Flexible Learning Models Utilizing Different Neural Plasticities

The existing models for vestibulo-ocular reflex (VOR) and optokinetic response (OKR) learning utilize neural circuit structure and capture a few characteristics of these two learning systems. However, it remains unclear how the error signals guide these learning processes. Here, we propose novel dynamic learning models using error feedback in a flexible fashion to account for both VOR and OKR learning. We first used a feedback modulation model (FMM) and found the error signals play an essential guiding role in the gain compensation of wild-type mice. However, this FMM cannot accurately reproduce gain changes during the recovery period. Therefore, we propose a nonuniform FMM using flexible plasticity learning rules of different memory sites to take into account the effect of classical linearity models in both training and recovery periods. To further study learning characteristics of gain reduction, we introduce a reversal-phase FMM and explore the contribution of synaptic plasticity to adaptive learning, in which characteristics and bidirectional synaptic plasticity in the VOR-decrease learning mode can be fully recovered. Taken together, our results suggest that, to explain VOR and OKR learning systems, one needs dynamical models with flexible and multiple components at different or same sites of neuronal circuits.