Design of Brain Computer Interface

1
Design of Brain Computer Interface
CSE 19991253 Seung hyun, Yang
2
Psychokinesis?

• Feature Extraction
• Classification

Raw data
ICA AR
HMM
Recognition
Feature Extraction
Pattern Classification
3
Feature Extraction

• AR Model (AutoRegressive Model)
• Zt random process,
• AR coefficient
• Xt is regressed on the past values of itself
• Independent Component Analysis
• Power Spectrum etc.
• Will be treated next semester

4
Class1 - Estimating Parameters

• The HMM needs to define the probability density
  function p(Z1, Z2, ... ,ZT )
• Baum-Welch algorithm
• (Forward-Backward algorithm)

5
Class1-Estimating Parameters (Cont.)
6
Class2 Training and Recognize

• Linear HMM

MAXIMIZE
7
Class2-Training and Recognize (Cont.)

• Switching State space model
• Factorial HMM, Tree-structured HMM

X(1)1
X(1)2
X(1)3
X(1)T
X(m)1
X(m)2
X(m)3
X(m)T
8
Implementation Tool

• MATLAB
• Best tool for handling matrices
• Probability between states can be easily
  expressed with matrices
• There are many papers and bare source at web
  sites studying HMM.

9
Reference

• An Introduction to HMM and BN
• Zoubin Ghaharamani
• Getting Started with MATLAB
• The MATHWORKS Inc.
• Statistical Modeling Using Gaussian Mixtures and
  HMMs with MATLAB
• Paul M. Baggenstoss