Sequential-decision ML researcher at NUS: bandits, Bayesian optimization, and LLM agents and post-training
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
17
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Insufficient data: no publication-volume score is available in this snapshot. This axis would measure overall research productivity — annual paper and preprint counts relative to field norms — and cannot be estimated from the current record.
Impact—
Insufficient data: no citation-impact score is available. This axis would measure scholarly influence — total citations, per-paper citation trajectories, and uptake by other researchers — and no citation counts were provided.
Focus100/100
Scores 100/100 for topical focus. The OpenAlex topic profile is tightly concentrated: Privacy-Preserving Technologies in Data (47 works), Advanced Bandit Algorithms Research (40), Topic Modeling (36), Adversarial Robustness in Machine Learning (25), and Machine Learning and Data Classification (25) — one coherent sequential-decision-and-learning core now applied to LLM agents and post-training.
Collaboration61/100
Scores 61/100. The OpenAlex record lists 20 distinct recent coauthors, and 12 recurring collaborators appear alongside the reviewed papers — including Zhiliang Chen, Sebastian Ament, David Eriksson, Maximilian Balandat, and Eytan Bakshy. Author placement on the three catalog papers (middle, last, middle; none corresponding) fits a senior contributor working within larger teams.
Activity100/100
Scores 100/100, the profile's strongest signal alongside focus. Recent output is high — 85 works in the last three years — and the OpenAlex annual series climbs steadily from 1 paper in 2013 and 7 in 2019 to 21 in 2021, 13 in 2022, 30 in 2024, and 35 in 2026, indicating sustained, accelerating productivity.
Reproducibility55/100
Scores 55/100 (reproMean), though this rests on a very small base — only one reproducibility-assessed paper (reproPapers: 1). Evolve's per-paper ratings echo the mixed picture: high readiness for the SFT-RL budget-allocation study, but medium for COBRA-Skills and PLES, with single-seed results, missing significance tests, and untested baselines flagged.
Research summary
Bryan Kian Hsiang Low is a researcher at the National University of Singapore whose work, as recorded in OpenAlex and the Evolve catalog, centers on sequential decision-making and machine learning: contextual and neural bandits, Bayesian optimization, and, most recently, LLM agents and post-training. His topic profile is highly concentrated — Privacy-Preserving Technologies in Data (47 works), Advanced Bandit Algorithms Research (40), Topic Modeling (36), Adversarial Robustness (25), and ML Classification (25). Activity is the standout axis: 85 works in the last three years, with the OpenAlex annual series rising from 1 (2013) to 21 (2021), 30 (2024), and 35 (2026). The Evolve-reviewed 2026 papers apply this decision-theoretic toolkit to modern LLM systems: COBRA-Skills uses contextual bandit-guided evolution for agent skill optimization (reported 55-58% lower cost than SkillOpt averages), a budget-allocation study finds near-optimal SFT-RL regions that transfer from small models up to 14B scale, and PLES accelerates hyperparameter scaling-law estimation through power-law entropy search. Bibliometric impact indicators in the indexed record are modest (h-index 17, i10-index 23), and no output or impact scores are available. Collaboration is broad — 20 distinct recent coauthors, 12 recurring. Reproducibility is rated 55/100 on a single assessed paper, with per-paper readiness ranging from medium to high.
Collaboration pattern
The OpenAlex record shows 20 distinct recent coauthors (collaboration dimension 61/100). Twelve recurring collaborators accompany the reviewed work, including Zhiliang Chen, Sebastian Ament, David Eriksson, Maximilian Balandat, Eytan Bakshy, and Jihao Andreas Lin. Across the three catalog papers Low appears in middle and last author positions (none corresponding), consistent with a senior contributor embedded in larger machine learning teams spanning bandits, Bayesian optimization, and LLM post-training.
Research keywords
Highlights
COBRA-Skills: bandit-guided evolution for agent skill optimization
Near-optimal SFT-RL budget allocation transfers from small to large LLMs
Power-Law Entropy Search cuts scaling-law estimation cost
Timeline
- 2026COBRA-Skills: contextual bandit-guided evolution for agent skill optimization (arXiv; middle author).
- 2026Scaling near-optimal SFT-RL annotation budget allocation study (arXiv; last author).
- 2026Power-Law Entropy Search (PLES) for hyperparameter scaling-law estimation (arXiv; middle author).
- 2026OpenAlex-recorded annual output reaches 35 works, the highest in the available 2013-2026 series.
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 (h-index 17, i10-index 23, annual counts, topics) come from an OpenAlex snapshot only and may not reflect the complete publication record.
- No output or impact scores exist (radar dimMissing); no citation counts were provided, so neither axis is scored or estimated.
- The reproducibility score (55/100) rests on a single assessed paper and is a small-sample estimate.
- Profile is name-resolved from public bibliographic data (OpenAlex, ORCID 0000-0003-2808-451X); Evolve catalog coverage is limited to three reviewed 2026 arXiv papers.