{"paper":{"title":"A Lightweight Transformer for Pain Recognition from Brain Activity","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A lightweight transformer fuses raw and spectral fNIRS signals through unified tokenization to recognize pain states while remaining compact enough for real-time use.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Christian Arzate Cruz, Giorgos Giannakakis, Lu Cao, Muhammad Umar Khan, Randy Gomez, Raul Fernandez Rojas, Stefanos Gkikas, Thomas Kassiotis, Yu Fang","submitted_at":"2026-04-13T13:25:19Z","abstract_excerpt":"Pain is a multifaceted and widespread phenomenon with substantial clinical and societal burden, making reliable automated assessment a critical objective. This paper presents a lightweight transformer architecture that fuses multiple fNIRS representations through a unified tokenization mechanism, enabling joint modeling of complementary signal views without requiring modality-specific adaptations or increasing architectural complexity. The proposed token-mixing strategy preserves spatial, temporal, and time-frequency characteristics by projecting heterogeneous inputs onto a shared latent repre"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The proposed lightweight transformer fuses multiple fNIRS representations through a unified tokenization mechanism, enabling joint modeling of complementary signal views without requiring modality-specific adaptations or increasing architectural complexity, while achieving competitive pain recognition performance on the AI4Pain dataset.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That projecting heterogeneous fNIRS inputs (raw waveform and power spectral density) onto a shared latent representation via structured segmentation preserves all spatial, temporal, and time-frequency information necessary for accurate pain classification without significant loss.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A lightweight transformer fuses multiple fNIRS signal views through shared tokenization to achieve competitive pain recognition on the AI4Pain dataset while staying computationally compact.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A lightweight transformer fuses raw and spectral fNIRS signals through unified tokenization to recognize pain states while remaining compact enough for real-time use.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3f747db88a0ea524082529aeae0b7a81113c7b8f4035b30fe9910aabef588d3a"},"source":{"id":"2604.16491","kind":"arxiv","version":4},"verdict":{"id":"950b894c-a98c-474d-b9f8-aaf820bae837","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T06:39:25.956084Z","strongest_claim":"The proposed lightweight transformer fuses multiple fNIRS representations through a unified tokenization mechanism, enabling joint modeling of complementary signal views without requiring modality-specific adaptations or increasing architectural complexity, while achieving competitive pain recognition performance on the AI4Pain dataset.","one_line_summary":"A lightweight transformer fuses multiple fNIRS signal views through shared tokenization to achieve competitive pain recognition on the AI4Pain dataset while staying computationally compact.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That projecting heterogeneous fNIRS inputs (raw waveform and power spectral density) onto a shared latent representation via structured segmentation preserves all spatial, temporal, and time-frequency information necessary for accurate pain classification without significant loss.","pith_extraction_headline":"A lightweight transformer fuses raw and spectral fNIRS signals through unified tokenization to recognize pain states while remaining compact enough for real-time use."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.16491/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6fb8f68efa845abfd49139f899ef9a8364181dac3c529beb5bedd28a420b51d6"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}