{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:2I2UMCCD677XIM42AAVHTW4BT5","short_pith_number":"pith:2I2UMCCD","schema_version":"1.0","canonical_sha256":"d235460843f7ff74339a002a79db819f77495d04071db325c4d03a1bb05c6f48","source":{"kind":"arxiv","id":"2104.08894","version":1},"attestation_state":"computed","paper":{"title":"The Intrinsic Dimension of Images and Its Impact on Learning","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.CV","authors_text":"Ahmed Abdelkader, Chen Zhu, Micah Goldblum, Phillip Pope, Tom Goldstein","submitted_at":"2021-04-18T16:29:23Z","abstract_excerpt":"It is widely believed that natural image data exhibits low-dimensional structure despite the high dimensionality of conventional pixel representations. This idea underlies a common intuition for the remarkable success of deep learning in computer vision. In this work, we apply dimension estimation tools to popular datasets and investigate the role of low-dimensional structure in deep learning. We find that common natural image datasets indeed have very low intrinsic dimension relative to the high number of pixels in the images. Additionally, we find that low dimensional datasets are easier for"},"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":"2104.08894","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2021-04-18T16:29:23Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"a7d7f1a00421db93de910e471e67060b79ac669a37a299cb9b8bef83239f787c","abstract_canon_sha256":"3efaabc9dbb010c896c7d4b7fc59d8118b7a66c4fec871b293c929ce6bddce13"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:32:56.571557Z","signature_b64":"HlIZGU2P0365oBy8T0Au2fKyiXWXJM45ckAFIl5hSymta8CrK38b20YC7U33y2xc9pmkBPQLO4/eq7LfYtF+Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d235460843f7ff74339a002a79db819f77495d04071db325c4d03a1bb05c6f48","last_reissued_at":"2026-07-05T02:32:56.571099Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:32:56.571099Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Intrinsic Dimension of Images and Its Impact on Learning","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.CV","authors_text":"Ahmed Abdelkader, Chen Zhu, Micah Goldblum, Phillip Pope, Tom Goldstein","submitted_at":"2021-04-18T16:29:23Z","abstract_excerpt":"It is widely believed that natural image data exhibits low-dimensional structure despite the high dimensionality of conventional pixel representations. This idea underlies a common intuition for the remarkable success of deep learning in computer vision. In this work, we apply dimension estimation tools to popular datasets and investigate the role of low-dimensional structure in deep learning. We find that common natural image datasets indeed have very low intrinsic dimension relative to the high number of pixels in the images. Additionally, we find that low dimensional datasets are easier for"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.08894","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2104.08894/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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":"2104.08894","created_at":"2026-07-05T02:32:56.571158+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.08894v1","created_at":"2026-07-05T02:32:56.571158+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.08894","created_at":"2026-07-05T02:32:56.571158+00:00"},{"alias_kind":"pith_short_12","alias_value":"2I2UMCCD677X","created_at":"2026-07-05T02:32:56.571158+00:00"},{"alias_kind":"pith_short_16","alias_value":"2I2UMCCD677XIM42","created_at":"2026-07-05T02:32:56.571158+00:00"},{"alias_kind":"pith_short_8","alias_value":"2I2UMCCD","created_at":"2026-07-05T02:32:56.571158+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23627","citing_title":"Diffusion Models Adapt to Low-Dimensional Structure Under Flexible Coefficient Choices","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21099","citing_title":"ShuffleFlow: Scalable Posterior Inference for Bayesian Inverse Imaging","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01987","citing_title":"Understanding Geometric Representations in Self-Supervised Vision Transformers via Subspace Intervention","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06882","citing_title":"Learning to Strategically Acquire Resources in Competition","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01443","citing_title":"UR-JEPA: Uniform Rectifiability as a Regularizer for Joint-Embedding Predictive Architectures","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10792","citing_title":"Implicit Neural Optimal Transport via Fixed-Point Optimization","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2407.08094","citing_title":"Density Estimation via Binless Multidimensional Integration","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2505.03205","citing_title":"Transformers for Learning on Noisy and Task-Level Manifolds: Approximation and Generalization Insights","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2602.00232","citing_title":"Complexity of Quantum Trajectories","ref_index":88,"is_internal_anchor":false},{"citing_arxiv_id":"2506.10959","citing_title":"Understanding In-Context Learning on Structured Manifolds: Bridging Attention to Kernel Methods","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2511.11934","citing_title":"A Systematic Analysis of Out-of-Distribution Detection Under Representation and Training Paradigm Shifts","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2510.01105","citing_title":"Geometric Analysis of Neural Regression Collapse via Intrinsic Dimension","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2602.03204","citing_title":"Sparsity is Combinatorial Depth: Quantifying MoE Expressivity via Tropical Geometry","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2602.20338","citing_title":"Emergent Manifold Separability during Reasoning in Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14274","citing_title":"CreFlow: Corrective Reflow for Sparse-Reward Embodied Video Diffusion RL","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13448","citing_title":"On the Limits of Latent Reuse in Diffusion Models","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26503","citing_title":"Delta Score Matters! Spatial Adaptive Multi Guidance in Diffusion Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10792","citing_title":"Implicit Neural Optimal Transport via Fixed-Point Optimization","ref_index":253,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5","json":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5.json","graph_json":"https://pith.science/api/pith-number/2I2UMCCD677XIM42AAVHTW4BT5/graph.json","events_json":"https://pith.science/api/pith-number/2I2UMCCD677XIM42AAVHTW4BT5/events.json","paper":"https://pith.science/paper/2I2UMCCD"},"agent_actions":{"view_html":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5","download_json":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5.json","view_paper":"https://pith.science/paper/2I2UMCCD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.08894&json=true","fetch_graph":"https://pith.science/api/pith-number/2I2UMCCD677XIM42AAVHTW4BT5/graph.json","fetch_events":"https://pith.science/api/pith-number/2I2UMCCD677XIM42AAVHTW4BT5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5/action/storage_attestation","attest_author":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5/action/author_attestation","sign_citation":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5/action/citation_signature","submit_replication":"https://pith.science/pith/2I2UMCCD677XIM42AAVHTW4BT5/action/replication_record"}},"created_at":"2026-07-05T02:32:56.571158+00:00","updated_at":"2026-07-05T02:32:56.571158+00:00"}