{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2016:MJDBDKNQ337SAA6LKHAOGQESAH","short_pith_number":"pith:MJDBDKNQ","canonical_record":{"source":{"id":"1610.01644","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2016-10-05T20:59:01Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"92c84a09dd32ca4757a4d79cae0417aaa14de4bb461d07657c772bb8ea8ea058","abstract_canon_sha256":"152bc9706aff7bec2fb3ee69bbef3d63361a1c95783a1885c16848a3ba3a56df"},"schema_version":"1.0"},"canonical_sha256":"624611a9b0deff2003cb51c0e3409201c70c0863c92ec3eff8cdecdff029ab31","source":{"kind":"arxiv","id":"1610.01644","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"1610.01644","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"arxiv_version","alias_value":"1610.01644v4","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1610.01644","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"pith_short_12","alias_value":"MJDBDKNQ337S","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"pith_short_16","alias_value":"MJDBDKNQ337SAA6L","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"pith_short_8","alias_value":"MJDBDKNQ","created_at":"2026-07-04T23:13:03Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2016:MJDBDKNQ337SAA6LKHAOGQESAH","target":"record","payload":{"canonical_record":{"source":{"id":"1610.01644","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2016-10-05T20:59:01Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"92c84a09dd32ca4757a4d79cae0417aaa14de4bb461d07657c772bb8ea8ea058","abstract_canon_sha256":"152bc9706aff7bec2fb3ee69bbef3d63361a1c95783a1885c16848a3ba3a56df"},"schema_version":"1.0"},"canonical_sha256":"624611a9b0deff2003cb51c0e3409201c70c0863c92ec3eff8cdecdff029ab31","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-04T23:13:03.801742Z","signature_b64":"lyVcWfMw+YWLwtt73UPFcbCp3eEz7UoglCDhPDMKc4I0TIMHlE1oH1m0gJh/LBJqquTvyYcX9H10LE5cvNXcDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"624611a9b0deff2003cb51c0e3409201c70c0863c92ec3eff8cdecdff029ab31","last_reissued_at":"2026-07-04T23:13:03.801103Z","signature_status":"signed_v1","first_computed_at":"2026-07-04T23:13:03.801103Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"1610.01644","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":"2026-07-04T23:13:03Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"DfRNiYmGY3zF+ghZeFL5licx4F0zVZhjUQLwsbUGwSi6ujINBrW9OroEW0f2shwXbJUdAS5Lfl7ICgcwn4NdCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-13T15:28:12.615853Z"},"content_sha256":"b02ac93e6615461c48b3137a32e0ad235637237a1fdac53bb340751ebb55cc4d","schema_version":"1.0","event_id":"sha256:b02ac93e6615461c48b3137a32e0ad235637237a1fdac53bb340751ebb55cc4d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2016:MJDBDKNQ337SAA6LKHAOGQESAH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Understanding intermediate layers using linear classifier probes","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Linear probes show that feature separability increases monotonically along the depth of neural networks.","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Guillaume Alain, Yoshua Bengio","submitted_at":"2016-10-05T20:59:01Z","abstract_excerpt":"Neural network models have a reputation for being black boxes. We propose to monitor the features at every layer of a model and measure how suitable they are for classification. We use linear classifiers, which we refer to as \"probes\", trained entirely independently of the model itself.\n  This helps us better understand the roles and dynamics of the intermediate layers. We demonstrate how this can be used to develop a better intuition about models and to diagnose potential problems.\n  We apply this technique to the popular models Inception v3 and Resnet-50. Among other things, we observe exper"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We observe experimentally that the linear separability of features increase monotonically along the depth of the model.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the accuracy of an independently trained linear probe on a layer's activations is a faithful measure of the information content and usefulness of those features for the downstream task.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Linear probes demonstrate that feature separability for classification increases monotonically with network depth in Inception v3 and ResNet-50.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Linear probes show that feature separability increases monotonically along the depth of neural networks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"fc6b89471f06ab31526f2d75e36752bd99b2074bfa1fdb294acc185b24a8d4e1"},"source":{"id":"1610.01644","kind":"arxiv","version":4},"verdict":{"id":"0325169f-55f5-4a0b-a782-ee1f4b74d2e6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T18:25:41.726397Z","strongest_claim":"We observe experimentally that the linear separability of features increase monotonically along the depth of the model.","one_line_summary":"Linear probes demonstrate that feature separability for classification increases monotonically with network depth in Inception v3 and ResNet-50.