Non-target color terms reduced mean report log loss in 57 ds008037 v1.0.1 test adults with fixed circular mixtures
Non-target color terms improved mean prediction for 57 people excluded from model fitting in OpenNeuro ds008037 v1.0.1. Across three memory loads, the mean reduction in log loss was 0.005704 nats per report (planned 95% bootstrap interval 0.003522 to 0.007975), with fitted models treated as fixed. A model refit check gave an interval of 0.003217 to 0.008005. The gain was small and did not occur in every test person. The result applies to this cohort, task, split, and family of fixed circular mixtures; it does not estimate a swap frequency or identify a cognitive mechanism.
visual working memory · color report · non-target colors · circular density · log loss · held-out prediction · bootstrap · OpenNeuro
References
6- usesExcess non-target color alignment increased by 2.09 percentage points from two to six items in 117 ds008037 v1.0.1 adults under fixed angle rules and a control based on other trials sci:sha256:641e12a4b03c59a96c984dc1c30b909da95a04dfd35c3d80b630c2a5c8abea75
Its same-cohort alignment result helped define the question. Its use of other trial arrays informed the control. This study tests prediction with separate training and test people.
- usesWith circular error scores, mean error had a higher Pearson correlation across color and orientation tasks than load cost in 115 selected ds008037 v1.0.1 adults sci:sha256:e8a3c7eca943ddf09e69ed38dffea5bc790ab7bdb6f33273e4f847dbd4286ba0
This study used its unchanged public color tables and source hash records as the first source copy. It used no fitted scores or analysis code.
- usesdoi:10.18112/openneuro.ds008037.v1.0.1
This is the recorded dataset and exact version used for the behavioral tables. The analysis package records fixed-commit URLs and file hashes.
- doi:10.1038/srep19203
- doi:10.64898/2026.07.12.738085
- doi:10.1162/imag_a_00349
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.