Effectiveness of Neural Network Learning Rules Generated by a Biophysical Model of Synaptic Plasticity

We describe our initial attempts to reconcile powerful neural network learning rules derived from computational principles with learning rules derived “bottom-up” from biophysical mechanisms. Using a biophysical model of synaptic plasticity (Shouval, Bear, and Cooper, 2002), we generated numerical synaptic learning rules and compared them to the performance of a Hebbian learning rule in a previously studied neural network model of self-organized learning. In general, the biophysically derived learning rules did not perform as well as the analytic rule, but their performance could be improved by adjusting various aspects of the biophysical model. These results show that some progress has been made in integrating our understanding of biological and artificial neural networks. Most artificial neural networks rely on analytic learning rules that are selected for their ability to successfully solve problems (e.g., Hopfield 1982, Oja 1982, Kohonen 1984), or that represent biologically plausible but highly idealized functions of the activity of the input and output neurons over a set of stimuli (e.g., Bienenstock et al. 1982). Conversely, many biophysical models of synaptic plasticity are developed primarily for the purpose of comparison with known experimental biological results (e.g., Shouval, Bear, and Cooper 2002, Froemke and Dan 2002, Sjostrom et al. 2001). In this research we attempt to bridge that gap by generating numerical “lookup table” learning rules for artificial neural networks using simulations of a biophysical model of synaptic plasticity. We applied those learning rules to a previously studied neural network model to ascertain their learning effectiveness as compared with idealized functions.