【干货】ICML2018论文公布！一文了解机器学习最新热议论文和研究热点

来源：新智元

【导读】ICML 2018上周公布了会议接受论文，各家组织机构和研究大牛们在Twitter上纷纷报喜，放出接受论文，恭喜！有Google Brain、DeepMind、Facebook、微软和各大高校等。我们整理了Twitter上的关注度比较热的一些论文，供大家了解，最新关于机器学习的一些热门研究方向！

1.  Differentiable Dynamic Programming for Structured Prediction and Attention

最热的就是这篇第一作者Arthur Mensch‏，来自法国Inria Parietal，也是scikit-learn 作者之一，论文关于结构性预测与注意力中的可微分动态编程。

作者重点指出：Sparsity and backprop in CRF-like inference layers using max-smoothing, application in text + time series (NER, NMT, DTW)。

Twitter上截止到现在 600赞。

论文网址：

http://www.zhuanzhi.ai/document/34c4176a60e002b524b56b5114db0e78

这位评价甚高！ one of the most innovative deep learning papers！

欢迎大家阅读！

2. WaveRNN、Parralel WaveNet 

来自DeepMind的两篇论文关于语音合成！

WaveRNN: http://arxiv.org/abs/1802.08435 

Parallel WaveNet: http://arxiv.org/abs/1711.10433 

WaveNet早已名声卓著，比原来快千倍，语音更自然，已经用在Google自家产品Google Assistant 里~

3. GAN性能表现分析

来自谷歌大脑GoodFellow团队，Is Generator Conditioning Causally Related to GAN Performance? find: 1. Spectrum of G's in/out Jacobian predicts Inception Score. 2. Intervening to change spectrum affects scores a lot

论文链接：https://t.co/cXQDEE2Uee

4.优化方法 Adam分析

Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients

论文地址：https://arxiv.org/abs/1705.07774

5. 图像转换器

论文地址：https://arxiv.org/abs/1802.05751

其他论文列表：

论文地址：

Bayesian Quadrature for Multiple Related Integrals 

https://arxiv.org/abs/1801.04153                  

Stein Points

https://arxiv.org/abs/1803.10161

Active Learning with Logged Data

https://arxiv.org/abs/1802.09069

Analyzing the Robustness of Nearest Neighbors to Adversarial Examples

https://arxiv.org/abs/1706.03922

Hierarchical Imitation and Reinforcement Learning

https://arxiv.org/abs/1803.00590

Analysis of Minimax Error Rate for Crowdsourcing and Its Application to Worker Clustering Model

https://arxiv.org/abs/1802.04551

Detecting and Correcting for Label Shift with Black Box Predictors

https://arxiv.org/abs/1802.03916

Yes, but Did It Work?: Evaluating Variational Inference

https://arxiv.org/abs/1802.02538

MAGAN: Aligning Biological Manifolds

https://arxiv.org/abs/1803.00385

Does Distributionally Robust Supervised Learning Give Robust Classifiers?

https://arxiv.org/abs/1611.02041

Knowledge Transfer with Jacobian Matching

https://arxiv.org/abs/1803.00443

Kronecker Recurrent Units

https://arxiv.org/abs/1705.10142

Entropy-SGD optimizes the prior of a PAC-Bayes bound: Generalization properties of Entropy-SGD and data-dependent priors

https://arxiv.org/abs/1712.09376

The Manifold Assumption and Defenses Against Adversarial Perturbations

https://arxiv.org/abs/1711.08001

Overcoming catastrophic forgetting with hard attention to the task

https://arxiv.org/abs/1801.01423

On the Opportunities and Pitfalls of Nesting Monte Carlo Estimators

https://arxiv.org/abs/1709.06181

Tighter Variational Bounds are Not Necessarily Better

https://arxiv.org/abs/1802.04537

LaVAN: Localized and Visible Adversarial Noise

https://arxiv.org/abs/1801.02608

Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples

https://arxiv.org/abs/1711.09576

Geometry Score: A Method For Comparing Generative Adversarial Networks

https://arxiv.org/abs/1802.02664

高质量延伸阅读

☞【学界】OpenPV：中科院研究人员建立开源的平行视觉研究平台

☞【征稿通知】IEEE IV 2018“智能车辆中的平行视觉”研讨会

☞【学界】基于平行视觉的特定场景下行人检测

☞【征稿】神经计算专刊Virtual Images for Visual Artificial Intelligence

☞【学界】ParallelEye：面向交通视觉研究构建的大规模虚拟图像集

☞【CFP】Virtual Images for Visual Artificial Intelligence

☞【最详尽的GAN介绍】王飞跃等：生成式对抗网络 GAN 的研究进展与展望

☞【智能自动化学科前沿讲习班第1期】王飞跃教授：生成式对抗网络GAN的研究进展与展望

☞【智能自动化学科前沿讲习班第1期】王坤峰副研究员：GAN与平行视觉

☞【重磅】平行将成为一种常态：从SimGAN获得CVPR 2017最佳论文奖说起

☞【平行讲坛】平行图像：图像生成的一个新型理论框架

☞【学界】IJCAI Oral：弱监督实现精确目标检测，上交大提出协同学习框架

☞【深度】CVPR 2018：阿里提出应用 LocalizedGAN 进行半监督训练

☞【深度】无需数学背景！谷歌研究员为你解密生成式对抗网络

☞【深度】Mobileye RSS理论模型：把人类驾驶常识总结为数学模型，教会每一辆无人车

☞【学界】MIT提出使用「深度强化学习」帮助智能体在运动中做出「动作决策」