Machine learning and heuristic search

点击翻译按钮获取中文摘要

My research aims to develop scientific understanding and improved algorithms in the areas of machine learning, and heuristic search, and to explore practical applications of techniques in these areas.A machine learning algorithm uses training data, or past experience, to improve its performance at a specific task.  In my research the task is playing poker against a single opponent, and machine learning will be used to identify the opponent's strengths and weaknesses and adapt its play accordingly in order to maximize its winnings against this specific opponent. This is very challenging because learning must be very rapid (only 100-200 hands will be played against this opponent in total), and the information available to the learning algorithm is noisy and incomplete.A heuristic search algorithm is a fast method for finding optimal paths between two nodes in a graph.  For example, if the graph is a road map, a heuristic search algorithm could be used to quickly find the shortest route from your current location to a particular destination.  Heuristic search is faster than standard methods because it uses an estimate of the distance to the goal, called a heuristic function.  My research investigates various ways of speeding up heuristic search.  One of the methods I am investigating is to decompose the heuristic function into the sum of a set of smaller heuristic functions. My aim is to understand when this is possible, and when it is superior to other methods.  I am also investigating techniques for predicting how much speedup will be produced by a given heuristic function.