{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:XKG5HDJH2XG2WSHL3JKFOVEGDJ","short_pith_number":"pith:XKG5HDJH","canonical_record":{"source":{"id":"2411.16073","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-25T03:52:47Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"e7581f2ed6b79d16e25d45d2b3648fc39fdb50e122bf029079f83f9b51c4a499","abstract_canon_sha256":"f31acb0b0abf2d3e5157b18a9effc475e2a54a029e7a1e40a2411717cd0ba219"},"schema_version":"1.0"},"canonical_sha256":"ba8dd38d27d5cdab48ebda545754861a5a6572ea559afa25cb6c370b0853c6b5","source":{"kind":"arxiv","id":"2411.16073","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2411.16073","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"arxiv_version","alias_value":"2411.16073v4","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.16073","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"pith_short_12","alias_value":"XKG5HDJH2XG2","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"pith_short_16","alias_value":"XKG5HDJH2XG2WSHL","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"pith_short_8","alias_value":"XKG5HDJH","created_at":"2026-07-22T01:22:09Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:XKG5HDJH2XG2WSHL3JKFOVEGDJ","target":"record","payload":{"canonical_record":{"source":{"id":"2411.16073","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-25T03:52:47Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"e7581f2ed6b79d16e25d45d2b3648fc39fdb50e122bf029079f83f9b51c4a499","abstract_canon_sha256":"f31acb0b0abf2d3e5157b18a9effc475e2a54a029e7a1e40a2411717cd0ba219"},"schema_version":"1.0"},"canonical_sha256":"ba8dd38d27d5cdab48ebda545754861a5a6572ea559afa25cb6c370b0853c6b5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-22T01:22:09.601260Z","signature_b64":"Zh9WitAd6Sk0utKHOV8+zmz2h2DJ3OrZUkJQmhKdxgvXsmCIABgDYq2WoWa0coACkPwkxOBpB8+OSAdAX1TUAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ba8dd38d27d5cdab48ebda545754861a5a6572ea559afa25cb6c370b0853c6b5","last_reissued_at":"2026-07-22T01:22:09.600324Z","signature_status":"signed_v1","first_computed_at":"2026-07-22T01:22:09.600324Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2411.16073","source_version":4,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-22T01:22:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"qLwXtCg9uLGCeuzOUQasxDff1KhKHJ/noEztQA2FIIN1lkfqqkJ4IAK4aEvg+OZb+pfW8FXzkCd4E/eGuvz2DA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-16T03:11:06.275585Z"},"content_sha256":"69cead4cf237899b858e0863a0e4ead49d1f8d9af73d49d350a5c2a4b9d4ae73","schema_version":"1.0","event_id":"sha256:69cead4cf237899b858e0863a0e4ead49d1f8d9af73d49d350a5c2a4b9d4ae73"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:XKG5HDJH2XG2WSHL3JKFOVEGDJ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Soft-TransFormers for Continual Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Task-specific multiplicative masks on attention projections enable continual learning in frozen pre-trained transformers with minimal added parameters.","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Chang D. Yoo, Haeyong Kang","submitted_at":"2024-11-25T03:52:47Z","abstract_excerpt":"Inspired by the Well-initialized Lottery Ticket Hypothesis (WLTH), we introduce Soft-TransFormers (Soft-TF), a continual learning framework that adapts a frozen pre-trained Transformer through task-specific soft subnetworks: real-valued multiplicative masks over the query, key, value, and output projections of selected self-attention layers. The masks are initialized at one, so optimization starts exactly at the pre-trained solution, and mask-space gradient descent is intrinsically biased toward modulating the backbone's dominant pathways; we prove that, under standard convex-Lipschitz assumpt"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across multiple continual learning benchmarks, Soft-TF achieves state-of-the-art performance, consistently outperforming prompt-based, adapter-based, and LoRA-style baselines while requiring minimal additional parameters.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The central claim depends on the premise that task-specific multiplicative masks applied to the key, query, value, and output projections in self-attention, together with a lightweight dual-prompt mechanism, will enable smooth task adaptation while preserving shared representations in the frozen pre-trained transformer.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Soft-TF learns task-specific multiplicative masks on transformer self-attention layers combined with dual prompts to achieve SOTA continual learning performance with minimal added parameters.