Can probabilistic population codes support probabilistic task transfer?

Determine whether Probabilistic Population Codes (PPCs) can implement the types of generalization required for probabilistic task transfer, specifically whether PPCs can support transferable marginalization and change-of-variables computations across tasks.

Background

PPCs posit that population activity can linearly encode posterior parameters, offering a neurally plausible scheme for uncertainty representation. While decodability of posterior parameters from PPCs has been shown, the paper emphasizes that this alone cannot establish probabilistic use without demonstrating transfer of distinctly probabilistic computations across tasks.

The authors highlight that it remains untested whether PPCs enable such transfer—i.e., generalizing learned probabilistic computations like marginalization and change of variables to new task settings without retraining.