Machine learning lecture 1 Tommi S Jaakkola MIT CSAIL tommi csail mit edu 6 867 Machine learning administrivia Instructor Prof Tommi Jaakkola tommi csail mit edu TA Jason Johnson jasonj mit edu General info lectures MW 2 30 4pm in 35 225 tutorials recitations time location tba website http www ai mit edu courses 6 867 Grading midterm 15 final 25 6 bi weekly problem sets 30 final project 30 Tommi Jaakkola MIT AI Lab 2 Machine learning Statistical machine learning principles methods and algorithms for learning and prediction on the basis of past experience already everywhere speech recognition hand written character recognition information retrieval operating systems compilers fraud detection security defense applications Tommi Jaakkola MIT AI Lab 3 Learning yes Tommi Jaakkola MIT AI Lab 4 Learning yes yes Tommi Jaakkola MIT AI Lab 5 Learning yes yes no Tommi Jaakkola MIT AI Lab oops 6 Learning Steps entertain a biased set of possibilities adjust predictions based on feedback rethink the set of possibilities Tommi Jaakkola MIT AI Lab 7 Learning Steps entertain a biased set of possibilities adjust predictions based on feedback rethink the set of possibilities Principles of learning are universal society e g scientific community animal e g human machine Tommi Jaakkola MIT AI Lab 8 Learning and prediction We make predictions all the time but rarely investigate the processes underlying our predictions In carrying out scientific research we are also governed by how theories are evaluated To automate the process of making predictions we need to understand in addition how we search and refine theories Tommi Jaakkola MIT AI Lab 9 Learning key steps data and assumptions what data is available for the learning task what can we assume about the problem Tommi Jaakkola MIT AI Lab 10 Learning key steps data and assumptions what data is available for the learning task what can we assume about the problem representation how should we represent the examples to be classified Tommi Jaakkola MIT AI Lab 11 Learning key steps data and assumptions what data is available for the learning task what can we assume about the problem representation how should we represent the examples to be classified method and estimation what are the possible hypotheses how do we adjust our predictions based on the feedback Tommi Jaakkola MIT AI Lab 12 Learning key steps data and assumptions what data is available for the learning task what can we assume about the problem representation how should we represent the examples to be classified method and estimation what are the possible hypotheses how do we adjust our predictions based on the feedback evaluation how well are we doing Tommi Jaakkola MIT AI Lab 13 Learning key steps data and assumptions what data is available for the learning task what can we assume about the problem representation how should we represent the examples to be classified method and estimation what are the possible hypotheses how do we adjust our predictions based on the feedback evaluation how well are we doing model selection can we rethink the approach to do even better Tommi Jaakkola MIT AI Lab 14 Data and assumptions yes yes is this a digit character image classification problem how are the examples generated labeled Tommi Jaakkola MIT AI Lab 15 Representation There are many ways of presenting the same information 0111111001110010000000100000001001111110111011111001110111110001 The choice of representation may determine whether the learning task is very easy or very difficult Tommi Jaakkola MIT AI Lab 16 Representation 0111111001110010000000100000001001111110111011111001110111110001 0001111100000011000001110000011001111110111111001111111100000011 1111111000000110000011000111111000000111100000111110001101111111 Tommi Jaakkola MIT AI Lab yes yes no 17 Representation 0111111001110010000000100000001001111110111011111001110111110001 0001111100000011000001110000011001111110111111001111111100000011 1111111000000110000011000111111000000111100000111110001101111111 yes yes no yes yes no Tommi Jaakkola MIT AI Lab 18 Method and estimation Examples binary vectors of length d 64 x 111111100 0000110001101111111 T Labels y 1 1 no yes A linear classifier hypotheses y sign x sign d X ixi i 1 where is a vector of parameters we have to learn Tommi Jaakkola MIT AI Lab 19 Method and estimation cont d x 0111111001110010000000100000001001111110111011111001110111110001 0001111100000011000001110000011001111110111111001111111100000011 1111111000000110000011000111111000000111100000111110001101111111 y 1 1 1 How do we adjust the parameters based on the labels y sign x Tommi Jaakkola MIT AI Lab 20 Method and estimation cont d x 0111111001110010000000100000001001111110111011111001110111110001 0001111100000011000001110000011001111110111111001111111100000011 1111111000000110000011000111111000000111100000111110001101111111 y 1 1 1 How do we adjust the parameters based on the labels y sign x For example we can simply refine update the parameters whenever we make a mistake yx when prediction was wrong Tommi Jaakkola MIT AI Lab 21 Evaluation How do we measure how well the method is working Tommi Jaakkola MIT AI Lab 22 Evaluation How do we measure how well the method is working 1 0 9 0 8 average error 0 7 0 6 0 5 0 4 0 3 0 2 0 1 0 0 200 400 600 800 number of examples 1000 1200 1400 For example average classification prediction error as a function of the number of examples seen so far Tommi Jaakkola MIT AI Lab 23 Model selection Can we rethink the approach to do even better our classifier is limited can we make it more flexible is there an entirely different type of classifier that would be more suitable Tommi Jaakkola MIT AI Lab 24 Types of learning problems A rough classification of learning problems Supervised learning explicit feedback e g labels Unsupervised learning no feedback emphasis on structure and organization Semi supervised learning partial feedback e g a few labels mostly unlabeled Reinforcement learning delayed feedback Tommi Jaakkola MIT AI Lab 25