It’s All About Confidence: System on Chip Verification and Building Trust into Neural Networks

How do we know that what our neural networks are telling us should be trusted? Can we build confidence into our neural networks so they can answer that for us? According to a new study out of MIT and Harvard, we can and it won’t break the computational bank! In this week’s Fish Fry podcast, we first check out a new way for deep learning neural networks to quickly estimate confidence levels in their output. Keeping with our verification theme, Moshik Rubin (Cadence Design Systems) also joins us to discuss the challenges of SoC verification, the increasing need for system level solutions today, and where he thinks chip level verification is headed in the future.