55 min
Seminar, Hosted by Simons Institute for the Theory of Computing
Networks of Spiking Neurons Learn to Learn and Remember
Graz University of Technology
Recording
Abstract
Wolfgang Maass considers how recurrent spiking networks with prior learning can adapt to new tasks. He shows how biologically motivated neuron models and network architectures can support working-memory modules analogous to LSTMs. Training such networks to learn enables complex new input-output behaviors through local learning rules and offers a model of working memory in spiking circuits.
Topics
- working memory
- recurrent spiking networks
- prior learning, broader topic: learning
- LSTMs
- biologically motivated neuron models, broader topic: neuron models
- network architectures
- local learning rules
- input-output behaviors, search World Wide
- spiking circuits, broader topic: circuits
- Learn to Learn, search World Wide