“机器学习是近20多年兴起的一门多领域交叉学科,涉及概率论、统计学、逼近论、凸分析、算法复杂度理论等多门学科。机器学习理论主要是设计和分析一些让 可以自动“ 学习”的算法。机器学习算法是一类从数据中自动分析获得规律,并利用规律对未知数据进行预测的算法。因为学习算法中涉及了大量的统计学理论,机器学习与统计推断学联系尤为密切,也被称为统计学习理论。算法设计方面,机器学习理论关注可以实现的,行之有效的学习算法。很多 推论问题属于 无程序可循难度,所以部分的机器学习研究是开发容易处理的近似算法。” ——中文维基百科

知识荟萃

机器学习课程 专知搜集

  1. cs229 机器学习 吴恩达
  2. 台大 李宏毅 机器学习
  3. 爱丁堡大学 机器学习与模式识别
  4. Courses on machine learning
  5. CSC2535 -- Spring 2013 Advanced Machine Learning
  6. Stanford CME 323: Distributed Algorithms and Optimization
  7. University at Buffalo CSE574: Machine Learning and Probabilistic Graphical Models Course
  8. Stanford CS229: Machine Learning Autumn 2015
  9. Stanford / Winter 2014-2015 CS229T/STATS231: Statistical Learning Theory
  10. CMU Fall 2015 10-715: Advanced Introduction to Machine Learning
  11. 2015 Machine Learning Summer School: Convex Optimization Short Course
  12. STA 4273H [Winter 2015]: Large Scale Machine Learning
  13. University of Oxford: Machine Learning: 2014-2015
  14. Computer Science 294: Practical Machine Learning [Fall 2009]
  1. Statistics, Probability and Machine Learning Short Course
  2. Statistical Learning
  3. Machine learning courses online
  4. Build Intelligent Applications: Master machine learning fundamentals in five hands-on courses
  5. Machine Learning
  6. Princeton Computer Science 598D: Overcoming Intractability in Machine Learning
  7. Princeton Computer Science 511: Theoretical Machine Learning
  8. MACHINE LEARNING FOR MUSICIANS AND ARTISTS
  9. CMSC 726: Machine Learning
  10. MIT: 9.520: Statistical Learning Theory and Applications, Fall 2015
  11. CMU: Machine Learning: 10-701/15-781, Spring 2011
  12. NLA 2015 course material
  13. CS 189/289A: Introduction to Machine Learning[with videos]
  14. An Introduction to Statistical Machine Learning Spring 2014 [for ACM Class]
  15. CS 159: Advanced Topics in Machine Learning [Spring 2016]
  16. Advanced Statistical Computing [Vanderbilt University]
  17. Stanford CS229: Machine Learning Spring 2016
  18. Machine Learning: 2015-2016
  19. CS273a: Introduction to Machine Learning
  20. Machine Learning CS-433
  21. Machine Learning Introduction: A machine learning course using Python, Jupyter Notebooks, and OpenML
  22. Advanced Introduction to Machine Learning
  23. STA 4273H [Winter 2015]: Large Scale Machine Learning
  24. Statistical Learning Theory and Applications [MIT]
  25. Regularization Methods for Machine Learning
  1. Convex Optimization: Spring 2015
  2. CMU: Probabilistic Graphical Models [10-708, Spring 2014]
  3. Advanced Optimization and Randomized Methods
  4. Machine Learning for Robotics and Computer Vision
  5. Statistical Machine Learning
  6. Probabilistic Graphical Models [10-708, Spring 2016]

数学基础

Calculus

  1. Khan Academy Calculus [https://www.khanacademy.org/math/calculus-home]

Linear Algebra

  1. Khan Academy Linear Algebra
  2. Linear Algebra MIT 目前最好的线性代数课程

Statistics and probability

  1. edx Introduction to Statistics [https://www.edx.org/course/introduction-statistics-descriptive-uc-berkeleyx-stat2-1x]
  2. edx Probability [https://www.edx.org/course/introduction-statistics-probability-uc-berkeleyx-stat2-2x]
  3. An exploration of Random Processes for Engineers [http://www.ifp.illinois.edu/~hajek/Papers/randomprocDec11.pdf]
  4. Information Theory [http://colah.github.io/posts/2015-09-Visual-Information/]

VIP内容

机器学习系统通常被认为是不透明的、不可预测的,和人类所接受的训练几乎没有任何共通之处。

难道,黑盒模型和可解释性的学习注定是两条路?

但最近有研究表明,至少在某些情况下,神经网络能够学习到一些人类可理解的表征!

例如分类器中的单个神经元可以表示一些语义信息,语言模型中也包含语法信息,在视觉和文本数据的对齐数据中也能发现一些复杂的概念表示,这些神经网络学到的概念都和人类接受的概念训练相关。

但还有一个问题,这些学习到的概念是通用的吗?我们是否也希望其他深度学习的系统具有类似的有意义的表示?

如果这些问题的答案都是没有的话,那么一些关于反映模型计算过程可解释性的研究将受到种种限制,并且很难找到其他合理的方法来解释。

虽然上面提到的几个例子能一定程度上能展现机器学习模型能够理解人类的语义,但本质上是因为它们只能接触到人类生成的数据,并且在分类任务中是将人类的类别概念强加给模型才导致它们能捕捉到类别语义。

或者说,这些任务也相对简单,解释起来也更加容易。

为了进一步测试机器学习模型是否真正获取到了人类可理解概念(human-understandable concepts),需要找到一个在没有使用人类标签数据的情况下,表现出超越人类表现的模型。

这不巧了吗?AlphaZero就同时满足这两个要求。

首先,AlphaZero是通过self-play的方式训练的,所以从未接触过人类数据,并且它在国际象棋,围棋和将棋(Shogi)这三项竞技游戏上借助蒙特卡洛树搜索成功战胜人类。

所以AlphaZero就成了研究机器学习模型和人类理解之间关系的一座重要桥梁,如果AlphaZero中能找到人类可理解的概念,那其他模型应该也会有!

说干就干!

DeepMind、Google Brain的研究人员携手国际象棋世界冠军共同打造了一篇长达69页的论文,主要研究了像AlphaZero这样的超越人类的神经网络模型正在学习什么,这是一个既科学又实用的问题。

在论文中研究人员证明了人类获取知识和AlphaZero在国际象棋中获得的知识都是相似的。并通过对大量人类关于国际象棋的概念的探索,还可以观察到其中一些概念在AlphaZero网络是如何表示的。

成为VIP会员查看完整内容
0
7

最新论文

Deploying Machine Learning (ML) algorithms within databases is a challenge due to the varied computational footprints of modern ML algorithms and the myriad of database technologies each with its own restrictive syntax. We introduce an Apache Spark-based micro-service orchestration framework that extends database operations to include web service primitives. Our system can orchestrate web services across hundreds of machines and takes full advantage of cluster, thread, and asynchronous parallelism. Using this framework, we provide large scale clients for intelligent services such as speech, vision, search, anomaly detection, and text analysis. This allows users to integrate ready-to-use intelligence into any datastore with an Apache Spark connector. To eliminate the majority of overhead from network communication, we also introduce a low-latency containerized version of our architecture. Finally, we demonstrate that the services we investigate are competitive on a variety of benchmarks, and present two applications of this framework to create intelligent search engines, and real-time auto race analytics systems.

0
0
下载
预览
Top