{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:RQHZAYWWV6IWIKPHD7XLFKSSAP","short_pith_number":"pith:RQHZAYWW","schema_version":"1.0","canonical_sha256":"8c0f9062d6af916429e71feeb2aa5203da4f5109a78c81c141926122bf5f2fee","source":{"kind":"arxiv","id":"2407.14435","version":3},"attestation_state":"computed","paper":{"title":"Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"JumpReLU sparse autoencoders deliver higher reconstruction fidelity than Gated or TopK SAEs at matched sparsity on Gemma 2 activations.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arthur Conmy, J\\'anos Kram\\'ar, Neel Nanda, Nicolas Sonnerat, Senthooran Rajamanoharan, Tom Lieberum, Vikrant Varma","submitted_at":"2024-07-19T16:07:19Z","abstract_excerpt":"Sparse autoencoders (SAEs) are a promising unsupervised approach for identifying causally relevant and interpretable linear features in a language model's (LM) activations. To be useful for downstream tasks, SAEs need to decompose LM activations faithfully; yet to be interpretable the decomposition must be sparse -- two objectives that are in tension. In this paper, we introduce JumpReLU SAEs, which achieve state-of-the-art reconstruction fidelity at a given sparsity level on Gemma 2 9B activations, compared to other recent advances such as Gated and TopK SAEs. We also show that this improveme"},"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":true},"canonical_record":{"source":{"id":"2407.14435","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-07-19T16:07:19Z","cross_cats_sorted":[],"title_canon_sha256":"ddba3bc248051a0a77b174db45bc08eacc7bda1175ce1108901b70d6287f634e","abstract_canon_sha256":"b50802e452220cb41cc41233ea1ba2f2115e3db7957ec175d305f65ac0749979"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:50.469156Z","signature_b64":"aJabJsw1FbP621eIm3t/OixhR8HgzTJx4Qf5VkPe8yrfKTJGfjkwWjdxiYBxfIp0oPTuzunkxp62HCfl4XzhAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8c0f9062d6af916429e71feeb2aa5203da4f5109a78c81c141926122bf5f2fee","last_reissued_at":"2026-05-17T23:38:50.468390Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:50.468390Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"JumpReLU sparse autoencoders deliver higher reconstruction fidelity than Gated or TopK SAEs at matched sparsity on Gemma 2 activations.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arthur Conmy, J\\'anos Kram\\'ar, Neel Nanda, Nicolas Sonnerat, Senthooran Rajamanoharan, Tom Lieberum, Vikrant Varma","submitted_at":"2024-07-19T16:07:19Z","abstract_excerpt":"Sparse autoencoders (SAEs) are a promising unsupervised approach for identifying causally relevant and interpretable linear features in a language model's (LM) activations. To be useful for downstream tasks, SAEs need to decompose LM activations faithfully; yet to be interpretable the decomposition must be sparse -- two objectives that are in tension. In this paper, we introduce JumpReLU SAEs, which achieve state-of-the-art reconstruction fidelity at a given sparsity level on Gemma 2 9B activations, compared to other recent advances such as Gated and TopK SAEs. We also show that this improveme"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"JumpReLU SAEs achieve state-of-the-art reconstruction fidelity at a given sparsity level on Gemma 2 9B activations, compared to other recent advances such as Gated and TopK SAEs, and this improvement does not come at the cost of interpretability.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That straight-through estimators applied to the discontinuous JumpReLU produce gradients that reliably optimize the intended sparse reconstruction objective rather than converging to a different local minimum.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"JumpReLU SAEs deliver higher reconstruction fidelity than Gated or TopK SAEs at fixed sparsity on Gemma 2 9B activations while preserving interpretability.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"JumpReLU sparse autoencoders deliver higher reconstruction fidelity than Gated or TopK SAEs at matched sparsity on Gemma 2 activations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f331e7f277e79dc657ba9058d62252d6fcd7671a063114430ac775d755258305"},"source":{"id":"2407.14435","kind":"arxiv","version":3},"verdict":{"id":"0db2304a-1b55-49b5-894e-af89d365e568","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T19:06:02.397883Z","strongest_claim":"JumpReLU SAEs achieve state-of-the-art reconstruction fidelity at a given sparsity level on Gemma 2 9B activations, compared to other recent advances such as Gated and TopK SAEs, and this improvement does not come at the cost of interpretability.","one_line_summary":"JumpReLU SAEs deliver higher reconstruction fidelity than Gated or TopK SAEs at fixed sparsity on Gemma 2 9B activations while preserving interpretability.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That straight-through estimators applied to the discontinuous JumpReLU produce gradients that reliably optimize the intended sparse reconstruction objective rather than converging to a different local minimum.","pith_extraction_headline":"JumpReLU sparse autoencoders deliver higher reconstruction fidelity than Gated or TopK SAEs at matched sparsity on Gemma 2 activations."