-critic\nmethods (authors claim only two published work us... | [
6,
4,
7,
-1,
-1,
-1,
-1,
-1
] | [
2,
4,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJk59JZ0b",
"iclr_2018_BJk59JZ0b",
"iclr_2018_BJk59JZ0b",
"B11ykcs-M",
"H1broNZ-G",
"rJwNjgqef",
"Hk6bJG9gf",
"iclr_2018_BJk59JZ0b"
] |
iclr_2018_ByOnmlWC- | Policy Optimization by Genetic Distillation | Genetic algorithms have been widely used in many practical optimization problems.
Inspired by natural selection, operators, including mutation, crossover
and selection, provide effective heuristics for search and black-box optimization.
However, they have not been shown useful for deep reinforcement l... | accepted-poster-papers | At least two of the reviewers found the proposed approach novel and interesting and worthy of publication at ICLR. The reviewers raised concerns regarding the paper's terminology, which may lead to some misunderstanding. I agree that upon a quick skim, a reader may think that the paper performs the crossover operation ... | train | [
"By_t2wVlG",
"ByvABDcxz",
"BySF6I2xz",
"BkPw48aXz",
"BJ7VGn2Qz",
"rkJBdVtmM",
"H1SlvnZ7z",
"HkV9S2-7z",
"SksMHhbXf",
"SJCy42ZQf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This is a highly interesting paper that proposes a set of methods that combine ideas from imitation learning, evolutionary computation and reinforcement learning in a novel way. It combines the following ingredients:\na) a population-based setup for RL\nb) a pair-selection and crossover operator\nc) a policy-gradi... | [
8,
6,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByOnmlWC-",
"iclr_2018_ByOnmlWC-",
"iclr_2018_ByOnmlWC-",
"BJ7VGn2Qz",
"SksMHhbXf",
"H1SlvnZ7z",
"By_t2wVlG",
"ByvABDcxz",
"BySF6I2xz",
"iclr_2018_ByOnmlWC-"
] |
iclr_2018_SkA-IE06W | When is a Convolutional Filter Easy to Learn? | We analyze the convergence of (stochastic) gradient descent algorithm for learning a convolutional filter with Rectified Linear Unit (ReLU) activation function. Our analysis does not rely on any specific form of the input distribution and our proofs only use the definition of ReLU, in contrast with previous works that ... | accepted-poster-papers | Dear authors,
The reviewers all appreciated your work and agree that this a very good first step in an interesting direction. | train | [
"SJA4C8_gG",
"Sk7b2-tlz",
"By3jcVfMM",
"HkqdxcuQM",
"Skg9Q7X7M",
"BkJpqU0fM",
"BJ_w9URzz",
"ryNVqLCff"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper studies the problem of learning a single convolutional filter using SGD. The main result is: if the \"patches\" of the convolution are sufficiently aligned with each other, then SGD with a random initialization can recover the ground-truth parameter of a convolutional filter (single filter, ReLU, averag... | [
6,
9,
8,
-1,
-1,
-1,
-1,
-1
] | [
3,
4,
3,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SkA-IE06W",
"iclr_2018_SkA-IE06W",
"iclr_2018_SkA-IE06W",
"Skg9Q7X7M",
"BkJpqU0fM",
"SJA4C8_gG",
"Sk7b2-tlz",
"By3jcVfMM"
] |
iclr_2018_BkrsAzWAb | Online Learning Rate Adaptation with Hypergradient Descent | We introduce a general method for improving the convergence rate of gradient-based optimizers that is easy to implement and works well in practice. We demonstrate the effectiveness of the method in a range of optimization problems by applying it to stochastic gradient descent, stochastic gradient descent with Nesterov... | accepted-poster-papers | All reviewers agreed that, despite the lack of novelty, the proposed method is sound and correctly linked to existing work. As the topic of automatically learning the stepsize is of great practical interest, I am glad to have this paper presented as a poster at ICLR. | train | [
"H1pbs28kG",
"r1jLC23Jf",
"BJ6v0V9ef",
"Hy8WTMFmf",
"S1WP2GFQz",
"HJcZnfFXM",
"H1PN17VXz",
"B1wQhWzGM",
"BydQzcHxf",
"S1sZVWMlz",
"rkaXMT-lz",
"r1HU3l1kf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"author",
"author",
"public",
"public"
] | [
"SUMMARY:\n\nThe authors reinvent a 20 years old technique for adapting a global or component-wise learning rate for gradient descent. The technique can be derived as a gradient step for the learning rate hyperparameter, or it can be understood as a simple and efficient adaptation technique.\n\n\nGENERAL IMPRESSION... | [
6,
7,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkrsAzWAb",
"iclr_2018_BkrsAzWAb",
"iclr_2018_BkrsAzWAb",
"H1pbs28kG",
"r1jLC23Jf",
"BJ6v0V9ef",
"B1wQhWzGM",
"iclr_2018_BkrsAzWAb",
"r1HU3l1kf",
"rkaXMT-lz",
"iclr_2018_BkrsAzWAb",
"iclr_2018_BkrsAzWAb"
] |
iclr_2018_HyWrIgW0W | Stochastic gradient descent performs variational inference, converges to limit cycles for deep networks | Stochastic gradient descent (SGD) is widely believed to perform implicit regularization when used to train deep neural networks, but the precise manner in which this occurs has thus far been elusive. We prove that SGD minimizes an average potential over the posterior distribution of weights along with an entropic regul... | accepted-poster-papers | Dear authors,
Based on the comments and your rebuttal, I am glad to accept your paper at ICLR. | val | [
"HkJG6iOlM",
"B1PK_0tgf",
"Bkic0BclM",
"S1_hIt67f",
"B11u4C3ZM",
"HJYXSAh-G",
"BJ-c702Zz",
"r1yXa6QWM",
"H1vZPmg-G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"The paper takes a closer look at the analysis of SGD as variational inference, first proposed by Duvenaud et al. 2016\nand Mandt et al. 2016. In particular, the authors point out that in general, SGD behaves quite differently from Langevin diffusion due to the multivariate nature of the Gaussian noise. As the auth... | [
8,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HyWrIgW0W",
"iclr_2018_HyWrIgW0W",
"iclr_2018_HyWrIgW0W",
"iclr_2018_HyWrIgW0W",
"HkJG6iOlM",
"B1PK_0tgf",
"Bkic0BclM",
"H1vZPmg-G",
"iclr_2018_HyWrIgW0W"
] |
iclr_2018_ByrZyglCb | Robustness of Classifiers to Universal Perturbations: A Geometric Perspective | Deep networks have recently been shown to be vulnerable to universal perturbations: there exist very small image-agnostic perturbations that cause most natural images to be misclassified by such classifiers. In this paper, we provide a quantitative analysis of the robustness of classifiers to universal perturbations, a... | accepted-poster-papers | The idea of universal perturbation is definitely interesting and well carried out in that paper. | train | [
"H17poxceM",
"SygwaSixG",
"ByJeL6EWz",
"S154AURbG",
"ryFLaI0ZM",
"Sy82jUCWz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper is written well and clear. The core contribution of the paper is the illustration that: under the assumption of flat, or curved decision boundaries with positive curvature small universal adversarial perturbations exist. \n\nPros: the intuition and geometry is rather clearly presented. \n\nCons: \nRe... | [
6,
5,
7,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_ByrZyglCb",
"iclr_2018_ByrZyglCb",
"iclr_2018_ByrZyglCb",
"H17poxceM",
"SygwaSixG",
"ByJeL6EWz"
] |
iclr_2018_B1hYRMbCW | On the regularization of Wasserstein GANs | Since their invention, generative adversarial networks (GANs) have become a popular approach for learning to model a distribution of real (unlabeled) data. Convergence problems during training are overcome by Wasserstein GANs which minimize the distance between the model and the empirical distribution in terms of a dif... | accepted-poster-papers | This paper proposes an interesting analysis of the limitations of WGANs as well as a solution to these limitations. I am not too convinced by the experimental part as, as some of the reviewers have mentioned, it relies on hyperparameters which can be hard to tune.
The more theoretical part, even if it could be written... | train | [
"ry2WdrtgM",
"r1aAjU_xG",
"ry1j9wpgz",
"BJIh7OpmM",
"Hy7NMda7f",
"BJJMfOp7M",
"r1z7Z_p7z",
"SyheZOTmz",
"r1MZlOaXM",
"HkEdTwa7z",
"r1iQxlMGM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper proposes a novel regularization scheme for Wasserstein GAN based on a relaxation of the constraints on the Lipschitz constant of 1. The proposed regularization penalize the critic function only when its gradient has a norm larger than one using some kind of squared hinge loss. The reasons for this choic... | [
7,
2,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
2,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_B1hYRMbCW",
"iclr_2018_B1hYRMbCW",
"iclr_2018_B1hYRMbCW",
"r1iQxlMGM",
"BJJMfOp7M",
"ry2WdrtgM",
"SyheZOTmz",
"r1MZlOaXM",
"r1aAjU_xG",
"ry1j9wpgz",
"iclr_2018_B1hYRMbCW"
] |
iclr_2018_Bk8ZcAxR- | Eigenoption Discovery through the Deep Successor Representation | Options in reinforcement learning allow agents to hierarchically decompose a task into subtasks, having the potential to speed up learning and planning. However, autonomously learning effective sets of options is still a major challenge in the field. In this paper we focus on the recently introduced idea of using repre... | accepted-poster-papers | This paper on automatic option discovery connects recent research on successor representations with eigenoptions. This is a solidly presented, conceptual paper with results in tabular and atari environments. | train | [
"BJlGRSOgf",
"BysvRfjez",
"HJagrMk-G",
"SJUvQG1Gf",
"Bk-5Mf1fG",
"rysyGMyff"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper extends the idea of eigenoptions, recently proposed by Machado et al. to domains with stochastic transitions and where state features are learned. An eigenoption is defined as an optimal policy for a reward function defined by an eigenvector of the matrix of successor representation (SR), which is an occ... | [
6,
9,
7,
-1,
-1,
-1
] | [
3,
5,
4,
-1,
-1,
-1
] | [
"iclr_2018_Bk8ZcAxR-",
"iclr_2018_Bk8ZcAxR-",
"iclr_2018_Bk8ZcAxR-",
"BJlGRSOgf",
"BysvRfjez",
"HJagrMk-G"
] |
iclr_2018_Bk9zbyZCZ | Neural Map: Structured Memory for Deep Reinforcement Learning | A critical component to enabling intelligent reasoning in partially observable environments is memory. Despite this importance, Deep Reinforcement Learning (DRL) agents have so far used relatively simple memory architectures, with the main methods to overcome partial observability being either a temporal convolution ov... | accepted-poster-papers | Biological memory systems are grounded in spatial representation and spatial memory, so neural methods for spatial memory are highly interesting. The proposed method is novel, well-designed and the empirical results are good on unseen environments, although the noise model may be too weak. Moreover, it would have been ... | train | [
"ByJWAeFxz",
"S1Ii7lcxz",
"H1E1RgqxM",
"rkoNnT9mG",
"BkFXGRy7M",
"SkXZVqqfG",
"Bkr07cqzz",
"H1cqz5cGz",
"H1ThWc5Mf",
"H1Mke9cMM",
"H1LfT3wZf",
"r151wyLbz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"The paper introduces a new memory mechanism specifically tailored for agent navigation in 2D environments. The memory consists of a 2D array and includes trainable read/write mechanisms. The RL agent's policy is a function of the context read, read, and next step write vectors (which are functions of the observati... | [
7,
9,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Bk9zbyZCZ",
"iclr_2018_Bk9zbyZCZ",
"iclr_2018_Bk9zbyZCZ",
"ByJWAeFxz",
"H1cqz5cGz",
"r151wyLbz",
"H1LfT3wZf",
"S1Ii7lcxz",
"H1E1RgqxM",
"ByJWAeFxz",
"iclr_2018_Bk9zbyZCZ",
"iclr_2018_Bk9zbyZCZ"
] |
iclr_2018_ry6-G_66b | Active Neural Localization | Localization is the problem of estimating the location of an autonomous agent from an observation and a map of the environment. Traditional methods of localization, which filter the belief based on the observations, are sub-optimal in the number of steps required, as they do not decide the actions taken by the agent. W... | accepted-poster-papers | The paper proposes a neural net based method for active localization in a known map using a learnt perception model (convnet) and a learnt control policy combined with a set belief state representation. The method compares well to baselines and has good accuracy in 2d and 3d envs. All three reviewers are in favor of ac... | train | [
"rJ74wm5xM",
"S1a6mx5xM",
"BJovaI9gf",
"S1Zvx0Jmf",
"SJsY_I3MG",
"r1tmOU3MM",
"HJphLU2MG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper describes a neural network-based approach to active localization based upon RGB images. The framework employs Bayesian filtering to maintain an estimate of the agent's pose using a convolutional network model for the measurement (perception) function. A convolutional network models the policy that govern... | [
6,
8,
7,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ry6-G_66b",
"iclr_2018_ry6-G_66b",
"iclr_2018_ry6-G_66b",
"SJsY_I3MG",
"S1a6mx5xM",
"rJ74wm5xM",
"BJovaI9gf"
] |
iclr_2018_B1al7jg0b | Overcoming Catastrophic Interference using Conceptor-Aided Backpropagation | Catastrophic interference has been a major roadblock in the research of continual learning. Here we propose a variant of the back-propagation algorithm, "Conceptor-Aided Backprop" (CAB), in which gradients are shielded by conceptors against degradation of previously learned tasks. Conceptors have their origin in reserv... | accepted-poster-papers | This paper is a timely application of linear algebra to propose a method for reducing catastrophic interference by training a new task in a subspace of the parameter space using conceptors. The conceptors are deployed in the backprop, making this a valuable alternative to recent continual learning methods such as EWC. ... | train | [
"rkY82b5lM",
"HkUpFYKeM",
"BJar6i8Vz",
"rkhi37_gz",
"H1hoeBkMG",
"HJ45oNJGf",
"Byej5NkGG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The paper leaves me guessing which part is a new contribution, and which one is already possible with conceptors as described in the Jaeger 2014 report. Figure (1) in the paper is identical to the one in the (short version of) the Jaeger report but is missing an explicit reference. Figure 2 is almost identical, ag... | [
7,
7,
-1,
7,
-1,
-1,
-1
] | [
5,
3,
-1,
3,
-1,
-1,
-1
] | [
"iclr_2018_B1al7jg0b",
"iclr_2018_B1al7jg0b",
"Byej5NkGG",
"iclr_2018_B1al7jg0b",
"rkY82b5lM",
"rkhi37_gz",
"HkUpFYKeM"
] |
iclr_2018_HyfHgI6aW | Memory Augmented Control Networks | Planning problems in partially observable environments cannot be solved directly with convolutional networks and require some form of memory. But, even memory networks with sophisticated addressing schemes are unable to learn intelligent reasoning satisfactorily due to the complexity of simultaneously learning to acces... | accepted-poster-papers | The authors have proposed an architecture that incorporates a VIN with a DNC to combine low level planning with high level memory-based optimization, resulting in a single policy for navigation and other similar problems that is trained end-to-end with sparse rewards. The reviews are mixed, but the authors did allay th... | train | [
"H1QljSQxz",
"HJBOB_oxf",
"r1IWuK2lf",
"ByRUdkx7f",
"r1pfOkgmf",
"Hk8IcbTbz",
"BJ7deMpZf",
"BJGZMyhWf",
"rkpxjJ-WM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"author"
] | [
"Summary:\n\nA method is proposed for robot navigation in partially observable scenarios. E.g. 2D navigation in a grid world from start to goal but the robot can only sense obstacles in a certain radius around it. A learning-based method is proposed here which takes the currently discovered partial map as input to ... | [
4,
6,
9,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
2,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HyfHgI6aW",
"iclr_2018_HyfHgI6aW",
"iclr_2018_HyfHgI6aW",
"BJGZMyhWf",
"r1IWuK2lf",
"H1QljSQxz",
"rkpxjJ-WM",
"iclr_2018_HyfHgI6aW",
"HJBOB_oxf"
] |
iclr_2018_B13njo1R- | Progressive Reinforcement Learning with Distillation for Multi-Skilled Motion Control | Deep reinforcement learning has demonstrated increasing capabilities for continuous control problems,
including agents that can move with skill and agility through their environment.
