Temporal contingency and informativeness in Laplace domain

Associative learning in conditioning protocols is traditionally thought to depend on temporal contiguity between the stimuli. However, recent proposals suggest that temporal contingency rather than temporal contiguity is responsible for associative learning. That is, associations to the unconditional stimulus develop whenever it provides meaningful information about the expected time of reinforcement. Such information-theoretic framework has recently been considered in the context of Pavlovian conditioning, operant conditioning, and behavioral neuroscience. Here we develop information-theoretic approach to the long-range temporal credit assignment subject to several cognitively-inspired constraints. First, we assume that the system cannot possibly measure the joint statistics of the past combinations of stimuli. Second, our method assumes that the approximate stimuli history is available in the form of the inverse Laplace transform, which makes the method time-local and scale-invariant and helps to overcome major drawbacks of traditional approaches to the long-range credit assignment problem. The result is a biologically-plausible, computationally efficient model of associative learning, one of the fundamental processes in neural function.