{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:37A4VU4L54T4SZ5MVE5DRGCPIF","short_pith_number":"pith:37A4VU4L","schema_version":"1.0","canonical_sha256":"dfc1cad38bef27c967aca93a38984f414efa0d672ee97f523474c6540706c874","source":{"kind":"arxiv","id":"2310.04861","version":2},"attestation_state":"computed","paper":{"title":"Uncovering hidden geometry in Transformers via disentangling position and context","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Jiajun Song, Yiqiao Zhong","submitted_at":"2023-10-07T15:50:26Z","abstract_excerpt":"Transformers are widely used to extract semantic meanings from input tokens, yet they usually operate as black-box models. In this paper, we present a simple yet informative decomposition of hidden states (or embeddings) of trained transformers into interpretable components. For any layer, embedding vectors of input sequence samples are represented by a tensor $\\boldsymbol{h} \\in \\mathbb{R}^{C \\times T \\times d}$. Given embedding vector $\\boldsymbol{h}_{c,t} \\in \\mathbb{R}^d$ at sequence position $t \\le T$ in a sequence (or context) $c \\le C$, extracting the mean effects yields the decompositi"},"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":"2310.04861","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-10-07T15:50:26Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"df62b2b290270a7a1b4d742d3b45f48832a60f88f7354ec8990496281435e99b","abstract_canon_sha256":"2426bcf0c935c97b920691bda7a37b00958d04d4ef6fd30feac0b274bc8819ec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:40:57.809864Z","signature_b64":"8z6f+Wqh8EmUAn3v5z0YGjteOg6+fuhakmp/48GUPb6H7MidvzpOoi0N/PaHnw13TFzBCiuHAPTdGT8/swnCDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dfc1cad38bef27c967aca93a38984f414efa0d672ee97f523474c6540706c874","last_reissued_at":"2026-07-05T07:40:57.809405Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:40:57.809405Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Uncovering hidden geometry in Transformers via disentangling position and context","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Jiajun Song, Yiqiao Zhong","submitted_at":"2023-10-07T15:50:26Z","abstract_excerpt":"Transformers are widely used to extract semantic meanings from input tokens, yet they usually operate as black-box models. In this paper, we present a simple yet informative decomposition of hidden states (or embeddings) of trained transformers into interpretable components. For any layer, embedding vectors of input sequence samples are represented by a tensor $\\boldsymbol{h} \\in \\mathbb{R}^{C \\times T \\times d}$. Given embedding vector $\\boldsymbol{h}_{c,t} \\in \\mathbb{R}^d$ at sequence position $t \\le T$ in a sequence (or context) $c \\le C$, extracting the mean effects yields the decompositi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.04861","kind":"arxiv","version":2},"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/2310.04861/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":"2310.04861","created_at":"2026-07-05T07:40:57.809471+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.04861v2","created_at":"2026-07-05T07:40:57.809471+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.04861","created_at":"2026-07-05T07:40:57.809471+00:00"},{"alias_kind":"pith_short_12","alias_value":"37A4VU4L54T4","created_at":"2026-07-05T07:40:57.809471+00:00"},{"alias_kind":"pith_short_16","alias_value":"37A4VU4L54T4SZ5M","created_at":"2026-07-05T07:40:57.809471+00:00"},{"alias_kind":"pith_short_8","alias_value":"37A4VU4L","created_at":"2026-07-05T07:40:57.809471+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.05076","citing_title":"High-Dimensional Statistics: Reflections on Progress and Open Problems","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30022","citing_title":"Give it Space! Explicit Disentangling of Positional and Semantic Representations in Encoders","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03780","citing_title":"Task Vector Geometry Underlies Dual Modes of Task Inference in Transformers","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05076","citing_title":"High-Dimensional Statistics: Reflections on Progress and Open Problems","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05115","citing_title":"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01148","citing_title":"Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19321","citing_title":"RDP LoRA: Geometry-Driven Identification for Parameter-Efficient Adaptation in Large Language Models","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF","json":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF.json","graph_json":"https://pith.science/api/pith-number/37A4VU4L54T4SZ5MVE5DRGCPIF/graph.json","events_json":"https://pith.science/api/pith-number/37A4VU4L54T4SZ5MVE5DRGCPIF/events.json","paper":"https://pith.science/paper/37A4VU4L"},"agent_actions":{"view_html":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF","download_json":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF.json","view_paper":"https://pith.science/paper/37A4VU4L","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.04861&json=true","fetch_graph":"https://pith.science/api/pith-number/37A4VU4L54T4SZ5MVE5DRGCPIF/graph.json","fetch_events":"https://pith.science/api/pith-number/37A4VU4L54T4SZ5MVE5DRGCPIF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF/action/storage_attestation","attest_author":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF/action/author_attestation","sign_citation":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF/action/citation_signature","submit_replication":"https://pith.science/pith/37A4VU4L54T4SZ5MVE5DRGCPIF/action/replication_record"}},"created_at":"2026-07-05T07:40:57.809471+00:00","updated_at":"2026-07-05T07:40:57.809471+00:00"}