Efe Mert Karag\"ozl\"u
Carnegie Mellon University · United States
CMU researcher combining FPGA-accelerated ML for tokamak plasma control with LLM-augmented Bayesian optimization
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
Output is not yet scored: this snapshot lacks the publication-volume record the axis requires. It would measure cumulative works and yearly productivity; the indexed counts (1 in 2025, 9 in 2026) hint at a short, fast-growing record.
Impact—
Impact is not yet scored: citation evidence is too thin in this snapshot (h-index 1, i10-index 0). The axis would measure citation uptake, influence on follow-up work, and recognition within the fusion-ML community.
Focus0/100
The focus score computes to 0, the floor of the scale, and the raw distribution explains why: OpenAlex splits this record evenly at 0.5 for plasma physics / machine learning systems (fusion control) and 0.5 for machine learning. Topic counts reinforce the breadth — magnetic confinement fusion (8), fusion and plasma physics studies (7), fusion materials (4), reinforcement learning in robotics (2), and multi-objective optimization (2) — so no single field yet dominates.
Collaboration53/100
Collaboration scores 53. The snapshot lists 13 distinct coauthors across 10 recent works, and both Evolve-reviewed papers sit in middle, non-corresponding authorship, indicating contribution within larger teams. The recurring partner roster spans both research threads — Barnabás Póczos, Jeff Schneider (also logged as 'Jeff Schnieder') and Tejus Gupta on the machine-learning side; Cong Hao, Daniel Gaytan-Villarreal, Aiken Xie, Tu Pham, Chiara Amendola, Matteo Cremonesi and Rohit Sonker on the fusion and hardware side.
Activity70/100
Activity scores 70 on a steep recent ramp: indexed output rose from 1 work in 2025 to 9 in 2026, totalling 10 works in the trailing three-year window. Both Evolve-reviewed entries are 2026 arXiv preprints, so the current burst is preprint-stage; sustained output beyond 2026 has not yet been observed.
Reproducibility38/100
Reproducibility scores 38, the mean readiness across the 2 papers Evolve has reviewed. The FPGA plasma-state prediction paper rates low: its 94.9% compression and 2.64 μs latency figures rest on timing simulation, with closed-loop performance degrading. The LLM-augmented Bayesian optimization paper rates medium, held back by unreported LLM identity, hyperparameters, and full-grid results.
Research summary
Efe Mert Karagözlü is a Carnegie Mellon University researcher working at the junction of machine learning and fusion science; OpenAlex splits the record evenly, 0.5 to plasma physics / machine learning systems (fusion control) and 0.5 to machine learning. The bibliometric record is early-stage: h-index 1, i10-index 0, 10 works in the trailing three-year window rising from 1 indexed work in 2025 to 9 in 2026, and 13 distinct coauthors. Topic counts concentrate on magnetic confinement fusion research (8), fusion and plasma physics studies (7), and fusion materials and technologies (4), with smaller strands in reinforcement learning for robotics (2) and multi-objective optimization (2). The Evolve catalog reviews two 2026 arXiv preprints, both in middle, non-corresponding authorship. The first compresses a DIII-D plasma recurrent probabilistic neural network by 94.9% for FPGA deployment, reaching 2.64 μs latency through a cautious deployment workflow, though hardware figures rest on timing simulation and closed-loop performance degrades. The second uses LLM-elicited priors and auxiliary observations to improve Bayesian optimization on HPOBench and fusion tasks. Reproducibility across the two reviews averages 38; output and impact dimensions remain unscored for lack of sufficient citation data, so the profile reads best as an early-career, fast-moving snapshot.
Collaboration pattern
The OpenAlex snapshot lists 13 distinct coauthors across 10 recent works (collaboration score 53), and both reviewed papers place Karagözlü in a middle, non-corresponding position within larger teams. The recurring roster spans both threads — machine learning: Barnabás Póczos, Jeff Schneider (also logged as 'Jeff Schnieder'), Tejus Gupta; fusion and hardware: Cong Hao, Daniel Gaytan-Villarreal, Aiken Xie, Tu Pham, Chiara Amendola, Matteo Cremonesi, Rohit Sonker. Coauthors are not mapped to individual papers in the snapshot.
Research keywords
Highlights
FPGA deployment of a DIII-D plasma RPNN at 2.64 μs latency
LLM-elicited priors and auxiliary observations for Bayesian optimization
Timeline
- 2025First indexed output in the OpenAlex record: 1 work appears.
- 2026Indexed output accelerates to 9 works — the most active year on record.
- 2026Two arXiv preprints enter the Evolve catalog — FPGA-accelerated plasma-state prediction and LLM-augmented Bayesian optimization — both in middle, non-corresponding authorship.
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.
- Bibliometrics are computed from OpenAlex only (identity: openalex); no ORCID is linked, so name resolution may merge or split records.
- Output and impact axes are unscored because citation data are insufficient; h-index 1 and i10-index 0 reflect a very early-career record.
- Both cataloged papers are arXiv preprints with middle, non-corresponding authorship; no peer-reviewed venue data are provided.
- Reproducibility aggregates only the 2 reviewed papers (mean 38) and may shift as catalog coverage grows.