Supplementary Materials A contrastive rule for meta-learning

Application to the top-down modulation model . . . . . . . . . . . . . . . . . . 3 S2 Review of implicit gradient methods for meta-learning 5 S3 Theoretical results 7 S3.1 Meta-gradient estimation error bound . . . . . . . . . . . . . . . . . . . . . . . We then simplify the two sides of the equation. Note that we use to denote partial derivatives and d to denote total derivatives. We can now derive the meta-learning rules for the complex synapse model of Section 4.1. Let us now decompose what this update means.