Cog Lunch: Alicia Chen

March 10, 2026
12pm
Location: 46-3189

Zoom: https://mit.zoom.us/j/91860197321​​​​​​​

Speaker: Ian Griffith

Affiliation: McDermott Lab

Title: A stimulus-computable framework for selective attention

Abstract: How does the brain select relevant information from a cluttered sensory environment and why does selection sometimes fail? We present a stimulus-computable model of selective attention, grounded in neurophysiologically motivated feature gains, and validate it using the cocktail party problem as a test domain. Though not trained to mimic humans, the model reproduces human-like patterns of attentional success and failure, predicts novel behavioral effects that we confirmed experimentally, and exhibits hallmarks of late selection seen in auditory cortex. The results suggest that the defining features of human attention naturally emerge from optimizing a system for selective listening, and that the classic neurophysiological motif of multiplicative gains is sufficient, when combined with appropriate features, to account for attention-driven behavior. They also show how stimulus-computable models can account for real-world attentional behavior. The approach is domain-general, offering a framework that could extend to other attentional paradigms and sensory modalities.