PDF 物理页索引 430
文件中的第 431 页 · 印刷页码:未识别
PDF 物理页索引 431
文件中的第 432 页 · 印刷页码:未识别
30 同上。
31 同上。
32 同上。
33 M. Dowd,“Elon Musk's Billion-Dollar Crusade to Stop the A. I. Apocalypse”, Vanity Fair , March 26, 2017.
34 L. Grossman,“2045: The Year Man Becomes Immortal”, Time , Feb. 10, 2011.
35 Singularity University website, accessed Dec. 4, 2018.
36 Kurzweil, Singularity Is Near , 316.
37 R. Kurzweil, The Age of Spiritual Machines: When Computers Exceed Human Intelligence (New York: Viking Press, 1999), 170.
38 D. R. Hofstadter,“Moore's Law, Artificial Evolution, and the Fate of Humanity”, in Perspectives on Adaptation in Natural and Artificial Systems , ed. L. Booker et al. (New York: Oxford University Press, 2005), 181.
PDF 物理页索引 432
文件中的第 433 页 · 印刷页码:未识别
39 Kurzweil, Age of Spiritual Machines , 169- 170.
40 Hofstadter,“Moore's Law, Artificial Evolution, and the Fate of Humanity”, 182.
41 Kurzweil,“Wager on the Turing Test”.
42 M. Kapor,“Why I Think I Will Win”, Kurzweil AI , April 9, 2002.
43 同上。
44 参见R. Kurzweil, Virtual Humans , by P. M.Plantec (New York: AMACOM, 2004)一书的前言。
45 Grossman,“2045”.
04 何人<sub>,</sub>何物<sub>,</sub>何时<sub>,</sub>何地<sub>,</sub>为何
1 S. A. Papert,“The Summer Vision Project”, MIT Artificial Intelligence Group Vision Memo 100 (July 7, 1966).
2 D. Crevier, AI: The Tumultuous History of the Search for Artificial Intelligence (New York: Basic Books, 1993), 88.
PDF 物理页索引 433
文件中的第 434 页 · 印刷页码:未识别
3 K. Fukushima,“Cognitron: A Self-Organizing Multilayered Neural Network Model”, Biological Cybernetics 20, no. 3-4 (1975): 121-136; K. Fukushima,“Neocognitron: A Hierarchical Neural Network Capable of Visual Pattern Recognition”, Neural Networks 1, no. 2 (1988): 119-130.
4 此处的分类模块指的是ConvNets的全连接层(fully connected layer)。
5 在输入网络之前,需要将图像缩放至固定大小——与网络的第1层大小相同。
关于大脑如何执行某项任务的大多数论述必须附带多条说明,我刚刚概述的这个故事也不例外。虽然我所描述的大致准确,但大脑却极度复杂,我所概述的这些发现只是早期研究的一小部分,科学家仍未完全理解大脑执行任务的内在机理。
7 与每个激活特征图关联的权重数组称为卷积过滤器或卷积核。
我对ConvNets的描述省去了许多细节。例如,要计算其特征激活值,卷积层中的一个单元会先进行卷积,然后对卷积结果使用非线性激活函数运算。ConvNets通常还具有其他类型的层,例如“池化层”。有关详细信息,请 参 见 I. Goodfellow, Y. Bengio, and A. Courville,Deep Learning (Cambridge, Mass.: MIT Press, 2016)。
PDF 物理页索引 434
文件中的第 435 页 · 印刷页码:未识别
9 在撰写本书时,谷歌的图片搜索引擎可通过单击搜索框中的小相机图标进行访问。
05 ConvNets和ImageNet<sub>,</sub>现代AI系统的基石
1 事实上,反向传播算法是由几个不同的研究群体独立发现的。并且,非常具有讽刺意味的是,反向传播函数是一种“置信度分配”(credit-assignment)算法,而发现该算法的功绩的分配则一直是神经网络研究者长期的“战争”。
2 D. Hernandez,“Facebook's Quest to Build an Artificial Brain Depends on This Guy”, Wired , Aug. 14, 2014.
3 还有另外的“检测”竞赛,其中参赛模型还必须对图像中多种类别的物体进行定位,并接受其他特定任务的挑战。此处我着重讨论的是分类任务。
4 D. Gershgorn,“The Data that Transformed AI Research—and Possibly the World”, Quartz , July 26, 2017.
PDF 物理页索引 435
文件中的第 436 页 · 印刷页码:未识别
5 “About Amazon Mechanical Turk”, www.mturk.com/help.
6 L. Fei-Fei and J. Deng,“ImageNet: Where Have We Been? Where Are We Going?”.
7 A. Krizhevsky, I. Sutskever, and G. E. Hinton,“ImageNet Classification with Deep Convolutional Neural Networks”, Advances in Neural Information Processing Systems 25 (2012): 1097- 1105.
8 T. Simonite,“Teaching Machines to Understand Us”, MIT Technology Review , Aug. 5, 2015.
9 Gershgorn,“The Data That Transformed AI Research”.
10 Hernandez,“Facebook's Quest to Build an Artificial Brain Depends on This Guy”.
11 B. Agüeray Arcas,“Inside the Machine Mind: Latest Insights on Neuroscience and Computer Science from Google”(lecture video), Oxford Martin School, May 10, 2016.
PDF 物理页索引 436
文件中的第 437 页 · 印刷页码:未识别
12 A. Linn,“Microsoft Researchers Win ImageNet Computer Vision Challenge”, AI Blog , Microsoft, Dec.10, 2015.
13 A. Hern,“Computers Now Better than Humans at Recognising and Sorting Images”, Guardian , May 13, 2015; T. Benson,“Microsoft Has Developed a Computer System That Can Identify Objects Better than Humans”, UPI, Feb. 14, 2015.
14 A. Karpathy,“What I Learned from Competing Against a ConvNets on ImageNet”, Sept. 2, 2014.
15 S. Lohr,“A Lesson of Tesla Crashes? Computer Vision Can't Do It All Yet”, New York Times , Sept. 19, 2016.
06 人类与机器学习的关键差距
1 关注2016年美国总统大选的读者,能够看出这是伯尼·桑德斯(Bernie Sanders)的支持者们的口号“感受伯尼”(Feel the Bern)的双关语。
2 E. Brynjolfsson and A. McAfee,“The Business of Artificial Intelligence”, Harvard Business Review , July 2017.
PDF 物理页索引 437
文件中的第 438 页 · 印刷页码:未识别
PDF 物理页索引 438
文件中的第 439 页 · 印刷页码:未识别
PDF 物理页索引 439
文件中的第 440 页 · 印刷页码:未识别
Applications (2017), 352-358; R. Geirhos et al., “Generalisation in Humans and Deep Neural Networks”, Advances in Neural Information Processing Systems 31 (2018): 7549-7561; M. Alcorn et al.,“Strike (with) a Pose: Neural Networks Are Easily Fooled by Strange Poses of Familiar Objects”, arXiv: 1811.11553(2018).
17 M. Orcutt,“Are Face Recognition Systems Accurate? Depends on Your Race”, MIT Technology Review , July 6, 2016.
18 J. Zhao et al.,“Men Also Like Shopping: Reducing Gender Bias Amplification Using Corpus-Level Constraints”, in Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (2017).
20 C. Szegedy et al.,“Intriguing Properties of Neural Networks”, in Proceedings of the International Conference on Learning Representations (2014).