A neurally plausible model for online recognition and postdiction in a dynamical environment

Abstract

Humans and other animals are frequently near-optimal in their ability to integrate noisy and ambiguous sensory data to form robust percepts, which are informed both by sensory evidence and by prior experience about the causal structure of the environment. It is hypothesized that the brain establishes these structures using an internal model of how the observed patterns can be generated from relevant but unobserved causes. In dynamic environments, such integration often takes the form of postdiction, wherein later sensory evidence affects inferences about earlier percepts. As the brain must operate in current time, without the luxury of acausal propagation of information, how does such postdictive inference come about? Here, we propose a general framework for neural probabilistic inference in dynamic models based on the distributed distributional code (DDC) representation of uncertainty, naturally extending the underlying encoding to incorporate implicit probabilistic beliefs about both present and past. We show that, as in other uses of the DDC, an inferential model can be learned efficiently using samples from an internal model of the world. Applied to stimuli used in the context of psychophysics experiments, the framework provides an online and plausible mechanism for inference, including postdictive effects.Behavioral [3,5,23,30,50] and physiological [8,9,15] findings suggest that the brain acquires an internal model of how relevant states of the world evolve in time, and how they give rise to the stream of sensory evidence. Recognition is then formally a process of statistical inference to form perceptual beliefs about the trajectory of latent causes given observations in time. While this type of 33rd

Learning neural codes for perceptual uncertainty

“…If the set of encoding functions is rich enough, the DDC expectations provide sufficient information to carry out probabilistic computation. Previous studies have shown that DDC representations provide an effective substrate to learn and to make inferences in deep generative models [8], to build successor representations within partially observable Markov decision processes [9], and to carry and resolve uncertainty over time [10]. In these studies, the set of encoding functions was chosen arbitrarily, and remained fixed throughout.…”