Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 46: Statistical Physics of Biological Systems II (joint session DY/BP)

DY 46.2: Talk

Friday, March 21, 2025, 12:00–12:15, H43

Mean transient drift of synaptic weights in feed-forward spiking neural networks with spike-timing-dependent plasticity — •Jakob Stubenrauch and Benjamin Lindner — BCCN Berlin and Physics Department HU Berlin, Germany

Spike-timing dependent plasticity (STDP) [1] is a phenomenological model for the dynamics of single synaptic weights. This concise microscopic (single-synapse) description allows for the derivation of macroscopic network theories, capturing for instance learning, forgetting, and representational drift.

For the development of such theories it is important to characterize the stochastic process of synaptic weights. Early attempts capture this process for Poissonian presynaptic spikes and conditionally Poissonian postsynaptic spikes [2]. However, since STDP depends on fine spike-timing differences below 20 ms [1], it is important to characterize the synaptic dynamics for neuron models that describe the fast response mechanistically.

Leveraging a recent theory [3] as well as established results for the leaky integrate-and-fire neuron [4,5], we analytically compute the drift and diffusion of feed-forward synapses in a setup where a layer of presynaptic Poisson processes feeds into a recurrent network of leaky integrate-and-fire neurons.

[1] Bi and Poo, J. Neurisci. (1998) [2] Kempter et al., Phys. Rev. E (1999) [3] Stubenrauch and Lindner, Phys. Rev. X (2024) [4] Brunel et al., Phys. Rev. Lett. (2001) [5] Lindner and Schimansky-Geier, Phys. Rev. Lett. (2001)

Keywords: Spike-timing dependent synaptic plasticity; Stochastic processes; Integrate-and-fire neurons; Neural networks