SLIDE 1
VERIFICATION OF DEEP NEURAL NETWORKS
SOURADEEP DUTTA*, SUSMIT JHA#, SRIRAM SANKARANARAYANAN*, ASHISH TIWARI # * UNIVERSITY OF COLORADO, BOULDER
# SRI INTERNATIONAL, MENLO PARK
PROBLEM STATEMENT
A much simpler problem….
VERIFICATION OF DEEP NEURAL NETWORKS SOURADEEP DUTTA*, SUSMIT JHA # , - - PowerPoint PPT Presentation
VERIFICATION OF DEEP NEURAL NETWORKS SOURADEEP DUTTA*, SUSMIT JHA # , - - PowerPoint PPT Presentation
VERIFICATION OF DEEP NEURAL NETWORKS SOURADEEP DUTTA*, SUSMIT JHA # , SRIRAM SANKARANARAYANAN*, ASHISH TIWARI # * UNIVERSITY OF COLORADO, BOULDER # SRI INTERNATIONAL, MENLO PARK PROBLEM STATEMENT A much simpler problem. ? PROBLEM STATEMENT
SLIDE 2
SLIDE 3
LET’S FIND THE MAXIMA ( MINIMA )
?
A much simpler problem….
PROBLEM STATEMENT
SLIDE 4
THE OVERALL IDEA
Random Sample
SLIDE 5
THE OVERALL IDEA
Random Sample
SLIDE 6
THE OVERALL IDEA
Random Sample
SLIDE 7
THE OVERALL IDEA
Random Sample
SLIDE 8
THE OVERALL IDEA
Random Sample
SLIDE 9
THE OVERALL IDEA
Random Sample
LOCAL MAXIMA
SLIDE 10
THE OVERALL IDEA
Random Sample
BETTER POINT
SLIDE 11
THE OVERALL IDEA
Random Sample
SLIDE 12
THE OVERALL IDEA
Random Sample
SLIDE 13
THE OVERALL IDEA
Random Sample
SLIDE 14
THE OVERALL IDEA
Random Sample
SLIDE 15
THE OVERALL IDEA
Random Sample
SLIDE 16
THE OVERALL IDEA
Random Sample
SLIDE 17
THE OVERALL IDEA
Random Sample
Cannot Improve further STOP !
SLIDE 18
THE OVERALL IDEA
KEY COMPONENTS :
- 1. Gradient search technique
- 2. Counter Example Generation
SLIDE 19