From molecular computation to adaptive behavior: Across level modeling of memory computation in the mushroom bodies

The insect mushroom body is an ideal model system to study fundamental processes of memory formation, -consolidation and -retrieval that underly adaptive behaviors. Existing computational models of olfactory learning in insects typically rely on simple abstract plasticity rules governed by a single adaptive time scale. However, the full complexity of memory computations in the mushroom bodies involves processing across the molecular, cellular and circuit level on distinct time scales that span several decades. We here propose a three-fold approach to a comprehensive model view of memory computation in the MB: (1) In close collaboration with the experimental partners in this consortium we aim at establishing vertical models of deep biological realism, that capture specific processes as experimentally addressed in individual projects. (2) Utilizing our model achievements during the first funding period, we will step-by-step integrate these vertical model components into the horizontal topology of our functional neural network architectures to simulate adaptive sensory-motor behavior. This will allow for model validation and experimentally testable predictions at all levels. (3) From a theoretical perspective it will be of high interest to infer mathematical and/or algorithmic descriptions of complex learning rules that can serve as building blocks in computational neuroscience and may drive brain-inspired advances in deep learning and artificial intelligence.