Talk at Bern University
Invited by W. Senn, Raoul gives a talk on “Learning versatile computations with recurrent spiking neural networks” at Bern University.
Yaroslav Felipe Kalle Kossio joins the group
Yaroslav Felipe Kalle Kossio, physics major from Hongkong University of Science and Technology, joins the group as a PhD student. Welcome!
Oihane Horno joins the group
Oihane Horno, physics major from Dartmouth College, Hannover, NH, joins the group for a four months internship to model network changes during epileptogenesis in a collaborative project with experimentally working scientists at Bonn University (H. Beck). Welcome!
Anna Hellfritzsch joins the group
Anna Hellfritzsch, physics major from Frankfurt University, joins the group as a Master’s student. Welcome!
Group moves to Frankfurt
As of September 2016, the group is located at the FIAS, Frankfurt Institute of Advanced Studies. A great place for Theoretical Neuroscience with an excellent environment composed of institutions such as the MPI Brain and the Ernst Strüngmann Institute.
Cooperation and competition of gamma oscillation mechanisms
Oscillations of neuronal activity in different frequency ranges are thought to reflect important aspects of cortical network dynamics. In our article “Cooperation and competition of gamma oscillation mechanisms” we investigate how various mechanisms that contribute to oscillations in neuronal networks may interact. This article appeared in its final version today in Journal of Neurophysiology. We focus on two prominent gamma oscillation generating mechanisms: interneuron gamma (ING) and pyramidal-interneuron gamma (PING). In ING, an externally driven subnetwork of inhibitory interneurons alone generates the oscillations, while PING requires interactions between local interneurons and pyramidal cells. What type of oscillation will a network generate that could in principle generate oscillations by both the ING and PING mechanism? We find that ING and PING oscillations compete: The mechanism generating the higher oscillation frequency “wins”; it determines the frequency of the network oscillation and suppresses the other mechanism. In our study, we further work out specific differences between networks where the interneurons belong to the class of type I or of type II neurons and we suggest experimental approaches to decide to what extent oscillatory activity in networks of interacting excitatory and inhibitory neurons is dominated by ING or PING oscillations and of which class the participating interneurons are.
Talk at Stanford University
Upon invitation of K. Boahen, Raoul gives a talk on “Learning versatile computations with recurrent spiking neural networks” at Stanford University.
Versatile computations with spikes, mental exploration
Animals and humans can learn versatile computations such as the generation of complicated activity patterns to steer movements or the generation of appropriate outputs in response to inputs. Such learning must be accomplished by networks of nerve cells in the brain, which communicate with short electrical impulses, so-called spikes. In our article “Learning Universal computations with Spikes”, which has appeared today in PLoS Computational Biology, we show how such spiking neural networks may perform the learning. We track their ability back to experimentally found nonlinearities in the couplings between nerve cells and to a network connectivity that complies with constraints. We show that the spiking networks are able to learn difficult tasks such as the generation of desired chaotic activity and the prediction of the impact of actions on the environment. The latter allows to compute optimal actions by mental exploration.
A unifying view of functional networks of spiking model neurons
Most of the networks used by computer scientists and many of those studied by modelers in neuroscience represent unit activities as continuous variables. Neurons, however, communicate primarily through discontinuous spiking. In our review “Building functional networks of spiking model neurons”, which has appeared today in Nature Neuroscience, we derive a unifying view of the current methods for transferring our ability to construct functional networks, from continuous to more realistic spiking network models.
Talk and two posters at CoSyne 2016
Raoul will talk about “Learning versatile computations with recurrent spiking neural networks” in the workshop on recurrent spiking neural networks. In the main meeting, he will present the hippocampal dynamics poster 447, “A unified dynamic model for learning, replay and sharp-wave/ripples” (together with S. Jahnke and M. Timme), and he contributes to poster 595 “Full-rank regularized learning in recurrently connected firing rate networks” (together with B. DePasquale, C. Cueva, L.F. Abbott, and S. Escola) on improving reservoir computing by the introduction of a driven teacher network.