An open problem in this setting is that of developing good strategies for integrating or merging policies
for multiple... | accepted-poster-papers | The authors propose an architecture that uses a curriculum and multi-task distillation to gain higher performance without forgetting. The paper is largely a smart composition of known methods, and it requires keeping data from all tasks to do the distillation, so it is not truly a scalable continual learning approach. ... | test | [
"H1RMUzgBM",
"ByMViwPef",
"SJmZPJd4z",
"H1BAcZ9eG",
"ByFImz_ZG",
"BkSxcvaQM",
"HkY9Ln-7G",
"B1pGI2Zmf",
"ryByL2WmG"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"public",
"public"
] | [
"We updated figure 1 in the paper to show a diagram for the other added baselines in this work (fine-tuning).\nThe figures in the paper have been updated, showing the results over 5 runs of each method. The findings are very similar.\nTable 1 has also been updated with data for the MultiTasker. This table now bette... | [
-1,
7,
-1,
7,
5,
-1,
-1,
-1,
-1
] | [
-1,
4,
-1,
4,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_B13njo1R-",
"iclr_2018_B13njo1R-",
"ByMViwPef",
"iclr_2018_B13njo1R-",
"iclr_2018_B13njo1R-",
"iclr_2018_B13njo1R-",
"ByMViwPef",
"H1BAcZ9eG",
"ByFImz_ZG"
] |
iclr_2018_B1hcZZ-AW | N2N learning: Network to Network Compression via Policy Gradient Reinforcement Learning | While bigger and deeper neural network architectures continue to advance the state-of-the-art for many computer vision tasks, real-world adoption of these networks is impeded by hardware and speed constraints. Conventional model compression methods attempt to address this problem by modifying the architecture manually ... | accepted-poster-papers | This is a meta-learning approach to model compression which trains 2 policies using RL to reduce the capacity (computational cost) of a trained network while maintaining performance, such that it can be effectively transferred to a smaller student network. The approach has similarities to recently proposed methods for ... | train | [
"S1dzeGtxz",
"rJO3m40ef",
"B1MNG0Z-f",
"Sy0kfZPGz",
"SyKQ--DGG",
"SJlie-wMz",
"BkE3lbvff"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper proposes to use reinforcement learning instead of pre-defined heuristics to determine the structure of the compressed model in the knowledge distillation process.\n\nThe draft is well-written, and the method is clearly explained. However, I have the following concerns for this draft:\n\n1. The technical... | [
5,
9,
4,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_B1hcZZ-AW",
"iclr_2018_B1hcZZ-AW",
"iclr_2018_B1hcZZ-AW",
"B1MNG0Z-f",
"rJO3m40ef",
"S1dzeGtxz",
"S1dzeGtxz"
] |
iclr_2018_SJJQVZW0b | Hierarchical and Interpretable Skill Acquisition in Multi-task Reinforcement Learning | Learning policies for complex tasks that require multiple different skills is a major challenge in reinforcement learning (RL). It is also a requirement for its deployment in real-world scenarios. This paper proposes a novel framework for efficient multi-task reinforcement learning. Our framework trains agents to emplo... | accepted-poster-papers | This method has a lot of strong points, but the reviewers had concerns about baselines, comparisons, and hand-engineered aspects of the method. The authors gave a strong rebuttal and made substantial updates to the paper to address the concerns. I think that this has saved the submission and tipped the balance towards ... | test | [
"Sye2eNDxM",
"rJWf00wEf",
"S1hvkpKxf",
"HJQ8haFxz",
"H1lZ5eo7z",
"rJScuximM",
"H1GX9limf",
"BkURFxoQz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper aims to learn hierarchical policies by using a recursive policy structure regulated by a stochastic temporal grammar. The experiments show that the method is better than a flat policy for learning a simple set of block-related skills in minecraft (find, get, put, stack) and generalizes better to a modif... | [
6,
-1,
6,
6,
-1,
-1,
-1,
-1
] | [
4,
-1,
3,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJJQVZW0b",
"Sye2eNDxM",
"iclr_2018_SJJQVZW0b",
"iclr_2018_SJJQVZW0b",
"S1hvkpKxf",
"iclr_2018_SJJQVZW0b",
"Sye2eNDxM",
"HJQ8haFxz"
] |
iclr_2018_rJwelMbR- | Divide-and-Conquer Reinforcement Learning | Standard model-free deep reinforcement learning (RL) algorithms sample a new initial state for each trial, allowing them to optimize policies that can perform well even in highly stochastic environments. However, problems that exhibit considerable initial state variation typically produce high-variance gradient estimat... | accepted-poster-papers | This paper proposes a specific architecture for training an ensemble of separate policies on a family of easier tasks with the goal of obtaining a single policy that can perform well on a harder task. There are significant similarities to the recently published Distral algorithm, but I am convinced that this work offer... | train | [
"r1A2hMtgz",
"HJNRVMqez",
"rycTQSqgG",
"B1k6v2lQG",
"HkH_PhlXM",
"HkX5HhgmG",
"H1KBH2x7z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper presents a reinforcement learning method for learning complex tasks by dividing the state space into slices, learning local policies within each slice, while ensuring that they don't deviate too far from each other, while simultaneously learning a central policy that works across the entire state space ... | [
7,
7,
4,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rJwelMbR-",
"iclr_2018_rJwelMbR-",
"iclr_2018_rJwelMbR-",
"rycTQSqgG",
"rycTQSqgG",
"HJNRVMqez",
"r1A2hMtgz"
] |
iclr_2018_B1e5ef-C- | A Compressed Sensing View of Unsupervised Text Embeddings, Bag-of-n-Grams, and LSTMs | Low-dimensional vector embeddings, computed using LSTMs or simpler techniques, are a popular approach for capturing the “meaning” of text and a form of unsupervised learning useful for downstream tasks. However, their power is not theoretically understood. The current paper derives formal understanding by looking at th... | accepted-poster-papers | sadly, none of the reviewers seem to have been able to fully appreciate and check the proofs.
but in the words of even the least positive reviewer:
In general, I find many of the observations in this paper interesting. However, this paper is not strong enough as a theory paper; rather, the value lies perhaps in its fr... | train | [
"SkzuQ_dlG",
"SyURgFFlG",
"H1kgYgogM",
"HyxkZZhXM",
"BkyahGuzM",
"rytGHUTWM",
"BkX26PhZf",
"SkhuaD3Zf",
"BJi42DnZM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"author",
"author"
] | [
"The main insight in this paper is that LSTMs can be viewed as producing a sort of sketch of tensor representations of n-grams. This allows the authors to design a matrix that maps bag-of-n-gram embeddings into the LSTM embeddings. They then show that the result matrix satisfies a restricted isometry condition. C... | [
7,
7,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
1,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_B1e5ef-C-",
"iclr_2018_B1e5ef-C-",
"iclr_2018_B1e5ef-C-",
"iclr_2018_B1e5ef-C-",
"rytGHUTWM",
"SkhuaD3Zf",
"SkzuQ_dlG",
"H1kgYgogM",
"SyURgFFlG"
] |
iclr_2018_BkSDMA36Z | A New Method of Region Embedding for Text Classification | To represent a text as a bag of properly identified “phrases” and use the representation for processing the text is proved to be useful. The key question here is how to identify the phrases and represent them. The traditional method of utilizing n-grams can be regarded as an approximation of the approach. Such a method... | accepted-poster-papers | despite not amazing scores, this is a solid paper.
it created a lot of discussion and was found to be reproducible.
we should accept it to let the iclr community partake in the discussion and learn about this method of n-gram embeddings
| train | [
"ryjxrEwlM",
"r1sXToOgf",
"Sy6ClHqef",
"B1b7od97z",
"B1_2OO9QG",
"rJGB_OqXz",
"Sy29H_5Qz",
"HyE1mxEGf",
"SJphme4Mz",
"BkQabg4MM",
"HJvnFyVMG",
"ryfTd1Nfz",
"BkA08yEGG",
"SklnG7Mff",
"SyUEJzzfM",
"Hk992WfGz",
"B1QjRyMGG",
"Sy-OR3-fM",
"B1hPEvKlM",
"H1cW4Kulf",
"HJl5GqvlG",
"... | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"public",
"public",
"public",
"public",
"official_reviewer",
"public",
"official_reviewer",... | [
"The authors propose a mechanism for learning task-specific region embeddings for use in text classification. Specifically, this comprises a standard word embedding an accompanying local context embedding. \n\nThe key idea here is the introduction of a (h x c x v) tensor K, where h is the embedding dim (same as the... | [
6,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
5,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkSDMA36Z",
"iclr_2018_BkSDMA36Z",
"iclr_2018_BkSDMA36Z",
"r1sXToOgf",
"Sy6ClHqef",
"ryjxrEwlM",
"iclr_2018_BkSDMA36Z",
"B1QjRyMGG",
"Sy-OR3-fM",
"Hk992WfGz",
"SyUEJzzfM",
"SklnG7Mff",
"iclr_2018_BkSDMA36Z",
"iclr_2018_BkSDMA36Z",
"iclr_2018_BkSDMA36Z",
"iclr_2018_BkSDMA36Z"... |
iclr_2018_S1Dh8Tg0- | Fix your classifier: the marginal value of training the last weight layer | Neural networks are commonly used as models for classification for a wide variety of tasks. Typically, a learned affine transformation is placed at the end of such models, yielding a per-class value used for classification. This classifier can have a vast number of parameters, which grows linearly with the number of po... | accepted-poster-papers | This paper proposes an interesting new idea which creates an interesting discussion.
| train | [
"HyGVuO0ez",
"rJX0wjUVz",
"rJ3ZYFtxM",
"S1kGhTKez",
"Bkvocznzf",
"SyPO9M2zz",
"SJOW5M2fz",
"By_g-D-xf",
"SkwPDJbxz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"official_reviewer"
] | [
"Revised Review:\n\nThe authors have largely addressed my concerns with the revised manuscript. I still have some doubts about the C > N setting (the new settings of C / N of 4 and 2 aren't C >> N, and the associated results aren't detailed clearly in the paper), but I think the paper warrants acceptance.\n\nOrigin... | [
6,
-1,
6,
6,
-1,
-1,
-1,
-1,
-1
] | [
4,
-1,
5,
3,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1Dh8Tg0-",
"Bkvocznzf",
"iclr_2018_S1Dh8Tg0-",
"iclr_2018_S1Dh8Tg0-",
"rJ3ZYFtxM",
"S1kGhTKez",
"HyGVuO0ez",
"SkwPDJbxz",
"iclr_2018_S1Dh8Tg0-"
] |
iclr_2018_HyRnez-RW | Multi-Mention Learning for Reading Comprehension with Neural Cascades | Reading comprehension is a challenging task, especially when executed across longer or across multiple evidence documents, where the answer is likely to reoccur. Existing neural architectures typically do not scale to the entire evidence, and hence, resort to selecting a single passage in the document (either via trunc... | accepted-poster-papers | The authors did a good job addressing reviewer concerns and analyzing and testing their model on interesting datasets with convincing results. | val | [
"rkKuj7zgz",
"rJa0zH9xf",
"B1jA_O5xM",
"BJghy1hmz",
"By5HYFnZM",
"H1whTGi-f",
"rJHomqP-M",
"SkOozcw-f",
"r15HG5DZM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"public",
"author",
"author",
"author"
] | [
"The authors present a scalable model for questioning answering that is able to train on long documents. On the TriviaQA dataset, the proposed model achieves state of the art results on both domains (wikipedia and web). The formulation of the model is straight-forward, however I am skeptical about whether the resul... | [
7,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HyRnez-RW",
"iclr_2018_HyRnez-RW",
"iclr_2018_HyRnez-RW",
"iclr_2018_HyRnez-RW",
"H1whTGi-f",
"rJa0zH9xf",
"rkKuj7zgz",
"rJa0zH9xf",
"B1jA_O5xM"
] |
iclr_2018_r1SnX5xCb | Deep Sensing: Active Sensing using Multi-directional Recurrent Neural Networks | For every prediction we might wish to make, we must decide what to observe (what source of information) and when to observe it. Because making observations is costly, this decision must trade off the value of information against the cost of observation. Making observations (sensing) should be an active choice. To solve... | accepted-poster-papers | This paper is well written, addresses and interesting problem, and provides an interesting solution. | train | [
"rkoifQKEM",
"Bk4-YGplf",
"HJyXsRtef",
"HJg15Lhgz",
"rJ4VNC1Mf",
"ryg27A1ff",
"ryzBXC1Gf",
"B1lmXCJGf",
"BJCe7A1GG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"I have raised my score from 6 to 8 to reflect the authors' thorough responses and significant improvements in the revised manuscript.\n\nFrom my own review, the authors addressed the following points in their revision:\n\n- expanded related work to cover active learning and submodular optimization\n- added a discu... | [
-1,
8,
6,
7,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
3,
4,
-1,
-1,
-1,
-1,
-1
] | [
"Bk4-YGplf",
"iclr_2018_r1SnX5xCb",
"iclr_2018_r1SnX5xCb",
"iclr_2018_r1SnX5xCb",
"HJyXsRtef",
"HJg15Lhgz",
"Bk4-YGplf",
"Bk4-YGplf",
"Bk4-YGplf"
] |
iclr_2018_HkZy-bW0- | Temporally Efficient Deep Learning with Spikes | The vast majority of natural sensory data is temporally redundant. For instance, video frames or audio samples which are sampled at nearby points in time tend to have similar values. Typically, deep learning algorithms take no advantage of this redundancy to reduce computations. This can be an obscene waste of energy... | accepted-poster-papers | This paper provides an interesting synthesis of ideas. Although the results could be improved, this is a good paper. | train | [
"ryHEjLtgz",
"Hy9zmitlG",
"Syrb8hW-G",
"SyWo3T5mf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author"
] | [
"The principal problem that the paper addresses is how to integrate error-backpropagation learning in a network of spiking neurons that use a form of sigma-delta coding. The main observation is that static sigma-delta coding as proposed in OConnor and Welling (2016b), is not correct when the weights change during t... | [
7,
6,
8,
-1
] | [
5,
4,
4,
-1
] | [
"iclr_2018_HkZy-bW0-",
"iclr_2018_HkZy-bW0-",
"iclr_2018_HkZy-bW0-",
"iclr_2018_HkZy-bW0-"
] |
iclr_2018_ry-TW-WAb | Variational Network Quantization | In this paper, the preparation of a neural network for pruning and few-bit quantization is formulated as a variational inference problem. To this end, a quantizing prior that leads to a multi-modal, sparse posterior distribution over weights, is introduced and a differentiable Kullback-Leibler divergence approximation ... | accepted-poster-papers | The paper presents a variational Bayesian approach for quantising neural network weights and makes interesting and useful steps in this increasingly popular area of deep learning. | train | [
"By5wsMNxM",
"S10EfvFxM",
"BkIy2pKxf",
"r1X08SpmM",
"SkbsUrTQM",
"S1ytLST7G",
"SyWjrHa7f",
"ryZUHSTXf",
"Hyg1BgzJM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper proposes to use a mixture of continuous spikes propto 1/abs(w_ij-c_k) as prior for a Bayesian neural network and demonstrates good performance with relatively sparsified convnets for minist and cifar-10. The paper is building quite a lot upon Kingma et al 2015 and Molchanov et al 2017. \n\nThe paper is... | [
7,
7,
7,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ry-TW-WAb",
"iclr_2018_ry-TW-WAb",
"iclr_2018_ry-TW-WAb",
"By5wsMNxM",
"S1ytLST7G",
"S10EfvFxM",
"BkIy2pKxf",
"iclr_2018_ry-TW-WAb",
"iclr_2018_ry-TW-WAb"
] |
iclr_2018_SJJySbbAZ | Training GANs with Optimism | We address the issue of limit cycling behavior in training Generative Adversarial Networks and propose the use of Optimistic Mirror Decent (OMD) for training Wasserstein GANs. Recent theoretical results have shown that optimistic mirror decent (OMD) can enjoy faster regret rates in the context of zero-sum games. WGANs ... | accepted-poster-papers | The reviewers thought the paper provides an interesting line of research. | train | [
"H1NffmKgz",
"Syhxg_jgf",
"S138CtnWf",
"SkcwGOpQz",
"rJ78fOT7z",
"rkWJrG5QM",
"rk8tNz5XM",
"BJw8Efc7G",
"BJhRfg8-z",
"HJZkcmSJf",
"rJ99FMSkf",
"SJxehjl1G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"public"
] | [
"This paper proposes the use of optimistic mirror descent to train Wasserstein Generative Adversarial Networks (WGANS). The authors remark that the current training of GANs, which amounts to solving a zero-sum game between a generator and discriminator, is often unstable, and they argue that one source of instabili... | [
7,
8,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJJySbbAZ",
"iclr_2018_SJJySbbAZ",
"iclr_2018_SJJySbbAZ",
"rJ78fOT7z",
"H1NffmKgz",
"Syhxg_jgf",
"BJhRfg8-z",
"S138CtnWf",
"iclr_2018_SJJySbbAZ",
"rJ99FMSkf",
"SJxehjl1G",
"iclr_2018_SJJySbbAZ"
] |
iclr_2018_SJA7xfb0b | Sobolev GAN | We propose a new Integral Probability Metric (IPM) between distributions: the Sobolev IPM. The Sobolev IPM compares the mean discrepancy of two distributions for functions (critic) restricted to a Sobolev ball defined with respect to a dominant measure mu. We show that the Sobolev IPM compares two distributions in high... | accepted-poster-papers | The paper provides a useful analysis of the role of gradient penalties and the performance of the proposed approach in semi-supervised cases. | val | [
"HyJ_7LFlG",
"Hk4rkBRlz",
"SJYrKxQ-f",
"BJIPz57bG",
"HJ5YP_p7f",
"SkSg7XTmM",
"rkeWfhZ7M",
"rkANbHuzf",
"HyWBqmufz",
"H1stO7_zM",
"H1pY73PGG",
"HJrd-2wGz",
"ByNaV2wGz",
"rJsEWhDGM",
"HJ4rCjPff",
"B1e-piPfG",
"SkWlhswff",
"r1rajiwGG",
"HJfPlp0WG",
"r1Tvj8EZM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"public",
"public",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"The paper deals with the increasingly popular GAN approach to constructing generative models. Following the first formulation of GANs in 2014, it was soon realized that the training dynamics was highly unstable, leading to significant difficulties in achieving stable results. The paper by Arjovsky et al (2017) pr... | [
8,
6,
6,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJA7xfb0b",
"iclr_2018_SJA7xfb0b",
"iclr_2018_SJA7xfb0b",
"iclr_2018_SJA7xfb0b",
"SkSg7XTmM",
"rkeWfhZ7M",
"HJ4rCjPff",
"HyWBqmufz",
"HJ4rCjPff",
"SkWlhswff",
"HyJ_7LFlG",
"rJsEWhDGM",
"iclr_2018_SJA7xfb0b",
"Hk4rkBRlz",
"SJYrKxQ-f",
"BJIPz57bG",
"r1Tvj8EZM",
"HJfPlp0WG"... |
iclr_2018_H1sUHgb0Z | Learning From Noisy Singly-labeled Data | Supervised learning depends on annotated examples, which are taken to be the ground truth. But these labels often come from noisy crowdsourcing platforms, like Amazon Mechanical Turk. Practitioners typically collect multiple labels per example and aggregate the results to mitigate noise (the classic crowdsourcing probl... | accepted-poster-papers | This paper provides an important discussion about the relationship between training efficiency and label redundancy. The updates to the paper will improve the paper further. Reviewers found the paper interesting, well written, and addresses and important problem. | train | [
"BJhUuZDgf",
"rJRdVJPNf",
