view MIT-media-projects.org @ 344:dd1d8697e54f

move file.
author Robert McIntyre <rlm@mit.edu>
date Sat, 21 Jul 2012 23:59:42 -0500
parents c264ebf683b4
children
line wrap: on
line source
1 *Machine Learning and Pattern Recognition with Multiple
2 Modalities Hyungil Ahn and Rosalind W. Picard
4 This project develops new theory and algorithms to enable
5 computers to make rapid and accurate inferences from
6 multiple modes of data, such as determining a person's
7 affective state from multiple sensors--video, mouse behavior,
8 chair pressure patterns, typed selections, or
9 physiology. Recent efforts focus on understanding the level
10 of a person's attention, useful for things such as
11 determining when to interrupt. Our approach is Bayesian:
12 formulating probabilistic models on the basis of domain
13 knowledge and training data, and then performing inference
14 according to the rules of probability theory. This type of
15 sensor fusion work is especially challenging due to problems
16 of sensor channel drop-out, different kinds of noise in
17 different channels, dependence between channels, scarce and
18 sometimes inaccurate labels, and patterns to detect that are
19 inherently time-varying. We have constructed a variety of
20 new algorithms for solving these problems and demonstrated
21 their performance gains over other state-of-the-art methods.
23 http://affect.media.mit.edu/projectpages/multimodal/