Efficient Representation, Learning, and Planning Through Abstraction - Clustering Cognitive Spaces Into Submaps

Explore the concept of efficient representation, learning, and planning through abstraction in this 58-minute lecture by Ila Fiete from MIT. Delve into the theory of online and real-time generation of fragmented representations and contextual re-anchoring from continuous experience. Examine how these fragmentations match neural recordings in animals navigating complex environments. Discover the benefits of map fragmentation as a form of state abstraction, including improved representation fidelity, flexible learning through reuse of existing fragments, and enhanced planning abilities in complex environments. Learn about memory architecture, neural mechanisms, submaps, and experimental data supporting this theory. Gain insights into the partitioning hypothesis, long-term memory, real-time clustering, and the reuse of submaps in various contexts.

Intro
Partitioning
Hypothesis
Memory architecture
Memory structure
Memory networks
Fragmented maps
Online fragmentation
Longterm memory
Test drive
Realtime clustering
Experimental data
Neural mechanisms
Submaps
Planning
Neural circuits
Reusing submaps
Motivation
Experiments
Environments
Environment trajectories
Animation
Preliminary results
Conclusion