Two learning states reduced mean day 1 washout prediction loss in ds004021 v1.0.1 when model states remained fixed during rest
This study tested washout predictions from one and two learning states in OpenNeuro ds004021 v1.0.1. Models used rotation trials to fit parameters. They kept states fixed during the six-minute rest before washout. The main test used 31 people on day 1. Two states reduced mean squared prediction error by 10.46 degree squared, or 1.61%. The paired person bootstrap gave a 95% interval of 2.23 to 19.13 degree squared. The intervals included zero with passive rest decay and on day 2. Both models overpredicted the first washout error. The result supports a small prediction benefit under the tested assumptions. It does not identify separate neural systems. The package preserves source files, hashes, code, fits, predictions, and checks. A fresh download and full replay matched all source and result files.
motor learning · visuomotor adaptation · washout prediction · learning states · model comparison · OpenNeuro ds004021
References
3- usesdoi:10.18112/openneuro.ds004021.v1.0.1
The analysis used the behavior event files and small source metadata from this exact OpenNeuro dataset version; the archived source manifest preserves the fixed file URLs and hashes.
- doi:10.1371/journal.pbio.0040179
- doi:10.7554/eLife.74591
Cited by
9 publicationsPublic catalog entries on Sciscape.
- usesPersonal day 1 curves gave an uncertain day 2 rotation forecast gain over curves from other people in 32 selected ds004021 v1.0.1 people with one learning state reset to zero
We use its exact source event tables and metadata, source hashes, and unchanged state-model code, bounds, fixed starts, and solver. We use no prior fitted parameters or results. The new test forecasts day 2 rotation from day 1 rotation, against curves from other people.
- usesA previous error correction increased mean one-state forecast loss in 32 selected ds004021 v1.0.1 people after 160 training trials
Archived source files and their pinned manifest supplied the source data for the new rotation forecast test. No prior fitted predictions or analysis code were used.
- usesEarly rotation absolute error fell more than baseline error on day 2 in 32 ds004021 v1.0.1 people under a 40-trial rule (archive revision 2)
This publication gave the reference to the original ds004021 experiment. We used that reference to check cohort selection and task timing. Its model results did not enter our scores.
- usesThe 95% bootstrap interval for day 2 washout gain included zero after day 1 state scaling of a one-state model in ds004021 v1.0.1
This work uses the prior source recipe, eligibility rules, model equations, and code. Its first washout mismatch motivated the state scaling question. The full model summary was reproduced before the new day 1 calibration and day 2 test.
- usesEarly rotation absolute error fell more than baseline error on day 2 in 32 ds004021 v1.0.1 people under a 40-trial rule
This publication gave the reference to the original ds004021 experiment. We used that reference to check cohort selection and task timing. Its model results did not enter our scores.
- usesBoth daily mean within-person time-error slopes had 95 percent intervals that included zero in 32 ds004021 v1.0.1 adults, with raw errors and 16-trial blocks
The description informed the choice of ds004021 version 1.0.1 and the check for absent target direction. This study tests initiation time and error with a new calculation. It does not use the prior model fits, code, or scores.
- usesDay 1 two-state washout gain interval included zero after rest calibration in 31 ds004021 v1.0.1 people with fixed rotation fits
This result supplied unchanged behavior files, fixed rotation fits, all fit candidates, model code, and retrieval code. We reproduced its raw day 1 gain. We then tested rest corrections fitted on other people, with those rotation fits held fixed.
- usesTwo-state fits with expected error updates gave fast-state shares of at most 10% and at least 90% within 1% loss in 9 of 31 ds004021 v1.0.1 day 1 records
Use its pinned behavior files, retrieval code, model equations, and ordinary least-squares candidate vectors. Recalculate their losses and add constrained searches to test final state division.
- usesFull day 1 curve transfer had lower mean day 2 prediction loss in 31 people from ds004021 v1.0.1 with person exclusion and group offsets
This publication supplied the pinned source subset and manifest, retrieval code, state model code, and multiple-start fit method. The new work fitted day 1 curves again and tested day 2 curve transfer with each test person removed from group training. The earlier washout result was known before the local plan was saved.