Lecture 1: Course outline and logistics What is Machine Learning - PowerPoint PPT Presentation

Why Study Machine Learning? 
 Engineering Better Computing Systems • Develop systems • too di ffi cult/expensive to construct manually • because they require specific detailed skills/knowledge • knowledge engineering bottleneck • Develop systems • that adapt and customize themselves to individual users. • Personalized news or mail filter • Personalized tutoring • Discover new knowledge from large databases • Medical text mining (e.g. migraines to calcium channel slide by Dhruv Batra blockers to magnesium) • data mining 63

Why Study Machine Learning? 
 Cognitive Science • Computational studies of learning may help us understand learning in humans • and other biological organisms. • Hebbian neural learning • “Neurons that fire together, wire together.” slide by Dhruv Batra 64

Why Study Machine Learning? 
 The Time is Ripe • Algorithms • Many basic e ff ective and e ffi cient algorithms available. • Data • Large amounts of on-line data available. • Computing • Large amounts of computational resources available. slide by Ray Mooney 65

Where does ML fit in? slide by Fei Sha 66

A Brief History of AI slide by Dhruv Batra 67

adopted from Dhruv Batra 68

AI Predictions: Experts slide by Dhruv Batra Image Credit: http://intelligence.org/files/PredictingAI.pdf 69

AI Predictions: Non-Experts slide by Dhruv Batra Image Credit: http://intelligence.org/files/PredictingAI.pdf 70

AI Predictions: Failed slide by Dhruv Batra Image Credit: http://intelligence.org/files/PredictingAI.pdf 71

Why is AI hard? slide by Dhruv Batra 72 Image Credit: http://karpathy.github.io/2012/10/22/state-of-computer-vision/

73 What humans see slide by Larry Zitnick

What computers see 243 239 240 225 206 185 188 218 211 206 216 225 242 239 218 110 67 31 34 152 213 206 208 221 243 242 123 58 94 82 132 77 108 208 208 215 235 217 115 212 243 236 247 139 91 209 208 211 233 208 131 222 219 226 196 114 74 208 213 214 232 217 131 116 77 150 69 56 52 201 228 223 232 232 182 186 184 179 159 123 93 232 235 235 232 236 201 154 216 133 129 81 175 252 241 240 235 238 230 128 172 138 65 63 234 249 241 245 237 236 247 143 59 78 10 94 255 248 247 251 234 237 245 193 55 33 115 144 213 255 253 251 248 245 161 128 149 109 138 65 47 156 239 255 190 107 39 102 94 73 114 58 17 7 51 137 23 32 33 148 168 203 179 43 27 17 12 8 17 26 12 160 255 255 109 22 26 19 35 24 slide by Larry Zitnick 74

“I saw her duck” slide by Liang Huang Image Credit: Liang Huang 75

“I saw her duck” slide by Liang Huang Image Credit: Liang Huang 76

“I saw her duck” slide by Liang Huang Image Credit: Liang Huang 77

We’ve come a long way… IBM Watson • What is Jeopardy? • http://youtu.be/Xqb66bdsQlw?t=53s • Challenge: • http://youtu.be/_429UIzN1JM • Watson Demo: • http://youtu.be/WFR3lOm_xhE?t=22s • IBM Watson wins on Jeopardy (February 2011) • Explanation • Watson provides cancer treatment options to doctors in seconds (February 2013) • http://youtu.be/d_yXV22O6n4?t=4s • Future: Automated operator, doctor assistant, slide by Liang Huang finance 78

Why are things working today? • More compute power Better • More data • Better algorithms/ Accuracy models slide by Dhruv Batra Amount of Training Data Figure Credit: Banko & Brill, 2011 79

Machine Learning 
 (by examples)

81 Pose Estimation slide by Alex Smola

Collaborative Filtering Don’t mix preferences on Netflix! Amazon books slide by Alex Smola 82

Imitation Learning in Games Avatar learns from your behavior Black & White slide by Alex Smola Lionsgate Studios 83

Reinforcement Learning slide by Alex Smola https://www.youtube.com/watch?v=lleRKHsJBJ0 84

Spam Filtering ham spam slide by Alex Smola 85

Cheque Reading segment image recognize handwriting slide by Alex Smola 86

Image Layout • Raw set of images from several cameras slide by Alex Smola • Joint layout based on image similarity 87

Search Ads why these ads? slide by Alex Smola 88

Google Self-Driving Cars slide by Alex Smola Google’s self-driving car passes 300,000 miles (Forbes, 8/15/2012) • 89

Speech Recognition Given an audio waveform, robustly extract & recognize any spoken words • Statistical models can be used to 
 - Provide greater robustness to noise - Adapt to accent of different speakers 
 - Learn from training 90

Natural Language Processing I need to hide a body noun, verb, preposition, … 91

Face Detection Sudhakar et al., Multi-view Face Detection Using Deep Convolutional Neural Networks, 2015 92

Face Detection Yang et al., From Facial Parts Responses to Face Detection: A Deep Learning Approach, ICCV 2015 93

Topic Models of Text Documents Topic Models of Text Documents slide by Eric Sudderth 94

Visual Scene Understanding dome sky skyscraper sky slide by Eric Sudderth buildings trees temple bell 95

Learning - revisited 



prior
 Learning
 knowledge
 knowledge
 data
 slide by Stuart Russell 96

Learning - revisited 



prior
 Learning
 knowledge
 knowledge
 data
 slide by Stuart Russell 97

Programming with Data • Want adaptive robust and fault tolerant systems • Rule-based implementation is (often) - di ffi cult (for the programmer) - brittle (can miss many edge-cases) - becomes a nightmare to maintain explicitly - often doesn’t work too well (e.g. OCR) 
 • Usually easy to obtain examples of what we want 
 IF x THEN DO y • Collect many pairs (x i , y i ) slide by Mehryar Mohri • Estimate function f such that f(x i ) = y i (supervised learning) • Detect patterns in data (unsupervised learning) 98

Objectives of Machine Learning • Algorithms: design of efficient, accurate, and general learning algorithms to – deal with large-scale problems. – make accurate predictions (unseen examples). – handle a variety of different learning problems. • Theoretical questions: – what can be learned? Under what conditions? – what learning guarantees can be given? – what is the algorithmic complexity? slide by Mehryar Mohri 99

Definitions and Terminology • Example: an object, instance of the data used. • Features: the set of attributes, often represented as a vector, associated to an example (e.g., height and weight for gender prediction). • Labels: in classification, category associated to an object (e.g., positive or negative in binary classification); in regression real value. slide by Mehryar Mohri • Training data: data used for training learning algorithm (often labeled data). 100