{"paper":{"title":"Entity-Centric World Models: Interaction-Aware Masking for Causal Video Prediction","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Motion-centric masking in self-supervised video models enables better learning of causal physical dynamics by focusing on interactions rather than static patches.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Santosh Kumar Paidi","submitted_at":"2026-05-14T23:10:04Z","abstract_excerpt":"Learning predictive world models from unlabelled video is a foundational challenge in artificial intelligence. While Joint Embedding Predictive Architectures (JEPA) have set new benchmarks in semantic classification, they often remain physics-blind, failing to capture the causal dynamics necessary for downstream reasoning. We hypothesize that this stems from standard patch-based masking strategies, which prioritize visual texture over rare but informative kinematic events. We propose Interaction-Aware JEPA (IA-JEPA), which utilizes a self-supervised motion-centric masking strategy to prioritiz"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"IA-JEPA achieves 14.26% accuracy on causal reasoning tasks on CLEVRER, a significant lead over the 3.22% achieved by standard patch-masked baselines, while inducing a higher-entropy latent space that linearizes physical energy (R²=0.43).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that self-supervised motion-centric masking targeting collisions and momentum transfers will force reconstruction of latent trajectories rather than static background features, as stated in the hypothesis section of the abstract.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"IA-JEPA applies interaction-aware masking to JEPA, raising causal reasoning accuracy on CLEVRER from 3.22% to 14.26% while producing a higher-entropy latent space that better aligns with physical energy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Motion-centric masking in self-supervised video models enables better learning of causal physical dynamics by focusing on interactions rather than static patches.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"eb3cc5b199f81656c0618002dbdd3df67fb05cdc726dc5c7cd1a295953f24e08"},"source":{"id":"2605.15466","kind":"arxiv","version":1},"verdict":{"id":"82815f3e-9255-4357-a695-6b5a70fbe473","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T14:47:01.208588Z","strongest_claim":"IA-JEPA achieves 14.26% accuracy on causal reasoning tasks on CLEVRER, a significant lead over the 3.22% achieved by standard patch-masked baselines, while inducing a higher-entropy latent space that linearizes physical energy (R²=0.43).","one_line_summary":"IA-JEPA applies interaction-aware masking to JEPA, raising causal reasoning accuracy on CLEVRER from 3.22% to 14.26% while producing a higher-entropy latent space that better aligns with physical energy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that self-supervised motion-centric masking targeting collisions and momentum transfers will force reconstruction of latent trajectories rather than static background features, as stated in the hypothesis section of the abstract.","pith_extraction_headline":"Motion-centric masking in self-supervised video models enables better learning of causal physical dynamics by focusing on interactions rather than static patches."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15466/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"cited_work_retraction","ran_at":"2026-05-19T15:22:27.318179Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:17.597392Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:00:38.852451Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.092944Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T13:49:41.413784Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.666693Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"d7d9139d469b48e2826300510ae414636852894f389560e4a3d2147f66bfe59b"},"references":{"count":35,"sample":[{"doi":"","year":2024,"title":"Video generation models as world simulators.OpenAI Blog, 2024","work_id":"4f22b49c-b820-4a3f-9f6f-5273805df4d1","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning","work_id":"a9c28401-f16a-4933-89f0-788e2f94e52b","ref_index":2,"cited_arxiv_id":"2506.09985","is_internal_anchor":true},{"doi":"","year":1910,"title":"CLEVRER: CoLlision Events for Video REpresentation and Reasoning","work_id":"595fdbf7-89d5-4593-8f57-24e8b469a43c","ref_index":3,"cited_arxiv_id":"1910.01442","is_internal_anchor":true},{"doi":"","year":2017,"title":"something something","work_id":"a40b0c21-8c68-4e98-9adb-1f90135432f9","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Phyre: A new benchmark for physical reasoning","work_id":"d2c80b4d-2fe5-40a1-9b4f-0efe0581edc2","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":35,"snapshot_sha256":"cfa31b16a6c8aeb1d985d1a0aa47a20a879ccad37daafb1b8ac1cb962ee4b18f","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d4576378eac17aeb1491e00efd95af7b4d716cb6c2e88f048a7cf1ca5910117c"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}