Sander Bohte Staff Researcher NeuroInformatics Lab Life Sciences Group CWI Amsterdam
Email: S.M.Bohte [at] cwi [dot] nl Phone: tel. (+31) (0)20 592 4074 Fax: tel (+31) (0)20 592 4199 Mailing Address: Dr Sander M. Bohte CWI, MAC4 Science Park 123 NL-1098XG Amsterdam The Netherlands
Research overview
I develop computational models to help understand the mechanisms that underly information processing in networks of - mainly - spiking neurons. I strongly believe that neural networks and computational neuroscience models should "compute"; the challenge is to develop insights from neuroscience into usefully computing neural networks, and to bring machine learning insights into models of how neurons in the brain compute. I have particularly focused on encoding information with timed spikes, supervised neural learning, and general reinforcement learning methods.
- Recently, we suggested a straightforward neural spike coding and decoding framework, where we observe that a spiketrain can be the fractional derivative of a signal. We believe this is both a simple and elegant neural coding paradigm, that can account for a number of experimental observations. Current work focuses on applications of this paradigm to standard learning theory and the link with biology.
- With Pieter Roelfsema and Arjen van Ooyen, we are working on biologically plausible policy gradient reinforcement learning in spiking neurons. Elaborating on the AGREL idea, the research is yielding some surprising results on the (in)sufficiency of standard policy-gradient reinforcement learning.
- Other machine learning efforts have focused on distributed learning paradigms, mainly within the Multi-Agent Learning paradigm, with such applications as hospital patient scheduling and energy distribution in smart grids.
Talented students are always welcome to come and do their MSc-thesis work at CWI on NeuroInformatics, either based on their own proposed ideas, or based some ready-made potential MSc thesis projects. I especially encourage ideas that link (spiking) neural networks to biology, and also on work that explores (probabilistic) learning in spiking neural networks (SNNs), for example in connection with reservoir computing. I would also like to further explore ways of efficiently simulating large-scale SNNs, with for example GPUs. Feel free to contact me for more information.