Impression learning Online representation learning with synaptic plasticity Appendices

Our derivation of the update for IL (Eq. 3) is based on an expansion of log Note that this is not a truncated Taylor series approximation - we are instead using Taylor's theorem, and the second term provides an exact expression for the bias. Thus, for our particular choice of neural model, the bias is proportional to B, which vanishes as performance improves. Note that the update term in Eq. (S1) is O(| In this section, we explore the relationships between impression learning (IL) and other stochastic learning algorithms. B.1 Neural Variational Inference Neural variational inference is a learning algorithm for neural networks that optimizes the evidence lower bound (ELBO) objective function. Here, we modify the algorithm by incorporating our novel loss (Eq.