{"id":"W4410949880","doi":"10.1109/access.2025.3575440","title":"STAA: Spatio-Temporal Attention Attribution for Real-Time Interpreting Transformer-Based AI Video Models","year":2025,"lang":"en","type":"article","venue":"IEEE Access","topic":"Human Pose and Action Recognition","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Transformer; Attribution; Artificial intelligence; Speech recognition; Psychology; Voltage; Electrical engineering; Social psychology; Engineering","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003327743,0.0001408386,0.000168214,0.0002375185,0.0002510643,0.0004343071,0.0004833093,0.00008995963,0.00003172403],"category_scores_gemma":[0.0000131962,0.0001448606,0.0001359616,0.0003639533,0.0000231423,0.002304427,0.00002448401,0.0001040347,0.00002554074],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001097151,"about_ca_system_score_gemma":0.0001204293,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001102372,"about_ca_topic_score_gemma":0.00005340626,"domain_scores_codex":[0.9988002,0.00006164501,0.0003420117,0.0003730942,0.0001752863,0.0002477725],"domain_scores_gemma":[0.9992378,0.00009573335,0.0001209117,0.0002400017,0.0002574474,0.00004814662],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001150011,0.001487904,0.006328891,0.001581679,0.0004241445,0.00002146891,0.001304341,0.03849396,0.08950882,0.069571,0.08071671,0.7094111],"study_design_scores_gemma":[0.0009988594,0.00008964623,0.0005404988,0.0002472127,0.00003135997,9.477594e-7,0.000007808653,0.9101663,0.06095384,0.02471772,0.002008283,0.000237479],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1050942,0.000007619689,0.8908743,0.001517237,0.0005512876,0.0004566102,0.00002669699,0.0002338627,0.001238188],"genre_scores_gemma":[0.9955802,0.000006953007,0.002908014,0.0008149178,0.00007404265,0.0001390041,0.0001592338,0.000009202521,0.0003084498],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.890486,"threshold_uncertainty_score":0.5907242,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03212404749350114,"score_gpt":0.3268996864187976,"score_spread":0.2947756389252965,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}