1a_Advances in Financial Machine Learning Lecture-

文件名称: 1a_Advances in Financial Machine Learning Lecture-1-9.pdf

所属分类: 机器学习

开发工具:

文件大小: 12mb

下载次数: 0

上传时间: 2019-08-31

提供者: geg****

下载 (12mb)

不能下载？报告错误

详细说明：notes for Advances_in_Financial_Machine_LearningSECTION The history of Machine Learning What Is Machine Learning? An ML algorithm learns complex patterns in a high-dimensional space without being specifically directed Marcos Lopez de prado, Advances in Financial Machine Learning(2018, p. 15) 1.0 Let's break this statement into its components 0.8 learns. without being specifically directed": Unlike with other 0.6 empirical tools, researchers do not impose a particular structure on the 驅噩里 0.4 data Instead, researchers let the data speak 0.2 learns complex patterns": The ML algorithm may find a pattern that cannot be easily represented with a finite set of equations 0.0 黑围 "learns.. in a high-dimensional space Solutions often involve a large 0.2 number of variables and the interactions between them 0.4 HH任 0.6 Suppose that you have a 1000x 1000 correlation matrix 0.8 1.0 What Is Machine Learning? An ML algorithm learns complex patterns in a high-dimensional space without being specifically directed Marcos Lopez de prado, Advances in Financial Machine Learning(2018, p. 15) 1.0 Lets break this statement into its components 0.8 “ earns∴ without being specifically directed": Unlike with other 0.6 empirical tools, researchers do not impose a particular structure on the 0.4 data. Instead researchers let the data speak 0.2 learns complex patterns": The ML algorithm may find a pattern that cannot be easily represented with a finite set of equations 0.0 learns .. in a high-dimensional space". Solutions often involve a large 0.2 number of variables and the interactions between them 0.4 0.6 Suppose that you have a 1000x1000 correlation matrix. A clustering 0.8 algorithm finds that there are 3 blocks: Highly correlated, low correlated 1.0 uncorrelated Timeline(1794-1950) Approx. 1794 The least squares method is invented independently by adrien Marie Legendre and by Carl Friedrich Gauss. Besides OLS, another critical contribution by gauss is setting the foundations for the Gauss-Markov theorem BlUe 1812 Pierre simon de laplace defines bayes theorem in its current form this theorem is the basic learning"method by which prior information(beliefs)can be updated with new information(observations) 1950s Alan turing proposes the idea of machines that can learn autonomously. While at Cornell, Frank rosenblatt invents the" perceptron", a type of linear binary classifier Evelyn Fix and J.L. Hodges introduce the k-Nearest Neighbor algorithm Portraits from Wikipedia commons Timeline(1960s-1990s 1960s-1970s DARPA The First Al Winter: Lack of progress in machine translation and limitations in neural networks and the perceptron approaches lead to cutbacks in public research funding. As a result research progress slows down 1980s David Rumelhart geoff Hinton and ronald j Williams introduce backpropagation as a method for training a neural network Terry sejnowski develops nettalk, a program that learns to pronounce words the same way a baby does. In 1987, dARPA cuts funding again, triggering a Second Al Winter 1990s Tin Kam ho, leo breiman and adele cutler introduce random forests Corinna cortes and Vladimir Vapnik introduce Support Vector Machines. Sepp hochreiter and Jurgen Schmidhuber invent Long Short-Term Memory ( lSTm). In 1997, IBM's Deep blue defeats garry Kasparov Portraits from Wikipedia commons Timeline(2000s- 2017 gaggle 20005 Private corporations offset DARPA's cuts, ending the second al Winter. a team of ML researchers wins the Netflix Prize. The website Kaggle is launched to host mL competitions 2010-2015 I BM's Watson defeats two human champions at the jeopardy competition Google brain s200 develops a neural network that learns to recognize cats in videos Facebooks deep Face NATSON Maxwells silver hamner recognizes faces with 97. 35% accuracy O AlphaS .KBA 2015-2017 In 2015, an algorithm developed by deep mind learns to play breakout, and within 4 hours finds a winning strategy unknown to humans. In 2016, Google's alpha go defeats Lee Sedol. Move 78 in the 4th game is considered superhuman In 2017, alphaGo Zero becomes a master within 21 days by playing against itself, without any training from human games Portraits from Wikipedia commons Machine Learning Today ML is all around us: Consumer products: Google, Amazon, Facebook, Netflix, Apple, Microsoft, Uber, etc Industrial services: Supply-chain, flight systems, agriculture, quality control, financial ratings, credit scores, etc Research: Drug development, genome research, new materials, physics research, etc The greatest mathematical breakthroughs occurred in the 1990s.. So why now and not earlier? Computing power has finally caught up with the mathematical needs of the 1990s Immense amounts of data The future is bright ML is no longer in incubation It is a source of revenue(no need for public funding 90% of all data has been created over the past 2 years, and 80% of all available data is unstructured Quantum computing will take ml to an entirely new level SECTION Machine Learning vs Econometrics

(系统自动生成,下载前可以参看下载内容)