|
|
|
|
Safety_Critical_Machine_Learning |
|
|
|
L_Anti-Spoofing Norm Engineered Recognizer (LASER™) is the result of a research started to give answers to some issues raised in the DARPA GARD (Guaranteeing AI Robustness against Deception) program. The LASER™ classifier can be used as a plug-in for a CNN or in a context where features extraction is executed in a different way. |
|
|
|
|
|
|
|
Recognition with a state of the
art CNN |
Recognition with the same CNN after a targeted attack with our GAN for generation of Physically Feasible Deception Patterns (PFDP) |
|
Recognition with a state of the art CNN and a final LASER™ classification layer |
With
LASER™ technology the GAN failed to converge in the generation of PFDP and in
the generation of any spoofing pixels pattern in the full image |
|
|
|
|
|
FREE DECEPTION PATTERNS (easy task for GAN) |
PHYSICALLY FEASIBLE DECEPTION PATTERNS (linear increase in complexity) |
PHYSICALLY CONSTRAINED DECEPTION PATTERNS (exponential increase in complexity) |
|
|
|
General Synaptics Aerospace_and_Defence_Machine_Learning_Company VAT NUMBER:_IT02670700992 REA NUMBER: GE-503104 Email:_luca.marchese@synaptics.org |
|
|
|
Copyright© 2026 General Synaptics |
