Multi-dimensional synaptic plasticity: from local learning rules to network dynamics

• Applicant: Professor Dr. Robert Gütig
• Subject Area: Cognitive, Systems and Behavioural Neurobiology
• Term: from 2004 to 2006
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 5416068

Recent experimental findings challenge the validity of conventional models of synaptic plasticity. Establishing a novel framework, the proposed research project aims at advancing our understanding of the computational principles that underlie activity-driven learning in neural networks. In particular, we will study the state-dependence of temporally asymmetric spike-timing dependent synaptic plasticity and clarify the role of bi-stable synaptic dynamics for neuronal feature-map and memory formation. We will derive optimal state-dependent learning rules from information theoretic principles. Advancing beyond scalar synaptic eflicacies, we will take recent experimental evidence into account and develop multi-dimensional models of long-term plasticity. In addition to investigating interactions between learning dynamics of several synaptic variables, we will study metaplasticity by introducing "hidden" synaptic variables that modulate plasticity without directly changing synaptic transmission. We will reach beyond established learning objectives based on equilibrium states of synaptic distributions. Combining multi-dimensional, state-dependent plasticity on different time scales, we finally intend to address the learning of network dynamics and clarify how short- and long-term plasticity cooperate.

• DFG Programme: Research Grants