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"22e9ca61f233f2b0009b65db182ea2ec5ab6674f1046ac505fe93e30d7bfd138"},"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":"2407.14435","created_at":"2026-05-17T23:38:50.468680+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.14435v3","created_at":"2026-05-17T23:38:50.468680+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.14435","created_at":"2026-05-17T23:38:50.468680+00:00"},{"alias_kind":"pith_short_12","alias_value":"RQHZAYWWV6IW","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"RQHZAYWWV6IWIKPH","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"RQHZAYWW","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":29,"internal_anchor_count":29,"sample":[{"citing_arxiv_id":"2605.21849","citing_title":"Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16372","citing_title":"SwordBench: Evaluating Orthogonality of Steering Image Representations","ref_index":42,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18882","citing_title":"To Call or Not to Call: Diagnosing Intrinsic Over-Calling Bias in LLM Agents","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2509.03738","citing_title":"Mechanistic Interpretability with Sparse Autoencoder Neural Operators","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2509.18127","citing_title":"Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2511.01680","citing_title":"Making Interpretable Discoveries from Unstructured Data: A High-Dimensional Multiple Hypothesis Testing Approach","ref_index":42,"is_internal_anchor":true},{"citing_arxiv_id":"2601.14004","citing_title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","ref_index":253,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12055","citing_title":"Do Language Models Encode Knowledge of Linguistic Constraint Violations?","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.04946","citing_title":"Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12770","citing_title":"WriteSAE: Sparse Autoencoders for Recurrent State","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2403.19647","citing_title":"Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12055","citing_title":"Do Language Models Encode Knowledge of Linguistic Constraint Violations?","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08611","citing_title":"The Echo Amplifies the Knowledge: Somatic Marker Analogues in Language Models via Emotion Vector Re-Injection","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07922","citing_title":"Tree SAE: Learning Hierarchical Feature Structures in Sparse Autoencoders","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09438","citing_title":"fmxcoders: Factorized Masked Crosscoders for Cross-Layer Feature Discovery","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08740","citing_title":"Causal Dimensionality of Transformer Representations: Measurement, Scaling, and Layer Structure","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06494","citing_title":"From Token Lists to Graph Motifs: Weisfeiler-Lehman Analysis of Sparse Autoencoder Features","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06610","citing_title":"SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01829","citing_title":"GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13304","citing_title":"Can Cross-Layer Transcoders Replace Vision Transformer Activations? An Interpretable Perspective on Vision","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08846","citing_title":"Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs","ref_index":86,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06610","citing_title":"SoftSAE: Dynamic Top-K Selection for Adaptive Sparse Autoencoders","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07922","citing_title":"Tree SAE: Learning Hierarchical Feature Structures in Sparse Autoencoders","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07990","citing_title":"Tool Calling is Linearly Readable and Steerable in Language Models","ref_index":67,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP","json":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP.json","graph_json":"https://pith.science/api/pith-number/RQHZAYWWV6IWIKPHD7XLFKSSAP/graph.json","events_json":"https://pith.science/api/pith-number/RQHZAYWWV6IWIKPHD7XLFKSSAP/events.json","paper":"https://pith.science/paper/RQHZAYWW"},"agent_actions":{"view_html":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP","download_json":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP.json","view_paper":"https://pith.science/paper/RQHZAYWW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.14435&json=true","fetch_graph":"https://pith.science/api/pith-number/RQHZAYWWV6IWIKPHD7XLFKSSAP/graph.json","fetch_events":"https://pith.science/api/pith-number/RQHZAYWWV6IWIKPHD7XLFKSSAP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP/action/storage_attestation","attest_author":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP/action/author_attestation","sign_citation":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP/action/citation_signature","submit_replication":"https://pith.science/pith/RQHZAYWWV6IWIKPHD7XLFKSSAP/action/replication_record"}},"created_at":"2026-05-17T23:38:50.468680+00:00","updated_at":"2026-05-17T23:38:50.468680+00:00"}