Overview of Machine Learning RPI Robotics Lab Spring 2011 Kane Hadley
Agenda What is Machine Learning? Some techniques Simple Implementations Implementations for complex problems
A computer program learns from an experience E with respect to task T and some performance measure P if its performance on T as measured on P improves with experience E. ~Tom Mitchell
Supervised Learning Aims to find a function f(x) -> y Learns by correcting itself to match that function Examples – Support Vector Machines – Artificial Neural Networks
Support Vector Machine
Artificial Neural Network
Unsupervised Learning Attempts to find a good representation for a given data set Examples – K-Means Clustering – Self Organizing Maps
K-Means Clustering Tries to find K clusters for a data set. Clusters are found by approximating centroids for each cluster.
Self Organizing Maps Attempts to fix the space of the map to a given data set.
Reinforcement Learning Goal is to maximize a given reward function. Reward is calculated using utilities given to each state in the world.
Genetic Algorithms Form of optimization. Starts with a population and fitness function At each time step evaluate the fitness of each member, remove the lowest fitness member, breed the two members with the highest fitness and mutate.
Videos Stanford Copter Little Dog
Criticisms Slow Requires lots of data Not necessarily optimal
References creatures.html creatures.html ndex.html ndex.html