Practice Reshapes the Geometry and Dynamics of Task-tailored Representations DOI Creative Commons
Atsushi Kikumoto, Kazuhisa Shibata,

Takahiro Nishio

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Сен. 15, 2024

Extensive practice makes task performance more efficient and precise, leading to automaticity. However, theories of automaticity differ on which levels representations (e.g., low-level features, stimulus-response mappings, or high-level conjunctive memories individual events) change with practice, despite predicting the same pattern improvement power law practice). To resolve this controversy, we built recent theoretical advances in understanding computations through neural population dynamics. Specifically, hypothesized that optimizes representational geometry minimally separate highest-level contingencies needed for successful performance. This involves efficiently reaching states integrate task-critical features nonlinearly while abstracting over non-critical dimensions. test hypothesis, human participants (n = 40) engaged extensive a simple, context-dependent action selection 3 days recording EEG. During initial rapid performance, highest-level, context-specific conjunctions task-features were enhanced as function number episodes. Crucially, only enhancement these representations, not lower-order predicted power-law Simultaneously, sessions, became stable earlier time aligned, redundant correlated offline gain reducing switch costs. Thus, dynamic task-tailored tesselate space, taming their high-dimensionality.

Язык: Английский

Practice Reshapes the Geometry and Dynamics of Task-tailored Representations DOI Creative Commons
Atsushi Kikumoto, Kazuhisa Shibata,

Takahiro Nishio

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Сен. 15, 2024

Extensive practice makes task performance more efficient and precise, leading to automaticity. However, theories of automaticity differ on which levels representations (e.g., low-level features, stimulus-response mappings, or high-level conjunctive memories individual events) change with practice, despite predicting the same pattern improvement power law practice). To resolve this controversy, we built recent theoretical advances in understanding computations through neural population dynamics. Specifically, hypothesized that optimizes representational geometry minimally separate highest-level contingencies needed for successful performance. This involves efficiently reaching states integrate task-critical features nonlinearly while abstracting over non-critical dimensions. test hypothesis, human participants (n = 40) engaged extensive a simple, context-dependent action selection 3 days recording EEG. During initial rapid performance, highest-level, context-specific conjunctions task-features were enhanced as function number episodes. Crucially, only enhancement these representations, not lower-order predicted power-law Simultaneously, sessions, became stable earlier time aligned, redundant correlated offline gain reducing switch costs. Thus, dynamic task-tailored tesselate space, taming their high-dimensionality.

Язык: Английский

Процитировано

1