{"paper":{"title":"Symmetry Reveals Layerwise Dynamics: How Transformers Perform In-Context Classification","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"By enforcing equivariance under feature and label permutations, transformer layers yield an explicit recursive update rule for in-context classification.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aditya Gangrade, Arjun Chandra, Patrick Lutz, Themistoklis Haris, Venkatesh Saligrama","submitted_at":"2026-04-13T15:20:41Z","abstract_excerpt":"Transformers can perform in-context classification from a few labeled examples, yet the inference-time algorithm remains opaque. We study multi-class linear classification in the hard no-margin regime and make the computation identifiable by enforcing feature- and label-permutation equivariance at every layer. This enables interpretability while maintaining functional equivalence and yields highly structured weights. From these models we extract an explicit depth-indexed recursion: an end-to-end identified, emergent update rule inside a softmax transformer, to our knowledge the first of its ki"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"From these models we extract an explicit depth-indexed recursion: an end-to-end identified, emergent update rule inside a softmax transformer, to our knowledge the first of its kind. Attention matrices formed from mixed feature-label Gram structure drive coupled updates of training points, labels, and the test probe. The resulting dynamics implement a geometry-driven algorithmic motif, which can provably amplify class separation and yields robust expected class alignment.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Enforcing feature- and label-permutation equivariance at every layer maintains functional equivalence to the original transformer while enabling interpretability and yielding highly structured weights from which the recursion can be extracted.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Enforcing feature- and label-permutation equivariance in transformers for in-context classification yields an identifiable emergent update rule driven by mixed feature-label Gram matrices that amplifies class separation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"By enforcing equivariance under feature and label permutations, transformer layers yield an explicit recursive update rule for in-context classification.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"561e6a605abf43197d17f8a1e223739c36d2f3722d6bbcba401f3e99b0a59dbb"},"source":{"id":"2604.11613","kind":"arxiv","version":3},"verdict":{"id":"6ec81c0c-c62c-464b-bc28-e8ae142e7a83","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T15:02:13.402151Z","strongest_claim":"From these models we extract an explicit depth-indexed recursion: an end-to-end identified, emergent update rule inside a softmax transformer, to our knowledge the first of its kind. Attention matrices formed from mixed feature-label Gram structure drive coupled updates of training points, labels, and the test probe. The resulting dynamics implement a geometry-driven algorithmic motif, which can provably amplify class separation and yields robust expected class alignment.","one_line_summary":"Enforcing feature- and label-permutation equivariance in transformers for in-context classification yields an identifiable emergent update rule driven by mixed feature-label Gram matrices that amplifies class separation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Enforcing feature- and label-permutation equivariance at every layer maintains functional equivalence to the original transformer while enabling interpretability and yielding highly structured weights from which the recursion can be extracted.","pith_extraction_headline":"By enforcing equivariance under feature and label permutations, transformer layers yield an explicit recursive update rule for in-context classification."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.11613/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"}