Summer research reports: Brain image analysis; Computer Go
Date: November 14 2012 3:30pm - 4:30pm Location: Howard 255
Talk 1: Brain Image Analysis: Mathematics as the Great Equalizer
We will have a conceptual discussion of the mathematics involved in group analysis of fMRI measures. Presentation will focus around the processes executed by a MATLAB toolbox called Data Processing Assistant for Resting-State fMRI (DPARSF). These processes are necessary for data comparison and much more extensive than you might expect.
Talk 2: Automatically Learning Heuristic Patterns for Computer Go
Go is an ancient two-player board game that originated in China
over three thousand years ago. It plays an important role in
artificial intelligence research, being one of the few remaining games
in which human players still dominate the top computer programs. Top
Go programs all use Monte-Carlo sampling. Many simulated games are
played, after which the move most often leading to a win is selected.
The moves within the simulated games are semi-random but biased by
both the results of previous simulations and heuristic knowledge.
This presentation focuses on a pattern-recognition heuristic.
Certain local configurations of stones are recognized as good (or bad)
by human Go players. We extracted patterns from a collection of games
played by human experts. Playing moves suggested by our
automatically-learned good patterns performed as well as expert hand
crafted patterns. Avoiding moves suggested by bad patterns does not