"SJpS_JYgz",
"ry8emW9gf",
"r18iu8g4M",
"BJnLnZyEG",
"S17KhF2Xz",
"H1IVZo4XG",
"Bk0D6FNQM",
"rJkQat4mM",
"BJ0EnF4mf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"public",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper proposes a method for learning from noisy labels, particularly focusing on the case when data isn't redundantly labeled (i.e. the same sample isn't labeled by multiple non-expert annotators). The authors provide both theoretical and experimental validation of their idea. \n\nPros:\n+ The paper is genera... | [
7,
-1,
6,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
-1,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1sUHgb0Z",
"H1IVZo4XG",
"iclr_2018_H1sUHgb0Z",
"iclr_2018_H1sUHgb0Z",
"BJnLnZyEG",
"iclr_2018_H1sUHgb0Z",
"BJ0EnF4mf",
"BJhUuZDgf",
"SJpS_JYgz",
"ry8emW9gf",
"iclr_2018_H1sUHgb0Z"
] |
iclr_2018_H1Y8hhg0b | Learning Sparse Neural Networks through L_0 Regularization | We propose a practical method for L0 norm regularization for neural networks: pruning the network during training by encouraging weights to become exactly zero. Such regularization is interesting since (1) it can greatly speed up training and inference, and (2) it can improve generalization. AIC and BIC, well-known mod... | accepted-poster-papers | The results in the paper are interesting, and the modifications improve the paper further. Reviewers found teh paper interesting and potentailly applicable to many models. | train | [
"rJUkD7vgf",
"BJF5RpKgG",
"S1tmowoeG",
"rkDib-FQf",
"ryTQxbK7f",
"ryTrTgK7z",
"rkNx5xYXG",
"rJeSs7YZz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper presents a continuous surrogate for the ell_0 norm and focuses on its applications in regularized empirical regularized minimization. The proposed continuous relaxation scheme allows for gradient based-stochastic optimization for binary discrete variables under the reparameterization trick, and extends ... | [
6,
6,
7,
-1,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H1Y8hhg0b",
"iclr_2018_H1Y8hhg0b",
"iclr_2018_H1Y8hhg0b",
"rJUkD7vgf",
"BJF5RpKgG",
"S1tmowoeG",
"rJeSs7YZz",
"iclr_2018_H1Y8hhg0b"
] |
iclr_2018_BkQqq0gRb | Variational Continual Learning | This paper develops variational continual learning (VCL), a simple but general framework for continual learning that fuses online variational inference (VI) and recent advances in Monte Carlo VI for neural networks. The framework can successfully train both deep discriminative models and deep generative models in compl... | accepted-poster-papers | The paper addresses the problem of continual learning and solutions based on variational inference. Updates to the paper have improved it and addresses many of the concerns raised by the reviewers during the discussion period. | train | [
"SyF0odSef",
"H1T4epKeM",
"BkgsE19xz",
"Hy67jKomG",
"rkgk9tsmG",
"S1GDuYi7G",
"SyQOwYi7M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Overall, the idea of this paper is simple but interesting. Via performing variational inference in a kind of online manner, one can address continual learning for deep discriminative or generative networks with considerations of model uncertainty.\n\nThe paper is written well, and literature review is sufficient. ... | [
6,
6,
6,
-1,
-1,
-1,
-1
] | [
3,
4,
2,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkQqq0gRb",
"iclr_2018_BkQqq0gRb",
"iclr_2018_BkQqq0gRb",
"SyF0odSef",
"H1T4epKeM",
"BkgsE19xz",
"iclr_2018_BkQqq0gRb"
] |
iclr_2018_H1-nGgWC- | Gaussian Process Behaviour in Wide Deep Neural Networks | Whilst deep neural networks have shown great empirical success, there is still much work to be done to understand their theoretical properties. In this paper, we study the relationship between Gaussian processes with a recursive kernel definition and random wide fully connected feedforward networks with more than one h... | accepted-poster-papers | A clearly written paper. While the practical relevance that came up in the review remains, the analysis and discussion is important for a deeper understanding of the deeper connections between these two important areas of machine learning. | train | [
"Hk4JEb5eM",
"BykFw7cxz",
"rJVpI8hgz",
"HywWHWKGM",
"Sy4PN-KGz",
"Hk08QWFff"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The authors study the limiting behaviour for wide Bayesian neural networks, comparing to Gaussian processes. \n\nThe paper is well written, and the experiments are enlightening. This work is a nice follow up to Neal (1994), and recent work considering similar results for neural networks with more than one hidden l... | [
6,
6,
6,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_H1-nGgWC-",
"iclr_2018_H1-nGgWC-",
"iclr_2018_H1-nGgWC-",
"Hk4JEb5eM",
"BykFw7cxz",
"rJVpI8hgz"
] |
iclr_2018_H135uzZ0- | Mixed Precision Training of Convolutional Neural Networks using Integer Operations | The state-of-the-art (SOTA) for mixed precision training is dominated by variants of low precision floating point operations, and in particular, FP16 accumulating into FP32 Micikevicius et al. (2017). On the other hand, while a lot of research has also happened in the domain of low and mixed-precision Integer training,... | accepted-poster-papers | Mixed precision application of CNNs is being explored for e.g. hardware implementations of networks trained at full precision. Mixed precision at training time is less common. This submission primarily concerns itself with the practical implementation details of training with mixed precision, and focuses primarily on... | train | [
"r1vIV-R1z",
"HyIm4t7xz",
"HJlhggcgM",
"SJ7rEp6Xf",
"Hykn4K6Wf",
"SJHIVYpbM",
"S1mqhBFWM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper describes an implementation of reduced precision deep learning using a 16 bit integer representation. This field has recently seen a lot of publications proposing various methods to reduce the precision of weights and activations. These schemes have generally achieved close-to-SOTA accuracy for small ne... | [
7,
7,
6,
-1,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_H135uzZ0-",
"iclr_2018_H135uzZ0-",
"iclr_2018_H135uzZ0-",
"iclr_2018_H135uzZ0-",
"HJlhggcgM",
"HyIm4t7xz",
"r1vIV-R1z"
] |
iclr_2018_SkFqf0lAZ | Memory Architectures in Recurrent Neural Network Language Models | We compare and analyze sequential, random access, and stack memory architectures for recurrent neural network language models. Our experiments on the Penn Treebank and Wikitext-2 datasets show that stack-based memory architectures consistently achieve the best performance in terms of held out perplexity. We also propos... | accepted-poster-papers | This paper provides a comparison of different types of a memory augmented models and extends some of them to beyond their simple form. Reviewers found the paper to be clearly written, saying it "nice introduction to the topic" and noting that they "enjoyed reading this paper". In general though there was a feeling that... | test | [
"SkiJh-5lM",
"SkTEzjqgf",
"H1gkDZaeM",
"Hy_dpvVZz",
"r1Wd8AC-z",
"SJS58CA-G",
"rkwnI0AZz",
"SJyC7vQgM",
"HJhJDmfxM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"author",
"author",
"author",
"author",
"public"
] | [
"The authors propose to compare three different memory architecture for recurrent neural network language models:\nvanilla LSTM, random access based on attention and continuous stack. The second main contribution of the paper is to propose an extension of continuous stacks, which allows to perform multiple pop oper... | [
6,
5,
8,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SkFqf0lAZ",
"iclr_2018_SkFqf0lAZ",
"iclr_2018_SkFqf0lAZ",
"iclr_2018_SkFqf0lAZ",
"SkiJh-5lM",
"SkTEzjqgf",
"Hy_dpvVZz",
"HJhJDmfxM",
"iclr_2018_SkFqf0lAZ"
] |
iclr_2018_ry_WPG-A- | On the Information Bottleneck Theory of Deep Learning | The practical successes of deep neural networks have not been matched by theoretical progress that satisfyingly explains their behavior. In this work, we study the information bottleneck (IB) theory of deep learning, which makes three specific claims: first, that deep networks undergo two distinct phases consisting of ... | accepted-poster-papers |
This submission explores recent theoretical work by Shwartz-Ziv and Tishby on explaining the generalization ability of deep networks. The paper gives counter-examples that suggest aspects of the theory might not be relevant for all neural networks.
There is some uncertainty surrounding the results where mutual inform... | train | [
"Bkzy2_YeG",
"rJzOv7qxG",
"rJdaeccgf",
"ByOQJGsXf",
"HkHYpbjXG",
"HymkpZs7z",
"rkeo2Zi7z",
"BkGKibjmM",
"rkdv9WsXf",
"Skp7LbP-f",
"ryJ6FjclG",
"rJ53OMckM",
"HJbi_G5Jf",
"SkbXdzckf",
"BksedMckM",
"Byn5PG9yG",
"S12WZqNyz",
"S1lBxcE1z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"author",
"author",
"author",
"author",
"public",
"public"
] | [
"This paper presents a study on the Information Bottleneck (IB) theory of deep learning, providing results in contrasts to the main theory claims. According to the authors, the IB theory suggests that the network generalization is mainly due to a ‘compression phase’ in the information plane occurring after a ‘fitti... | [
6,
7,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
3,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ry_WPG-A-",
"iclr_2018_ry_WPG-A-",
"iclr_2018_ry_WPG-A-",
"ryJ6FjclG",
"Bkzy2_YeG",
"rkeo2Zi7z",
"rJzOv7qxG",
"rJdaeccgf",
"iclr_2018_ry_WPG-A-",
"S1lBxcE1z",
"iclr_2018_ry_WPG-A-",
"HJbi_G5Jf",
"S12WZqNyz",
"BksedMckM",
"Byn5PG9yG",
"S1lBxcE1z",
"S1lBxcE1z",
"iclr_2018_... |
iclr_2018_BJJ9bz-0- | Reinforcement Learning from Imperfect Demonstrations | Robust real-world learning should benefit from both demonstrations and interaction with the environment. Current approaches to learning from demonstration and reward perform supervised learning on expert demonstration data and use reinforcement learning to further improve performance based on reward from the environm... | workshop-papers | I appreciate the experimental results, which includes a comparison against several baselines, however, I echo some of the concerns raised by the reviewers that the formulation is unclear and hard to follow. Moreover, the novelty over [Nachum, 2017] and [Haarnoja, 2017] seems small. Especially because [Nachum, 2017] als... | train | [
"rynqOnBez",
"ryE3gOulz",
"B1ornw9xG",
"Hk_MXdpQM",
"r1mUz_TXf",
"HJA2l_67z",
"BJL7Cv6mM",
"Hym3W3ref"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"public",
"public",
"public",
"official_reviewer"
] | [
"Thanks for all the explanations on my review and the other comments. While I can now clearly see the contributions of the paper, the minimal revisions in the paper do not make the contributions clear yet (in my opinion that should already be clear after having read the introduction). The new section \"intuitive an... | [
5,
6,
5,
-1,
-1,
-1,
-1,
-1
] | [
3,
5,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJJ9bz-0-",
"iclr_2018_BJJ9bz-0-",
"iclr_2018_BJJ9bz-0-",
"rynqOnBez",
"ryE3gOulz",
"B1ornw9xG",
"iclr_2018_BJJ9bz-0-",
"iclr_2018_BJJ9bz-0-"
] |
iclr_2018_HJjvxl-Cb | Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor | Model-free deep reinforcement learning (RL) algorithms have been demonstrated on a range of challenging decision making and control tasks. However, these methods typically suffer from two major challenges: very high sample complexity and brittle convergence properties, which necessitate meticulous hyperparameter tuning... | workshop-papers | The reviewers agree that the results are promising and there are some interesting and novel aspect to the formulation. However, two of the reviews have raised concerns regarding the exposition and the discussion of previous work. The paper benefits from a detailed description of soft Q-learning, PCL, and off-policy act... | test | [
"HJr5IZ5gM",
"Bk2K_F9lz",
"SJ1QyNJZf",
"BJ8AjRxNz",
"BJHwgKj7G",
"SJ0JxFsXM",
"Sk2hJKsXz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper proposes a soft actor-critic method aiming at lowering sample complexity and achieving a new convergence guarantee. However, the current paper has some correctness issues, is missing some related work and lacks a clear statement of innovation. \n\nThe first issue is that augmenting reward by adding an e... | [
3,
7,
5,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HJjvxl-Cb",
"iclr_2018_HJjvxl-Cb",
"iclr_2018_HJjvxl-Cb",
"SJ0JxFsXM",
"HJr5IZ5gM",
"Bk2K_F9lz",
"SJ1QyNJZf"
] |
iclr_2018_rk6H0ZbRb | Intriguing Properties of Adversarial Examples | It is becoming increasingly clear that many machine learning classifiers are vulnerable to adversarial examples. In attempting to explain the origin of adversarial examples, previous studies have typically focused on the fact that neural networks operate on high dimensional data, they overfit, or they are too linear. H... | workshop-papers | I am somewhat of two minds from the paper. The authors show empirically that adversarial perturbation error follows power law and looks for a possible explanation. The tie in with generalization is not clear to me and makes me wonder how to evaluate the significance of the finding of the power law distribution.. On th... | test | [
"Bkm7cMvgf",
"B17JC8dlf",
"B1yz5XhgM",
"H11B-xo7f",
"ryQEy8JGG",
"HJSmxwCbG",
"r1DxkwA-G",
"BJV6080bf",
"SJTNVQ_1M",
"H1mlAzf1f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper insists that adversarial error for small adversarial perturbation follows power low as a function of the perturbation size, and explains the cause by the logit-difference distributions using mean-field theory.\nThen, the authors propose two methods for improving adversarial robustness (entropy regulariz... | [
5,
8,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rk6H0ZbRb",
"iclr_2018_rk6H0ZbRb",
"iclr_2018_rk6H0ZbRb",
"iclr_2018_rk6H0ZbRb",
"B17JC8dlf",
"Bkm7cMvgf",
"BJV6080bf",
"B1yz5XhgM",
"H1mlAzf1f",
"iclr_2018_rk6H0ZbRb"
] |
iclr_2018_BJy0fcgRZ | Capturing Human Category Representations by Sampling in Deep Feature Spaces | Understanding how people represent categories is a core problem in cognitive science, with the flexibility of human learning remaining a gold standard to which modern artificial intelligence and machine learning aspire. Decades of psychological research have yielded a variety of formal theories of categories, yet valid... | workshop-papers | This paper introduces a GAN-based framework for inferring human category representations. The reviewers agree that the idea is interesting and well-motivated, and the results are promising. The technical contribution is not significant, but nevertheless the paper combines existing ideas in an interesting way. The revie... | train | [
"r1i3YtHgG",
"H1jrxgclM",
"Bk01UWclf",
"H1cX123mM",
"BJ5FaihmM",
"ByxrasnQf",
"HJp9njnQM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Quality\n\nThis paper demonstrates that human category representations can be inferred by sampling deep feature spaces. The idea is an extension of the earlier developed MCMC with people approach where samples are drawn in the latent space of a DCGAN and a BiGAN. The approach is thoroughly validated using two onli... | [
6,
5,
5,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJy0fcgRZ",
"iclr_2018_BJy0fcgRZ",
"iclr_2018_BJy0fcgRZ",
"iclr_2018_BJy0fcgRZ",
"r1i3YtHgG",
"H1jrxgclM",
"Bk01UWclf"
] |
iclr_2018_Sy4c-3xRW | DropMax: Adaptive Stochastic Softmax | We propose DropMax, a stochastic version of softmax classifier which at each iteration drops non-target classes with some probability, for each instance. Specifically, we overlay binary masking variables over class output probabilities, which are learned based on the input via regularized variational inference. This st... | workshop-papers | This paper proposes a general regularization algorithm which builds on the dropout idea. This is a very significant topic. The overall motivation is good, but the specific design choices are less well motivated over, for example, ad-hoc choices. Some concerns remain after the post-rebuttal discussion with the reviewers... | train | [
"rkBcQHBgG",
"ryl6gl5xM",
"BkQfMZLNz",
"Bkp1F5OlG",
"ry0kYb6Xf",
"Syd-ObpXM",
"HkjrLl2Gf",
"ry49UlnGz",
"SJs2_xhGz",
"BJbUYx3GM",
"ryj2c77MG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper propose an adaptive dropout strategy for class logits. They learn a distribution q(z | x, y) that randomly throw class logits. By doing so they ensemble predictions of the models between different set of classes, and focuses on more difficult discrimination tasks. They learn the dropout distribution by ... | [
6,
6,
-1,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
4,
-1,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Sy4c-3xRW",
"iclr_2018_Sy4c-3xRW",
"SJs2_xhGz",
"iclr_2018_Sy4c-3xRW",
"ryj2c77MG",
"iclr_2018_Sy4c-3xRW",
"ryl6gl5xM",
"ryl6gl5xM",
"Bkp1F5OlG",
"rkBcQHBgG",
"iclr_2018_Sy4c-3xRW"
] |
iclr_2018_SJUX_MWCZ | Predict Responsibly: Increasing Fairness by Learning to Defer | When machine learning models are used for high-stakes decisions, they should predict accurately, fairly, and responsibly. To fulfill these three requirements, a model must be able to output a reject option (i.e. say "``I Don't Know") when it is not qualified to make a prediction. In this work, we propose learning to de... | workshop-papers | This work is proposing an approach for ensuring classification fairness through models that encapsulate deferment criteria. On the positive side, the paper provides ideas which are conceptually interesting and novel. On the other hand, the reviewers find the technical contribution to be limited and, in some cases, chal... | train | [
"r1RTd8hgG",
"HJMCQAFgf",
"HJkk4w6lM",
"BkIG-LYzf",
"Bk8LlIYff",
"BkLnyItzG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The proposed method is a classifier that is fair and works in collaboration with an unfair (but presumably accurate model). The novel classifier is the result of the optimisation of a loss function (composed of a part similar to a logistic regression model and a part being the disparate impact). Hence, it can be i... | [
5,
4,
6,
-1,
-1,
-1
] | [
3,
5,
3,
-1,
-1,
-1
] | [
"iclr_2018_SJUX_MWCZ",
"iclr_2018_SJUX_MWCZ",
"iclr_2018_SJUX_MWCZ",
"HJMCQAFgf",
"r1RTd8hgG",
"HJkk4w6lM"
] |
iclr_2018_r1hsJCe0Z | Semantic Code Repair using Neuro-Symbolic Transformation Networks | We study the problem of semantic code repair, which can be broadly defined as automatically fixing non-syntactic bugs in source code. The majority of past work in semantic code repair assumed access to unit tests against which candidate repairs could be validated. In contrast, the goal here is to develop a strong stati... | workshop-papers | To summarize the pros and cons:
Pro:
* Interesting application
* Impressive results on a difficult task
* Nice discussion of results and informative examples
* Clear presentation, easy to read.
