Flexible compression of sensory evidence makes human decisions more resilient to noisy neural computations

Humans performing a visual categorization task were more accurate and made less noisy inferences when they accumulated sensory evidence in compressed category space, as shown by behavioral results.

• Why it matters: The work addresses whether ambiguous evidence is accumulated in native sensory space or in an abstract compressed category space, and how that choice affects robustness to noisy neural computations.
• What they did: Participants completed a task where revealing decision categories before or after a stimulus sequence determined the evidence space used, while concurrent MEG measured accuracy-predictive compressed representations and RNNs were trained on the same setup.
• The result: MEG showed flexible, accuracy-predictive evidence representations compressed onto decision categories, and RNNs using the same strategy produced decisions more resilient to noisy computations.