{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:M3LDA2HUNCSK7G3BMUES2KNZHA","short_pith_number":"pith:M3LDA2HU","schema_version":"1.0","canonical_sha256":"66d63068f468a4af9b6165092d29b93832c10d520f18cd4899bc73fb0dd10211","source":{"kind":"arxiv","id":"2202.05306","version":3},"attestation_state":"computed","paper":{"title":"Characterizing and overcoming the greedy nature of learning in multi-modal deep neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Krzysztof J. Geras, Kyunghyun Cho, Nan Wu, Stanis{\\l}aw Jastrz\\k{e}bski","submitted_at":"2022-02-10T20:11:21Z","abstract_excerpt":"We hypothesize that due to the greedy nature of learning in multi-modal deep neural networks, these models tend to rely on just one modality while under-fitting the other modalities. Such behavior is counter-intuitive and hurts the models' generalization, as we observe empirically. To estimate the model's dependence on each modality, we compute the gain on the accuracy when the model has access to it in addition to another modality. We refer to this gain as the conditional utilization rate. In the experiments, we consistently observe an imbalance in conditional utilization rates between modali"},"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":"2202.05306","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-02-10T20:11:21Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"704454b66d98f74196f97515bd3b70336056ceafdfd72dd03b678c82f37cad6a","abstract_canon_sha256":"315792bc8b0f4a5f32dc8c39067e3fae0f0744622563e2aa1050521466e5161d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:58:15.254024Z","signature_b64":"5QEL5fxNHL8MuO3D/tO8ff6/Xk8WXMu4n2OD4Nc68iQ1PlBUhB1+XqCIvhluTGBH5f8+xxMtYGTG6Au6eYLeDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"66d63068f468a4af9b6165092d29b93832c10d520f18cd4899bc73fb0dd10211","last_reissued_at":"2026-07-05T04:58:15.253510Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:58:15.253510Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Characterizing and overcoming the greedy nature of learning in multi-modal deep neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Krzysztof J. Geras, Kyunghyun Cho, Nan Wu, Stanis{\\l}aw Jastrz\\k{e}bski","submitted_at":"2022-02-10T20:11:21Z","abstract_excerpt":"We hypothesize that due to the greedy nature of learning in multi-modal deep neural networks, these models tend to rely on just one modality while under-fitting the other modalities. Such behavior is counter-intuitive and hurts the models' generalization, as we observe empirically. To estimate the model's dependence on each modality, we compute the gain on the accuracy when the model has access to it in addition to another modality. We refer to this gain as the conditional utilization rate. In the experiments, we consistently observe an imbalance in conditional utilization rates between modali"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2202.05306","kind":"arxiv","version":3},"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/2202.05306/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":"2202.05306","created_at":"2026-07-05T04:58:15.253580+00:00"},{"alias_kind":"arxiv_version","alias_value":"2202.05306v3","created_at":"2026-07-05T04:58:15.253580+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2202.05306","created_at":"2026-07-05T04:58:15.253580+00:00"},{"alias_kind":"pith_short_12","alias_value":"M3LDA2HUNCSK","created_at":"2026-07-05T04:58:15.253580+00:00"},{"alias_kind":"pith_short_16","alias_value":"M3LDA2HUNCSK7G3B","created_at":"2026-07-05T04:58:15.253580+00:00"},{"alias_kind":"pith_short_8","alias_value":"M3LDA2HU","created_at":"2026-07-05T04:58:15.253580+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.08519","citing_title":"SeBA: Semi-supervised few-shot learning via Separated-at-Birth Alignment for tabular data","ref_index":119,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA","json":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA.json","graph_json":"https://pith.science/api/pith-number/M3LDA2HUNCSK7G3BMUES2KNZHA/graph.json","events_json":"https://pith.science/api/pith-number/M3LDA2HUNCSK7G3BMUES2KNZHA/events.json","paper":"https://pith.science/paper/M3LDA2HU"},"agent_actions":{"view_html":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA","download_json":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA.json","view_paper":"https://pith.science/paper/M3LDA2HU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2202.05306&json=true","fetch_graph":"https://pith.science/api/pith-number/M3LDA2HUNCSK7G3BMUES2KNZHA/graph.json","fetch_events":"https://pith.science/api/pith-number/M3LDA2HUNCSK7G3BMUES2KNZHA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA/action/storage_attestation","attest_author":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA/action/author_attestation","sign_citation":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA/action/citation_signature","submit_replication":"https://pith.science/pith/M3LDA2HUNCSK7G3BMUES2KNZHA/action/replication_record"}},"created_at":"2026-07-05T04:58:15.253580+00:00","updated_at":"2026-07-05T04:58:15.253580+00:00"}