Con:
* The method appears to be highly specialized to the four bug types. It is not clear how generalizable it will be to mo... | test | [
"rJvBlRteG",
"SkSfxq9xM",
"SynixA1WM",
"r1XRhmeXf",
"r1LW37xmM",
"HkIRj7g7z",
"rJtTxEgXf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper presents a neural network architecture consisting of the share, specialize and compete parts for repairing code in four cases, i.e., VarReplace, CompReplace, IsSwap, and ClassMember. Experiments on the source codes from Github are conducted and the performance is evaluated against one sequence-to-sequen... | [
4,
6,
6,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1hsJCe0Z",
"iclr_2018_r1hsJCe0Z",
"iclr_2018_r1hsJCe0Z",
"rJvBlRteG",
"SkSfxq9xM",
"SynixA1WM",
"iclr_2018_r1hsJCe0Z"
] |
iclr_2018_rk3pnae0b | Topic-Based Question Generation | Asking questions is an important ability for a chatbot. This paper focuses on question generation. Although there are existing works on question generation based on a piece of descriptive text, it remains to be a very challenging problem. In the paper, we propose a new question generation problem, which also requires t... | workshop-papers | The pros and cons of the paper under consideration can be summarized below:
Pros:
* Reviewers thought the underlying model is interesting and intuitive
* Main contributions are clear
Cons:
* There is confusion between keywords and topics, which is leading to a somewhat confused explanation and lack of clear compariso... | train | [
"rk27E6PlG",
"HyrNUBYlz",
"B1chwjFlz",
"r18wTowMG",
"S1ZZ5jvff",
"ByHq_sPMG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper presents a neural network-based approach to generate topic-specific questions with the motivation that topical questions are more meaningful in practical applications like real-world conversations. Experiments and evaluation have been conducted on the AQAD corpus to show the effectiveness of the approac... | [
3,
4,
8,
-1,
-1,
-1
] | [
5,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_rk3pnae0b",
"iclr_2018_rk3pnae0b",
"iclr_2018_rk3pnae0b",
"rk27E6PlG",
"HyrNUBYlz",
"B1chwjFlz"
] |
iclr_2018_SJDJNzWAZ | Time-Dependent Representation for Neural Event Sequence Prediction | Existing sequence prediction methods are mostly concerned with time-independent sequences, in which the actual time span between events is irrelevant and the distance between events is simply the difference between their order positions in the sequence. While this time-independent view of sequences is applicable for da... | workshop-papers | I've summarized the pros and cons of the reviews below:
Pros:
* The method for time representation in event sequences is novel and well founded
* It shows improvements on several but not all datasets that may have real-world applications
Cons:
* Gains are somewhat small
* The task is also not of huge interest to ICLR... | train | [
"Sye5BLIyG",
"S1edG-sxf",
"rk0LDknlz",
"ByyJ4OT7M",
"SJ5WB_a7f",
"S1jyLOpmG",
"rknWGOaXM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Quality above threshold.\nClarity above threshold.\nOriginality slightly below threshold.\nSignificance slightly below threshold.\n\nPros:\nThis paper proposed a RNN for event sequence prediction. It provides two constructed choices for combining time(duration) information to event. Experiments on various datasets... | [
4,
4,
5,
-1,
-1,
-1,
-1
] | [
4,
5,
3,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SJDJNzWAZ",
"iclr_2018_SJDJNzWAZ",
"iclr_2018_SJDJNzWAZ",
"Sye5BLIyG",
"rk0LDknlz",
"iclr_2018_SJDJNzWAZ",
"S1edG-sxf"
] |
iclr_2018_SkHl6MWC- | Regularization Neural Networks via Constrained Virtual Movement Field | We provide a novel thinking of regularization neural networks. We smooth the objective of neural networks w.r.t small adversarial perturbations of the inputs. Different from previous works, we assume the adversarial perturbations are caused by the movement field. When the magnitude of movement field approaches 0, we ca... | workshop-papers | R1 thought the proposed method was novel and the idea interesting. However, he/she raised concerns with consistency in the experimental validation, the trade-off between accuracy and running time, and the positioning/motivation, specifically the claim about interpretability. The authors responded to these concerns, and... | val | [
"HJY_d5iEG",
"HJ6Svr8Ez",
"H1g6cQsxM",
"Sy_2ES9lG",
"ryfzqnhxz",
"S1cdKD7Gz",
"SJenYvmfM",
"B16EYDXGM"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Thanks for your higher score.\n\nWe have tried two ways of setting wi's:\n (1) Fixed values, i.e. wi = 1/3.\n (2) Sampling wi' s from uniform distribution then normalizing them to a vector with unit length.\nAnd we find (2) is better than (1). \n\nHere are potential reasons: \nFirst, when we linearly combine diffe... | [
-1,
-1,
5,
5,
6,
-1,
-1,
-1
] | [
-1,
-1,
4,
4,
4,
-1,
-1,
-1
] | [
"HJ6Svr8Ez",
"S1cdKD7Gz",
"iclr_2018_SkHl6MWC-",
"iclr_2018_SkHl6MWC-",
"iclr_2018_SkHl6MWC-",
"H1g6cQsxM",
"Sy_2ES9lG",
"ryfzqnhxz"
] |
iclr_2018_rJWrK9lAb | Autoregressive Generative Adversarial Networks | Generative Adversarial Networks (GANs) learn a generative model by playing an adversarial game between a generator and an auxiliary discriminator, which classifies data samples vs. generated ones. However, it does not explicitly model feature co-occurrences in samples. In this paper, we propose a novel Autoregressive G... | workshop-papers | The reviewers (all experts in this area) appreciated the novelty of the idea, though they felt that the experimental results (samples and Inception scores) did not provide convincing evidence value of this method over already established techniques. The authors responded to the concerns but were not able to address the... | train | [
"S1klbTulM",
"BkuDb6tgf",
"S13bO3cez",
"ByWsMXCfz",
"S1Ma2AnmM",
"rJWGLRJmz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper proposes a new GAN model whereby the discriminator (rather than being a binary classifier) consists of an encoding network followed by an autoregressive model on the encoded features. The discriminator is trained to maximize the probability of the true data and minimize the probability of the generated ... | [
5,
3,
5,
-1,
-1,
-1
] | [
4,
5,
5,
-1,
-1,
-1
] | [
"iclr_2018_rJWrK9lAb",
"iclr_2018_rJWrK9lAb",
"iclr_2018_rJWrK9lAb",
"BkuDb6tgf",
"S1klbTulM",
"S13bO3cez"
] |
iclr_2018_Hy1d-ebAb | Learning Deep Generative Models of Graphs | Graphs are fundamental data structures required to model many important real-world data, from knowledge graphs, physical and social interactions to molecules and proteins. In this paper, we study the problem of learning generative models of graphs from a dataset of graphs of interest. After learning, these models can b... | workshop-papers | Predicting graphs is an interesting and important direction, and there exist essentially no (effective) general-purpose techniques for this problem. The idea of predicting nodes one by one, though not entirely surprising, is interesting and the approach makes sense. Unfortunately, I (and some of reviewers) less convin... | train | [
"Sy6ZK8IEM",
"S1crSKYgM",
"ByZ8Tx9ez",
"HkCWGa0xM",
"SJ0xpMM4f",
"HJH-GmaXz",
"Hk9Ol76mG",
"Hk8ilROQG",
"B1GMUvUGf",
"Bk1RrPUMz",
"r1kajbsyG",
"HkzAQO_kM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"official_reviewer",
"author",
"author",
"author",
"public"
] | [
"Thanks for adding this comparison against the Grammar VAE model. I think it certainly allows for a better placement of your proposed model w.r.t related work.\n\nWhile I hoped this comparison would tell a clear story about whether a) decoding with a grammar or b) decoding directly into a graph representation would... | [
-1,
5,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
3,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"HJH-GmaXz",
"iclr_2018_Hy1d-ebAb",
"iclr_2018_Hy1d-ebAb",
"iclr_2018_Hy1d-ebAb",
"S1crSKYgM",
"Hk8ilROQG",
"iclr_2018_Hy1d-ebAb",
"HkCWGa0xM",
"Bk1RrPUMz",
"iclr_2018_Hy1d-ebAb",
"HkzAQO_kM",
"iclr_2018_Hy1d-ebAb"
] |
iclr_2018_r1lfpfZAb | Learning to Write by Learning the Objective | Recurrent Neural Networks (RNNs) are powerful autoregressive sequence models for learning prevalent patterns in natural language. Yet language generated by RNNs often shows several degenerate characteristics that are uncommon in human language; while fluent, RNN language production can be overly generic, repetitive, ... | workshop-papers | I (and some of the reviewers) find the general motivation quite interesting (operationalizing the Gricean maxims in order to improve language generation). However, we are not convinced that the actual model encodes these maxims in a natural and proper way. Without this motivation, the approach can be regarded as a se... | train | [
"HkN9lyRxG",
"ByGHp-S4M",
"ByWqV4YlG",
"BJFJrHcgz",
"rks9NupQG",
"SJrPVdp7f",
"HkMz4Oamz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper proposes to bring together multiple inductive biases that hope to correct for inconsistencies in sequence decoding. Building on previous works that utilize modified objectives to generate sequences, this work proposes to optimize for the parameters of a pre-defined combination of various sub-objectives.... | [
6,
-1,
5,
4,
-1,
-1,
-1
] | [
5,
-1,
4,
5,
-1,
-1,
-1
] | [
"iclr_2018_r1lfpfZAb",
"rks9NupQG",
"iclr_2018_r1lfpfZAb",
"iclr_2018_r1lfpfZAb",
"BJFJrHcgz",
"ByWqV4YlG",
"HkN9lyRxG"
] |
iclr_2018_ryZ283gAZ | Beyond Finite Layer Neural Networks: Bridging Deep Architectures and Numerical Differential Equations | Deep neural networks have become the state-of-the-art models in numerous machine learning tasks. However, general guidance to network architecture design is still missing. In our work, we bridge deep neural network design with numerical differential equations. We show that many effective networks, such as ResNet, PolyN... | workshop-papers | The reviewers agree that the proposed architecture is novel. However, there are issues in terms of the motivation. It would be helpful in future drafts to strengthen the argument about why the architecture is expected to be better than others. Most importantly, the gains at this stage are still incremental. A larger im... | train | [
"B15nEv7Sf",
"Bk421ULNf",
"ByjXORWlG",
"ryMdpXref",
"ByZSRnteG",
"Byn56OfWM",
"r1FRx5fWG",
"HyG_XDfWM"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Thanks for taking your time reviewing our manuscript and all your comments. Here are our responses to the \"cons\".\n\nThe reason behind making the current title is to indicate the potential benefit of thinking beyond finite layer neural networks by looking at the continuum, i.e. the underlying dynamic system. Thi... | [
-1,
7,
6,
5,
5,
-1,
-1,
-1
] | [
-1,
4,
1,
3,
1,
-1,
-1,
-1
] | [
"Bk421ULNf",
"iclr_2018_ryZ283gAZ",
"iclr_2018_ryZ283gAZ",
"iclr_2018_ryZ283gAZ",
"iclr_2018_ryZ283gAZ",
"ryMdpXref",
"ByjXORWlG",
"ByZSRnteG"
] |
iclr_2018_B14uJzW0b | No Spurious Local Minima in a Two Hidden Unit ReLU Network | Deep learning models can be efficiently optimized via stochastic gradient descent, but there is little theoretical evidence to support this. A key question in optimization is to understand when the optimization landscape of a neural network is amenable to gradient-based optimization. We focus on a simple neural network... | workshop-papers | This submission is a continuation of a line of theoretical work that seeks to characterize optimization landscapes of neural networks by the presence or absence of spurious local minima. As the number of critical points grows combinatorially for larger networks, it is very challenging to show such results. The presen... | train | [
"By_zofR1z",
"rky-kk_eG",
"HykMheuxG",
"SJh8JW3Gz",
"BkkE1bhfG",
"BJBbkWhfM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary: \nThe paper considers the problem of a single hidden layer neural network, with 2 RELU units (this is what I got from the paper - as I describe below, it was not clear at all the setting of the problem - if I'm mistaken, I will also wait for the rest of the reviews to have a more complete picture of the p... | [
4,
6,
6,
-1,
-1,
-1
] | [
4,
3,
2,
-1,
-1,
-1
] | [
"iclr_2018_B14uJzW0b",
"iclr_2018_B14uJzW0b",
"iclr_2018_B14uJzW0b",
"HykMheuxG",
"rky-kk_eG",
"By_zofR1z"
] |
iclr_2018_Bki4EfWCb | Inference Suboptimality in Variational Autoencoders | Amortized inference has led to efficient approximate inference for large datasets. The quality of posterior inference is largely determined by two factors: a) the ability of the variational distribution to model the true posterior and b) the capacity of the recognition network to generalize inference over all datapoint... | workshop-papers | Thank you for submitting you paper to ICLR. This paper provides an informative analysis of the approximation contributions from the various assumptions made in variational auto-encoders. The revision has demonstrated the robustness of the paper’s conclusions, however these conclusions are arguably unsurprising. Althoug... | train | [
"HyXty1qlM",
"rJ5VMfcxG",
"r1_Ulf9gz",
"rJ201f3QG",
"By5gZbnmf",
"Byip9x2XG",
"S14LJgnmG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"* EDIT: Increased score from 5 to 6 to reflect improvements made in the revision.\n\nThe authors break down the \"inference gap\" in VAEs (the slack in the variational lower bound) into two components: 1. the \"amortization gap\", measuring what part of the slack is due to amortizing inference using a neural net e... | [
6,
6,
6,
-1,
-1,
-1,
-1
] | [
5,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Bki4EfWCb",
"iclr_2018_Bki4EfWCb",
"iclr_2018_Bki4EfWCb",
"HyXty1qlM",
"r1_Ulf9gz",
"rJ5VMfcxG",
"iclr_2018_Bki4EfWCb"
] |
iclr_2018_B1lMMx1CW | THE EFFECTIVENESS OF A TWO-LAYER NEURAL NETWORK FOR RECOMMENDATIONS | We present a personalized recommender system using neural network for recommending
products, such as eBooks, audio-books, Mobile Apps, Video and Music.
