Learning across space and time in spiking neural networks

Institutskolloquium

  • Datum: 16.11.2018
  • Uhrzeit: 10:30
  • Vortragende(r): Prof. Robert Gütig
  • Charité Berlin and the Berlin Institute of Health
  • Ort: Garching und Greifswald
  • Raum: HGW S1 (Übertragung Hörsaal D2)
  • Gastgeber: IPP
The brain routinely discovers sensory clues that predict opportunities or dangers. However, it is unclear how neural learning processes can bridge the typically long delays between sensory clues and behavioral outcomes. Here, I introduce a learning concept, aggregate-label learning, that enables biologically plausible model neurons to solve this temporal credit assignment problem. Aggregate-label learning matches a neuron’s number of output spikes to a feedback signal that is proportional to the number of clues but carries no information about their timing. Aggregate-label learning outperforms stochastic reinforcement learning at identifying predictive clues and is able to solve unsegmented speech-recognition tasks. Furthermore, it allows unsupervised neural networks to discover reoccurring constellations of sensory features even when they are widely dispersed across space and time.

Biography:

Robert Gütig is a professor for "Mathematical modeling of Neural Learning" at the Charité Berlin and the Berlin Institute of Health. The goal of his research is to identify the algorithms and computational principles that underlie information processing and learning in the brain. Robert Gütig studied physics at the Free University of Berlin (Germany) and received his master's degree from the University of Cambridge (United Kingdom) in 1997 for work on quantum vacuum fluctuations. He pursued his Ph.D. in computational neuroscience at the University of Freiburg (Germany) in the laboratory of Prof. Aertsen, where he received his Ph.D. in 2002 for a unified description of competitive and cooperative learning dynamics in neural networks with spike-timing dependent synaptic plasticity. Robert Gütig received his postdoctoral training in the laboratory of Prof. Sompolinsky at the Hebrew University of Jerusalem (Israel) between 2005 and 2011. Together with Prof. Sompolinsky, Robert Gütig pioneered the tempotron model of supervised learning in spiking neural networks.

Zur Redakteursansicht