{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HIQ6DGUX56AUYPTLVC6VJVCBS7","short_pith_number":"pith:HIQ6DGUX","schema_version":"1.0","canonical_sha256":"3a21e19a97ef814c3e6ba8bd54d44197f90e099733e106bf0366b2789945b30c","source":{"kind":"arxiv","id":"2409.05929","version":6},"attestation_state":"computed","paper":{"title":"M3-JEPA: Multimodal Alignment via Multi-gate MoE based on the Joint-Embedding Predictive Architecture","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dan Wang, Hongyang Lei, Huazhen Huang, Kun Fan, Luo Ji, Qingqing Gu, Qi Qin, Xiaolong Cheng, Yetao Wu, Yong Chen, Zhonglin Jiang","submitted_at":"2024-09-09T10:40:50Z","abstract_excerpt":"Current multimodal learning strategies primarily optimize in the original token space. Such a framework is easy to incorporate with the backbone of pretrained language model, but might result in modality collapse. To alleviate such issues, we leverage the Joint-Embedding Predictive Architecture (JEPA) on the multimodal tasks, which converts the input embedding into the output embedding space by a predictor and then conducts the cross-modal alignment on the latent space. We implement this predictor by a Multi-Gate Mixture of Experts (MMoE) and name the framework as M3-JEPA, accordingly. The gat"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2409.05929","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-09T10:40:50Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d527195e0bed7d7d313e97665d2541ae8db8c47369b092cc842649e57f7a785e","abstract_canon_sha256":"7bcdc0326893ab43d0acde9453729c8a71ce303cca81e4e52b68d3211d297f13"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:23:18.086068Z","signature_b64":"Jw/Cdn0hu0ts7hLDhC0GOAhevbUex+mGpIWdu4Par999wfyF2CLqhENTfrbdXPIazdZ7u/nq55SyRBG2xxX5Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a21e19a97ef814c3e6ba8bd54d44197f90e099733e106bf0366b2789945b30c","last_reissued_at":"2026-07-05T11:23:18.085597Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:23:18.085597Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"M3-JEPA: Multimodal Alignment via Multi-gate MoE based on the Joint-Embedding Predictive Architecture","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dan Wang, Hongyang Lei, Huazhen Huang, Kun Fan, Luo Ji, Qingqing Gu, Qi Qin, Xiaolong Cheng, Yetao Wu, Yong Chen, Zhonglin Jiang","submitted_at":"2024-09-09T10:40:50Z","abstract_excerpt":"Current multimodal learning strategies primarily optimize in the original token space. Such a framework is easy to incorporate with the backbone of pretrained language model, but might result in modality collapse. To alleviate such issues, we leverage the Joint-Embedding Predictive Architecture (JEPA) on the multimodal tasks, which converts the input embedding into the output embedding space by a predictor and then conducts the cross-modal alignment on the latent space. We implement this predictor by a Multi-Gate Mixture of Experts (MMoE) and name the framework as M3-JEPA, accordingly. The gat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.05929","kind":"arxiv","version":6},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2409.05929/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":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2409.05929","created_at":"2026-07-05T11:23:18.085657+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.05929v6","created_at":"2026-07-05T11:23:18.085657+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.05929","created_at":"2026-07-05T11:23:18.085657+00:00"},{"alias_kind":"pith_short_12","alias_value":"HIQ6DGUX56AU","created_at":"2026-07-05T11:23:18.085657+00:00"},{"alias_kind":"pith_short_16","alias_value":"HIQ6DGUX56AUYPTL","created_at":"2026-07-05T11:23:18.085657+00:00"},{"alias_kind":"pith_short_8","alias_value":"HIQ6DGUX","created_at":"2026-07-05T11:23:18.085657+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10305","citing_title":"SARM2: Multi-Task Stage Aware Reward Modeling for Self Improving Robotic Manipulation","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00573","citing_title":"BrainFIBRE: A Foundation Model via Information Decomposition for Brain Microstructure","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27431","citing_title":"Tackling Multimodal Learning Challenges with Mixture-of-Expert: A Survey","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04943","citing_title":"DART: A Vision-Language Foundation Model for Comprehensive Rope Condition Monitoring","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7","json":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7.json","graph_json":"https://pith.science/api/pith-number/HIQ6DGUX56AUYPTLVC6VJVCBS7/graph.json","events_json":"https://pith.science/api/pith-number/HIQ6DGUX56AUYPTLVC6VJVCBS7/events.json","paper":"https://pith.science/paper/HIQ6DGUX"},"agent_actions":{"view_html":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7","download_json":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7.json","view_paper":"https://pith.science/paper/HIQ6DGUX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.05929&json=true","fetch_graph":"https://pith.science/api/pith-number/HIQ6DGUX56AUYPTLVC6VJVCBS7/graph.json","fetch_events":"https://pith.science/api/pith-number/HIQ6DGUX56AUYPTLVC6VJVCBS7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7/action/storage_attestation","attest_author":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7/action/author_attestation","sign_citation":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7/action/citation_signature","submit_replication":"https://pith.science/pith/HIQ6DGUX56AUYPTLVC6VJVCBS7/action/replication_record"}},"created_at":"2026-07-05T11:23:18.085657+00:00","updated_at":"2026-07-05T11:23:18.085657+00:00"}