{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DPE52BLBS3G6PCOV2BPZSMF54O","short_pith_number":"pith:DPE52BLB","schema_version":"1.0","canonical_sha256":"1bc9dd056196cde789d5d05f9930bde3aa7b25c0764d5354712ca8b80c3cbf3f","source":{"kind":"arxiv","id":"2406.04341","version":3},"attestation_state":"computed","paper":{"title":"Interpreting the Second-Order Effects of Neurons in CLIP","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Alexei A. Efros, Jacob Steinhardt, Yossi Gandelsman","submitted_at":"2024-06-06T17:59:52Z","abstract_excerpt":"We interpret the function of individual neurons in CLIP by automatically describing them using text. Analyzing the direct effects (i.e. the flow from a neuron through the residual stream to the output) or the indirect effects (overall contribution) fails to capture the neurons' function in CLIP. Therefore, we present the \"second-order lens\", analyzing the effect flowing from a neuron through the later attention heads, directly to the output. We find that these effects are highly selective: for each neuron, the effect is significant for <2% of the images. Moreover, each effect can be approximat"},"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":"2406.04341","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-06-06T17:59:52Z","cross_cats_sorted":[],"title_canon_sha256":"aed9a69839a06b46f9dd8e63178e534594abaefa03e72c5690b607e84b8c1d8c","abstract_canon_sha256":"7e20db686cb46c0b3a74ae2b4cbcce17a60ed261d32c3735711ba0fef178877e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:13:26.437959Z","signature_b64":"o4h2X4piQrgp7XGbNJ9JZMty5VJGTn0RQQYRraZuF7M9W296LmlXbsgBV+cbUIOrNcT2MXslNL9juUvzwgvlCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1bc9dd056196cde789d5d05f9930bde3aa7b25c0764d5354712ca8b80c3cbf3f","last_reissued_at":"2026-07-05T10:13:26.437442Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:13:26.437442Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting the Second-Order Effects of Neurons in CLIP","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Alexei A. Efros, Jacob Steinhardt, Yossi Gandelsman","submitted_at":"2024-06-06T17:59:52Z","abstract_excerpt":"We interpret the function of individual neurons in CLIP by automatically describing them using text. Analyzing the direct effects (i.e. the flow from a neuron through the residual stream to the output) or the indirect effects (overall contribution) fails to capture the neurons' function in CLIP. Therefore, we present the \"second-order lens\", analyzing the effect flowing from a neuron through the later attention heads, directly to the output. We find that these effects are highly selective: for each neuron, the effect is significant for <2% of the images. Moreover, each effect can be approximat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.04341","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/2406.04341/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":"2406.04341","created_at":"2026-07-05T10:13:26.437503+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.04341v3","created_at":"2026-07-05T10:13:26.437503+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.04341","created_at":"2026-07-05T10:13:26.437503+00:00"},{"alias_kind":"pith_short_12","alias_value":"DPE52BLBS3G6","created_at":"2026-07-05T10:13:26.437503+00:00"},{"alias_kind":"pith_short_16","alias_value":"DPE52BLBS3G6PCOV","created_at":"2026-07-05T10:13:26.437503+00:00"},{"alias_kind":"pith_short_8","alias_value":"DPE52BLB","created_at":"2026-07-05T10:13:26.437503+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02494","citing_title":"Towards Robustness against Typographic Attack with Training-free Concept Localization","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07802","citing_title":"Latent Anomaly Knowledge Excavation: Unveiling Sparse Sensitive Neurons in Vision-Language Models","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O","json":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O.json","graph_json":"https://pith.science/api/pith-number/DPE52BLBS3G6PCOV2BPZSMF54O/graph.json","events_json":"https://pith.science/api/pith-number/DPE52BLBS3G6PCOV2BPZSMF54O/events.json","paper":"https://pith.science/paper/DPE52BLB"},"agent_actions":{"view_html":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O","download_json":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O.json","view_paper":"https://pith.science/paper/DPE52BLB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.04341&json=true","fetch_graph":"https://pith.science/api/pith-number/DPE52BLBS3G6PCOV2BPZSMF54O/graph.json","fetch_events":"https://pith.science/api/pith-number/DPE52BLBS3G6PCOV2BPZSMF54O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O/action/storage_attestation","attest_author":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O/action/author_attestation","sign_citation":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O/action/citation_signature","submit_replication":"https://pith.science/pith/DPE52BLBS3G6PCOV2BPZSMF54O/action/replication_record"}},"created_at":"2026-07-05T10:13:26.437503+00:00","updated_at":"2026-07-05T10:13:26.437503+00:00"}