Deep learning’s explanatory power beyond vision

Determine whether deep learning can lead to improved understanding of cognitive processes beyond visual object recognition, by evaluating if deep learning models can serve as explanatory algorithmic accounts for non-visual cognitive functions in biological brains.

Background

The paper highlights that deep learning has successfully modeled aspects of visual object recognition by aligning hierarchical representations in artificial and biological systems. However, the authors note that such alignment relies on suitable training data, clear task objectives, and rich neuroscientific comparison data—conditions that are less established for non-visual domains.

They explicitly flag uncertainty about extending this success to other cognitive processes, motivating a need to test whether deep learning can provide mechanistic insights into cognitive functions outside vision, where objectives, datasets, and neuroscientific benchmarks are less clear.