It produces recommendations based on customer’s implicit feedback history such
as purchases, listens or watches. Our key contribution is to formulate ... | workshop-papers | Meta score: 6
This is a thorough empirical paper, demonstrating the effectiveness of a relatively simple model for recommendations:
Pros:
- strong experiments
- always good to see simple models pushed to perform well
- presumably of interest to practioners in the area
Cons:
- quite oriented to the recommendation... | train | [
"H1RrugAbf",
"SJqZdRi1f",
"r1rOlgOlz",
"B1RdHXTef",
"B1WDuZAWz",
"SJ3gtkUWz",
"SJOFHTH-M"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Thank you for your feedback\nIn addition to your comments we would like to highlight several points:\n1. Two methods of integrating time decay of purchases into the learning framework were proposed:\n1.1 Convolutional layer for exploring the shape of the time decay function.\n1.2 We explored properties of differen... | [
-1,
6,
6,
7,
-1,
-1,
-1
] | [
-1,
3,
4,
3,
-1,
-1,
-1
] | [
"r1rOlgOlz",
"iclr_2018_B1lMMx1CW",
"iclr_2018_B1lMMx1CW",
"iclr_2018_B1lMMx1CW",
"iclr_2018_B1lMMx1CW",
"SJqZdRi1f",
"B1RdHXTef"
] |
iclr_2018_rk8wKk-R- | Convolutional Sequence Modeling Revisited | This paper revisits the problem of sequence modeling using convolutional
architectures. Although both convolutional and recurrent architectures have a
long history in sequence prediction, the current "default" mindset in much of
the deep learning community is that generic sequence modeling is best h... | workshop-papers | meta score: 5
This paper gives a thorough experimental comparison of convolutional vs recurrent networks for a variety of sequence modelling tasks. The experimentation is thorough, but the main point of the paper, that convolutional networks are unjustly ignored for sequence modelling, is overstated as there are sev... | train | [
"SkdHpQDez",
"HkUwN_Ylf",
"HkTNLM5gM",
"rkR0McB7M",
"SyYmNNQzf",
"S1FZN4mzf",
"SyxkEEQfM",
"rkehXEQff",
"HkDIQN7ff",
"ByC7mNQGf",
"B1lh3uXbf",
"Bk2tcOm-f",
"BJ7tE5OCb"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"public"
] | [
"In this paper, the authors argue for the use of convolutional architectures as a general purpose tool for sequence modeling. They start by proposing a generic temporal convolution sequence model which leverages recent advances in the field, discuss the respective advantages of convolutional and recurrent networks,... | [
8,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rk8wKk-R-",
"iclr_2018_rk8wKk-R-",
"iclr_2018_rk8wKk-R-",
"iclr_2018_rk8wKk-R-",
"BJ7tE5OCb",
"SkdHpQDez",
"HkUwN_Ylf",
"HkTNLM5gM",
"Bk2tcOm-f",
"B1lh3uXbf",
"iclr_2018_rk8wKk-R-",
"iclr_2018_rk8wKk-R-",
"iclr_2018_rk8wKk-R-"
] |
iclr_2018_Hkfmn5n6W | Exponentially vanishing sub-optimal local minima in multilayer neural networks | Background: Statistical mechanics results (Dauphin et al. (2014); Choromanska et al. (2015)) suggest that local minima with high error are exponentially rare in high dimensions. However, to prove low error guarantees for Multilayer Neural Networks (MNNs), previous works so far required either a heavily modified MNN mod... | workshop-papers | The paper analyzes neural network with hidden layer of piecewise linear units, a single output, and a quadratic loss. The reviewers find the results incremental and not "surprising", and also complained about comparison with previous work. I think the topic is very pertinent, and definitely more relevant compared to st... | train | [
"BJG3EUIVz",
"ryA1wwKgz",
"HkgrJeEgM",
"Skq3uQKxG",
"SJtwo9DMf",
"r1k7o5vGz",
"HkRqc9DfM",
"HyQgq9wGf"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"After our revision, the only remaining major concern of the reviewer is that\n\"the main message is mostly covered by existing works.\"\n\nHowever, it is not clear to us what existing work the reviewer is referring to. If the reviewer is saying that similar results were proved earlier, then we disagree. In any cas... | [
-1,
5,
7,
6,
-1,
-1,
-1,
-1
] | [
-1,
3,
2,
3,
-1,
-1,
-1,
-1
] | [
"ryA1wwKgz",
"iclr_2018_Hkfmn5n6W",
"iclr_2018_Hkfmn5n6W",
"iclr_2018_Hkfmn5n6W",
"HkgrJeEgM",
"Skq3uQKxG",
"ryA1wwKgz",
"iclr_2018_Hkfmn5n6W"
] |
iclr_2018_ByxLBMZCb | Learning Deep Models: Critical Points and Local Openness | With the increasing interest in deeper understanding of the loss surface of many non-convex deep models, this paper presents a unifying framework to study the local/global optima equivalence of the optimization problems arising from training of such non-convex models. Using the "local openness" property of the underl... | workshop-papers | The paper nicely unifies previous results and develops the property of local openness. While interesting, I find the application to multi-layer linear networks extremely limiting. There appears to be a sub-field in theory now focusing on solely multi-layer linear networks which is meaningless in practice. I can appreci... | train | [
"BkL0g3a1f",
"rkimHPzbz",
"SJtc2C4bz",
"SkSlUunQf",
"HJImbOhmM",
"HkNlXO37M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary: The paper focuses on the characterization of the landscape of deep neural networks; i.e., when and why local minima are global, what are the conditions for saddle critical points, etc. The paper covers a somewhat wide range of deep nets (from shallow with linear activation to deeper with non-linear activa... | [
6,
5,
6,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_ByxLBMZCb",
"iclr_2018_ByxLBMZCb",
"iclr_2018_ByxLBMZCb",
"BkL0g3a1f",
"SJtc2C4bz",
"rkimHPzbz"
] |
iclr_2018_rJGY8GbR- | Deep Mean Field Theory: Layerwise Variance and Width Variation as Methods to Control Gradient Explosion | A recent line of work has studied the statistical properties of neural networks to great success from a {\it mean field theory} perspective, making and verifying very precise predictions of neural network behavior and test time performance.
In this paper, we build upon these works to explore two methods for tam... | workshop-papers | All the reviewers agree that this is an interesting paper but have concerns about readability and presentation. There is also concern that many results are speculative and not concretely tested. I recommend the authors to carefully investigate their claims with stronger experiments and submit it to another venue. I rec... | train | [
"rkDLp95lG",
"SJTc3MAgf",
"Bk8iCb0Wz",
"B17PSW0Qz",
"S1iqVbRmG",
"BkbKHuamf",
"HJtXHu6mM",
"H1sObu67z",
"S1pyeO67G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper further develops the research program using mean field theory to predict generalization performance of deep neural networks. As with all recent mean-field papers, the main query here is to what extent the assumptions (Axioms 1+2, which basically define the asymptotic parameters of interest to be the qua... | [
7,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
1,
3,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_rJGY8GbR-",
"iclr_2018_rJGY8GbR-",
"iclr_2018_rJGY8GbR-",
"S1iqVbRmG",
"HJtXHu6mM",
"SJTc3MAgf",
"Bk8iCb0Wz",
"rkDLp95lG",
"iclr_2018_rJGY8GbR-"
] |
iclr_2018_BygpQlbA- | Towards Provable Control for Unknown Linear Dynamical Systems | We study the control of symmetric linear dynamical systems with unknown dynamics and a hidden state. Using a recent spectral filtering technique for concisely representing such systems in a linear basis, we formulate optimal control in this setting as a convex program. This approach eliminates the need to solve the non... | workshop-papers | This paper studies the control of symmetric linear dynamical systems with unknown dynamics. While the reviewers agree that this is an interesting topic, there are concerns that the assumptions are not realistic. Lack of experiments also stands out. I recommend the paper to workshop track with the hope that it will fost... | train | [
"HynVA_vxG",
"SydMCJ9gz",
"ryr6tuv-G",
"ByvMH7zMz",
"HJGD4Qffz",
"BJE1VXfGz",
"H1hFmXMGf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"This paper proposes a new algorithm to generate the optimal control inputs for unknown linear dynamical systems (LDS) with known system dimensions.\n\nThe idea is exciting LDS by wave filter inputs and record the output and directly estimate the operator that maps the input to the output instead of estimating the ... | [
4,
7,
5,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BygpQlbA-",
"iclr_2018_BygpQlbA-",
"iclr_2018_BygpQlbA-",
"HynVA_vxG",
"SydMCJ9gz",
"ryr6tuv-G",
"iclr_2018_BygpQlbA-"
] |
iclr_2018_HJYQLb-RW | On the limitations of first order approximation in GAN dynamics | Generative Adversarial Networks (GANs) have been proposed as an approach to learning generative models. While GANs have demonstrated promising performance on multiple vision tasks, their learning dynamics are not yet well understood, neither in theory nor in practice. In particular, the work in this domain has been foc... | workshop-papers | All the reviewers agree that the paper is studying an important problem and makes a good first step towards understanding learning in GANs. But the reviewers are concerned that the setup is too simplistic and not relevant in practical settings. I recommend the authors to carefully go through reviews and to present it a... | train | [
"HkxWKlkgM",
"H1FyxrBgz",
"SyLbm9eWM",
"Hk-tVNhmM",
"rJTPEN2Qz",
"S1SS4437M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Although GAN recently has attracted so many attentions, the theory of GAN is very poor. This paper tried to make a new insight of GAN from theories and I think their approach is a good first step to build theories for GAN. \n\nHowever, I believe this paper is not enough to be accepted. The main reason is that the ... | [
4,
5,
7,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_HJYQLb-RW",
"iclr_2018_HJYQLb-RW",
"iclr_2018_HJYQLb-RW",
"HkxWKlkgM",
"H1FyxrBgz",
"SyLbm9eWM"
] |
iclr_2018_H1YynweCb | Kronecker Recurrent Units | Our work addresses two important issues with recurrent neural networks: (1) they are over-parameterized, and (2) the recurrent weight matrix is ill-conditioned. The former increases the sample complexity of learning and the training time. The latter causes the vanishing and exploding gradient problem. We present a flex... | workshop-papers | I tend to agree with the most positive reviewer who characterizes the work with the following statements:
"Kronecker factorization was introduced for Convolutional networks (citation is in the paper). Soft unitary constraints also have been introduced in earlier work (citations are also in the paper). Nevertheless, sh... | train | [
"ByhgguzeM",
"Hy28Xy9lM",
"HkZmXGcxf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper presents a method to parametrize unitary matrices in an RNN as a Kronecker product of smaller matrices. Given N inputs and output, this method allows one to specify a linear transformation with O(log(N)) parameters, and perform a forward and backward pass in O(Nlog(N)) time. \nIn addition a relaxation i... | [
6,
5,
7
] | [
5,
4,
3
] | [
"iclr_2018_H1YynweCb",
"iclr_2018_H1YynweCb",
"iclr_2018_H1YynweCb"
] |
iclr_2018_rk4Fz2e0b | Graph Partition Neural Networks for Semi-Supervised Classification | We present graph partition neural networks (GPNN), an extension of graph neural networks (GNNs) able to handle extremely large graphs. GPNNs alternate between locally propagating information between nodes in small subgraphs and globally propagating information between the subgraphs. To efficiently partition graphs, we ... | workshop-papers | This paper was perceived as being well written, but the technical contribution was seen as being incremental and somewhat heuristic in nature. Some important prior work was not discussed and more extensive experimentation was recommended.
However, the proposed approach of partitioning the graph into sub graphs and a sc... | train | [
"HkaZrhuez",
"r1Q8qCdgf",
"Hkk48Xg-f",
"ByBxb1Imf",
"rytTg1I7z",
"BJ2TNUQXM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Graph Neural Networks are methods using NNs to deal with graph data (each data point has some features, and there is some known connectivity structure among nodes) for problems such as semi-supervised classification. They can also be viewed as an abstraction and generalizations of RNNs to arbitrary graphs. As such... | [
6,
5,
6,
-1,
-1,
-1
] | [
3,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_rk4Fz2e0b",
"iclr_2018_rk4Fz2e0b",
"iclr_2018_rk4Fz2e0b",
"HkaZrhuez",
"Hkk48Xg-f",
"r1Q8qCdgf"
] |
iclr_2018_HymuJz-A- | Not-So-CLEVR: Visual Relations Strain Feedforward Neural Networks | The robust and efficient recognition of visual relations in images is a hallmark of biological vision. Here, we argue that, despite recent progress in visual recognition, modern machine vision algorithms are severely limited in their ability to learn visual relations. Through controlled experiments, we demonstrate that... | workshop-papers | This paper studies an important problem (visual relationship detection and generalization capabilities existing networks for this task). Unfortunately, all reviewers raise concerns (e.g. limited relations studied) and are largely on the fence about this paper. While this paper does not propose solutions, it does presen... | train | [
"B1pcOYBlG",
"rkFUZ2uxf",
"r1AAFH5xG",
"rJptTYrMG",
"HkDmAFBMG",
"Hyay0YrMz",
"HyRn6trMM",
"H1VUptHzz",
"ryfIDWSff"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Quality\n\nThis paper demonstrates that convolutional and relational neural networks fail to solve visual relation problems by training networks on artificially generated visual relation data. This points at important limitations of current neural network architectures where architectures depend mainly on rote mem... | [
6,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HymuJz-A-",
"iclr_2018_HymuJz-A-",
"iclr_2018_HymuJz-A-",
"iclr_2018_HymuJz-A-",
"B1pcOYBlG",
"rkFUZ2uxf",
"r1AAFH5xG",
"ryfIDWSff",
"iclr_2018_HymuJz-A-"
] |
iclr_2018_SyunbfbAb | FigureQA: An Annotated Figure Dataset for Visual Reasoning | We introduce FigureQA, a visual reasoning corpus of over one million question-answer pairs grounded in over 100,000 images. The images are synthetic, scientific-style figures from five classes: line plots, dot-line plots, vertical and horizontal bar graphs, and pie charts. We formulate our reasoning task by generating ... | workshop-papers | This paper was reviewed by 3 expert reviews. While they all see value in the new task and dataset, they raise concerns (templated language, unclear what exactly are the new challenges posed by this task and dataset, etc) that this AC agrees with. To be clear, the lack of a fundamentally new model is not a problem (or a... | train | [
"BJskuZ9ez",
"r1y8KxiEf",
"ByXaZpqEG",
"r1PPfGuNM",
"HJ2y6-5gz",
"Hk2Kgd3gM",
"SytTB_TXG",
"BkI24OTQz",
"SyJIEO6XG",
"BkjM4uamG",
"BkNKQ_amG",
"B1Lr7dp7M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper introduces a new dataset called FigureQA. The images are synthetic scientific style figures and questions are generated from 15 templates which concern various relationships between plot elements. The authors experiment with the proposed dataset with 3 baselines. Text-only baseline, which only considers... | [
6,
-1,
-1,
-1,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
-1,
-1,
-1,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SyunbfbAb",
"BkI24OTQz",
"BkNKQ_amG",
"SyJIEO6XG",
"iclr_2018_SyunbfbAb",
"iclr_2018_SyunbfbAb",
"iclr_2018_SyunbfbAb",
"BJskuZ9ez",
"BkjM4uamG",
"HJ2y6-5gz",
"B1Lr7dp7M",
"Hk2Kgd3gM"
] |
iclr_2018_SylJ1D1C- | PDE-Net: Learning PDEs from Data | Partial differential equations (PDEs) play a prominent role in many disciplines such as applied mathematics, physics, chemistry, material science, computer science, etc. PDEs are commonly derived based on physical laws or empirical observations. However, the governing equations for many complex systems in modern appli... | workshop-papers | This paper studies the approximation and integration of partial differential equations using convolutional neural networks. By constraining CNN filters to have prescribed vanishing moments, the authors interpret CNN-based temporal prediction in terms of 'pde discovery'. The method is demonstrated on simple convection-d... | train | [
"B1JeHrDlf",
"rJVcvUYlz",
"SJt5gvplM",
"Sy9QBDw7z",
"HyYNMZ7Zz",
"H18xu-XZf",
"rky1xeQ-G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The paper explores the use of deep learning machinery for the purpose of identifying dynamical systems specified by PDEs.\n\nThe paper advocates the following approach:\nOne assumes a dynamic PDE system involving differential operators up to a given order. Each differential operator term is approximated by a filte... | [
7,
8,
5,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_SylJ1D1C-",
"iclr_2018_SylJ1D1C-",
"iclr_2018_SylJ1D1C-",
"iclr_2018_SylJ1D1C-",
"rJVcvUYlz",
"B1JeHrDlf",
"SJt5gvplM"
] |
iclr_2018_S1TgE7WR- | Covariant Compositional Networks For Learning Graphs | Most existing neural networks for learning graphs deal with the issue of permutation invariance by conceiving of the network as a message passing scheme, where each node sums the feature vectors coming from its neighbors. We argue that this imposes a limitation on their representation power, and instead propose a new g... | workshop-papers | This is a good contribution, with the potential to become extremely good and significant if presentation is substantially improved.
All reviewers comment on the lack of clarity of the paper, especially concerning its central contributions (Section 4 and 5), as illustrated also by the relatively low confidence scores.
R... | train | [
"S1MHyoFgf",
"H14mK9iNG",
"HkfwgoYef",
"BJxhdf9lf",
"ryrCjA5mz",
"HywtoC97f",
"HJmIyK5QG",
"SkSEy4q7z",
"HkiEKl5Xz",
"HyqaFy9XG",
"SyP9nCt7z",
"rkNRStmzG",
"Hyx1wVQzf",
"Hk0MpZ2WG",
"r1wHO8EWM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public",
"public"
] | [
"Thank you for your contribution to ICLR. The paper covers a very interesting topic and presents some though-provoking ideas. \n\nThe paper introduces \"covariant compositional networks\" with the purpose of learning graph representations. An example application also covered in the experimental section is graph cla... | [
5,
-1,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
-1,
2,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_S1TgE7WR-",
"HJmIyK5QG",
"iclr_2018_S1TgE7WR-",
"iclr_2018_S1TgE7WR-",
"SkSEy4q7z",
"SkSEy4q7z",
"SkSEy4q7z",
"HkiEKl5Xz",
"S1MHyoFgf",
"BJxhdf9lf",
"HkfwgoYef",
"Hyx1wVQzf",
"iclr_2018_S1TgE7WR-",
"r1wHO8EWM",
"iclr_2018_S1TgE7WR-"
] |
iclr_2018_SyUkxxZ0b | Adversarial Spheres | State of the art computer vision models have been shown to be vulnerable to small adversarial perturbations of the input. In other words, most images in the data distribution are both correctly classified by the model and are very close to a visually similar misclassified image. Despite substantial research interes... | workshop-papers | This paper studies the interplay between adversarial examples and generalization in the uniform setting (not specific assumptions on the architecture) in a toy high-dimensional setting. In particular, the authors show a fundamental tradeoff between generalization error and the average distance of adversarial examples.
