Cognitive Load Traces as Symbolic and Visual Accounts of Deep Model Cognition

Dong Liu, Yanxuan Yu · 2025

Read Cognitive Load Traces as Symbolic and Visual Accounts of Deep Model Cognition by Dong Liu, Yanxuan Yu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-15980.txt.

Description

Dong Liu and Yanxuan Yu propose reading a deep model's effort the way psychologists read human cognitive load. Their Cognitive Load Traces are symbolic, time-varying functions that track how a model allocates internal resources while it works. The idea borrows the three-way split of Cognitive Load Theory into intrinsic, extraneous, and germane load.

Each load is measured through concrete proxies such as attention entropy, cache miss ratio, representation spread, and decoding stability. Symbolic formulas and pictures, including load curves and simplex diagrams, make the traces readable as accounts of reasoning dynamics. On reasoning and planning benchmarks, the traces predict where errors are about to start and reveal which strategy a model is following. Interventions guided by the loads improve reasoning efficiency by a solid margin while holding accuracy, the pair report.

Catalog record

CREATORDong Liu, Yanxuan Yu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE13 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-15980

Browse the full library · About the Tower of Babel