ICML2017 優化領域有什麼值得關注的論文？

接受的論文名單已經出來了，先佔個樓，請各位大牛不吝賜教。

對於優化的熱點問題或者今後發展方向大家有什麼想法。

非常推薦nonconvex的新分析：[1703.00887] How to Escape Saddle Points Efficiently

然後厚顏無恥的提一下自己的灌水文：[1702.07944] Stochastic Variance Reduction Methods for Policy Evaluation

推薦一下

Li Q, Tai C, Weinan E ,. Dynamics of Stochastic Gradient Algorithms[J]. Computer Science, 2015.

用SDE建模SGD這方面工作有很多

2017COLT

Best Paper. Yuchen Zhang, Percy Liang and Moses Charikar. A Hitting Time Analysis of Stochastic Gradient Langevin Dynamics

Maxim Raginsky, Alexander Rakhlin and Matus Telgarsky. Non-Convex Learning via Stochastic Gradient Langevin Dynamics: A Nonasymptotic Analysis

然後最推薦這篇

Pratik Chaudhari, Adam Oberman, Stanley Osher, Stefano Soatto, and Guillame Carlier, Deep Relaxation: Partial Differential Equations for Optimizing Deep Neural Networks, April 2017 (revised (June 2017)

用hamilton jacobi來建模

我做一個粗粗整理類的事情吧（沒有細看，先粗粗的分類一下好了），

我瞄了瞄我個人關注的領域，Non-Convex, Convex, Stochastic Non-Convex, Stochastic Convex, Frank Wolfe 衍生演算法 。所以Online ，Distributed， Coordinate Descent ,以及Application 之類的我就沒有加入進來。 另外還有許多別的很好的optimization演算法，但因為我之前沒有接觸過我也沒有放進來，如果還遺漏了別的paper也請見諒。

Non-Convex:

1&> [1703.00887] How to Escape Saddle Points Efficiently

2&> [1703.02628] Global optimization of Lipschitz functions

3&> [1705.04925] Convergence Analysis of Proximal Gradient with Momentum for Nonconvex Optimization

Non-Convex + Stochastic:

1&> [1705.05933] Sub-sampled Cubic Regularization for Non-convex Optimization

2&> Faster Non-Convex Stochastic Optimization Via Strongly Non-Convex Parameter

Convex:

1&> [1611.04982] Oracle Complexity of Second-Order Methods for Finite-Sum Problems

2&> [1705.00772] A Semismooth Newton Method for Fast, Generic Convex Programming

3&> [1702.08124] A Unifying Framework for Convergence Analysis of Approximate Newton Methods

Convex + Stochastic:

1&> [1607.01027] Accelerated Stochastic Subgradient Methods under Local Error Bound Condition

2&> A Novel Method for Machine Learning Problems Using Stochastic Recursive Gradient

Frank Wolfe 衍生演算法:

1&> [1610.05120] Lazifying Conditional Gradient Algorithms

2&> [1703.05840] Conditional Accelerated Lazy Stochastic Gradient Descent

也許這篇文章可以幫到你：

ICML 2017論文精選#1 用影響函數(Influence Functions)理解機器學習中的黑盒預測（Best paper award 最佳論文獎@斯坦福）