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Task-specific multiplicative masks on attention projections enable continual learning in frozen pre-trained transformers with minimal added parameters.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c4ff11323261ab90278de6465da1485fce8de7d6da76c6ef7ef88fdc6cd48316"},"source":{"id":"2411.16073","kind":"arxiv","version":4},"verdict":{"id":"6f61dbe1-c6ba-4b2a-b130-f5d32a313076","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-23T08:11:08.840576Z","strongest_claim":"Across multiple continual learning benchmarks, Soft-TF achieves state-of-the-art performance, consistently outperforming prompt-based, adapter-based, and LoRA-style baselines while requiring minimal additional parameters.","one_line_summary":"Soft-TF learns task-specific multiplicative masks on transformer self-attention layers combined with dual prompts to achieve SOTA continual learning performance with minimal added parameters.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The central claim depends on the premise that task-specific multiplicative masks applied to the key, query, value, and output projections in self-attention, together with a lightweight dual-prompt mechanism, will enable smooth task adaptation while preserving shared representations in the frozen pre-trained transformer.","pith_extraction_headline":"Task-specific multiplicative masks on attention projections enable continual learning in frozen pre-trained transformers with minimal added parameters."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2411.16073/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"},"verdict_id":"6f61dbe1-c6ba-4b2a-b130-f5d32a313076"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-22T01:22:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fnZA2lYL604krtw5x2cVD+cgOb7VscCHkJfxQvd9n0N2AlMo9UugUTurgRd3A4cpB1WSj8dpXZh6CZkHHYk1Dg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-16T03:11:06.276309Z"},"content_sha256":"5d12f097d7061667656421a2d52d4b05aa4a15163c41de81153a9a169bab69f6","schema_version":"1.0","event_id":"sha256:5d12f097d7061667656421a2d52d4b05aa4a15163c41de81153a9a169bab69f6"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ/bundle.json","state_url":"https://pith.science/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-16T03:11:06Z","links":{"resolver":"https://pith.science/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ","bundle":"https://pith.science/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ/bundle.json","state":"https://pith.science/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/XKG5HDJH2XG2WSHL3JKFOVEGDJ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:XKG5HDJH2XG2WSHL3JKFOVEGDJ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f31acb0b0abf2d3e5157b18a9effc475e2a54a029e7a1e40a2411717cd0ba219","cross_cats_sorted":["cs.AI","cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-25T03:52:47Z","title_canon_sha256":"e7581f2ed6b79d16e25d45d2b3648fc39fdb50e122bf029079f83f9b51c4a499"},"schema_version":"1.0","source":{"id":"2411.16073","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2411.16073","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"arxiv_version","alias_value":"2411.16073v4","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.16073","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"pith_short_12","alias_value":"XKG5HDJH2XG2","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"pith_short_16","alias_value":"XKG5HDJH2XG2WSHL","created_at":"2026-07-22T01:22:09Z"},{"alias_kind":"pith_short_8","alias_value":"XKG5HDJH","created_at":"2026-07-22T01:22:09Z"}],"graph_snapshots":[{"event_id":"sha256:5d12f097d7061667656421a2d52d4b05aa4a15163c41de81153a9a169bab69f6","target":"graph","created_at":"2026-07-22T01:22:09Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Across multiple continual learning benchmarks, Soft-TF achieves state-of-the-art performance, consistently outperforming prompt-based, adapter-based, and LoRA-style baselines while requiring minimal additional parameters."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The central claim depends on the premise that task-specific multiplicative masks applied to the key, query, value, and output projections in self-attention, together with a lightweight dual-prompt mechanism, will enable smooth task adaptation while preserving shared representations in the frozen pre-trained transformer."