26 min
Seminar, Hosted by Simons Institute for the Theory of Computing
Oscillator associative memories for high-capacity, compositional inference
Bruno Olshausen and Chris Kymn
UC Berkeley
Recording
Abstract
A continuous-time oscillator network implements high-capacity associative memory for factorizing compositional representations. This talk studies applications to perceptual inference and combinatorial problems such as subset sum, compares factorization performance with existing approaches, and analyses convergence to explain the network's capacity advantages and efficient implementation.
Topics
- continuous-time oscillator network, search World Wide
- associative memory
- high-capacity associative memory, broader topic: associative memory
- factorizing compositional representations, broader topic: representations
- perceptual inference
- combinatorial problems, search World Wide
- subset sum, search World Wide
- factorization performance, broader topic: performance
- convergence