... | val | [
"r1LAwb9xz",
"rJOiq-clf",
"HkKWeUCef",
"By64zFzVG",
"rJFzMBpmz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author"
] | [
"The idea of analyzing a simple synthetic data set to get insights into open issues about adversarial examples has merit. However, the results reported here are not sufficiently significant for ICLR.\n\nThe authors make a big deal throughout the paper about how close to training data the adversarial examples they c... | [
4,
5,
3,
-1,
-1
] | [
4,
3,
3,
-1,
-1
] | [
"iclr_2018_SyUkxxZ0b",
"iclr_2018_SyUkxxZ0b",
"iclr_2018_SyUkxxZ0b",
"rJFzMBpmz",
"iclr_2018_SyUkxxZ0b"
] |
iclr_2018_ryZ8sz-Ab | Fast and Accurate Text Classification: Skimming, Rereading and Early Stopping | Recent advances in recurrent neural nets (RNNs) have shown much promise in many applications in natural language processing. For most of these tasks, such as sentiment analysis of customer reviews, a recurrent neural net model parses the entire review before forming a decision. We argue that reading the entire input is... | workshop-papers | this is an interesting approach that applies the idea of dynamically controlling the amount of information from the input fed into the classifier (some of the earlier approaches have used this idea for, e.g., parsing, real-time translation, online speech recognition, and so on...) this is also related to some of the re... | train | [
"rkWJ_hbrM",
"SkDsWQqgz",
"HkN4OZr4G",
"HyAwBpKeG",
"rk_3IZjlM",
"BkTK6oNmG",
"Bk7JAsNmM",
"HkrNaoE7z",
"Hkq1piVQM",
"BkOOxwFxz",
"ry89PEKgM",
"Hkc-XfHA-"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"public",
"public"
] | [
"Thanks for the insightful reviews, here are our answers to these questions:\n1- Concerning the sequence2sequence and sequence2scalar problem: We have a typo here that writing the PoS tagging as \"seq2scalar\", this task should be classification problem. We are only addressing the classification problem in this wor... | [
-1,
5,
-1,
5,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
-1,
3,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"HkN4OZr4G",
"iclr_2018_ryZ8sz-Ab",
"BkTK6oNmG",
"iclr_2018_ryZ8sz-Ab",
"iclr_2018_ryZ8sz-Ab",
"SkDsWQqgz",
"HyAwBpKeG",
"rk_3IZjlM",
"iclr_2018_ryZ8sz-Ab",
"iclr_2018_ryZ8sz-Ab",
"iclr_2018_ryZ8sz-Ab",
"iclr_2018_ryZ8sz-Ab"
] |
iclr_2018_B1KJJf-R- | Neural Program Search: Solving Data Processing Tasks from Description and Examples | We present a Neural Program Search, an algorithm to generate programs from natural language description and a small number of input / output examples. The algorithm combines methods from Deep Learning and Program Synthesis fields by designing rich domain-specific language (DSL) and defining efficient search algorithm g... | workshop-papers | the reviewers all found the problem to be important, the proposed approach to be interesting, but the manuscript to be preliminary. i agree with them. | train | [
"SkJQ6oUgG",
"rJmyxltgz",
"BJNUHb5lz",
"SJdDNQvGf",
"SyyJBfwGz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author"
] | [
"This paper presents a seq2Tree model to translate a problem statement in natural \nlanguage to the corresponding functional program in a DSL. The model uses\nan RNN encoder to encode the problem statement and uses an attention-based\ndoubly recurrent network for generating tree-structured output. The learnt model ... | [
4,
5,
7,
-1,
-1
] | [
4,
4,
4,
-1,
-1
] | [
"iclr_2018_B1KJJf-R-",
"iclr_2018_B1KJJf-R-",
"iclr_2018_B1KJJf-R-",
"SkJQ6oUgG",
"rJmyxltgz"
] |
iclr_2018_Sy3XxCx0Z | Natural Language Inference with External Knowledge | Modeling informal inference in natural language is very challenging. With the recent availability of large annotated data, it has become feasible to train complex models such as neural networks to perform natural language inference (NLI), which have achieved state-of-the-art performance. Although there exist relatively... | workshop-papers | the reviewers seem to agree that this submission could be much more strengthened if more investigation is done in two directions: (1) the effect of different, available resources (e.g., in the comment, the authors mentioned WikiData didn't improve, and this raises a question of what kind of properties of external resou... | train | [
"SkGeeKygf",
"rkumgRdxM",
"SkkxKWJbM",
"Sy3Z4By-z",
"S1uNWq3QM",
"r1VZWqhmf",
"rknhgcn7G",
"BytbgcnQG",
"HJCviIs1G",
"rJWNCT9kz",
"SJ-i-Fy1z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public",
"author",
"public"
] | [
"Update:\n\nThe response addressed all my major concerns, and I think the paper is sound. (I'm updating my confidence to a 5.) So, the paper makes an empirically *very* small step in an interesting line of language understanding research. This paper should be published in some form, but my low-ish score is due simp... | [
6,
5,
3,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
5,
4,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_Sy3XxCx0Z",
"iclr_2018_Sy3XxCx0Z",
"iclr_2018_Sy3XxCx0Z",
"iclr_2018_Sy3XxCx0Z",
"SkGeeKygf",
"rkumgRdxM",
"SkkxKWJbM",
"Sy3Z4By-z",
"rJWNCT9kz",
"SJ-i-Fy1z",
"iclr_2018_Sy3XxCx0Z"
] |
iclr_2018_Sk4w0A0Tb | Rotational Unit of Memory | The concepts of unitary evolution matrices and associative memory have boosted the field of Recurrent Neural Networks (RNN) to state-of-the-art performance in a variety of sequential tasks. However, RNN still has a limited capacity to manipulate long-term memory. To bypass this weakness the most successful applicatio... | workshop-papers | although the authors argue that their experiments were selected from the earlier work from which major comparing approaches were taken, the reviewers found the empirical result to be weak. why not some real tasks (i do not believe bAbI nor PTB could be considered real) that could clearly reveal the superiority of the p... | test | [
"Bkq3EZcxM",
"B1yLVH5lM",
"Byf4Vs5gM",
"Byjpfc2Qz",
"r1CbMch7G",
"BJepZqh7G",
"SJdf-qn7f",
"rJDVFq8mM",
"HJsj51iMz",
"HJvHqzeGM",
"HJUg-BVWz",
"H1MfgBVWM",
"SJLOkB4-G",
"H1qQJHEZz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"author",
"author",
"author",
"author"
] | [
"The authors of this paper propose a new type of RNN architecture that modifies the reset gate of GRU with a rotational operator, where this rotational operator serves as an associative memory of their RNN model. The idea is sound, and the way they conduct experiments also make sense. The motivation and the details... | [
4,
6,
5,
-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
] | [
"iclr_2018_Sk4w0A0Tb",
"iclr_2018_Sk4w0A0Tb",
"iclr_2018_Sk4w0A0Tb",
"H1MfgBVWM",
"rJDVFq8mM",
"H1qQJHEZz",
"iclr_2018_Sk4w0A0Tb",
"SJLOkB4-G",
"HJvHqzeGM",
"HJUg-BVWz",
"iclr_2018_Sk4w0A0Tb",
"Bkq3EZcxM",
"B1yLVH5lM",
"Byf4Vs5gM"
] |
iclr_2018_Byd-EfWCb | Decoding Decoders: Finding Optimal Representation Spaces for Unsupervised Similarity Tasks | Experimental evidence indicates that simple models outperform complex deep networks on many unsupervised similarity tasks. Introducing the concept of an optimal representation space, we provide a simple theoretical resolution to this apparent paradox. In addition, we present a straightforward procedure that, without an... | workshop-papers | this submission has two results; (1) it defines what it means for the optimal representation is, although this is rather uninteresting that it simply says that if the representation from a model is going to be used based on some given metric, the cost function should directly reflect it, and (2) it shows that different... | train | [
"BJFJeNPHf",
"H1QnkNwSM",
"S1GVQk5gG",
"HJFbzhFeM",
"SkOj779lM",
"rkjz4_a7z",
"rkwqhGkzz",
"Sk9jGf1fG",
"B1Fzd-yzM"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Dear Reviewer,\n\nAgain thank you for your the time you took to initially assess our work. \n\nWe found the comments you made informative, and hope that the newer manuscript manages to address the issues you had with the paper. The reviewer that has currently re-reviewed the manuscript has concluded that the paper... | [
-1,
-1,
6,
5,
4,
-1,
-1,
-1,
-1
] | [
-1,
-1,
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"HJFbzhFeM",
"SkOj779lM",
"iclr_2018_Byd-EfWCb",
"iclr_2018_Byd-EfWCb",
"iclr_2018_Byd-EfWCb",
"iclr_2018_Byd-EfWCb",
"S1GVQk5gG",
"HJFbzhFeM",
"SkOj779lM"
] |
iclr_2018_rkxY-sl0W | Tree-to-tree Neural Networks for Program Translation | Program translation is an important tool to migrate legacy code in one language into an ecosystem built in a different language. In this work, we are the first to consider employing deep neural networks toward tackling this problem. We observe that program translation is a modular procedure, in which a sub-tree of the ... | workshop-papers | the problem is interesting, and the approach is also interesting. however, the reviewers have found that this manuscript would benefit from more experiments, potentially involving some real data (even at least for evaluation) in addition to largely synthetic data sets used in the submission. i also agree with them and ... | val | [
"HJsu29V4G",
"BkLxPNweG",
"r1-4-eYlf",
"Bk_WgJqgM",
"r1Kak22Xz",
"BJ8e0oh7M",
"rkGWRHCMM",
"BkIEnSAff",
"B1LYcrRMz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"- Linearization\nMake sense, and I recommend to add at least 1 conversion example to the paper to guarantee reproducibility, because some accidental style errors in linearized texts may affect the results.\n\n- \"Meaningless\" test sets\nThe point of my concern is the problem which possibly not-few auto-generated ... | [
-1,
6,
4,
4,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
3,
4,
-1,
-1,
-1,
-1,
-1
] | [
"rkGWRHCMM",
"iclr_2018_rkxY-sl0W",
"iclr_2018_rkxY-sl0W",
"iclr_2018_rkxY-sl0W",
"B1LYcrRMz",
"iclr_2018_rkxY-sl0W",
"Bk_WgJqgM",
"r1-4-eYlf",
"BkLxPNweG"
] |
iclr_2018_BkoXnkWAb | Shifting Mean Activation Towards Zero with Bipolar Activation Functions | We propose a simple extension to the ReLU-family of activation functions that allows them to shift the mean activation across a layer towards zero. Combined with proper weight initialization, this alleviates the need for normalization layers. We explore the training of deep vanilla recurrent neural networks (RNNs) with... | workshop-papers | the reviewers were not fully convinced of the setting under which the proposed bipolar activation function was found by the authors to be preferable, and neither am i. | train | [
"SkGpJUL4G",
"SkBvfy5lz",
"rJC1hZqxf",
"B1qbgHcxz",
"SJiFf4jZG",
"SJNVMViZG",
"Sy3ZW4sZz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Thanks for your answer,\n\n I like the general idea of bipolar activation, but I think the empirical evaluation still need to be improved. Although authors show that bipolar activation improve the trainability of deep-stack RNN and simple convolutional networks, their approach tends to underperform other methods... | [
-1,
4,
5,
5,
-1,
-1,
-1
] | [
-1,
4,
5,
3,
-1,
-1,
-1
] | [
"SJNVMViZG",
"iclr_2018_BkoXnkWAb",
"iclr_2018_BkoXnkWAb",
"iclr_2018_BkoXnkWAb",
"SkBvfy5lz",
"rJC1hZqxf",
"B1qbgHcxz"
] |
iclr_2018_By3v9k-RZ | LEARNING TO ORGANIZE KNOWLEDGE WITH N-GRAM MACHINES | Deep neural networks (DNNs) had great success on NLP tasks such as language modeling, machine translation and certain question answering (QA) tasks. However, the success is limited at more knowledge intensive tasks such as QA from a big corpus. Existing end-to-end deep QA models (Miller et al., 2016; Weston et al., 201... | workshop-papers | i am a big fan of this idea, but i agree with the reviewers that evaluating this idea on bAbI (which was originally created from a small set of rules and primitives) discounts quite a bit of what is being claimed here. one of the future directions mentioned by the authors ("investigating whether the proposed n-gram rep... | train | [
"rJg3uzqlG",
"BkonMcoef",
"HyywCCnef",
"B11RkcSGz",
"B1trkcBfM",
"B1VMJcHMM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The authors propose the N-Gram machine to answer questions over long documents. The model first encodes the document via tuple extraction. An autoencoder objective is used to produce meaningful tuples. Then, the model generates a program, based on the extracted tuple collection and the question, to find an answer.... | [
4,
5,
4,
-1,
-1,
-1
] | [
4,
4,
3,
-1,
-1,
-1
] | [
"iclr_2018_By3v9k-RZ",
"iclr_2018_By3v9k-RZ",
"iclr_2018_By3v9k-RZ",
"rJg3uzqlG",
"BkonMcoef",
"HyywCCnef"
] |
iclr_2018_HkcTe-bR- | Exploring Deep Recurrent Models with Reinforcement Learning for Molecule Design | The design of small molecules with bespoke properties is of central importance to drug discovery. However significant challenges yet remain for computational methods, despite recent advances such as deep recurrent networks and reinforcement learning strategies for sequence generation, and it can be difficult to compar... | workshop-papers | The paper creates a dataset for exploration of RL for molecular design and I think this makes it a strong contribution to the community at the intersection of the two. For a methods focussed conference such as ICLR however, it may not be the best fit. Hence I would recommend submitting to a workshop track or targeting ... | val | [
"ryhZvNRNM",
"HJlpDGpEG",
"rkUfmabyM",
"rkejdYtxz",
"S1ZlQfqeM",
"r19_YHamz",
"HyIP4S6QM",
"Hkqv7raQz",
"H1P3mSpQf"
] | [
"author",
"public",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The first version our paper was uploaded to openreview on 27th Oct 2017.\n\nThe paper by Popova et al. was put on arXiv on 29th Nov 2017, that is about a month later than our paper.",
"There is a recently published paper on molecular design with deep reinforcement learning which is not addressed in your work:\nP... | [
-1,
-1,
4,
7,
6,
-1,
-1,
-1,
-1
] | [
-1,
-1,
2,
4,
3,
-1,
-1,
-1,
-1
] | [
"HJlpDGpEG",
"iclr_2018_HkcTe-bR-",
"iclr_2018_HkcTe-bR-",
"iclr_2018_HkcTe-bR-",
"iclr_2018_HkcTe-bR-",
"rkUfmabyM",
"rkejdYtxz",
"iclr_2018_HkcTe-bR-",
"S1ZlQfqeM"
] |
iclr_2018_SkBYYyZRZ | Searching for Activation Functions | The choice of activation functions in deep networks has a significant effect on the training dynamics and task performance. Currently, the most successful and widely-used activation function is the Rectified Linear Unit (ReLU). Although various hand-designed alternatives to ReLU have been proposed, none have managed to... | workshop-papers | The author's propose to use swish and show that it performs significantly better than Relus on sota vision models. Reviewers and anonymous ones counter that PRelus should be doing quite well too. Unfortunately, the paper falls in the category where it is hard to prove the utility of the method through one paper alone, ... | train | [
"BkIXIiLNG",
"Sy-QnQHef",
"Hy7GD19gM",
"HylYITVZG",
"HJ5pEygNM",
"Skfsiap7G",
"r1a4oTTmz",
"rJMj2S57z",
"rkQoM7wmM",
"rk32mXwXz",
"SkVAW7PXM",
"HkC-JdjkG",
"S1jZrPjyG",
"S1UnJvoyz",
"B1sGYLokG",
"SkQHfvoA-"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"public",
"author",
"author",
"official_reviewer",
"author",
"author",
"author",
"public",
"public",
"public",
"public",
"public"
] | [
"1. Novelty \n\nThe methodology of searching has been used in Genetic Programming for a long time. The RNN controller has been used in many paper from Google Brain. This paper's contribution is using RL to search in a GP flavor. Although it is new in activation function search field, in methodology view, it is not ... | [
-1,
4,
5,
7,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"rkQoM7wmM",
"iclr_2018_SkBYYyZRZ",
"iclr_2018_SkBYYyZRZ",
"iclr_2018_SkBYYyZRZ",
"B1sGYLokG",
"rJMj2S57z",
"iclr_2018_SkBYYyZRZ",
"rk32mXwXz",
"Hy7GD19gM",
"Sy-QnQHef",
"HylYITVZG",
"iclr_2018_SkBYYyZRZ",
"iclr_2018_SkBYYyZRZ",
"iclr_2018_SkBYYyZRZ",
"iclr_2018_SkBYYyZRZ",
"iclr_2018_... |
iclr_2018_ByQZjx-0- | Faster Discovery of Neural Architectures by Searching for Paths in a Large Model | We propose Efficient Neural Architecture Search (ENAS), a faster and less expensive approach to automated model design than previous methods. In ENAS, a controller learns to discover neural network architectures by searching for an optimal path within a larger model. The controller is trained with policy gradient to se... | workshop-papers | First off, this was a difficult paper to decide on. There was some vigorous discussion on the paper centering around the choices that were available to the conv-nets. The author's strongly emphasized the improvements on the PTB task.
