Output details
11 - Computer Science and Informatics
University of Surrey
Learning Temporally Precise Spiking Patterns through Reward Modulated Spike-Timing-Dependent Plasticity
<24>Learning to produce precisely timed spikes is an important aspect of cognitive and neural processing. With a novel combination of reinforcement learning and spike-timing dependent plasticity, in this work we train networks of two different types of escape-noise neurons: the commonly used exponential model and the biological more plausible but mathematically intricate Arrhenius & Current model. Surprisingly the paper demonstrates that the Arrhenius & Current model exhibits better performance and also better resilience to noise. The work explores how neural learning in the brain forms cognitive processes and is a substantial building block to (re)construct biologically plausible learning algorithms.