NYU researcher pairing LLM reasoning with machine-guided math discovery; earlier record rooted in quantum computing.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
39
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: the aggregated publication-volume records this axis needs are incomplete in the current snapshot. Output would measure total works and per-year production; the 32 recent three-year works are only a partial signal.
Impact—
Not yet scored: the citation-linked records this axis requires are incomplete in the snapshot. Impact would measure per-paper citations and field-weighted influence; the h-index (39) is a lifetime aggregate, not a computed impact score.
Focus0/100
The focus score is 0: the current field distribution is an exact 50/50 split between machine learning / NLP (LLM reasoning) and machine learning for automated mathematical discovery (formal theorem proving), so no single field dominates recent work. Historical topic tags do concentrate — quantum computing algorithms and architecture leads with 74 works and quantum information and cryptography follows at 49 — but the present program deliberately spans two adjacent directions.
Collaboration46/100
Collaboration scores 46, reflecting a compact network: 32 works over the last three years against only 9 distinct coauthors, a low coauthor-per-work ratio. Recurring partners include Ismail Labiad, Matthieu Kowalski, Marc Schoenauer, Rémi Munos, Niket Patel, Ahmad Rammal and Amaury Hayat, pointing to small stable teams on the recent LLM papers rather than a broad rotating cast.
Activity100/100
Activity reaches the maximum score of 100. The counts-by-year record shows a sharp recent acceleration — 1 work in 2020, 1 in 2021, 4 in 2022, 5 in 2023, 16 in 2024 (the record's highest annual count), 7 in 2025 and 9 already in 2026 — totalling 32 works over the last three years, after sparse years between 2013 and 2019.
Reproducibility38/100
The reproducibility score is 38 (reproMean), computed from 2 papers assessed so far. The two Evolve-reviewed works split unevenly: 'Beyond Repeated Sampling' rates medium readiness, with strong experimental controls but noted judge bias and training cost, while 'Learning to Discover Interesting Mathematics' rates low, with judge-based, variance-free evidence. Coverage is thin, so treat this as an early, provisional signal.
Research summary
Julia Kempe is a New York University researcher whose OpenAlex-derived record spans 1999 to 2026, with an h-index of 39 and an i10-index of 62. Historical topic mass concentrates in quantum computing algorithms and architecture (74 works) and quantum information and cryptography (49), with complexity in graphs (27) and computability, logic and AI algorithms (19) behind. The recent field distribution, however, is an even 50/50 split between machine learning / NLP (LLM reasoning) and machine learning for automated mathematical discovery (formal theorem proving). Output is high: 32 works in the last three years and 16 recorded in 2024, the largest annual figure in the record, for a maximal activity score of 100. Collaboration is compact — 9 distinct recent coauthors, with recurring partners including Ismail Labiad, Matthieu Kowalski, Marc Schoenauer and Rémi Munos. Two 2026 Evolve-reviewed papers, both with Kempe as last author, develop learned search policies for LLM reasoning and an interestingness-guided Lean theorem discovery loop; reproducibility readiness is modest (38, from two papers rated low and medium). Impact and aggregate output scores await fuller citation data.
Collaboration pattern
Collaboration is compact and team-based: 32 works over the last three years with only 9 distinct coauthors. The recurring roster — Ismail Labiad, Matthieu Kowalski, Marc Schoenauer, Rémi Munos, Niket Patel, Ahmad Rammal, Amaury Hayat — suggests small stable groups on the recent LLM papers, where Kempe appears as last author on both Evolve-reviewed entries. The provided list also contains the researcher's own name and an accent-variant duplicate, an entity-resolution artifact.
Research keywords
Highlights
Learned search policies that go beyond repeated sampling
An 'interestingness' objective for machine-led mathematical discovery
A sharp output surge in 2024
The counts-by-year record peaks at 16 works in 2024 — the largest annual figure in the snapshot — following sparse output from 2013 to 2019, and is followed by 7 works in 2025 and 9 so far in 2026, for 32 works across the last three years.
Timeline
- 1999First works appear in the bibliometric record, with 1 publication recorded this year.
- 2001Early peak: 14 publications recorded in a single year.
- 2020Return to steady publishing after a gap in which 2013 and 2015, and 2017 through 2019, have no recorded works; 1 work recorded this year.
- 2024Annual output peaks at 16 recorded works, the highest single-year figure in the record.
- 2026Two Evolve-reviewed papers appear, both with Kempe as last author: learned search policies for LLM reasoning and interestingness-guided Lean theorem discovery; 9 works recorded so far this year.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric figures come from the OpenAlex-derived snapshot only; the output and impact radar axes are unscored because the underlying aggregation and citation-linked records are incomplete.
- Identity is name-resolved from public bibliographic data (OpenAlex; ORCID 0009-0006-1984-0060). The record mixes historical quantum-computing topic tags with a recent LLM-focused field split, which may reflect a merged or evolving profile.
- The frequentCoauthors list contains the researcher's own name and an accent-variant duplicate (Rémi/Remi Munos), indicating entity-resolution noise.
- The reproducibility score (38) rests on only 2 assessed papers and should be read as provisional.