This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision | ||
keynote:lesson01 [2010/04/26 13:17] 20921038 replace equation with jsmath |
keynote:lesson01 [2023/08/19 21:02] (current) |
||
---|---|---|---|
Line 47: | Line 47: | ||
===== 1.2 机器学习单元概况 ===== | ===== 1.2 机器学习单元概况 ===== | ||
- | <note important> Call for editing </note> | ||
Line 77: | Line 76: | ||
**但是,当前它通常被当做一个黑盒来使用** | **但是,当前它通常被当做一个黑盒来使用** | ||
- | **确定性 VS 几率性** | + | **确定性 VS 机率性** |
===数据驱动模型=== | ===数据驱动模型=== | ||
{{:keynote:81.jpg?500*320}} | {{:keynote:81.jpg?500*320}} | ||
Line 162: | Line 161: | ||
* Reinforcement learning: A survey | * Reinforcement learning: A survey | ||
<note important> Edit by Xinyuan Luo(骆歆远), <wisp@zju.edu.cn> </note> | <note important> Edit by Xinyuan Luo(骆歆远), <wisp@zju.edu.cn> </note> | ||
+ | |||
+ | ===未来体系结构对机器学习的支持=== | ||
+ | * 通用图灵机模型与冯诺依曼架构在并行处理上的局限[1] | ||
+ | * 冯诺依曼架构源自通用图灵机模型(Universal Turing Machine) | ||
+ | * 优点:便于建造、编程 | ||
+ | * 缺点:“解析指令->执行指令->处理数据”这种循环存在瓶颈,不利于并行处理 | ||
+ | * 设计与演化:生物神经元系统的达尔文主义解释[2] | ||
+ | * 生物的神经元网络是基于达尔文进化论的模式演化出来的 | ||
+ | * 知识和技能都是通过后天的不断学习和训练获得。在这种学习和训练中,大脑中神经元网络的结构不断完善 | ||
+ | * 不论是对一个物种的长期进化而言,还是单个个体的一生成长而言,其大脑的学习过程都是进化论(Darwinism)的 | ||
+ | * 区别“知识”与“技能”,即"Knowledge" vs "Skill" | ||
+ | * 机器学习形成的知识,其应用过程的算法效率未必最优 | ||
+ | * 等价的知识,存在多种表达和应用方式 | ||
+ | * 对某一知识,更高效的表达和应用算法,可以称之为技能(Skill),以区别于知识(Knowledge) | ||
+ | * 技能可以通过设计获得,也可以通过演化获得。 | ||
+ | * 演化思想在计算模式与体系结构设计中的萌芽 | ||
+ | * 自动并行化研究[3,4,5] | ||
+ | * 代码级别的知识学习与存储[6] | ||
+ | * 未来发展 | ||
+ | * 更高效的核间通讯:NoC、片上光通信 | ||
+ | * 新的计算模型:函数式编程语言、并行计算语言、GPGPU | ||
+ | * 自主进化(可重构)的众核处理器架构 | ||
+ | * 参考文献 | ||
+ | * [1] John Backus, Can programming be liberated from the von Neumann style?: a functional style and its algebra of programs, Commun. ACM, Vol. 21, No. 8. (August 1978), pp. 613-641. doi:10.1145/359576.359579 | ||
+ | * [2] Gerald Edelman, "Bright Air and Brilliant Fire: on the matter of mind", New York: Basic Books, 1992. | ||
+ | * [3] Gurindar S. Sohi, Scott E. Breach, T. N. Vijaykumar, Multiscalar Processors, SIGARCH Comput. Archit. News In ISCA '95: Proceedings of the 22nd annual international symposium on Computer architecture, Vol. 23 (May 1995), pp. 414-425. doi:10.1145/223982.224451 | ||
+ | * [4] L. Hammond, B. A. Hubbert, M. Siu, M. K. Prabhu, M. Chen, K. Olukolun, The Stanford Hydra CMP, IEEE Micro, Vol. 20, No. 2. (March 2000), pp. 71-84. doi:10.1109/40.848474 | ||
+ | * [5] Guilherme Ottoni, Ram Rangan, Adam Stoler, David I. August, Automatic Thread Extraction with Decoupled Software Pipelining,Microarchitecture, 2005. MICRO-38. Proceedings. 38th Annual IEEE/ACM International Symposium on In Proceedings of the 38th annual IEEE/ACM International Symposium on Microarchitecture (2005), pp. 105-118. doi:10.1109/MICRO.2005.13 | ||
+ | * [6] Yushi Kamiya, Tomoaki Tsumura, Hiroshi Matsuo, Yasuhiko Nakashima, A Speculative Technique for Auto-Memoization Processor with Multithreading, (December 2009), pp. 160-166. doi:10.1109/PDCAT.2009.67 | ||
+ | |||
+ | <note important> Edit by Ye Minjiao (叶敏娇, 11021036), <johnye@zju.edu.cn> </note> | ||
+ | |||
===== 1.4 点估计 ===== | ===== 1.4 点估计 ===== | ||
最大似然, 最大化后验估计, 贝叶斯估计, 回归方法与过拟合问题 | 最大似然, 最大化后验估计, 贝叶斯估计, 回归方法与过拟合问题 |