AI BriefingAtlas.scime.ai Maps 109k ML Papers into 1,260 Topic Pages with Benchmark Records
·2026.11.03 14:09
Key point
The free tool extracts results tables and ablations from 79k papers across ICML, NeurIPS, ICLR, CVPR, and AAAI since 2016 to track benchmark records and baselines.
Details
- Atlas.scime.ai processes approximately 109,000 accepted papers from ICML, NeurIPS, ICLR, CVPR, and AAAI since 2016, grouping them into roughly 1,260 topics.
- Each topic page displays best reported numbers per benchmark over time, identifying the specific paper that set each record.
- The platform extracts results tables and ablations from the full text of 79,000 papers, including negative results and recent arXiv preprints.
- Example data for compact/efficient Gaussian splatting shows growth from 29 papers in 2023-24 to 240 in 2025-26, with Deep Blending PSNR improving slightly from 30.42 to 30.45.
- The service is free and requires no signup, though extraction is imperfect and numbers from different papers are not always directly comparable.
- Users can view recent baselines used by papers, such as 3DGS, 2DGS, Scaffold-GS, and Mip-Splatting, and access BibTeX files for key papers.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Why LLM Research Agents Generalize Without Benchmark Overfitting: 'Compressibility' Is Key
MUSINSA develops DeciML and NeXA to move beyond prediction-centric personalization toward decision-making and discovery
Balestriero and LeCun Prove Platonic Representation Hypothesis Using JEPAs
Nathan Lambert argues Artificial Analysis v5 benchmarks favor closed models over open-weight alternativesArtificial Analysis Intelligence Index v5: Claude Opus 5.5 Leads as Open Weights Gap Widens to 12 Points