A fixed test did not confirm shared ResMem prediction gains in 14 iEEG patients and six fMRI adults
We tested whether hidden ResMem features add new image forecasts after output score and object category controls. Separate public studies supplied human depth EEG and BOLD fMRI. We tested selected ventral temporal contacts and face-area MRI vertices. Development used three patients and two MRI adults. A local code lock preceded tests in 14 other patients and six other adults. Mean added gains were -0.210 and +0.687 percentage points of reference error. Their 95% person intervals were [-0.369, -0.051] and [+0.296, +1.078]. Gains were positive in 3/14 and 6/6 people. The shared positive mean and majority rule failed, p=0.9935303. The electrical feature files needed a unique score-based row join. All 500 matches passed without neural outcomes. An official readout and source hashes checked the inputs. An independent calculation checked one outer fold in all 25 development and test people. The package preserves plans, source versions and terms, code, every result, uncertainty draws and figures. Original neural arrays, photos and model weights stay external. Clinical coverage, different cohorts and tasks, sparse single-image electrical responses, source selection and different measurement filters limit scope. No positive cross-modal gain or common neural cause is established. The result does not show absent memory information.
human iEEG · fMRI · memorability · encoding models · independent test · negative result · data provenance
References
9- usesdoi:10.1371/journal.pbio.3003614
We reuse its public ImageNet depth-electrode summaries and model features.
- usesdoi:10.1007/s42113-022-00126-5
We use the fixed ResMem model and its public weights.
- usesdoi:10.1038/s41593-021-00962-x
NSD is the original human fMRI study.
- usesdoi:10.48550/arXiv.2301.03198
We use the Algonauts source preparation through a fixed public derivative.
- usesdoi:10.3389/fninf.2025.1515873
Net2Brain supplies the public shared-subset download route.
- usesFixed tests did not confirm a positive relation between image memory scores and pattern stability in eight EEG adults and three fMRI adults sci:sha256:c81bd331212c58abb7bb7bc43666250411322b3988976d321ce12ddb313a0dab
We read its memory-norm distinction and used an acquisition helper during source inspection. None of its neural arrays is reused.
- Fixed tests of six posterior EEG channels and V1/V2 fMRI did not show shared spread ranks or mean pair relations for 51 videos sci:sha256:ede6a6964998375c701dd538e1a57ab47c134587eb6879b7c2a3ba26d3373f7a
We screened this video study pair before choosing the final question.
- doi:10.1016/j.conb.2019.04.002
- doi:10.7554/eLife.82566
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesA fixed test did not confirm taxonomic geometry beyond animacy and visual controls in 14 SEEG patients and five fMRI adults
Its source audit and first-session extraction informed input discovery and order checks. No neural arrays or fitted forecasts were reused.
- usesA fixed late hippocampal memory test failed in 25 patients and 21 fMRI adults
Its fixed prediction test and interpretation limits helped define the scope. We reuse no code or neural array.
- usesA fixed class covariance test failed in 14 iEEG patients and six fMRI adults
Its source discovery, version and data-route notes informed input inspection; no neural prediction losses or model features are reused.
- usesA fixed name prior reduced error with one class removed from each fit in 14 iEEG patients and six FFA-1 fMRI adults; electrical total skill was uncertain
We follow its documented source route and first-session anatomical eligibility, with a different class-exclusion endpoint.