{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2015:SEHA43DNJ6IWC7X7BKWCAPVWBZ","short_pith_number":"pith:SEHA43DN","canonical_record":{"source":{"id":"1511.04561","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2015-11-14T14:04:51Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"bea3b7dea49207127e2c002d3df7590ac8ed7931805504dc4521e4b5757aa6b1","abstract_canon_sha256":"4aed1b670886631ee7beb542fe04aa3593e70abdc52d83907aa402b886761c8b"},"schema_version":"1.0"},"canonical_sha256":"910e0e6c6d4f91617eff0aac203eb60e4885ec7dee3f2dbf62aec05268a66151","source":{"kind":"arxiv","id":"1511.04561","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"1511.04561","created_at":"2026-05-18T01:20:18Z"},{"alias_kind":"arxiv_version","alias_value":"1511.04561v4","created_at":"2026-05-18T01:20:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1511.04561","created_at":"2026-05-18T01:20:18Z"},{"alias_kind":"pith_short_12","alias_value":"SEHA43DNJ6IW","created_at":"2026-05-18T12:29:39Z"},{"alias_kind":"pith_short_16","alias_value":"SEHA43DNJ6IWC7X7","created_at":"2026-05-18T12:29:39Z"},{"alias_kind":"pith_short_8","alias_value":"SEHA43DN","created_at":"2026-05-18T12:29:39Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2015:SEHA43DNJ6IWC7X7BKWCAPVWBZ","target":"record","payload":{"canonical_record":{"source":{"id":"1511.04561","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2015-11-14T14:04:51Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"bea3b7dea49207127e2c002d3df7590ac8ed7931805504dc4521e4b5757aa6b1","abstract_canon_sha256":"4aed1b670886631ee7beb542fe04aa3593e70abdc52d83907aa402b886761c8b"},"schema_version":"1.0"},"canonical_sha256":"910e0e6c6d4f91617eff0aac203eb60e4885ec7dee3f2dbf62aec05268a66151","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T01:20:18.863541Z","signature_b64":"zFonkPsBH9h+X+0Xhu1OnkPuY9LgUbdqXc+N2WTEaXDYP1LAwhmc4BGqE5KELaY2xPugXUEhUKE+8FqDI0OZBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"910e0e6c6d4f91617eff0aac203eb60e4885ec7dee3f2dbf62aec05268a66151","last_reissued_at":"2026-05-18T01:20:18.862998Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T01:20:18.862998Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"1511.04561","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":"2026-05-18T01:20:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2eGvxK/7/t9lOxOGAo3cdekh/Pty6pgXJRI7pkOwIEU9UT+gA1mm3Jnk3nW/MCaFs5peqewIVdZsj3EYw1DmAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-23T21:38:01.688611Z"},"content_sha256":"5be82d21aae826cc87b7742489151b4ac1d21bad26246446307f08a4bb706770","schema_version":"1.0","event_id":"sha256:5be82d21aae826cc87b7742489151b4ac1d21bad26246446307f08a4bb706770"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2015:SEHA43DNJ6IWC7X7BKWCAPVWBZ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"8-Bit Approximations for Parallelism in Deep Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.NE","authors_text":"Tim Dettmers","submitted_at":"2015-11-14T14:04:51Z","abstract_excerpt":"The creation of practical deep learning data-products often requires parallelization across processors and computers to make deep learning feasible on large data sets, but bottlenecks in communication bandwidth make it difficult to attain good speedups through parallelism. Here we develop and test 8-bit approximation algorithms which make better use of the available bandwidth by compressing 32-bit gradients and nonlinear activations to 8-bit approximations. We show that these approximations do not decrease predictive performance on MNIST, CIFAR10, and ImageNet for both model and data paralleli"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1511.04561","kind":"arxiv","version":4},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"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":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T01:20:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"5zxQlz9iTOdNaS7ePtapEUq6Wp32TO2ax9l3O50MrWZgPICtecitmLY2UvuV4P6Jl5ocMm2ifwnQ7lLGkUnXAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-23T21:38:01.689309Z"},"content_sha256":"94159aeb97920493a9b678cc5e3d3c63e196f841cd622a5df51dda027687ee61","schema_version":"1.0","event_id":"sha256:94159aeb97920493a9b678cc5e3d3c63e196f841cd622a5df51dda027687ee61"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ/bundle.json","state_url":"https://pith.science/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ/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":"2026-05-23T21:38:01Z","links":{"resolver":"https://pith.science/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ","bundle":"https://pith.science/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ/bundle.json","state":"https://pith.science/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/SEHA43DNJ6IWC7X7BKWCAPVWBZ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2015:SEHA43DNJ6IWC7X7BKWCAPVWBZ","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":"4aed1b670886631ee7beb542fe04aa3593e70abdc52d83907aa402b886761c8b","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2015-11-14T14:04:51Z","title_canon_sha256":"bea3b7dea49207127e2c002d3df7590ac8ed7931805504dc4521e4b5757aa6b1"},"schema_version":"1.0","source":{"id":"1511.04561","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"1511.04561","created_at":"2026-05-18T01:20:18Z"},{"alias_kind":"arxiv_version","alias_value":"1511.04561v4","created_at":"2026-05-18T01:20:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1511.04561","created_at":"2026-05-18T01:20:18Z"},{"alias_kind":"pith_short_12","alias_value":"SEHA43DNJ6IW","created_at":"2026-05-18T12:29:39Z"},{"alias_kind":"pith_short_16","alias_value":"SEHA43DNJ6IWC7X7","created_at":"2026-05-18T12:29:39Z"},{"alias_kind":"pith_short_8","alias_value":"SEHA43DN","created_at":"2026-05-18T12:29:39Z"}],"graph_snapshots":[{"event_id":"sha256:94159aeb97920493a9b678cc5e3d3c63e196f841cd622a5df51dda027687ee61","target":"graph","created_at":"2026-05-18T01:20:18Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"The creation of practical deep learning data-products often requires parallelization across processors and computers to make deep learning feasible on large data sets, but bottlenecks in communication bandwidth make it difficult to attain good speedups through parallelism. Here we develop and test 8-bit approximation algorithms which make better use of the available bandwidth by compressing 32-bit gradients and nonlinear activations to 8-bit approximations. We show that these approximations do not decrease predictive performance on MNIST, CIFAR10, and ImageNet for both model and data paralleli","authors_text":"Tim Dettmers","cross_cats":["cs.LG"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2015-11-14T14:04:51Z","title":"8-Bit Approximations for Parallelism in Deep Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1511.04561","kind":"arxiv","version":4},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:5be82d21aae826cc87b7742489151b4ac1d21bad26246446307f08a4bb706770","target":"record","created_at":"2026-05-18T01:20:18Z","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":"4aed1b670886631ee7beb542fe04aa3593e70abdc52d83907aa402b886761c8b","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.NE","submitted_at":"2015-11-14T14:04:51Z","title_canon_sha256":"bea3b7dea49207127e2c002d3df7590ac8ed7931805504dc4521e4b5757aa6b1"},"schema_version":"1.0","source":{"id":"1511.04561","kind":"arxiv","version":4}},"canonical_sha256":"910e0e6c6d4f91617eff0aac203eb60e4885ec7dee3f2dbf62aec05268a66151","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"910e0e6c6d4f91617eff0aac203eb60e4885ec7dee3f2dbf62aec05268a66151","first_computed_at":"2026-05-18T01:20:18.862998Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T01:20:18.862998Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"zFonkPsBH9h+X+0Xhu1OnkPuY9LgUbdqXc+N2WTEaXDYP1LAwhmc4BGqE5KELaY2xPugXUEhUKE+8FqDI0OZBw==","signature_status":"signed_v1","signed_at":"2026-05-18T01:20:18.863541Z","signed_message":"canonical_sha256_bytes"},"source_id":"1511.04561","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:5be82d21aae826cc87b7742489151b4ac1d21bad26246446307f08a4bb706770","sha256:94159aeb97920493a9b678cc5e3d3c63e196f841cd622a5df51dda027687ee61"],"state_sha256":"6ef2321ff76591c74ecdc5b839dfcdfa4004ddf62e0852a0eef3ef9140935a9a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2gK2ajVjQY+jOOT6OY7RqyI5FJlyBoUuUhGZu1BU6oW6vvI4VBsXE01QlOyyyOcPlKvEfx+Xa9sh6DcQWlshCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-23T21:38:01.692946Z","bundle_sha256":"612dfb693287c9264cb88a9100888ef5bc8274b41fc73469fa8b72dfe87384bd"}}