{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:GZJPF47WITQTVEJSITO3CZKPRV","short_pith_number":"pith:GZJPF47W","schema_version":"1.0","canonical_sha256":"3652f2f3f644e13a913244ddb1654f8d7f2e4c28eda926db3f16cc4be2853a66","source":{"kind":"arxiv","id":"2103.09762","version":1},"attestation_state":"computed","paper":{"title":"Gradient Projection Memory for Continual Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Gobinda Saha, Isha Garg, Kaushik Roy","submitted_at":"2021-03-17T16:31:29Z","abstract_excerpt":"The ability to learn continually without forgetting the past tasks is a desired attribute for artificial learning systems. Existing approaches to enable such learning in artificial neural networks usually rely on network growth, importance based weight update or replay of old data from the memory. In contrast, we propose a novel approach where a neural network learns new tasks by taking gradient steps in the orthogonal direction to the gradient subspaces deemed important for the past tasks. We find the bases of these subspaces by analyzing network representations (activations) after learning e"},"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":"2103.09762","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-03-17T16:31:29Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"55284b3e0592a68b045fc03d549ad1bbb415e00d771291e388781532acf2a761","abstract_canon_sha256":"ef82253febebf5d423ce3f600fde0c3d1c5dbcc58d21e5ed0a1f1dbf9b289f7f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:24:07.122146Z","signature_b64":"p7ZCa+TFQDWWrwi8lLmAdo/eqqkCUIvyX7266eB9Q7JjFjYqt9oeMPGUalXB0jH68X0j5TLLLGYYaCGD6xwnDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3652f2f3f644e13a913244ddb1654f8d7f2e4c28eda926db3f16cc4be2853a66","last_reissued_at":"2026-07-05T02:24:07.121740Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:24:07.121740Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Projection Memory for Continual Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Gobinda Saha, Isha Garg, Kaushik Roy","submitted_at":"2021-03-17T16:31:29Z","abstract_excerpt":"The ability to learn continually without forgetting the past tasks is a desired attribute for artificial learning systems. Existing approaches to enable such learning in artificial neural networks usually rely on network growth, importance based weight update or replay of old data from the memory. In contrast, we propose a novel approach where a neural network learns new tasks by taking gradient steps in the orthogonal direction to the gradient subspaces deemed important for the past tasks. We find the bases of these subspaces by analyzing network representations (activations) after learning e"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.09762","kind":"arxiv","version":1},"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/2103.09762/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":"2103.09762","created_at":"2026-07-05T02:24:07.121799+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.09762v1","created_at":"2026-07-05T02:24:07.121799+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.09762","created_at":"2026-07-05T02:24:07.121799+00:00"},{"alias_kind":"pith_short_12","alias_value":"GZJPF47WITQT","created_at":"2026-07-05T02:24:07.121799+00:00"},{"alias_kind":"pith_short_16","alias_value":"GZJPF47WITQTVEJS","created_at":"2026-07-05T02:24:07.121799+00:00"},{"alias_kind":"pith_short_8","alias_value":"GZJPF47W","created_at":"2026-07-05T02:24:07.121799+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21307","citing_title":"Task-Differentiated Atomic Skill Expansion and Routing for Continual Learning Across Highly Heterogeneous Tasks","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18627","citing_title":"PACT: Preserving Anchored Cores in Task-vectors for Model Merging","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.24901","citing_title":"LLM Evolution as an Industry-Scale Ecosystem: A Lifecycle Perspective on Continual Learning","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11391","citing_title":"Recursive Binding on a Budget: Subspace Carving in Order-p Tensor Memories","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28745","citing_title":"FreqOrtho-SR: Frequency-Guided Orthogonal Expert Learning for Real-World Image Super-Resolution","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29164","citing_title":"Invariant Reasoning Directions in Latent Trajectories of Language Models","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15775","citing_title":"Continual Learning of Domain-Invariant Representations","ref_index":105,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19042","citing_title":"Interference-Aware Multi-Task Unlearning","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08949","citing_title":"Muon-OGD: Muon-based Spectral Orthogonal Gradient Projection for LLM Continual Learning","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2603.20410","citing_title":"SLE-FNO: Single-Layer Extensions for Task-Agnostic Continual Learning in Fourier Neural Operators","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14176","citing_title":"The Devil Is in Gradient Entanglement: Energy-Aware Gradient Coordinator for Robust Generalized Category Discovery","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08949","citing_title":"Muon-OGD: Muon-based Spectral Orthogonal Gradient Projection for LLM Continual Learning","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV","json":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV.json","graph_json":"https://pith.science/api/pith-number/GZJPF47WITQTVEJSITO3CZKPRV/graph.json","events_json":"https://pith.science/api/pith-number/GZJPF47WITQTVEJSITO3CZKPRV/events.json","paper":"https://pith.science/paper/GZJPF47W"},"agent_actions":{"view_html":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV","download_json":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV.json","view_paper":"https://pith.science/paper/GZJPF47W","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.09762&json=true","fetch_graph":"https://pith.science/api/pith-number/GZJPF47WITQTVEJSITO3CZKPRV/graph.json","fetch_events":"https://pith.science/api/pith-number/GZJPF47WITQTVEJSITO3CZKPRV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV/action/storage_attestation","attest_author":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV/action/author_attestation","sign_citation":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV/action/citation_signature","submit_replication":"https://pith.science/pith/GZJPF47WITQTVEJSITO3CZKPRV/action/replication_record"}},"created_at":"2026-07-05T02:24:07.121799+00:00","updated_at":"2026-07-05T02:24:07.121799+00:00"}