120 likes | 265 Views
A man-machine human interface for a special device of the pervasive computing world. B. Apolloni, S. Bassis, A. Brega, S. Gaito, D. Malchiodi, A.M. Zanaboni DSI - University of Milano (I). Outline. A procedure detecting attention states in car driving
E N D
A man-machine human interface for a special device of the pervasive computing world B. Apolloni, S. Bassis, A. Brega, S. Gaito, D. Malchiodi, A.M. Zanaboni DSI - University of Milano (I)
Outline • A procedure detecting attention states in car driving • Fed by biologic input supplied through non invasive sensors • Explains its output through a possibly interpretable rule
The data • One subject using a car driver simulator • Subjected to alternate attention demanding manoeuvres (fast lane exchange, pedestrian avoidance) and relaxed driving • 4 signals traced by a Biopac device (SKT, GSR, ECG, RSP) • Collected by the School of Psychology, Queen’s University Belfast.
Preprocessing • Extracted features • 8 conventional (from medical knowledge) • FFT processing for ECG • Drift of the ECG signal from a neural prediction • SKT not considered (constant)
Feature processing I t-1 t t+1 • 15 Boolean values are extracted from a time-window of width 3 • Result of a Boolean ICA through minimization of empirical entropy • 253 connections (after pruning)
Feature processing II • Boolean values interpreted as propositional variables • Minimal DNF and DNF on variables interpreted as symbolic wavelets Begin DNF=ø; for each positive exampleu; DNF = DNF{m}; return DNF; End
An obtained CNF (x1+x3+x6+x8+x10+x14)(x1+x2+x5+x6+x7+x9+x11+x12+x13)(x1+x2+x6+x7+x9+x10+x12+x14)(x1+x2+x5+x7+x8+x12+x13+x15)(x1+x3+x5+x11+x13+x15)(x3+x8+x10+x11+x13+x14+x15)(x1+x2+x4+x6+x7+x11+x12+x13+x14+x15)(x1+x2+x4+x8+x12+x13+x14+x15)(x1+x2+x3+x7+x8+x10+x13+x14+x15)(x3+x4+x8+x10+x13+x14+x15)(x1+x3+x6+x7+x8+x13+x14)(x3+x5+x7+x8+x13)(x2+x3+x6+x7+x10+x11+x12+x13+x14)(x1+x5+x6+x8+x9+x11+x12+x15)(x1+x5+x6+x8+x9+x11+x14+x15)(x3+x5+x7+x9+x10+x11)(x1+x3+x4+x7+x13+x14+x15)(x1+x2+x3+x8+x11+x13+x14+x15)(x2+x5+x6+x7+x9+x11+x12+x15)(x2+x3+x4+x6+x9+x10+x11+x12+x14)(x2+x4+x5+x7+x8+x13+x15)(x3+x4+x6+x7+x10+x13)(x3+x4+x6+x8+x10+x14)(x1+x4+x7+x8+x13+x15)(x1+x4+x5+x7+x8+x9+x15)(x4+x7+x8+x10+x15)(x4+x7+x8+x10+x14)(x2+x4+x5+x6+x7+x9+x10+x12+x13+x15)(x1+x3+x5+x8+x13+x15)(x1+x3+x6+x9+x10+x11+x14)(x1+x6+x9+x10+x11+x12+x14)(x1+x6+x8+x9+x11+x12+x14)(x1+x2+x4+x6+x7+x10+x11+x12+x13+x14)(x2+x4+x5+x8+x9+x10+x12+x13+x15)(x2+x4+x5+x7+x9+x10+x11+x12)(x1+x3+x6+x7+x8+x10+x13)(x3+x4+x5+x6+x8+x9+x10+x11)(x2+x5+x6+x8+x9+x10+x11+x12)(x2+x5+x6+x7+x8+x9+x11+x12)(x1+x2+x4+x5+x6+x7+x9+x12+x13)(x3+x4+x5+x7+x10+x11)(x1+x2+x5+x8+x10+x12+x13+x15)(x1+x2+x6+x7+x9+x10+x11)(x1+x2+x5+x7+x8+x9+x14)(x3+x4+x5+x8+x9+x10+x13+x15)(x3+x8+x9+x10+x11+x13)(x3+x9+x10+x11+x13+x14)(x1+x6+x7+x8+x9+x10+x12)(x1+x2+x3+x6+x11+x13+x14+x15)(x2+x4+x5+x6+x9+x11+x12+x14+x15)(x2+x6+x7+x10+x11+x12+x13+x15)(x2+x4+x6+x10+x11+x12+x14+x15)(x2+x5+x6+x9+x10+x11+x12+x14+x15)(x3+x4+x6+x8+x11+x14+x15)
Post processing • Simplification of the learnt rules through stochastic optimization of the cost • L: rule length, r:rule radius, n:disregarded points
A simplified CNF (x6+x11+x1+x13)(x10+x12+x9+x6)(x1+x13+x11+x5)(x3+x8+x6)(x4+x1+x13)(x12+x6+x7+x13)(x13+x8+x9+x4)(x1+x6+x8)(x12+x6+x7+x8)(x1+x8+x5+x7)(x4+x6+x7+x13)(x7+x9+x10+x11+x3)(x1+x8+x13+x15)(x3+x8+x13+x15)(x4+x7+x8)(x1+x6+x9+x10+x11)(x2+x6+x11+x12+x15)(x3+x5+x8+x13)(x4+x5+x7+x10+x11)(x3+x9+x10+x11+x13) • From 403 to 81 literals
Performance I • 50 cross-validation test • FP: false positives; FN: false negatives