Spiking allows neurons to estimate their causal effect

Abstract

Neural learning aims at the maximization of reward and the typical gradient descent learning is an approximate causal estimator. However real neurons spike based on presynaptic drive, creating discontinuities. The regression discontinuity method, popularized by economics, uses such discontinuities to estimate causal effects. Here we show how the spiking discontinuity can thus reveal the influence of a neuron's activity on reward, producing a deep link between simple learning rules and quasi-experimental causal inference.