A Unified Framework for Dopamine Signals across Timescales

摘要: Rapid phasic activity of midbrain dopamine neurons is thought to signal reward prediction errors (RPEs), resembling temporal difference used in machine learning. However, recent studies describing slowly increasing signals have instead proposed that they represent state values and arise independent from somatic spiking activity. Here we developed experimental paradigms using virtual reality disambiguate RPEs values. We examined circuit at various stages, including spiking, calcium somata axons, striatal concentrations. Our results demonstrate ramping are consistent with rather than value, this observed all stages examined. Ramping can be driven by a dynamic stimulus indicates gradual approach reward. provide unified computational understanding rapid signals: perform derivative-like computation over on moment-by-moment basis.