CS480/680 Machine Learning Lecture 1: January 7 th , 2020 Course - PowerPoint PPT Presentation

CS480/680 Machine Learning Lecture 1: January 7 th , 2020 Course Introduction Zahra Sheikhbahaee CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 1

Outline • Introduction to Machine Learning • Course website and details: https://cs.uwaterloo.ca/~zsheikhb/CS480-winter2020.html# • Learn (Assignment, grades) https://learn.uwaterloo.ca/ CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 2

Instructor Who am I? Dr. Zahra Sheikhbahaee Postdoctoral Researcher PhD in Astrophysics CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 3

The Team for CS480/680 • TA’s § Gaurav Gupta g27gupta@uwaterloo.ca § Zeou Hu z97hu@uwaterloo.ca § Arash Mollajafari Sohi amollaja@uwaterloo.ca § Zahra Rezapour Siahgourabi zrezapou@uwaterloo.ca § Colin Vandenhof cm5vande@uwaterloo.ca CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 4

Prerequisites of this Course • Programming :python • Probability: distributions • Calculus: partial derivatives • Linear algebra: vector/matrix manipulations, properties • Statistics: mean, median, mode, standard deviation CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 5

Exam & Evaluation • Midterm 25% Ø Feb 28 Ø Start/end time: 8:30-10:00pm • Assignment 35% • Final 40% Ø Grad students : Project with a submitted proposal by 10 th of February (6 pages and written in the format of a paper, Novel and innovative method ) Ø Under grad student: Optional either exam or a project CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 6

Machine Learning • Traditional computer science – Program computer for every task • New paradigm – Provide examples to machine – Machine learns to accomplish a task based on the examples CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 7

Definitions • Arthur Samuel (1959): Machine learning is the field of study that gives computers the ability to learn without being explicitly programmed. • Tom Mitchell (1998): A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E. • Ethem Alpaydın: Machine learning is programming computers to optimize a performance criterion using example data or past experience. We need learning in cases where we cannot directly write a computer program to solve a given problem, but need example data or experience. In statistics, going from particular observations to general descriptions is called inference and learning is called estimation and classification is called discriminant analysis . CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 8

Three Categories Supervised learning Ø Classification Ø Regression Reinforcement learning Unsupervised learning Ø Clustering Ø reducing dimensionality CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 9

Supervised Learning • Classification: e.g. digit recognition (postal code) 𝑔: ℝ $ ⟶ {1, … , 𝑙} • Simplest approach: memorization CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 10

Supervised Learning • Nearest neighbour: It can be used to solve both classification and regression problems. CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 11

Definition of Supervised Learning • Inductive learning or inferring general rules for a limited set of examples: – Given a training set of examples of the form (𝑦, 𝑔(𝑦)) • 𝑦 is the input, 𝑔(𝑦) is the output – Return a function ℎ that approximates 𝑔 • ℎ is called the hypothesis CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 12

Prediction • Find function ℎ that fits 𝑔 at instances 𝑦 CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 13

Prediction • Find function ℎ that fits 𝑔 at instances 𝑦 CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 14

Prediction • Find function ℎ that fits 𝑔 at instances 𝑦 CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 15

Prediction • Find function ℎ that fits 𝑔 at instances 𝑦 CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 16

Prediction • Find function ℎ that fits 𝑔 at instances 𝑦 CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 17

Generalization • Key: a good hypothesis will generalize well (i.e. predict unseen examples correctly) • The Occam’s razor: prefer the simplest hypothesis consistent with data • Capacity is a measure of complexity and measures the expressive power, richness or flexibility of a set of functions (low capacity: struggle to fit the training set, high capacity: overfit by memorizing properties of the training set). • The Vapnik-Chervonenkis dimension: A dataset containing N points can be labeled in 2 N ways as positive and negative and 2 N different learning problems can be defined by N data points. If for any of these problems, we can find a hypothesis h ∈ H that separates the positive examples from the negative, then we say H shatters N points. The maximum number of points that can be arranged so that classifier H can shatter them and it is called the Vapnik- Chervonekis (VC) dimension of H , is denoted as VC ( H ), and measures the capacity of the hypothesis class H . CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 18

ImageNet Classification • 1000 classes • 1 million images • Deep neural networks (supervised learning) CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 19

Unsupervised Learning • An output is not given as part of training set • Find model that explains the data – Clustering: e.g. K-mean clustering – Compressed representation, features, generative model: CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 20

Unsupervised Feature Generation • Encoder trained on large number of images to build a face detector from only unlabeled images https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/38115.pdf CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 21

Reinforcement Learning Agent State Action Reward Environment When the output of the system is a sequence of actions . In such a case, a single action is not important; what is important is the policy that is the sequence of correct actions to reach the goal . The reward is a numerical signal which indicates how good actions are. Goal: Learn to choose actions that maximize rewards CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 22

Game Playing • Example: Go (one of the oldest and hardest board games) • Agent: player • Environment: opponent • State: board configuration • Action: next stone location • Reward: +1 win / -1 loose • 2016: AlphaGo defeats top player Lee Sedol (4-1) – Game 2 move 37: AlphaGo plays unexpected move (odds 1/10,000) CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 23

Reinforcement Learning The theories that incorporate constraints on the information processing capacities of an agent are called theories of bounded rationality (Herbert Simon). • Perfect rationality: the agent can determine the best course of action, without taking into account its limited computational resources. • Bounded rationality : the rationality of a realistic agent is limited by resources such as time, access to information, capacity for information, and processing power and can only be rational to a certain extent. Agents modeled with unbounded rationality act to maximize utility, while agents modeled with bounded rationality can only aim for some satisfactory amount of utility (a regularized expected utility known as the free energy, where the regularizer is given by the information divergence from a prior to a posterior policy). CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 24

Applications of Machine Learning • Speech recognition – Siri, Cortana • Natural Language Processing – Machine translation, question answering, dialog systems • Computer vision – Image and video analysis • Robotic Control – Autonomous vehicles • Intelligent assistants – Activity recognition, recommender systems • Computational finance – Stock trading, credit scoring, fraud detection CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 25

This course • Supervised and unsupervised machine learning • But not reinforcement learning CS480/680 Winter 2020 Zahra Sheikhbahaee University of Waterloo 26