For my part, I think the method is very compelling -- sharing weights for all the mo... | train | [
"BkrqNswgf",
"rJaD7VugM",
"Bkj-7CYef",
"BJI7WwsQz",
"rJE-zQ9mG",
"Sksb-TtQM",
"rkMLW6K7z",
"H1uV2IG7M",
"B1BcnLGXG",
"S1KqcLMmz",
"BJREcLfXM",
"H1wfqUMXf",
"HkAyq8MQf",
"HkpjFLfQM",
"ryc0uUz7M",
"H1ThH37-f",
"By51thRez",
"BJi-csZgM",
"rJpVtZbgG",
"rJWmCYxyM",
"r1SrSDy1f"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"public",
"author",
"public"
] | [
"In the paper titled \"Faster Discovery of Neural Architectures by Searching for Paths in a Large Model\", the authors proposed an efficient algorithm which can be used for efficient (less resources and time) and faster architecture design for neural networks. The motivation of the new algorithm is by sharing param... | [
6,
5,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
2,
3,
2,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_ByQZjx-0-",
"iclr_2018_ByQZjx-0-",
"iclr_2018_ByQZjx-0-",
"rJE-zQ9mG",
"Sksb-TtQM",
"rJaD7VugM",
"Sksb-TtQM",
"By51thRez",
"H1ThH37-f",
"iclr_2018_ByQZjx-0-",
"BkrqNswgf",
"HkAyq8MQf",
"HkpjFLfQM",
"rJaD7VugM",
"Bkj-7CYef",
"rJaD7VugM",
"iclr_2018_ByQZjx-0-",
"rJpVtZbgG"... |
iclr_2018_r1pW0WZAW | Analyzing and Exploiting NARX Recurrent Neural Networks for Long-Term Dependencies | Recurrent neural networks (RNNs) have achieved state-of-the-art performance on many diverse tasks, from machine translation to surgical activity recognition, yet training RNNs to capture long-term dependencies remains difficult. To date, the vast majority of successful RNN architectures alleviate this problem using nea... | workshop-papers | I think the model itself is not very novel, as pointed by the reviewers and the analysis is not very insightful either. However, the results themselves are interesting and quite good (on the copy task and pMnist, but not so much the other datasets presented (timit etc) where it not clear that long term dependencies wou... | train | [
"BJMTxiOlG",
"H1OSO2dlz",
"rycLSbcgf",
"BJMxrXomf",
"SyaF_Qomf",
"SJkNAGiQM",
"Bka_1mi7M"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"The followings are my main critics of the paper: \n1. Analysis does not provide any new insights. \n2. Similar work (recurrent skip coefficient and the corresponding architecture in [1]) has been done, but has not been mentioned. \n3. The experimental results are not convincing. This includes 1. the choices of tas... | [
3,
7,
6,
-1,
-1,
-1,
-1
] | [
4,
5,
4,
-1,
-1,
-1,
-1
] | [
"iclr_2018_r1pW0WZAW",
"iclr_2018_r1pW0WZAW",
"iclr_2018_r1pW0WZAW",
"H1OSO2dlz",
"BJMTxiOlG",
"iclr_2018_r1pW0WZAW",
"rycLSbcgf"
] |
iclr_2018_S1lN69AT- | To Prune, or Not to Prune: Exploring the Efficacy of Pruning for Model Compression | Model pruning seeks to induce sparsity in a deep neural network's various connection matrices, thereby reducing the number of nonzero-valued parameters in the model. Recent reports (Han et al., 2015; Narang et al., 2017) prune deep networks at the cost of only a marginal loss in accuracy and achieve a sizable reduction... | workshop-papers | The authors present a thorough exploration of large-sparse models that are pruned down to a target size and show that these models can perform better than small dense models. Results are shown on a variety of datasets with as conv models and seq2seq. The authors even go so far as to release the code. I think the author... | train | [
"rJPMO6YxG",
"BkE3cW5gG",
"r1L5FHqeG",
"HySffdp7z",
"S1kV-OamM",
"ryDNsFvAb"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"public"
] | [
"This paper presents a comparison of model sizes and accuracy variation for pruned version of over-parameterized deep networks and smaller but dense models of the same size. It also presents an algorithm for gradual pruning of small magnitude weight to achieve a pre-determined level of sparsity. The paper demonstra... | [
5,
5,
5,
-1,
-1,
-1
] | [
4,
4,
5,
-1,
-1,
-1
] | [
"iclr_2018_S1lN69AT-",
"iclr_2018_S1lN69AT-",
"iclr_2018_S1lN69AT-",
"iclr_2018_S1lN69AT-",
"r1L5FHqeG",
"iclr_2018_S1lN69AT-"
] |
iclr_2018_SkRsFSRpb | GeoSeq2Seq: Information Geometric Sequence-to-Sequence Networks | The Fisher information metric is an important foundation of information geometry, wherein it allows us to approximate the local geometry of a probability distribution. Recurrent neural networks such as the Sequence-to-Sequence (Seq2Seq) networks that have lately been used to yield state-of-the-art performance on speech... | workshop-papers | The reviewers found the paper meaningful but noted that they were not convinced by the experiments as they stand and the presentation was dense for them. | train | [
"HJlgLNYxf",
"rkU2dUDlf",
"HknoNEWZz",
"rJLKk-67G",
"SkSV0xp7f",
"HyNaTgpmf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"==== UPDATE AFTER REVIEWER RESPONSE\n\nI apologize to the authors for my late response.\n\nI appreciate the reviewer responses, and they are helpful on a number of\nfronts. Still, there are several problematic points.\n\nFirst, as the authors anticipated, I question whether the geometric encoding\noperations can b... | [
5,
4,
5,
-1,
-1,
-1
] | [
2,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_SkRsFSRpb",
"iclr_2018_SkRsFSRpb",
"iclr_2018_SkRsFSRpb",
"rkU2dUDlf",
"HJlgLNYxf",
"HknoNEWZz"
] |
iclr_2018_ryALZdAT- | Feature Incay for Representation Regularization | Softmax-based loss is widely used in deep learning for multi-class classification, where each class is represented by a weight vector and each sample is represented as a feature vector. Different from traditional learning algorithms where features are pre-defined and only weight vectors are tunable through training, fe... | workshop-papers | + An intriguing novel regularization method: encouraging larger norms for the feature vector input to the last softmax layer of a classifier.
+ Resonably extensive experimental validation shows that it improves test accuracy to some degree.
- While a motivation is given, the formal analysis of what is really going o... | train | [
"SkNxPOYlf",
"BkEcWHKlf",
"ryRBHPFxz",
"H1ASOQr-f",
"HkjBl2jZf",
"HkbH3Pobz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The analyses of this paper (1) increasing the feature norm of correctly-classified examples induce smaller training loss, (2) increasing the feature norm of mis-classified examples upweight the contribution from hard examples, are interesting. The reciprocal norm loss seems to be reasonable idea to improve the CNN... | [
6,
6,
6,
-1,
-1,
-1
] | [
3,
2,
4,
-1,
-1,
-1
] | [
"iclr_2018_ryALZdAT-",
"iclr_2018_ryALZdAT-",
"iclr_2018_ryALZdAT-",
"BkEcWHKlf",
"ryRBHPFxz",
"SkNxPOYlf"
] |
iclr_2018_HkmaTz-0W | Visualizing the Loss Landscape of Neural Nets | Neural network training relies on our ability to find ````````"good" minimizers of highly non-convex loss functions. It is well known that certain network architecture designs (e.g., skip connections) produce loss functions that train easier, and well-chosen training parameters (batch size, learning rate, optimizer) pr... | workshop-papers | This work proposes an improved visualisation techniques for ReLU networks that compensates for filter scale symmetries/invariances, thus allowing a more meaningful comparison of low-dimensional projected optimization landscapes between different network architectures.
- the visualisation techniques are a small variati... | train | [
"SkB16fKxf",
"ByfiU65gf",
"S1O4Hinlf",
"rJkCfR2Gz",
"SJ6ZPA3zz",
"SJJJIAhzz",
"BJwKFTtMM",
"r14kg0WfG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"public"
] | [
"This paper provides visualizations of different deep network loss surfaces using 2D contour plots, both at minima and along optimization trajectories. They mention some subtle details that must be taken into account, such as scaling the plot axes by the filter magnitudes, in order to obtain correctly scaled plots.... | [
5,
4,
5,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HkmaTz-0W",
"iclr_2018_HkmaTz-0W",
"iclr_2018_HkmaTz-0W",
"ByfiU65gf",
"SkB16fKxf",
"S1O4Hinlf",
"r14kg0WfG",
"iclr_2018_HkmaTz-0W"
] |
iclr_2018_SJTB5GZCb | Extending the Framework of Equilibrium Propagation to General Dynamics | The biological plausibility of the backpropagation algorithm has long been doubted by neuroscientists. Two major reasons are that neurons would need to send two different types of signal in the forward and backward phases, and that pairs of neurons would need to communicate through symmetric bidirectional connections.
... | workshop-papers | + interesting novel extension of equilibrium propagation, as a biologically more plausible alternative to backpropagation, with encouraging initial experimental validation.
- currently lacks theoretical guarantees regarding convergence of the algorithm to a meaningful result
- experimental study should be more exte... | test | [
"rJNtVBtgM",
"ryzHMf9gz",
"HJqm1ilzG",
"SJcRZdpXz",
"rkyMfdaQz",
"BJnjZdp7G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The manuscript discusses a learning algorithm that is based on the equilibrium propagation method, which can be applied to networks with asymmetric connections. This extension is interesting, but the results seem to be incomplete and missing necessary additional analyses. Therefore, I do not recommend acceptance o... | [
4,
3,
6,
-1,
-1,
-1
] | [
4,
4,
2,
-1,
-1,
-1
] | [
"iclr_2018_SJTB5GZCb",
"iclr_2018_SJTB5GZCb",
"iclr_2018_SJTB5GZCb",
"ryzHMf9gz",
"rJNtVBtgM",
"HJqm1ilzG"
] |
iclr_2018_SkmiegW0b | Challenges in Disentangling Independent Factors of Variation |
We study the problem of building models that disentangle independent factors of variation. Such models encode features that can efficiently be used for classification and to transfer attributes between different images in image synthesis. As data we use a weakly labeled training set, where labels indicate what s... | workshop-papers | The paper proposes a method to disentangle style from content (two factor disentanglement) using weak labels (information about the common factor for a pair of images). It is similar to an earlier work by Mathieu et al (2016) with main novelty being in the use of the discriminator which operates with pairs of images in... | train | [
"rycISJNgz",
"H1P_fBdeM",
"HJbE6CKlM",
"rJvWy-Cbf",
"B1e06xCZM",
"rycS6xC-G"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Quality\nThe method description, particularly about reference ambiguity, I found difficult to follow. The experiments and analysis look solid, although it would be nice to see experiments on more challenging natural image datasets. \n\nClarity\n“In general this is not possible… “ - you are saying it is not possibl... | [
6,
5,
5,
-1,
-1,
-1
] | [
4,
3,
3,
-1,
-1,
-1
] | [
"iclr_2018_SkmiegW0b",
"iclr_2018_SkmiegW0b",
"iclr_2018_SkmiegW0b",
"rycISJNgz",
"H1P_fBdeM",
"HJbE6CKlM"
] |
iclr_2018_HkpYwMZRb | Gradients explode - Deep Networks are shallow - ResNet explained | Whereas it is believed that techniques such as Adam, batch normalization and, more recently, SeLU nonlinearities ``solve'' the exploding gradient problem, we show that this is not the case and that in a range of popular MLP architectures, exploding gradients exist and that they limit the depth to which networks can be ... | workshop-papers | The paper sets out to analyze the problem of exploding gradients in deep nets which is of fundamental importance. Reviewers largely acknowledge the novelty of the main ideas in the paper towards this goal, however it is also strongly felt that the writing/presentation of the paper needs significant improvement to make ... | train | [
"B1E_6q8NM",
"Byt6QYOxf",
"Skuwdz5eM",
"rJ4WHpjgM",
"HyYweyxzf",
"B15gl1xMz",
"H1-9y1efM",
"ByqOJylff",
"BkZGR0JzM",
"B1TiCL51z",
"HkbTdUcyf",
"Bk46DI9yG",
"B1EqOIqJf",
"B1HYDI5kz"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"Dear Reviewer,\n\nThank you for your response. You criticize that our revised version contained re-orderings of the text rather than substantive changes. I apologize if I was not able to address your criticisms in the way you wanted in the revised version.\n\nAs far as I could tell, your criticisms of the original... | [
-1,
3,
5,
8,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
2,
4,
1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"Byt6QYOxf",
"iclr_2018_HkpYwMZRb",
"iclr_2018_HkpYwMZRb",
"iclr_2018_HkpYwMZRb",
"Byt6QYOxf",
"Byt6QYOxf",
"Skuwdz5eM",
"Skuwdz5eM",
"rJ4WHpjgM",
"iclr_2018_HkpYwMZRb",
"iclr_2018_HkpYwMZRb",
"iclr_2018_HkpYwMZRb",
"iclr_2018_HkpYwMZRb",
"iclr_2018_HkpYwMZRb"
] |
iclr_2018_Syr8Qc1CW | DNA-GAN: Learning Disentangled Representations from Multi-Attribute Images | Disentangling factors of variation has always been a challenging problem in representation learning. Existing algorithms suffer from many limitations, such as unpredictable disentangling factors, bad quality of generated images from encodings, lack of identity information, etc. In this paper, we proposed a supervised a... | workshop-papers | The method proposed in the paper for latent disentanglement and attribute-conditional image generation is novel to the best of my understanding but reviewers (Anon1 and Anon3) have expressed concerns on the quality of results (CelebA images) as well as on the technical presentation and claims in the paper.
Given the n... | val | [
"Bk6FQyPVG",
"ryz0obDxM",
"BJlAlKOgM",
"rkqvQmKxM",
"H19Hbk-bz",
"ryFXlRgZG",
"SyD336yZf",
"Hy-aBlxbM",
"BJ3U_JlWG",
"BkCwgT1Zz",
"SJbRMqk-G",
"HJuqGBnef",
"SkEBPDExM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Thanks for the rebuttal. It clears up some confusion but my score remains slightly negative. ",
"This paper proposes to disentangle attributes by forcing a representation where individual components of this representation account for individual attributes. \n\nPros: \n+ The idea of forcing different parts of the... | [
-1,
4,
5,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
4,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"Hy-aBlxbM",
"iclr_2018_Syr8Qc1CW",
"iclr_2018_Syr8Qc1CW",
"iclr_2018_Syr8Qc1CW",
"ryFXlRgZG",
"BJ3U_JlWG",
"BkCwgT1Zz",
"ryz0obDxM",
"SyD336yZf",
"BJlAlKOgM",
"rkqvQmKxM",
"SkEBPDExM",
"iclr_2018_Syr8Qc1CW"
] |
iclr_2018_ryDNZZZAW | Multiple Source Domain Adaptation with Adversarial Learning | While domain adaptation has been actively researched in recent years, most theoretical results and algorithms focus on the single-source-single-target adaptation setting. Naive application of such algorithms on multiple source domain adaptation problem may lead to suboptimal solutions. We propose a new generalization b... | workshop-papers | Pros
-- Lays out bounds for multi-domain adaptation based on earlier work on a single source-target domain pair.
-- Shows gains over choosing the best source domain for a target domain, or naively combining domains.
Cons
-- The reviewers agree that the extensions are relatively straightforward extensions to single sou... | val | [
"ryY-CNPEG",
"SJ0ahZwEf",
"rkT23LLEz",
"S1yZTj_lz",
"HyxIlUFlz",
"SkmS_5--z",
"Hkuk_f7ZG",
"rkwtwWmNG",
"B1x2PDezz",
"S1nuuvxzf",
"ryXLuDgfz",
"r1-VOPlff",
"rkpWuPlfG"
] | [
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"I thank the authors for their responsiveness. It seems that we reach a common ground. The authors added a comment about the possibility to combine k single-source bounds to obtain a possibly tighter bound (I appreciate the honesty). Intuitively, the fact that we loosen the bound to obtain a more desirable trade-of... | [
-1,
-1,
-1,
6,
6,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
-1,
-1,
3,
5,
5,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"SJ0ahZwEf",
"rkT23LLEz",
"ryXLuDgfz",
"iclr_2018_ryDNZZZAW",
"iclr_2018_ryDNZZZAW",
"iclr_2018_ryDNZZZAW",
"iclr_2018_ryDNZZZAW",
"S1yZTj_lz",
"iclr_2018_ryDNZZZAW",
"SkmS_5--z",
"HyxIlUFlz",
"Hkuk_f7ZG",
"S1yZTj_lz"
] |
iclr_2018_H1I3M7Z0b | WSNet: Learning Compact and Efficient Networks with Weight Sampling | We present a new approach and a novel architecture, termed WSNet, for learning compact and efficient deep neural networks. Existing approaches conventionally learn full model parameters independently and then compress them via \emph{ad hoc} processing such as model pruning or filter factorization. Alternatively, WSNet... | workshop-papers | The paper received generally positive reviews, but the reviewers also had some concerns about the evaluations.
Pros:
-- An improvement over HashNet, the model ties weights more systematically, and gets better accuracy.
Cons:
-- Tying weights to compress models already tried before.
-- Tasks are all small and/or audio ... | train | [
"Bkc2TkFlG",
"S1xBMQtgG",
"rJRJeMoxz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"In this work, the authors propose a technique to compress convolutional and fully-connected layers in a network by tying various weights in the convolutional filters: specifically within a single channel (weight sampling) and across channels (channel sampling). When combined with quantization, the proposed approac... | [
6,
6,
5
] | [
4,
3,
5
] | [
"iclr_2018_H1I3M7Z0b",
"iclr_2018_H1I3M7Z0b",
"iclr_2018_H1I3M7Z0b"
] |
iclr_2018_rybAWfx0b | COLD FUSION: TRAINING SEQ2SEQ MODELS TOGETHER WITH LANGUAGE MODELS | Sequence-to-sequence (Seq2Seq) models with attention have excelled at tasks which involve generating natural language sentences such as machine translation, image captioning and speech recognition. Performance has further been improved by leveraging unlabeled data, often in the form of a language model. In this work, w... | workshop-papers | Pros
-- A novel way to incorporate LM into an end-to-end model, with good adaptation results.
Cons
-- Lacks results on public corpora or results are not close to SOTA (e.g., for Librispeech).
Given the reviews, it is clear that the experimental evaluations can be improved. But the presented approach is novel and inte... | train | [
"Sy0xMaHlG",
"ryGQ4uugM",
"B10RWItgz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer"
] | [
"The paper proposes a novel approach to integrate a language model (LM) to a seq2seq based speech recognition system (ASR). The LM is pretrained on separate data (presumably larger, potentially not the same exact distribution). It has a similar flavor as DeepFusion (DF), a previous work which also integrated an LM ... | [
5,
6,
5
] | [
5,
5,
5
] | [
"iclr_2018_rybAWfx0b",
"iclr_2018_rybAWfx0b",
"iclr_2018_rybAWfx0b"
] |
iclr_2018_S1pWFzbAW | Weightless: Lossy Weight Encoding For Deep Neural Network Compression | The large memory requirements of deep neural networks strain the capabilities of many devices, limiting their deployment and adoption. Model compression methods effectively reduce the memory requirements of these models, usually through applying transformations such as weight pruning or quantization. In this paper, we ... | workshop-papers | Pros:
-- Use of Bloomier filters for lossy compression of nets is novel and well motivated, with interesting compression performance.
