{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:KWR3RFXI5LAPRMTISHZ34OC3MP","short_pith_number":"pith:KWR3RFXI","schema_version":"1.0","canonical_sha256":"55a3b896e8eac0f8b26891f3be385b63f4d784b59e1a81afeef9dcab15b01298","source":{"kind":"arxiv","id":"2605.13265","version":1},"attestation_state":"computed","paper":{"title":"LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LightSplit uses a fixed orthogonal random projection at the cut layer to cut transmitted dimensionality by up to 32 times while retaining more than 95% of baseline accuracy in split learning.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ahmad-Reza Sadeghi, Alessandro Pegoraro, Antonino Nocera, Mert Cihangiroglu, Phillip Rieger","submitted_at":"2026-05-13T09:45:49Z","abstract_excerpt":"Split learning (SL) enables collaborative training by partitioning a neural network across clients and a central server, but the cut-layer interface introduces a key challenge: high-dimensional activations incur substantial communication overhead while exposing representations vulnerable to reconstruction attacks. Existing approaches typically address efficiency or privacy in isolation, relying on additional mechanisms such as sparsification, quantization, or noise injection. We propose LightSplit, which limits information exposure and reduces communication overhead by applying a lightweight f"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.13265","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T09:45:49Z","cross_cats_sorted":[],"title_canon_sha256":"a1595972fd3564df05e5d2a8a5afd78df77d3709383f59f2ff1df16250ed3f94","abstract_canon_sha256":"43bc4468d8efa64bc047a2b1c89fa529f5a80989e9a3944abc7a1c7de90affa9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:49.320941Z","signature_b64":"g615/n97VecI+8kW39/ImJDD5A5ue236W4UugReHrRr1zOdha5BN6cgfvHYPu2sBXN82HB00HQq7abFfwpriDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"55a3b896e8eac0f8b26891f3be385b63f4d784b59e1a81afeef9dcab15b01298","last_reissued_at":"2026-05-18T02:44:49.320494Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:49.320494Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LightSplit uses a fixed orthogonal random projection at the cut layer to cut transmitted dimensionality by up to 32 times while retaining more than 95% of baseline accuracy in split learning.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ahmad-Reza Sadeghi, Alessandro Pegoraro, Antonino Nocera, Mert Cihangiroglu, Phillip Rieger","submitted_at":"2026-05-13T09:45:49Z","abstract_excerpt":"Split learning (SL) enables collaborative training by partitioning a neural network across clients and a central server, but the cut-layer interface introduces a key challenge: high-dimensional activations incur substantial communication overhead while exposing representations vulnerable to reconstruction attacks. Existing approaches typically address efficiency or privacy in isolation, relying on additional mechanisms such as sparsification, quantization, or noise injection. We propose LightSplit, which limits information exposure and reduces communication overhead by applying a lightweight f"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our results show that the method retains more than 95% of the baseline accuracy at up to 32x reduction in transmitted dimensionality while maintaining stable training dynamics.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a fixed orthogonal random projection alone sufficiently restricts instance-specific information to prevent reconstruction attacks across varying projection dimensions and client scales without requiring additional mechanisms such as sparsification or noise.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LightSplit uses non-invertible orthogonal projections as an information bottleneck in split learning to reduce transmitted dimensionality by 32x while retaining more than 95% accuracy and limiting reconstruction risk.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LightSplit uses a fixed orthogonal random projection at the cut layer to cut transmitted dimensionality by up to 32 times while retaining more than 95% of baseline accuracy in split learning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"086abe4a0abc7537889a9c94c365b4fa8407decf160bb012b06e64a07c632408"},"source":{"id":"2605.13265","kind":"arxiv","version":1},"verdict":{"id":"a42467e1-7ef4-4d20-ae4b-e17c4d05dc2f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:42:52.093627Z","strongest_claim":"Our results show that the method retains more than 95% of the baseline accuracy at up to 32x reduction in transmitted dimensionality while maintaining stable training dynamics.","one_line_summary":"LightSplit uses non-invertible orthogonal projections as an information bottleneck in split learning to reduce transmitted dimensionality by 32x while retaining more than 95% accuracy and limiting reconstruction risk.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a fixed orthogonal random projection alone sufficiently restricts instance-specific information to prevent reconstruction attacks across varying projection dimensions and client scales without requiring additional mechanisms such as sparsification or noise.","pith_extraction_headline":"LightSplit uses a fixed orthogonal random projection at the cut layer to cut transmitted dimensionality by up to 32 times while retaining more than 95% of baseline accuracy in split learning."},"references":{"count":67,"sample":[{"doi":"","year":2016,"title":"Deep learning with differential privacy","work_id":"661d5a82-ddfd-4b60-b514-61c0abb5389b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"CONTRA: Defending against poisoning attacks in federated learning","work_id":"f0ba6c5d-8028-405c-ab32-b794a573c865","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2012,"title":"The Johnson–Lindenstrauss transform itself preserves differential privacy","work_id":"a5785994-d438-49e5-8bae-0ce12f6f4cf0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Ashish Bora, Ajil Jalal, Eric Price, and Alexandros G. Dimakis. Com- pressed sensing using generative models. InICML, 2017. 13","work_id":"80e34f3d-a9a6-48ca-9cb9-8b90729f25cd","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"FLTrust: Byzantine-robust federated learning via trust bootstrapping","work_id":"94f11d46-340d-450c-b6b5-bbb5b9de8dfb","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":67,"snapshot_sha256":"9de08cc000085ae22e5744aa58ba20d605c84334510012a225a6904595326c83","internal_anchors":3},"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":"2605.13265","created_at":"2026-05-18T02:44:49.320574+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.13265v1","created_at":"2026-05-18T02:44:49.320574+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13265","created_at":"2026-05-18T02:44:49.320574+00:00"},{"alias_kind":"pith_short_12","alias_value":"KWR3RFXI5LAP","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"KWR3RFXI5LAPRMTI","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"KWR3RFXI","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP","json":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP.json","graph_json":"https://pith.science/api/pith-number/KWR3RFXI5LAPRMTISHZ34OC3MP/graph.json","events_json":"https://pith.science/api/pith-number/KWR3RFXI5LAPRMTISHZ34OC3MP/events.json","paper":"https://pith.science/paper/KWR3RFXI"},"agent_actions":{"view_html":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP","download_json":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP.json","view_paper":"https://pith.science/paper/KWR3RFXI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.13265&json=true","fetch_graph":"https://pith.science/api/pith-number/KWR3RFXI5LAPRMTISHZ34OC3MP/graph.json","fetch_events":"https://pith.science/api/pith-number/KWR3RFXI5LAPRMTISHZ34OC3MP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP/action/storage_attestation","attest_author":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP/action/author_attestation","sign_citation":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP/action/citation_signature","submit_replication":"https://pith.science/pith/KWR3RFXI5LAPRMTISHZ34OC3MP/action/replication_record"}},"created_at":"2026-05-18T02:44:49.320574+00:00","updated_at":"2026-05-18T02:44:49.320574+00:00"}