For fixed sounds, separate category maps failed a shared gain test in 21 iEEG patients and 24 fMRI adults

We tested whether separate acoustic maps for five sound categories lower held-out prediction error beyond a common nonlinear map and category means. Separate human studies supplied signed iEEG field means and BOLD fMRI. Four patients and four MRI adults supplied development. A local code lock preceded tests in 21 other patients and 16 other adults. An added MRI study supplied eight adults. Mean added gains were -0.041, +0.465 and -0.338 error points. Pointwise 95% person Student limits were [-0.731,+0.648], [+0.249,+0.681] and [-0.745,+0.069]. Both fixed claim rules failed after Holm correction, p=1. One B2021 estimate had a negative mean gain. The averaged-response sound resample crossed zero. No positive shared gain was confirmed. Later one-sided 99% Student upper limits were 0.795, 0.729 and 0.178 points. These bounds require an adequate independent-person mean model and fixed sounds. They do not bound category information or all models. The upper-limit analysis is exploratory. Sparse clinical samples, separate cohorts and sounds, tasks, source screens and measurement filters limit scope. All negative results remain. The package gives code, licensed compact inputs, source pins and terms, all scores, figures and rerun steps. No common neural cause or causal effect was established.

human neuroscience · intracranial EEG · fMRI · auditory cortex · natural sounds · kernel regression · category models · negative result · prediction uncertainty