{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:QRWNU2EHR3OHX4Q65JJTPFSRUW","short_pith_number":"pith:QRWNU2EH","schema_version":"1.0","canonical_sha256":"846cda68878edc7bf21eea53379651a5818939343c6af685a257de7cba1f9214","source":{"kind":"arxiv","id":"2206.04496","version":1},"attestation_state":"computed","paper":{"title":"Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Adri\\'an Javaloy, Isabel Valera, Maryam Meghdadi","submitted_at":"2022-06-09T13:29:25Z","abstract_excerpt":"A number of variational autoencoders (VAEs) have recently emerged with the aim of modeling multimodal data, e.g., to jointly model images and their corresponding captions. Still, multimodal VAEs tend to focus solely on a subset of the modalities, e.g., by fitting the image while neglecting the caption. We refer to this limitation as modality collapse. In this work, we argue that this effect is a consequence of conflicting gradients during multimodal VAE training. We show how to detect the sub-graphs in the computational graphs where gradients conflict (impartiality blocks), as well as how to l"},"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":"2206.04496","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-06-09T13:29:25Z","cross_cats_sorted":[],"title_canon_sha256":"b51fc7c7e70da776e899bb305dd2439b5bd588e8a2b62a1bed64c52d8c957a61","abstract_canon_sha256":"2a6efd00a4c34bc41c145e761bbd770aacf1ef30cae36efe7caa424e1d71cde7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:30:34.454873Z","signature_b64":"sUxnR6lFDdq45aiCuveHcjfQY/J+CgHAU10Ve2F3J28sJTYhlnvTuiDMz4qgJUYN+ePi8WigEx318xOX4ofVCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"846cda68878edc7bf21eea53379651a5818939343c6af685a257de7cba1f9214","last_reissued_at":"2026-07-05T04:30:34.454443Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:30:34.454443Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Adri\\'an Javaloy, Isabel Valera, Maryam Meghdadi","submitted_at":"2022-06-09T13:29:25Z","abstract_excerpt":"A number of variational autoencoders (VAEs) have recently emerged with the aim of modeling multimodal data, e.g., to jointly model images and their corresponding captions. Still, multimodal VAEs tend to focus solely on a subset of the modalities, e.g., by fitting the image while neglecting the caption. We refer to this limitation as modality collapse. In this work, we argue that this effect is a consequence of conflicting gradients during multimodal VAE training. We show how to detect the sub-graphs in the computational graphs where gradients conflict (impartiality blocks), as well as how to l"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.04496","kind":"arxiv","version":1},"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/2206.04496/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":"2206.04496","created_at":"2026-07-05T04:30:34.454511+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.04496v1","created_at":"2026-07-05T04:30:34.454511+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.04496","created_at":"2026-07-05T04:30:34.454511+00:00"},{"alias_kind":"pith_short_12","alias_value":"QRWNU2EHR3OH","created_at":"2026-07-05T04:30:34.454511+00:00"},{"alias_kind":"pith_short_16","alias_value":"QRWNU2EHR3OHX4Q6","created_at":"2026-07-05T04:30:34.454511+00:00"},{"alias_kind":"pith_short_8","alias_value":"QRWNU2EH","created_at":"2026-07-05T04:30:34.454511+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.02492","citing_title":"Token-Efficient Multimodal Reasoning via Image Prompt Packaging","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW","json":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW.json","graph_json":"https://pith.science/api/pith-number/QRWNU2EHR3OHX4Q65JJTPFSRUW/graph.json","events_json":"https://pith.science/api/pith-number/QRWNU2EHR3OHX4Q65JJTPFSRUW/events.json","paper":"https://pith.science/paper/QRWNU2EH"},"agent_actions":{"view_html":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW","download_json":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW.json","view_paper":"https://pith.science/paper/QRWNU2EH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.04496&json=true","fetch_graph":"https://pith.science/api/pith-number/QRWNU2EHR3OHX4Q65JJTPFSRUW/graph.json","fetch_events":"https://pith.science/api/pith-number/QRWNU2EHR3OHX4Q65JJTPFSRUW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW/action/storage_attestation","attest_author":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW/action/author_attestation","sign_citation":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW/action/citation_signature","submit_replication":"https://pith.science/pith/QRWNU2EHR3OHX4Q65JJTPFSRUW/action/replication_record"}},"created_at":"2026-07-05T04:30:34.454511+00:00","updated_at":"2026-07-05T04:30:34.454511+00:00"}