{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2012:JDUSXM5GCXG7CAJDDJ3IHZ3ZH2","short_pith_number":"pith:JDUSXM5G","schema_version":"1.0","canonical_sha256":"48e92bb3a615cdf101231a7683e7793eadcaabc805ad936c1f139fb2735d83c0","source":{"kind":"arxiv","id":"1211.5063","version":2},"attestation_state":"computed","paper":{"title":"On the difficulty of training Recurrent Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Razvan Pascanu, Tomas Mikolov, Yoshua Bengio","submitted_at":"2012-11-21T15:40:11Z","abstract_excerpt":"There are two widely known issues with properly training Recurrent Neural Networks, the vanishing and the exploding gradient problems detailed in Bengio et al. (1994). In this paper we attempt to improve the understanding of the underlying issues by exploring these problems from an analytical, a geometric and a dynamical systems perspective. Our analysis is used to justify a simple yet effective solution. We propose a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem. We validate empirically our hypothesis and proposed so"},"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":"1211.5063","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2012-11-21T15:40:11Z","cross_cats_sorted":[],"title_canon_sha256":"48e1617522fe5a2a1f95afd7d6c97967a8ae53d60bf710b28ee425e1405d5a07","abstract_canon_sha256":"7b839f3beb92e596f8dd2d6ac2b875fa170be63d39b3aa5602ce967034926136"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:33:25.437085Z","signature_b64":"jIwD+WOuhX7U+IdfXN4NAvn/HEWA2xEoWQruVj22ZFL9k87jRet66ZJboQWfHcB2k3mNpGoI0wjm9/uA0l+ZAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"48e92bb3a615cdf101231a7683e7793eadcaabc805ad936c1f139fb2735d83c0","last_reissued_at":"2026-05-18T03:33:25.436642Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:33:25.436642Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the difficulty of training Recurrent Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Razvan Pascanu, Tomas Mikolov, Yoshua Bengio","submitted_at":"2012-11-21T15:40:11Z","abstract_excerpt":"There are two widely known issues with properly training Recurrent Neural Networks, the vanishing and the exploding gradient problems detailed in Bengio et al. (1994). In this paper we attempt to improve the understanding of the underlying issues by exploring these problems from an analytical, a geometric and a dynamical systems perspective. Our analysis is used to justify a simple yet effective solution. We propose a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem. We validate empirically our hypothesis and proposed so"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1211.5063","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":"1211.5063","created_at":"2026-05-18T03:33:25.436707+00:00"},{"alias_kind":"arxiv_version","alias_value":"1211.5063v2","created_at":"2026-05-18T03:33:25.436707+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1211.5063","created_at":"2026-05-18T03:33:25.436707+00:00"},{"alias_kind":"pith_short_12","alias_value":"JDUSXM5GCXG7","created_at":"2026-05-18T12:27:11.947152+00:00"},{"alias_kind":"pith_short_16","alias_value":"JDUSXM5GCXG7CAJD","created_at":"2026-05-18T12:27:11.947152+00:00"},{"alias_kind":"pith_short_8","alias_value":"JDUSXM5G","created_at":"2026-05-18T12:27:11.947152+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":7,"sample":[{"citing_arxiv_id":"1906.10198","citing_title":"Multimodal and Multi-view Models for Emotion Recognition","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"1907.02908","citing_title":"On Inductive Biases in Deep Reinforcement Learning","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"1907.06286","citing_title":"Autoencoding sensory substitution","ref_index":208,"is_internal_anchor":true},{"citing_arxiv_id":"2404.03099","citing_title":"Composite Bayesian Optimization In Function Spaces Using NEON -- Neural Epistemic Operator Networks","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2003.00295","citing_title":"Adaptive Federated Optimization","ref_index":192,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20250","citing_title":"Physics-informed convolutional neural networks for fluid flow through porous media","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2604.16334","citing_title":"Preventing overfitting in deep learning using differential privacy","ref_index":30,"is_internal_anchor":true},{"citing_arxiv_id":"1609.08144","citing_title":"Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09022","citing_title":"Inferring identified hadron production in $pp$ collisions with physics-informed machine learning at the LHC","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08661","citing_title":"Geometry-Induced Long-Range Correlations in Recurrent Neural Network Quantum States","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06678","citing_title":"A Wasserstein GAN-based climate scenario generator for risk management and insurance: the case of soil subsidence","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2204.02311","citing_title":"PaLM: Scaling Language Modeling with Pathways","ref_index":108,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19625","citing_title":"Coherent-State Propagation: A Computational Framework for Simulating Bosonic Quantum Systems","ref_index":155,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2","json":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2.json","graph_json":"https://pith.science/api/pith-number/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/graph.json","events_json":"https://pith.science/api/pith-number/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/events.json","paper":"https://pith.science/paper/JDUSXM5G"},"agent_actions":{"view_html":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2","download_json":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2.json","view_paper":"https://pith.science/paper/JDUSXM5G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1211.5063&json=true","fetch_graph":"https://pith.science/api/pith-number/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/graph.json","fetch_events":"https://pith.science/api/pith-number/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/action/storage_attestation","attest_author":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/action/author_attestation","sign_citation":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/action/citation_signature","submit_replication":"https://pith.science/pith/JDUSXM5GCXG7CAJDDJ3IHZ3ZH2/action/replication_record"}},"created_at":"2026-05-18T03:33:25.436707+00:00","updated_at":"2026-05-18T03:33:25.436707+00:00"}