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the accuracy of an independently trained linear probe on a layer's activations is a faithful measure of the information content and usefulness of those features for the downstream task.","pith_extraction_headline":"Linear probes show that feature separability increases monotonically along the depth of neural networks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/1610.01644/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":28,"sample":[{"doi":"","year":2016,"title":"Understanding intermediate layers using linear classifier probes","work_id":"bdc944db-4be2-44f7-950b-eaef12fab00e","ref_index":1,"cited_arxiv_id":"1610.01644","is_internal_anchor":true},{"doi":"","year":2017,"title":"Explaining Recurrent Neural Network Predictions in Sentiment Analysis","work_id":"7bb4c170-9593-4e95-bc56-a7766913fbfc","ref_index":2,"cited_arxiv_id":"1706.07206","is_internal_anchor":false},{"doi":"","year":2015,"title":"Bach, S., Binder, A., Montavon, G., Klauschen, F., M \\\"u ller, K.-R., and Samek, W. (2015). On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PloS one","work_id":"195d8045-c9c0-4315-a70a-1fd464b1adcb","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2013,"title":"Biggio, B., Corona, I., Maiorca, D., Nelson, B., S rndi \\'c , N., Laskov, P., Giacinto, G., and Roli, F. (2013). Evasion attacks against machine learning at test time. In Joint European Conference on ","work_id":"d5a1e20b-34d4-47be-a136-c6d09e472bfa","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Binder, A., Montavon, G., Lapuschkin, S., M \\\"u ller, K.-R., and Samek, W. (2016). Layer-wise relevance propagation for neural networks with local renormalization layers. In International Conference o","work_id":"84f6fbf9-05c1-423b-83ab-8f4ac75a5756","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":28,"snapshot_sha256":"9a870ce093f0c4551a90c3748d21a4638b0ffec862b7aab9ce15e414fa59338f","internal_anchors":4},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c004339899a6e1aca575113edd3508d4fdefe8b8e7c9c8b261bea8ee8459e617"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"0325169f-55f5-4a0b-a782-ee1f4b74d2e6"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-04T23:13:03Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Qp026VM7oXSDhkRqwbkgLmTWLVuu8MqLWgs6inLYeAyWdYE0CNLs2XWPdjP+iV6yVElGW9/MN8A+o2ij4qRhBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-13T15:28:12.616400Z"},"content_sha256":"af00b58f338aa7305de66c3feb0a399735bb4de427df08751c45caa00f82d020","schema_version":"1.0","event_id":"sha256:af00b58f338aa7305de66c3feb0a399735bb4de427df08751c45caa00f82d020"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/MJDBDKNQ337SAA6LKHAOGQESAH/bundle.json","state_url":"https://pith.science/pith/MJDBDKNQ337SAA6LKHAOGQESAH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/MJDBDKNQ337SAA6LKHAOGQESAH/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":"2026-07-13T15:28:12Z","links":{"resolver":"https://pith.science/pith/MJDBDKNQ337SAA6LKHAOGQESAH","bundle":"https://pith.science/pith/MJDBDKNQ337SAA6LKHAOGQESAH/bundle.json","state":"https://pith.science/pith/MJDBDKNQ337SAA6LKHAOGQESAH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/MJDBDKNQ337SAA6LKHAOGQESAH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2016:MJDBDKNQ337SAA6LKHAOGQESAH","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":"152bc9706aff7bec2fb3ee69bbef3d63361a1c95783a1885c16848a3ba3a56df","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2016-10-05T20:59:01Z","title_canon_sha256":"92c84a09dd32ca4757a4d79cae0417aaa14de4bb461d07657c772bb8ea8ea058"},"schema_version":"1.0","source":{"id":"1610.01644","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"1610.01644","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"arxiv_version","alias_value":"1610.01644v4","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1610.01644","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"pith_short_12","alias_value":"MJDBDKNQ337S","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"pith_short_16","alias_value":"MJDBDKNQ337SAA6L","created_at":"2026-07-04T23:13:03Z"},{"alias_kind":"pith_short_8","alias_value":"MJDBDKNQ","created_at":"2026-07-04T23:13:03Z"}],"graph_snapshots":[{"event_id":"sha256:af00b58f338aa7305de66c3feb0a399735bb4de427df08751c45caa00f82d020","target":"graph","created_at":"2026-07-04T23:13:03Z","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":"We observe experimentally that the linear separability of features increase monotonically along the depth of the model."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the accuracy of an independently trained linear probe on a layer's activations is a faithful measure of the information content and usefulness of those features for the downstream task."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Linear probes demonstrate that feature separability for classification increases monotonically with network depth in Inception v3 and ResNet-50."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Linear probes show that feature separability increases monotonically along the depth of neural networks."