Predictive plasticity in dendrites: from a computational principle to experimental data

Spicher, Dominik; Clopath, Claudia; Senn, Walter (25 February 2017). Predictive plasticity in dendrites: from a computational principle to experimental data (Unpublished). In: COSYNE. Salt Lake City. 23-26.02.2017.

Preview

Text Spicher_Clopath_Senn_COSYNE_2017.pdf - Other Available under License Publisher holds Copyright. Download (607kB) | Preview

Plasticity of excitatory cortical synapses is thought to be the main mediator of mammalian learning. Due to the selective advantage that the ability to adapt to a changing environment grants, it is rea- sonable to assume that the processes governing these changes in connection strength are in some sense optimal. Yet, this has been difficult to reconcile with an “embarrassment of riches” of the LTP and LTD phenomenology. Here, we present an attempt at bridging this gap by showing that a mathematically derived model can exhibit some of these experimentally observed e↵ects, while still retaining functional capabilities under diverse learning paradigms.
Our work is based on a published plasticity model [3] that postulates that learning is driven by an intraneuronal prediction error where the weights of “student inputs” onto a dendritic compartment change in order to reproduce voltage changes imposed on a somatic compartment by “teacher inputs.” We show here that this two-compartment model of a pyramidal neuron can be extended in some simple ways, such as using conductance-based instead of current-based inputs, bringing it closer to the biophysics of pyramidal neurons. This allows us to reproduce a diverse set of experimental observations on cortical plasticity, such as di↵erent characteristics of the spike-timing dependence of plasticity.
Additionally, we show within a simple setup of a pattern recognition task that the extended model, while being less analytically tractable, can still perform well under unsupervised and reinforcement learning paradigms. Therefore, a single learning rule derived from the optimization of a well-defined cost function can be brought into correspondence with a large body of experimental evidence on synaptic plasticity, while still providing a diverse set of relevant functionality.

Interest & Impact

Downloads

0 since deposited on 17 Jan 2024
0 in the past 12 months

Citations

5 Citations in Web of Science ®
6 Citations in Scopus

Search

in Google Scholar™

Services

Actions (login required)

Edit item

Actions (login required)

Edit item