{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2015:J2XRAV4ISRPJW4MLUWW26ENR4T","short_pith_number":"pith:J2XRAV4I","schema_version":"1.0","canonical_sha256":"4eaf105788945e9b718ba5adaf11b1e4ea2c55898a9617c3e7ccbafbd6ad9888","source":{"kind":"arxiv","id":"1506.02078","version":2},"attestation_state":"computed","paper":{"title":"Visualizing and Understanding Recurrent Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.NE"],"primary_cat":"cs.LG","authors_text":"Andrej Karpathy, Justin Johnson, Li Fei-Fei","submitted_at":"2015-06-05T22:33:04Z","abstract_excerpt":"Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal"},"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":"1506.02078","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2015-06-05T22:33:04Z","cross_cats_sorted":["cs.CL","cs.NE"],"title_canon_sha256":"3ee0e2b543ddef1c0a1d57a6c175a596a954b373620a71bbbb72270ab634c44e","abstract_canon_sha256":"c95e9ab2536bddf8156d0c73e428ab4ea20de4075bb45b274569cb9068a9ad98"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T01:26:50.297127Z","signature_b64":"At/5MJKScdqtbICTQVGCGL0Sh7AVhilgWImPiPXh2YMEO+2rpdx8Lwpw3FRzsQWrE+mePnSAdThiwDM1J4zBDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4eaf105788945e9b718ba5adaf11b1e4ea2c55898a9617c3e7ccbafbd6ad9888","last_reissued_at":"2026-05-18T01:26:50.296422Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T01:26:50.296422Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Visualizing and Understanding Recurrent Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.NE"],"primary_cat":"cs.LG","authors_text":"Andrej Karpathy, Justin Johnson, Li Fei-Fei","submitted_at":"2015-06-05T22:33:04Z","abstract_excerpt":"Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1506.02078","kind":"arxiv","version":2},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"1506.02078","created_at":"2026-05-18T01:26:50.296524+00:00"},{"alias_kind":"arxiv_version","alias_value":"1506.02078v2","created_at":"2026-05-18T01:26:50.296524+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1506.02078","created_at":"2026-05-18T01:26:50.296524+00:00"},{"alias_kind":"pith_short_12","alias_value":"J2XRAV4ISRPJ","created_at":"2026-05-18T12:29:27.538025+00:00"},{"alias_kind":"pith_short_16","alias_value":"J2XRAV4ISRPJW4ML","created_at":"2026-05-18T12:29:27.538025+00:00"},{"alias_kind":"pith_short_8","alias_value":"J2XRAV4I","created_at":"2026-05-18T12:29:27.538025+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":5,"sample":[{"citing_arxiv_id":"1906.12284","citing_title":"Widening the Representation Bottleneck in Neural Machine Translation with Lexical Shortcuts","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"1907.00820","citing_title":"Understanding Memory Modules on Learning Simple Algorithms","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"1907.09728","citing_title":"Interpretable and Steerable Sequence Learning via Prototypes","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16318","citing_title":"Investigating Action Encodings in Recurrent Neural Networks in Reinforcement Learning","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12485","citing_title":"Letting the neural code speak: Automated characterization of monkey visual neurons through human language","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12485","citing_title":"Letting the neural code speak: Automated characterization of monkey visual neurons through human language","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2209.10652","citing_title":"Toy Models of Superposition","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22885","citing_title":"Federated Cross-Modal Retrieval with Missing Modalities via Semantic Routing and Adapter Personalization","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T","json":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T.json","graph_json":"https://pith.science/api/pith-number/J2XRAV4ISRPJW4MLUWW26ENR4T/graph.json","events_json":"https://pith.science/api/pith-number/J2XRAV4ISRPJW4MLUWW26ENR4T/events.json","paper":"https://pith.science/paper/J2XRAV4I"},"agent_actions":{"view_html":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T","download_json":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T.json","view_paper":"https://pith.science/paper/J2XRAV4I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1506.02078&json=true","fetch_graph":"https://pith.science/api/pith-number/J2XRAV4ISRPJW4MLUWW26ENR4T/graph.json","fetch_events":"https://pith.science/api/pith-number/J2XRAV4ISRPJW4MLUWW26ENR4T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T/action/storage_attestation","attest_author":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T/action/author_attestation","sign_citation":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T/action/citation_signature","submit_replication":"https://pith.science/pith/J2XRAV4ISRPJW4MLUWW26ENR4T/action/replication_record"}},"created_at":"2026-05-18T01:26:50.296524+00:00","updated_at":"2026-05-18T01:26:50.296524+00:00"}