The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Sensory Prediction Error Signals in the Neocortex with Blake Richards - #331
Dec 24, 2019
Blake Richards, Assistant Professor at McGill University and Core Faculty Member at Mila, dives into the brain's learning abilities with a focus on sensory prediction error signals. He elaborates on two-photon calcium imaging studies revealing how the neocortex processes unexpected stimuli. Discussing predictive coding, he highlights its implications for both neuroscience and machine learning. The conversation also touches on integrating memory systems in reinforcement learning, showcasing how insights from biology can lead to more adaptive AI.
AI Snips
Chapters
Transcript
Episode notes
Hardwired Surprise
- Mice brains show strong responses to unexpected visual stimuli, like changes in flow.
- This suggests hardwired prediction mechanisms, potentially for predator avoidance.
Learned Surprise
- Mice learn to respond to other unexpected stimuli, like edge orientations, over time.
- This suggests the neocortex learns and updates its predictive model of the world.
Hierarchical Surprise
- Learned surprise signals appear in the apical dendrites, which receive top-down input.
- This suggests a hierarchical predictive model, where higher brain areas detect and communicate surprise.