Mila reinforcement learning researcher: dense reward shaping, plasticity theory, and agent-based clinical training
3
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
52
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—
Underlying output data is not yet available for scoring. This axis would measure total and per-year publication production, such as the 1996–2026 countsByYear series, benchmarked against field peers; no estimate is made here.
Impact—
Impact data is not yet sufficient for scoring. The axis would measure citation footprint and field-normalized influence; the record holds an unnormalized h-index of 52 and i10 of 160, but no score is assigned here.
Focus0/100
The focus score of 0 reflects near-total concentration: the OpenAlex field distribution splits 50/50 between two essentially identical labels, 'machine learning (reinforcement learning)' and 'reinforcement learning (machine learning)', with no other fields on record. Topic counts reinforce this: reinforcement learning in robotics leads at 299 works, ahead of AI-based planning (106), adaptive dynamic programming control (96), machine learning (80), and bandit algorithms (74).
Collaboration55/100
The collaboration score of 55 rests on 15 distinct coauthors in the last three years against 65 recent works — a compact core rather than a wide network. The frequent-coauthor list pairs reinforcement-learning collaborators (Xiao-Wen Chang, Jeremy Lucas, Shuyuan Zhang, Zihan Wang) with a separate cluster (Anita Slominska, Rennie Bimman, Elizabeth Di Flumeri, Amanda Mayappo-Neeposh, Conall Francoeur, Tamara Carver) alongside Esin Darici Haritaoglu.
Activity100/100
Activity scores 100, the maximum. The countsByYear series shows 65 works in the most recent three years, with annual output of 36 (2018), 40 (2019), 36 (2020), 34 (2021), 28 (2022), 21 (2023), 26 (2024), 20 (2025) and 19 indexed in 2026 — three decades after the first indexed work in 1996, the pipeline is still dense.
Reproducibility78/100
Reproducibility averages 78 across the two papers Evolve has assessed. The Bellman optimality equation for plasticity is rated high readiness, while G2QDR is medium — its reported gains over four backbones lack statistical testing, and the plasticity paper's convexity proof is flagged as flawed, both of which temper an otherwise strong openness signal.
Research summary
Doina Precup is a reinforcement learning researcher affiliated with Mila – Quebec Artificial Intelligence Institute (Canada), whose OpenAlex-resolved bibliographic record spans three decades, from a first indexed work in 1996 to an active 2026 pipeline. The record carries an h-index of 52 and 160 i10-indexed works, with a field distribution that is entirely reinforcement learning and machine learning (50/50 across two closely matching labels) — consistent with the radar's minimal focus dispersion. Topic masses are led by reinforcement learning in robotics (299 works), followed by AI-based problem solving and planning (106), adaptive dynamic programming control (96), machine learning and algorithms (80), and bandit algorithms (74). Activity remains high: 65 works in the last three years, with annual counts between 19 and 40 from 2018 onward and a peak of 40 in 2019. Evolve's catalog includes three 2026 papers: as last author, G2QDR — directed state-graph connectivity turned into dense auxiliary rewards for goal-conditioned hierarchical RL, with consistent gains over four backbones though no statistical testing — and a first Bellman optimality equation for directed-information plasticity, rated high for reproducibility readiness though its convexity proof is flagged as flawed; and as a middle author, SIC-Agents, an adaptive simulator benchmark for pediatric serious illness communication training. Reproducibility scoring averages 78 across the two assessed papers.
Collaboration pattern
The recent record shows a two-track pattern: a compact reinforcement-learning core — Xiao-Wen Chang, Jeremy Lucas, Shuyuan Zhang, Zihan Wang — alongside a larger applied group including Anita Slominska, Rennie Bimman, Elizabeth Di Flumeri, Amanda Mayappo-Neeposh, Conall Francoeur, Tamara Carver, and Esin Darici Haritaoglu, consistent with the pediatric communication-training project in the catalog. With 15 distinct coauthors against 65 works in three years, the network reads as selective rather than expansive.
Research keywords
Highlights
G2QDR: turning state-graph connectivity into dense rewards
A first Bellman optimality equation for plasticity
Agent-based simulator for pediatric serious-illness communication training
The 2026 SIC-Agents paper benchmarks and builds an adaptive simulator for pediatric serious illness communication training, with the researcher in a middle author position; Evolve has not yet published a verdict, keywords, or reproducibility rating for it.
Timeline
- 1996First work in the OpenAlex countsByYear record (1 publication) — the start of a three-decade publication trail.
- 2012An early output spike of 23 works, later surpassed by the sustained 2018–2021 stretch of 28–40 works per year.
- 2019Annual output peaks at 40 works, the highest single year in the 1996–2026 countsByYear series.
- 2026Three papers enter the Evolve catalog — G2QDR and the plasticity Bellman equation (last author) plus SIC-Agents (middle author); 19 works indexed 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 indicators (h-index 52, i10 160, countsByYear, topics, field distribution) come from an OpenAlex-resolved identity; the output and impact axes are unscored for lack of sufficient data.
- Only three 2026 catalog papers were reviewed; two carry Evolve verdicts, while SIC-Agents has no verdict, keywords, or reproducibility assessment yet.
- The reproducibility mean of 78 rests on just two assessed papers and should not be read as a career-wide measure.
- Beyond the stated Mila affiliation and country (CA), no biographical facts are covered by this data scope.