Fixed acoustic metrics did not pass a shared transfer test in 20 iEEG patients and 16 fMRI adults
We tested whether acoustic metrics learned from one human neural measurement track sound distances in the other. Separate studies supplied signed mean iEEG local field potential and BOLD fMRI. Five patients and four adults supplied development. A local code lock preceded tests in 20 other patients and 16 other adults. Each target used a metric fitted only to development people in the other modality. A shared 88-descriptor family linked the sound panels. The exact electrical extractor version is unknown. We removed source class means and controlled equal-feature distance and a source-design metric from ridge algebra. Mean partial correlations were 0.02115 in iEEG and 0.12548 in fMRI. The 95% person intervals were [-0.01013, 0.05243] and [0.08825, 0.16270]. Positive counts were 14/20 and 16/16. The shared mean and sign rule failed, p=0.08657. The MRI repeat-distance check kept a positive mean. Ranges that also resampled sounds included zero in both modalities. This does not confirm transfer in both modalities. A separate finite construction used dependent noise columns. It gave transfer correlation one after equal-feature control, with no source sound tuning. The source-design reference removes that predictor exactly. An independent-noise simulation gave mean 0.661 before this extra control and -0.021 after it at 1,024 source sites. These examples are not fitted human causes. The package gives the assay, proof, metrics, licensed electrical inputs, exact external source retrieval, all results and repeat steps. Clinical sampling, cohort differences, source selection, unequal class labels, measurement noise and descriptor assumptions limit scope.
iEEG · fMRI · auditory cortex · acoustic metric · representational similarity · ridge regression · negative result · measurement bias
References
11- usesAcoustic maps fitted to other sound classes reduced mean error for fixed music clips in human iEEG and fMRI sci:sha256:bb69205ba6952221ba99edf66c2f93e577a7ec38f3af18b0ceac7ea1c1fda7f5
Verified source of the unchanged licensed electrical derivative, MRI selection reader and source checks.
- One iEEG panel and eight fMRI adults showed small pattern transfer across labels for 49 sounds, but centered BOLD error exceeded baseline sci:sha256:fce5ce660523de796311505b1b20c40fdccf3a250aeb008d1ad8a101e2c3b62d
Related cross-label transfer benchmark; the present test transfers acoustic metrics between modalities with person holdouts.
- doi:10.1016/j.cub.2026.07.063
- doi:10.5281/zenodo.20599001
- doi:10.1152/jn.00588.2020
- doi:10.1371/journal.pbio.3002366
- doi:10.1016/j.neuron.2015.11.035
- doi:10.1371/journal.pcbi.1005508
- doi:10.1016/j.neuroimage.2022.119294
- doi:10.1109/TAFFC.2015.2457417
- doi:10.1016/j.cub.2022.01.069
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesIn exploratory tests, acoustic feature clipping lowered baseline error in 20 iEEG patients and 24 fMRI adults; a fixed slope test failed
Licensed electrical derivative, source reconstruction code and fixed audio metadata.
- usesA fixed acoustic distance rule failed a shared error rank test in 20 iEEG patients and 24 fMRI adults
Licensed electrical derivative, fixed source pins and MIT source readers.
- usesSixteen acoustic directions failed a shared prediction gain test in 20 iEEG patients and 16 fMRI adults
We extend its acoustic-metric transfer question and its source-design control to held-out response prediction and an exact class-shuffle expectation.
- usesA fixed acoustic field map did not pass a shared transfer test in 13 iEEG patients and 16 fMRI adults
We use its licensed electrical derivative, pinned source readers and ridge design-noise calculation. We test a new anatomical contrast.
- A fixed class relation rule failed a shared gain test in 20 iEEG patients and 16 fMRI adults
Prior fixed acoustic metric transfer failed a shared test; our class residual operator tests a different prediction.