Efficient Data Compression in Perception, Memory, and Attention
Thursday 2 March 2023 – 13:00 GMT
Department of Brain and Cognitive Sciences, University of Rochester, Rochester, NY (USA)
Efficient data compression is essential for capacity-limited systems, such as biological perception, memory, and attention. We hypothesize that the need for efficient compression shapes biological systems in many of the same ways it shapes engineered systems. If true, then the tools engineers use to analyze and design systems, namely rate-distortion theory (RDT), can profitably be used to understand human behaviour. In this talk, we’ll start by discussing three general principles for efficient data compression, and provide experimental data evaluating the implications of these principles for human performance. Next, we’ll note that exact RDT methods are often computationally intractable, and explore a deep neural network architecture that implements RDT in an approximate manner. Lastly, we’ll hypothesize that perception, sensory memory, short-term memory, and long-term memory lie along a continuum, differing primarily in terms of their capacity limitations.