Cognitive transforms in perception and memory

Abstract

Visual perception seems to provide a direct and immediate view onto the outside world. In reality, it is an active and adaptive process. Cognitive factors, such as our prior knowledge and goals, transform the information streaming in from the retina to create a reconstruction of our environment tailored to our needs. The study of these cognitive influences on perception and their underlying mechanisms falls within the purview of visual cognition. If the principles gleaned from these studies can inform our understanding of cognition more generally, however, it is necessary to test if these principles generalize to domains beyond visual perception (and, if not, to understand if these principles can at least provide a useful basis for comparison and understanding). Towards that end, the work in this thesis examines how expectation and attention influence visual perception and additionally interrogates these same processes in the context of working memory. In the realm of expectation, we show that percepts reflect a weighted average of sensory information and prior knowledge, biasing percepts towards expected values. We find that these biases persist in working memory, accumulating over time to counteract memory noise. In the realm of attention, we find that, once attended, both percepts and memories are represented using radically different (i.e. orthogonal) patterns of neural activity relative to their unattended state. Furthermore, in this new post-attentional subspace, perceptual and mnemonic codes are reorganized in a way that allows task-relevant features to be decoded and task-irrelevant features to be abstracted away. This transform may selectively gate the influence of perceptual and mnemonic representations on other cognitive processes. In both the case of expectation and attention, these common principles uniting perception and memory coexist with key differences. For instance, learning modifies the influence of expectations on memory faster than on perception, and attention biases competition between perceptual but not mnemonic representations. Together, these results suggest that while the cognitive transforms observed in perception do generalize to other domains, they may be actualized by distinct mechanisms.