Workshop “Recurrent spiking neural networks – dynamics, learning, computation” at CoSyne 2016
The CoSyne 2016 workshop “Recurrent spiking neural networks – dynamics, learning, computation” proposed and organized by Brian DePasquale, Ben Dongsung Huh, Omri Barak and Raoul has been accepted. The workshop seeks to integrate recent advances in understanding dynamics in both continuous variable and spiking recurrent neural networks and their relationship to learning and computation in these systems. The focus of the workshop will be on spiking networks, but relevant results from rate networks shall also be discussed. There will be eleven talks with subsequent discussions, all speakers have already confirmed their participation.
A unified dynamic model for learning, replay and sharp-wave/ripples
Our manuscript “A unifying model for learning, replay and sharp-wave/ripples” has been accepted for publication in The Journal of Neuroscience. It summarizes our recent results on the role of hippocampal sharp-wave/ripples in learning and memory processing. Sharp-wave/ripples occur together with replay of activity sequences reflecting previous behavior. Developing a unifying computational model, we propose that both phenomena are tightly linked, by mutually generating and supporting each other. The underlying mechanism depends on nonlinear amplification of synchronous inputs that has been prominently found in the hippocampus. We also find that – perhaps contrary to intuition – for a commonly assumed, standard STDP window the events tend to rather erase than enhance learned hippocampal network structures.
Talk and poster at the Bernstein conference Heidelberg
Raoul talks about “Learning computations with temporally precisely spiking neural networks” in M. Diamond and A. Fassihi’s workshop “How do time and sensory information interact in perceptual decision making?”. A poster in the main meeting summarizes the latest results of our group on “Learning universal computations with spikes”.
Vladimir returns to Krasnoyarsk
After six months of fruitful research in our group funded by the Eranet Mundus grant, Vladimir Zakhvataev now returns to Krasnoyarsk. We will keep in touch and continue our collaborative work on the impact of pathological changes in synapses on the dynamics of neural networks.
Raoul starts at Columbia
Raoul starts his time as visiting faculty at the Center for Theoretical Neuroscience of Columbia University funded by the grant of the Max Kade Society New York.
Talk at Bonn University
In the framework of an invited talk, Raoul gives an overview of our results on artificial and biological computation with spiking neural networks.
Talk at SISSA Trieste
Invited by M. Diamond to speak at the workshop “Working memory data and models — bridging the gap”, Raoul presents our recent findings on learning of working memory dependent computations.
Talk at Frankfurt University
Raoul is the speaker of today’s physics colloquium at the Goethe University Frankfurt. He gives a talk on “Learning precisely timed spikes”.
Han receives his Bachelor’s degree
Han Nauta receives his Bachelor’s degree, with a very good grade for his thesis on the influence of synaptic changes on neural network dynamics and for the exams (second reader/examinator: B. Kappen). Congratulations!
Marvin receives his Master’s degree
Marvin Uhlmann receives his Master’s degree, with a very good grade for his thesis on “Computation with spiking neural networks” and the exams (second reader/examinator: J. van Opstal). Congratulations!
Poster at the SNN Adaptive Intelligence Symposium Nijmegen
Dominik Thalmeier and Raoul present a poster on “Learning universal computations with spikes” at the SNN Adaptive Intelligence Symposium Nijmegen.
Poster at CoSyne 2015 Salt Lake City
Raoul presents a poster on our recent results on learning of computations with spiking neural networks in the main meeting of the Computational and Systems Neuroscience 2015 conference.
Raoul is awarded a Max Kade Fellowship
Raoul is awarded a Fellowship of the Max Kade Society New York for research at the Center for Theoretical Neuroscience at Columbia University. He will begin his time as visiting faculty as soon as possible and has requested a reduction of the stipend to ten month, to be able to start his Bernstein group on time.
Talk at the MPI for Dynamics of Complex Systems Dresden
Raoul gives an invited talk at the MPIPKS. He will present learning rules for spiking neurons and recurrent networks, highlight analytically and numerically derived expressions for the neurons’ memory capacity and its scaling, and present applications to data analysis problems.