}],"snapshot_sha256":"fc6b89471f06ab31526f2d75e36752bd99b2074bfa1fdb294acc185b24a8d4e1"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c004339899a6e1aca575113edd3508d4fdefe8b8e7c9c8b261bea8ee8459e617"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/1610.01644/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Neural network models have a reputation for being black boxes. We propose to monitor the features at every layer of a model and measure how suitable they are for classification. We use linear classifiers, which we refer to as \"probes\", trained entirely independently of the model itself.\n  This helps us better understand the roles and dynamics of the intermediate layers. We demonstrate how this can be used to develop a better intuition about models and to diagnose potential problems.\n  We apply this technique to the popular models Inception v3 and Resnet-50. Among other things, we observe exper","authors_text":"Guillaume Alain, Yoshua Bengio","cross_cats":["cs.LG"],"headline":"Linear probes show that feature separability increases monotonically along the depth of neural networks.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2016-10-05T20:59:01Z","title":"Understanding intermediate layers using linear classifier probes"},"references":{"count":28,"internal_anchors":4,"resolved_work":28,"sample":[{"cited_arxiv_id":"1610.01644","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Understanding intermediate layers using linear classifier probes","work_id":"bdc944db-4be2-44f7-950b-eaef12fab00e","year":2016},{"cited_arxiv_id":"1706.07206","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Explaining Recurrent Neural Network Predictions in Sentiment Analysis","work_id":"7bb4c170-9593-4e95-bc56-a7766913fbfc","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Bach, S., Binder, A., Montavon, G., Klauschen, F., M \\\"u ller, K.-R., and Samek, W. (2015). On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PloS one","work_id":"195d8045-c9c0-4315-a70a-1fd464b1adcb","year":2015},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Biggio, B., Corona, I., Maiorca, D., Nelson, B., S rndi \\'c , N., Laskov, P., Giacinto, G., and Roli, F. (2013). Evasion attacks against machine learning at test time. In Joint European Conference on ","work_id":"d5a1e20b-34d4-47be-a136-c6d09e472bfa","year":2013},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Binder, A., Montavon, G., Lapuschkin, S., M \\\"u ller, K.-R., and Samek, W. (2016). Layer-wise relevance propagation for neural networks with local renormalization layers. In International Conference o","work_id":"84f6fbf9-05c1-423b-83ab-8f4ac75a5756","year":2016}],"snapshot_sha256":"9a870ce093f0c4551a90c3748d21a4638b0ffec862b7aab9ce15e414fa59338f"},"source":{"id":"1610.01644","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-11T18:25:41.726397Z","id":"0325169f-55f5-4a0b-a782-ee1f4b74d2e6","model_set":{"reader":"grok-4.3"},"one_line_summary":"Linear probes demonstrate that feature separability for classification increases monotonically with network depth in Inception v3 and ResNet-50.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Linear probes show that feature separability increases monotonically along the depth of neural networks.","strongest_claim":"We observe experimentally that the linear separability of features increase monotonically along the depth of the model.","weakest_assumption":"That the accuracy of an independently trained linear probe on a layer's activations is a faithful measure of the information content and usefulness of those features for the downstream task."}},"verdict_id":"0325169f-55f5-4a0b-a782-ee1f4b74d2e6"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b02ac93e6615461c48b3137a32e0ad235637237a1fdac53bb340751ebb55cc4d","target":"record","created_at":"2026-07-04T23:13:03Z","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":"152bc9706aff7bec2fb3ee69bbef3d63361a1c95783a1885c16848a3ba3a56df","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2016-10-05T20:59:01Z","title_canon_sha256":"92c84a09dd32ca4757a4d79cae0417aaa14de4bb461d07657c772bb8ea8ea058"},"schema_version":"1.0","source":{"id":"1610.01644","kind":"arxiv","version":4}},"canonical_sha256":"624611a9b0deff2003cb51c0e3409201c70c0863c92ec3eff8cdecdff029ab31","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"624611a9b0deff2003cb51c0e3409201c70c0863c92ec3eff8cdecdff029ab31","first_computed_at":"2026-07-04T23:13:03.801103Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-04T23:13:03.801103Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"lyVcWfMw+YWLwtt73UPFcbCp3eEz7UoglCDhPDMKc4I0TIMHlE1oH1m0gJh/LBJqquTvyYcX9H10LE5cvNXcDw==","signature_status":"signed_v1","signed_at":"2026-07-04T23:13:03.801742Z","signed_message":"canonical_sha256_bytes"},"source_id":"1610.01644","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b02ac93e6615461c48b3137a32e0ad235637237a1fdac53bb340751ebb55cc4d","sha256:af00b58f338aa7305de66c3feb0a399735bb4de427df08751c45caa00f82d020"],"state_sha256":"e5e9b8fe6bbd011b7418d0fd4e26ca06753f876d03765db1e10cde9c6b0d6a41"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"TKObzUgZuFnXneXHwPEhD7DW8F46vxOK6TwGstA2GRX8O/uIiThCTn+TrF/MtqlnBbdSnlEHb77U78O4SjStCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-13T15:28:12.619156Z","bundle_sha256":"b959d9c24e860264a238f8ad90d612e01ad8fe392f91e41fca10c0890182e672"}}