{"paper":{"title":"From Clever Hans to Scientific Discovery: Interpreting EEG Foundational Transformers with LRP","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LRP on EEG foundation models can verify decisions and generate new biological hypotheses.","cross_cats":["q-bio.NC"],"primary_cat":"cs.AI","authors_text":"Bogdan Franczyk, Justus Meyer zu Bexten, Nico Scherf, Simon M. Hofmann","submitted_at":"2026-05-12T09:59:14Z","abstract_excerpt":"Emerging foundation models (FMs) in electroencephalography (EEG) promise a path to scale deep learning in diagnostics and brain-computer interfaces despite data scarcity, yet their opaque nature remains a barrier to wider adoption. We investigate attention-aware Layer-wise relevance propagation (LRP) as a post-hoc attribution method for EEG-FMs, extending LRP's use on convolutional neural network (CNN)-based EEG models to the Transformer architectures that current FMs are based on. We find that LRP can both verify EEG-FM decisions and surface novel, biologically plausible hypotheses from them."},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We find that LRP can both verify EEG-FM decisions and surface novel, biologically plausible hypotheses from them.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the LRP heatmaps correspond to biologically meaningful signals rather than model artifacts or post-hoc interpretation ambiguities in the complex EEG domain.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LRP on EEG transformers reveals Clever Hans artifacts in motor imagery tasks and a recurring central electrode cluster as a candidate sensorimotor signature of arousal.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LRP on EEG foundation models can verify decisions and generate new biological hypotheses.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ba1de7cdad7832d9a0a2b1385f7c0c089727b76c4e31eae4983413976bf6c7a7"},"source":{"id":"2605.11885","kind":"arxiv","version":1},"verdict":{"id":"569e63fe-d8d0-4c12-a8e2-df34db6c84e3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T06:18:42.674019Z","strongest_claim":"We find that LRP can both verify EEG-FM decisions and surface novel, biologically plausible hypotheses from them.","one_line_summary":"LRP on EEG transformers reveals Clever Hans artifacts in motor imagery tasks and a recurring central electrode cluster as a candidate sensorimotor signature of arousal.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the LRP heatmaps correspond to biologically meaningful signals rather than model artifacts or post-hoc interpretation ambiguities in the complex EEG domain.","pith_extraction_headline":"LRP on EEG foundation models can verify decisions and generate new biological hypotheses."},"integrity":{"clean":false,"summary":{"advisory":2,"critical":1,"by_detector":{"doi_compliance":{"total":3,"advisory":2,"critical":1,"informational":0}},"informational":0},"endpoint":"/pith/2605.11885/integrity.json","findings":[{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.48550/arXiv.1603.09382.arXiv:1603.09382) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":44,"audited_at":"2026-05-19T08:05:38.872408Z","detected_doi":"10.48550/arXiv.1603.09382.arXiv:1603.09382","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1038/s41586-023-05964-2.10) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":48,"audited_at":"2026-05-19T08:05:38.872408Z","detected_doi":"10.1038/s41586-023-05964-2.10","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"Identifier '10.1016/s0013-4694(96' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","detector":"doi_compliance","severity":"critical","ref_index":30,"audited_at":"2026-05-19T08:05:38.872408Z","detected_doi":"10.1016/s0013-4694(96","finding_type":"unresolvable_identifier","verdict_class":"cross_source","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T11:36:19.813915Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T09:01:17.982451Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:05:38.872408Z","status":"completed","version":"1.0.0","findings_count":3}],"snapshot_sha256":"af842796845c6c5f55ddbbe8de2917c310f3d29395860ea9cc8af1ddfaec003e"},"references":{"count":66,"sample":[{"doi":"10.48550/arxiv.2601.17883","year":2026,"title":"Dingkun Liu et al.EEG Foundation Models: Progresses, Benchmarking, and Open Problems. Feb. 5, 2026.DOI:10.48550/arXiv.2601.17883. arXiv:2601.17883 [cs]. Pre-published","work_id":"0f8b4eb4-4d7d-49b0-a5ed-20d1c6a90db0","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Large Brain Model for Learning Generic Represen- tations with Tremendous EEG Data in BCI","work_id":"370a7527-bcfa-4be6-84e7-c5ef4ffd0425","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.48550/arxiv.2412.07236","year":2025,"title":"Cbramod: A criss-cross brain foundation model for eeg decoding","work_id":"ebcabd2f-d1c5-42de-a39d-0974080f3619","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"An Accurate and Rapidly Calibrating Speech Neuroprosthesis","work_id":"e1c8c7ea-0a5f-4d80-859b-67f1cdd6374f","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1371/journal.pone.0130140","year":2015,"title":"On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer- Wise Relevance Propagation","work_id":"a2492a93-db2c-4bde-8872-6eaf0d7e310a","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":66,"snapshot_sha256":"11949ea8d11dc8b4026e019d67e81928b7b5d526eb880f6935a3c30c3bfe0ae4","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}