Cons:
-- Does lossy compression for transmission, doesn’t address FLOPS required for runtime execution. A lot of times, client devices do not have enough cpu to run large networks (title... | train | [
"HynYT5_xz",
"SyJvYjugz",
"Bk_dwGqeG",
"rJG33LYzf",
"BJ2whUKzM",
"H1J4o8KGG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"Summary: The paper addresses the actual problem of compression of deep neural networks. Authors propose to use another technique for sparse matrix storage. Namely, authors propose to use Bloomier filter for more efficient storage of sparse matrices obtained from Dynamic Network Surgery (DNS) method. Moreover, auth... | [
6,
6,
4,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1
] | [
"iclr_2018_S1pWFzbAW",
"iclr_2018_S1pWFzbAW",
"iclr_2018_S1pWFzbAW",
"HynYT5_xz",
"SyJvYjugz",
"Bk_dwGqeG"
] |
iclr_2018_S1Auv-WRZ | Data Augmentation Generative Adversarial Networks | Effective training of neural networks requires much data. In the low-data regime,
parameters are underdetermined, and learnt networks generalise poorly. Data
Augmentation (Krizhevsky et al., 2012) alleviates this by using existing data
more effectively. However standard data augmentation produces only... | workshop-papers | The paper based on cGAN developed a data augmentation GAN to deal with unseen classes of data. The paper developed new modifications to each component and designed network structure using ideas from state-of-the-art nets. As pointed out by reviewer 1 & 2, the technical contribution is not sufficient. We hence recommend... | train | [
"H1O8xnDlM",
"Hym3oxKlf",
"S1vTg99gz",
"SJxgSOxzG",
"SyMfNdxGG",
"HkpnsPxfz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"This paper proposes a conditional Generative Adversarial Networks that is used for data augmentation. In order to evaluate the performance of the proposed model, they use Omniglot, EMNIST, and VGG-Faces datasets and uses in the meta-learning task and standard classification task in the low-data regime. The paper i... | [
4,
9,
6,
-1,
-1,
-1
] | [
4,
5,
3,
-1,
-1,
-1
] | [
"iclr_2018_S1Auv-WRZ",
"iclr_2018_S1Auv-WRZ",
"iclr_2018_S1Auv-WRZ",
"Hym3oxKlf",
"H1O8xnDlM",
"S1vTg99gz"
] |
iclr_2018_rJrTwxbCb | Empirical Analysis of the Hessian of Over-Parametrized Neural Networks | We study the properties of common loss surfaces through their Hessian matrix. In particular, in the context of deep learning, we empirically show that the spectrum of the Hessian is composed of two parts: (1) the bulk centered near zero, (2) and outliers away from the bulk. We present numerical evidence and mathematica... | workshop-papers | Pros:
+ Builds in important ways on the work of Sagun et al., 2016.
Cons:
- The reviewers were very concerned that the assumption in the paper that the second term of Equation (6) is negligible was insufficiently supported, and this concern remained after the discussion and the revision.
- The paper needs to be more p... | train | [
"S1Zf_tBNM",
"ryIbx22yz",
"rJT6jEcgz",
"HkeyY0M-z",
"BJ1qruaQG",
"ByLNSd6XM",
"H1bMzuamM",
"SkTaZOamG",
"rJHmeupmz",
"ryoh5LEMM"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"public"
] | [
"Thanks for attempting to address my concerns. But the responses to Point 2 and Point 3 are not still convincing to me. In particular, the soundness of the assumption for the mathematical justification is still not addressed and the experimental setting comparing SB against LB is not well designed. Considering the ... | [
-1,
5,
4,
5,
-1,
-1,
-1,
-1,
-1,
-1
] | [
-1,
2,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"SkTaZOamG",
"iclr_2018_rJrTwxbCb",
"iclr_2018_rJrTwxbCb",
"iclr_2018_rJrTwxbCb",
"ryoh5LEMM",
"ryIbx22yz",
"iclr_2018_rJrTwxbCb",
"rJT6jEcgz",
"HkeyY0M-z",
"iclr_2018_rJrTwxbCb"
] |
iclr_2018_B1Z3W-b0W | Learning to Infer | Inference models, which replace an optimization-based inference procedure with a learned model, have been fundamental in advancing Bayesian deep learning, the most notable example being variational auto-encoders (VAEs). In this paper, we propose iterative inference models, which learn how to optimize a variational lowe... | workshop-papers | This paper is intersting but has a few flaws that still need to be addressed. As one reviewer noted, "the authors seems to have simply applied the method of Andrychowicz et al. If they added some discussion and experiments clearly showing why this is a better way to improve the existing inference methods, the paper mig... | train | [
"SkLbR5mSf",
"S1W-El7rG",
"rykU00Y4f",
"BkKZyyP4z",
"SkZMqqHEG",
"SkeA8hENf",
"rJuH-vKeG",
"Sy3-NV9xG",
"Bk8FeZjgf",
"BkIzESGGM",
"H1aYeSMGz",
"HJ5tTEGGf"
] | [
"author",
"official_reviewer",
"author",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"We’re glad that you find the revised version is an improvement and more clearly conveys the contributions of the paper.\n\nIterative inference model parameter gradients are obtained using the reparameterization trick, as with standard inference models. The difference is that these gradients are obtained and averag... | [
-1,
-1,
-1,
-1,
-1,
-1,
5,
6,
5,
-1,
-1,
-1
] | [
-1,
-1,
-1,
-1,
-1,
-1,
4,
5,
4,
-1,
-1,
-1
] | [
"S1W-El7rG",
"SkZMqqHEG",
"iclr_2018_B1Z3W-b0W",
"BkIzESGGM",
"SkeA8hENf",
"H1aYeSMGz",
"iclr_2018_B1Z3W-b0W",
"iclr_2018_B1Z3W-b0W",
"iclr_2018_B1Z3W-b0W",
"Sy3-NV9xG",
"rJuH-vKeG",
"Bk8FeZjgf"
] |
iclr_2018_HkCnm-bAb | Can Deep Reinforcement Learning solve Erdos-Selfridge-Spencer Games? | Deep reinforcement learning has achieved many recent successes, but our understanding of its strengths and limitations is hampered by the lack of rich environments in which we can fully characterize optimal behavior, and correspondingly diagnose individual actions against such a characterization.
Here we ... | workshop-papers | The paper introduces an interesting family of two-player zero-sum games with tunable complexity, called Erdos-Selfridge-Spencer games, as a new domain for RL. The authors report on extensive empirical results using a wide variety of training methods, including supervised learning and several flavors of RL (PPO, A2C, D... | train | [
"SJNicmVeM",
"rynBydweG",
"r16fdy3xG",
"ryyIUs7VM",
"HkBezd67G",
"H18csR2mz",
"SJreVx2Xf",
"SJK3Xl2mz",
"SJIqzlhXz",
"r1hM8u5Xz",
"r1aTS_qQM",
"ByYUrO5QM",
"Sk187Bmff",
"SJVXQS7Gf",
"H1KaWSmff",
"Sy39ZHQMf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"The paper presents Erdos-Selfridge-Spencer games as environments for investigating\ndeep reinforcement learning algorithms. The proposed games are interesting and clearly challenging, but I am not sure what they tell us about the algorithms chosen to test them. There are some clarity issues with the justification ... | [
5,
6,
6,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
3,
3,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_HkCnm-bAb",
"iclr_2018_HkCnm-bAb",
"iclr_2018_HkCnm-bAb",
"H1KaWSmff",
"SJNicmVeM",
"Sy39ZHQMf",
"SJK3Xl2mz",
"SJIqzlhXz",
"SJNicmVeM",
"SJNicmVeM",
"rynBydweG",
"r16fdy3xG",
"SJVXQS7Gf",
"SJNicmVeM",
"rynBydweG",
"r16fdy3xG"
] |
iclr_2018_Bya8fGWAZ | Value Propagation Networks | We present Value Propagation (VProp), a parameter-efficient differentiable planning module built on Value Iteration which can successfully be trained in a reinforcement learning fashion to solve unseen tasks, has the capability to generalize to larger map sizes, and can learn to navigate in dynamic environments. We eva... | workshop-papers | This paper and reviews makes for a difficult call. The reviewers appear to be in agreement that Value Propagation provides an interesting algorithmic advance over earlier work on Value Iteration networks. AnonReviewer1 gives a strong rationale why the advance is both original and significant. Their experiments also ... | train | [
"Sy5I_xKgM",
"S1I3_bqgM",
"rJRfJZKxf",
"By8Sseq7G",
"r1ewFxqXG",
"rkRPdgqQz"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author"
] | [
"The original value-iteration network paper assumed that it was trained on near-expert trajectories and used that information to learn a convolutional transition model that could be used to solve new problem instances effectively without further training.\n\nThis paper extends that work by\n- training from reinforc... | [
5,
7,
5,
-1,
-1,
-1
] | [
4,
3,
2,
-1,
-1,
-1
] | [
"iclr_2018_Bya8fGWAZ",
"iclr_2018_Bya8fGWAZ",
"iclr_2018_Bya8fGWAZ",
"Sy5I_xKgM",
"rJRfJZKxf",
"S1I3_bqgM"
] |
iclr_2018_BkfEzz-0- | Neuron as an Agent | Existing multi-agent reinforcement learning (MARL) communication methods have relied on a trusted third party (TTP) to distribute reward to agents, leaving them inapplicable in peer-to-peer environments. This paper proposes reward distribution using {\em Neuron as an Agent} (NaaA) in MARL without a TTP with two key ide... | workshop-papers |
The reviewers have significantly different views, with one strongly negative,
one strongly positive, and one borderline negative. However, all three
reviews seem to regard the NaaA framework as a very interesting and novel approach to training neural nets. They also concur that the major issue with the paper is very... | train | [
"Bk3zRoBGz",
"H12VRW9gM",
"HJSqWxjez",
"HkQCEwaXM",
"SyM-3W87f",
"BJgxl4WbM",
"Bykv9-bWG",
"H1mjFWW-G",
"BJxz0TA1f",
"BJTvnoCJG",
"BkTpRSRkf",
"SJPHs661G",
"rkdWpPEJz",
"ByRdDLV1z"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author",
"author",
"public",
"public",
"author",
"public"
] | [
"This paper proposed a novel framework Neuron as an Agent (NaaA) for training neural networks to perform various machine learning tasks, including classification (supervised learning) and sequential decision making (reinforcement learning). The NaaA framework is based on the idea of treating all neural network unit... | [
6,
7,
3,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
3,
5,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BkfEzz-0-",
"iclr_2018_BkfEzz-0-",
"iclr_2018_BkfEzz-0-",
"iclr_2018_BkfEzz-0-",
"Bk3zRoBGz",
"H12VRW9gM",
"HJSqWxjez",
"HJSqWxjez",
"BkTpRSRkf",
"SJPHs661G",
"iclr_2018_BkfEzz-0-",
"iclr_2018_BkfEzz-0-",
"ByRdDLV1z",
"iclr_2018_BkfEzz-0-"
] |
iclr_2018_Hk91SGWR- | Investigating Human Priors for Playing Video Games | What makes humans so good at solving seemingly complex video games? Unlike computers, humans bring in a great deal of prior knowledge about the world, enabling efficient decision making. This paper investigates the role of human priors for solving video games. Given a sample game, we conduct a series of ablation studi... | workshop-papers | This paper turned out to be quite difficult to call. My take on the pros/cons is:
1. The research topic, how and why humans can massively outperform DQN, is unanimously viewed as highly interesting by all participants.
2. The authors present an original human subject study, aiming to reveal whether human outperforma... | val | [
"HJfzeB4Hz",
"B1mN4AC4M",
"ry07SzQgG",
"S1sHPAWgz",
"BJZ52L6lf",
"Syy7rN67M",
"SJML7NamG",
"SkMvz4TQM",
"B1LEVNTXG"
] | [
"author",
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"We have revised our paper and added new experiments to address all of your previous concerns. It would be great if you can please find some time to look at our response and inform us of any other feedback or concerns. This would go a long way in helping us improve the paper further. Thank you so much once again!",... | [
-1,
-1,
4,
5,
7,
-1,
-1,
-1,
-1
] | [
-1,
-1,
3,
4,
4,
-1,
-1,
-1,
-1
] | [
"Syy7rN67M",
"B1LEVNTXG",
"iclr_2018_Hk91SGWR-",
"iclr_2018_Hk91SGWR-",
"iclr_2018_Hk91SGWR-",
"S1sHPAWgz",
"BJZ52L6lf",
"iclr_2018_Hk91SGWR-",
"ry07SzQgG"
] |
iclr_2018_rJk51gJRb | Adversarial Policy Gradient for Alternating Markov Games | Policy gradient reinforcement learning has been applied to two-player alternate-turn zero-sum games, e.g., in AlphaGo, self-play REINFORCE was used to improve the neural net model after supervised learning. In this paper, we emphasize that two-player zero-sum games with alternating turns, which have been previously for... | workshop-papers | The reviewers agree that the paper is below threshold for acceptance in the main track (one with very low confidence), but they favor submitting the paper to the workshop track.
The paper considers policy gradient methods for two-player zero-sum Alternating Markov games. They propose adversarial policy gradient (fair... | train | [
"SkLrUaZWG",
"rJFql_Nxz",
"ByzeYntef",
"SkNEyzcxG",
"SkURGVXVf",
"SkpoKqYzz",
"SyowYgBfz",
"rkJtIabWM",
"SkGAjuFzz"
] | [
"author",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author"
] | [
"Thanks for your comment. \n\nIn the revised paper, we have added our neural net model to search, the resulting program is stronger than MoHex 2.0 on board sizes 9x9 to 13x13. We have also included a comparison with ExIt. It appears that ExIt might not as strong as MoHex 2.0 (the ExIt paper was comparing their pla... | [
-1,
5,
5,
5,
-1,
-1,
-1,
-1,
-1
] | [
-1,
2,
4,
4,
-1,
-1,
-1,
-1,
-1
] | [
"ByzeYntef",
"iclr_2018_rJk51gJRb",
"iclr_2018_rJk51gJRb",
"iclr_2018_rJk51gJRb",
"SkGAjuFzz",
"rJFql_Nxz",
"iclr_2018_rJk51gJRb",
"ByzeYntef",
"SkNEyzcxG"
] |
iclr_2018_BJInEZsTb | Learning Representations and Generative Models for 3D Point Clouds | Three-dimensional geometric data offer an excellent domain for studying representation learning and generative modeling. In this paper, we look at geometric data represented as point clouds. We introduce a deep autoencoder (AE) network with excellent reconstruction quality and generalization ability. The learned repres... | workshop-papers | This paper compares autoencoder and GAN-based methods for 3D point cloud representation and generation, as well as new (and welcome) metrics for quantitatively evaluating generative models. The experiments form a good but still a bit too incomplete exploration of this topic. More analysis is needed to calibrate the n... | test | [
"SJyXoTtlG",
"B1Mvg-qlM",
"HJf1JQqez",
"H1n5Uv6QG",
"rJoOW2dfz",
"SyYI6idfz",
"S19u3suGf",
"HJfG2jOzG"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper introduces a generative approach for 3D point clouds. More specifically, two Generative Adversarial approaches are introduced: Raw point cloud GAN, and Latent-space GAN (r-GAN and l-GAN as referred to in the paper). In addition, a GMM sampling + GAN decoder approach to generation is also among the exper... | [
6,
8,
5,
-1,
-1,
-1,
-1,
-1
] | [
5,
5,
4,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJInEZsTb",
"iclr_2018_BJInEZsTb",
"iclr_2018_BJInEZsTb",
"iclr_2018_BJInEZsTb",
"iclr_2018_BJInEZsTb",
"SJyXoTtlG",
"B1Mvg-qlM",
"HJf1JQqez"
] |
iclr_2018_BJubPWZRW | Cross-View Training for Semi-Supervised Learning | We present Cross-View Training (CVT), a simple but effective method for deep semi-supervised learning. On labeled examples, the model is trained with standard cross-entropy loss. On an unlabeled example, the model first performs inference (acting as a "teacher") to produce soft targets. The model then learns from these... | workshop-papers | This paper combines ideas from student-teacher training and multi-view learning in a simple but clever way. There is not much novelty in the methods, but promising results are given across several tasks, including realistic NLP tasks. The improvements are not huge but are consistent. Considering the limited novelty,... | train | [
"Bkp-xJ5xf",
"HJhFVtqez",
"SkDHZacef",
"HkwzPvbmf",
"By4SmTOMz",
"S1b4QXkzf",
"ByvGWXyGf",
"BJdP17kzG",
"B12yJQkMf"
] | [
"official_reviewer",
"official_reviewer",
"official_reviewer",
"author",
"author",
"author",
"author",
"author",
"author"
] | [
"This paper presents a so-called cross-view training for semi-supervised deep models. Experiments were conducted on various data sets and experimental results were reported.\n\nPros:\n* Studying semi-supervised learning techniques for deep models is of practical significance.\n\nCons:\n* The novelty of this paper i... | [
2,
5,
7,
-1,
-1,
-1,
-1,
-1,
-1
] | [
4,
4,
4,
-1,
-1,
-1,
-1,
-1,
-1
] | [
"iclr_2018_BJubPWZRW",
"iclr_2018_BJubPWZRW",
"iclr_2018_BJubPWZRW",
"ByvGWXyGf",
"iclr_2018_BJubPWZRW",
"Bkp-xJ5xf",
"HJhFVtqez",
"SkDHZacef",
"iclr_2018_BJubPWZRW"
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.