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Soft-TF learns task-specific multiplicative masks on transformer self-attention layers combined with dual prompts to achieve SOTA continual learning performance with minimal added parameters."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Task-specific multiplicative masks on attention projections enable continual learning in frozen pre-trained transformers with minimal added parameters."}],"snapshot_sha256":"c4ff11323261ab90278de6465da1485fce8de7d6da76c6ef7ef88fdc6cd48316"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2411.16073/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Inspired by the Well-initialized Lottery Ticket Hypothesis (WLTH), we introduce Soft-TransFormers (Soft-TF), a continual learning framework that adapts a frozen pre-trained Transformer through task-specific soft subnetworks: real-valued multiplicative masks over the query, key, value, and output projections of selected self-attention layers. The masks are initialized at one, so optimization starts exactly at the pre-trained solution, and mask-space gradient descent is intrinsically biased toward modulating the backbone's dominant pathways; we prove that, under standard convex-Lipschitz assumpt","authors_text":"Chang D. Yoo, Haeyong Kang","cross_cats":["cs.AI","cs.CV"],"headline":"Task-specific multiplicative masks on attention projections enable continual learning in frozen pre-trained transformers with minimal added parameters.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-25T03:52:47Z","title":"Soft-TransFormers for Continual Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.16073","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-23T08:11:08.840576Z","id":"6f61dbe1-c6ba-4b2a-b130-f5d32a313076","model_set":{"reader":"grok-4.3"},"one_line_summary":"Soft-TF learns task-specific multiplicative masks on transformer self-attention layers combined with dual prompts to achieve SOTA continual learning performance with minimal added parameters.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Task-specific multiplicative masks on attention projections enable continual learning in frozen pre-trained transformers with minimal added parameters.","strongest_claim":"Across multiple continual learning benchmarks, Soft-TF achieves state-of-the-art performance, consistently outperforming prompt-based, adapter-based, and LoRA-style baselines while requiring minimal additional parameters.","weakest_assumption":"The central claim depends on the premise that task-specific multiplicative masks applied to the key, query, value, and output projections in self-attention, together with a lightweight dual-prompt mechanism, will enable smooth task adaptation while preserving shared representations in the frozen pre-trained transformer."}},"verdict_id":"6f61dbe1-c6ba-4b2a-b130-f5d32a313076"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:69cead4cf237899b858e0863a0e4ead49d1f8d9af73d49d350a5c2a4b9d4ae73","target":"record","created_at":"2026-07-22T01:22:09Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f31acb0b0abf2d3e5157b18a9effc475e2a54a029e7a1e40a2411717cd0ba219","cross_cats_sorted":["cs.AI","cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-25T03:52:47Z","title_canon_sha256":"e7581f2ed6b79d16e25d45d2b3648fc39fdb50e122bf029079f83f9b51c4a499"},"schema_version":"1.0","source":{"id":"2411.16073","kind":"arxiv","version":4}},"canonical_sha256":"ba8dd38d27d5cdab48ebda545754861a5a6572ea559afa25cb6c370b0853c6b5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ba8dd38d27d5cdab48ebda545754861a5a6572ea559afa25cb6c370b0853c6b5","first_computed_at":"2026-07-22T01:22:09.600324Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-22T01:22:09.600324Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Zh9WitAd6Sk0utKHOV8+zmz2h2DJ3OrZUkJQmhKdxgvXsmCIABgDYq2WoWa0coACkPwkxOBpB8+OSAdAX1TUAw==","signature_status":"signed_v1","signed_at":"2026-07-22T01:22:09.601260Z","signed_message":"canonical_sha256_bytes"},"source_id":"2411.16073","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:69cead4cf237899b858e0863a0e4ead49d1f8d9af73d49d350a5c2a4b9d4ae73","sha256:5d12f097d7061667656421a2d52d4b05aa4a15163c41de81153a9a169bab69f6"],"state_sha256":"a7ca2a79010f38d7ad828b6b9dbb214a3613fad8ea132753354ed27bb997b3c9"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1YBT6Rz2jTovI5FQNvJU4uDOQesJctwzxdu0fhdpcuYNOzJI7cLCpWCUdyRrhx7/k6blyL1nIoh6z6YHmh0MCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-16T03:11:06.280652Z","bundle_sha256":"f1d553c672e4d6d315af177d21b67520c74a054776fbe7a4dc97681f5446d623"}}