Ranked by citations in public submissions · Snapshot: October 11, 2026
Each paper is counted once per citing submission. Ranked from 3,604 candidate works.
Loading trends…
Every rank refers to this downloaded snapshot of ICLR 2027 submissions. It is not a global citation count, an acceptance decision, or a measure of research quality.
The site uses 43,013 public submissions downloaded through the OpenReview API. These are the citing papers. Their reference sections contain about 1.75 million entries. The entries are not deduplicated: the same paper appears with different capitalization, arXiv and venue versions, and truncated titles.
Rankings therefore cover a candidate set of frequently cited works, not every cited paper. Candidates are arXiv IDs cited explicitly by at least 30 submissions, 2025–2026 arXiv IDs cited by at least 15, and frequent titles from an earlier catalog. Of 3,604 candidates checked, 3,034 appear on the site: the top 1,000 across all years plus up to 1,000 per release year. Rarely cited papers, and papers popular only within a small area, may be missing.
Top 1,000 across all release years and up to 1,000 for each available year. Years with fewer indexed papers show all available entries. Candidate coverage is incomplete, especially for older years. Papers are counted once per distinct citing submission using bibliography titles, reviewed title variants and arXiv identifiers.
All extracted author names, without a frequent-paper cutoff. Counts combine explicit names with coauthors recovered from cached arXiv author metadata. People are ranked by distinct citing submissions. A submission counts once for a person, even if it cites several of their papers.
Catalog candidates require matching both an author name and a cited paper. Separate profiles can be namesakes or duplicate records for one person. Catalog affiliations are source-reported publication metadata, not verified current employment. Candidate details show their papers, OpenAlex record and available ORCID links; unresolved matches keep the existing name-level count.
Affiliations are reported by linked sources and may be historical. Direct Scholar links identify researchers whose work appears in the record; they do not verify every citation or resolve namesakes. “Find on Scholar” opens a name search, not a verified profile. “Affiliation not added yet” means this site has no sourced entry for that name; it does not imply the person’s affiliation is unknown.
Titles and abstracts are OpenReview metadata that the authors entered at submission time. Title patterns and abstract words are matched case-insensitively where shown. Model, method, optimizer, precision, compute, benchmark, dataset and library charts count a paper once if a term appears in its body text, references excluded, across 42,877 extracted PDFs. Mentions include baselines, judges and related work. GPU counts use the largest GPU count stated in each paper.
AI tools are counted from each paper’s AI Use Statement, in sentences that describe a use; sentences that negate a use are ignored. The “Citing area” filter on the Papers tab ranks papers by the citing submissions in one primary area; papers cited by fewer than two submissions in that area are omitted. Citation charts use the 3,034 indexed works described above.
“Citations by publication year” uses all parsed reference entries instead. The year is the first year that follows each reference’s title in the bibliography; 18.5% of entries have no detectable year and are excluded. “Citations by area” assigns each indexed cited paper the most common primary area among the submissions citing it, or Cross-area when that area accounts for under 30% of them. Large areas win more often, so small areas are under-represented.
The two rankings have different matching methods and coverage. A paper’s title may be detected even when its author list cannot be parsed. Name variants can split a person, namesakes can merge, and “et al.” can omit coauthors. Full author credit makes large-team reports prominent. Curated individual reports may have more complete counts.
Includes self-citations and public withdrawn or desk-rejected submissions. The year filter is the cited paper’s first public release year where known, not its conference year. Two PDFs were unreadable. Source references and OpenReview links are available in the detail panel.
Public endpoint. No login or API token required.