From the Statistical Physics of Disordered Systems to Neuroscience
This chapter studies the bridges and differences between the statistical physics of disordered systems, as developed notably in the context of spin glass theory, and problems in neuroscience. In a first contribution (Sec. 25.1), Nicolas Brunel, Rémi Monasson and Haim Sompolinsky first recall the main lines of the statistical physics approach to neural networks models as developed in the 1980s and 1990s. They then survey more recent developments at the interface between statistical physics and neuroscience, including the inference of synaptic plasticity rules and the statistics of synaptic connectivity. Finally they present the Tempotron model for learning temporal patterns. In a second contribution (Sec. 25.2), Leo van Hemmen discusses the difference between real spin glasses, neuronal networks (of real biological neurons) and neural networks (of artificial neurons); with illustrations ranging from site-disorder models of spin glasses to temporal coding in neuronal networks and unlearning.