A Compressive Sensing Approach to Inferring Cognitive Representations with Reverse Correlation

Uncovering high-level cognitive representations of categories such as faces is an elusive goal that has been frequently sought using reverse correlation, a technique employed in fields ranging from neurophysiology to cognitive psychology. In reverse correlation, subjects are asked to make perceptual judgments (e.g., "do you see a face?") about richly varying stimuli (e.g., white noise), and observed responses are then regressed against stimuli to yield reconstructions of the underlying cognitive representation. However, many thousands of stimulus-response pairs are frequently required, which severely limits the breadth of studies that are feasible using this powerful method. Techniques that are currently employed to improve efficiency, such as filtering the reconstruction, nevertheless bias the outcome. Here, we show that an advanced signal processing technique for improving sampling efficiency - compressive sensing - is directly compatible with reverse correlation. A trio of simulations are performed to demonstrate that compressive sensing can reduce the required stimulus-response pairs by up to 90% without biasing the reconstruction or can retrospectively improve the accuracy of the reconstructions on existing data. This work concludes by outlining the potential of compressive sensing to improve representation reconstruction throughout the field of neuroscience and beyond.