Algorithms for structural and dynamical polychronous groups detection

Regis Martinez (1) , Hélène Paugam-Moisy (2)

1 DM2L - Data Mining and Machine Learning
2 TAO - Machine Learning and Optimisation

Polychronization has been proposed as a possible way to investigate the notion of cell assemblies and to understand their role as memory supports for information coding. In a spiking neuron network, polychronous groups (PGs) are small subsets of neurons that can be activated in a chain reaction according to a specific time-locked pattern. PGs can be detected in a neural network with known connection delays and visualized on a spike raster plot. In this paper, we specify the definition of PGs, making a distinction between structural and dynamical polychronous groups. We propose two algortihms to scan for structural PGs supported by a given network topology, one based on the distribution of connection delays and the other taking into account the synaptic weight values. At last, we propose a third algorithm to scan for the PGs that are actually activated in the network dynamics during a given time window.

Mots clés: Spiking Neuron Networks; Polychronization; Algorithms

Domaines: Réseau de neurones [cs.NE]

Regis Martinez, Hélène Paugam-Moisy. Algorithms for structural and dynamical polychronous groups detection. ICANN'2009, International Conference on Artificial Neural Networks, IEEE - INNS, Sep 2009, Limassol, Cyprus. pp.75-84. ⟨inria-00425514⟩