MIT researcher bridging generative modeling and safe reinforcement learning through rigorous optimization theory.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
12
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 metrics are not present in this snapshot, so no score is given. This axis would measure career-wide and per-year publication counts and how they are distributed across venues, complementing the citation-based view of the record.
Impact—
Citation impact data is unavailable in the provided record, so the axis is left unscored. It would measure total citations, citation velocity, and uptake of the work beyond its immediate subfields.
Focus0/100
The focus score of 0 reflects a perfectly even split: the recent portfolio divides exactly 0.5 and 0.5 between machine learning (generative modeling) and safe reinforcement learning and control, while topic tags spread across five clusters led by stochastic gradient optimization (19 works) and domain adaptation and few-shot learning (14). The breadth is structural rather than drift.
Collaboration32/100
Score 32 aligns with a compact recent footprint: 4 distinct coauthors in the last three years of OpenAlex-indexed work, and a recurring partner set of four — Zeyang Li, Yunan Wang, Paolo Giaretta, and Sunbochen Tang. Both 2026 reviewed preprints place him as last author, consistent with small theory-oriented team supervision rather than a wide network.
Activity79/100
Score 79 rests on 14 works in the recent three-year window and a year-by-year record that climbs from 2 works (2015) to a peak of 10 (2023), then 6 (2024), 5 (2025), and 3 recorded so far in 2026 — a sustained and currently active publishing cadence.
Reproducibility55/100
The mean of 55 covers the 2 papers Evolve has reviewed, both rated medium readiness. Verdicts attribute the ceiling to evidence depth rather than tooling: the safe meta-RL study runs two HalfCheetah tasks with 3 seeds and thin baselines, and Newton Matching presents guarantees without empirical results shown.
Research summary
Navid Azizan is a researcher at the Massachusetts Institute of Technology whose work spans two strands captured in the provided bibliometric record: machine learning for generative modeling and safe reinforcement learning and control, each accounting for exactly half of the recent field distribution. The OpenAlex snapshot reports an h-index of 12 and an i10-index of 15, with 14 works in the most recent three-year window and a year-by-year record rising from 2 publications in 2015 to a peak of 10 in 2023, followed by 6 (2024), 5 (2025), and 3 recorded so far in 2026. Topic clustering places stochastic gradient optimization first with 19 tagged works, ahead of domain adaptation and few-shot learning (14), adversarial robustness (8), neural networks and applications (7), and machine learning and data classification (7). Evolve has reviewed two 2026 arXiv preprints, both with Azizan in the last author position: Safe Meta-Reinforcement Learning via Information Space Reachability, credited for clean contraction theory on belief-state safety but tested on only two HalfCheetah tasks with 3 seeds, and Newton Matching, a unified fine-tuning and sampling framework for diffusion and flow models with KL-descent and quadratic-convergence guarantees but no empirical results shown. The overall pattern is theory-first: rigorous guarantees, a small and stable collaboration circle, and empirical validation still to be deepened.
Collaboration pattern
The recent collaboration footprint is compact: 4 distinct coauthors across the last three years, and a recurring partner set of four — Zeyang Li, Yunan Wang, Paolo Giaretta, and Sunbochen Tang. Both 2026 Evolve-cataloged preprints carry him as last author without corresponding-author designation, consistent with a senior-supervision pattern on small, theory-heavy teams rather than large multi-institution consortia.
Research keywords
Highlights
Belief-space reachability for safe meta-reinforcement learning
Newton Matching unifies fine-tuning and sampling for generative models
Timeline
- 2015The OpenAlex record opens with 2 works, the earliest activity captured in this profile.
- 2023Annual output peaks at 10 works, the highest single-year count in the provided record.
- 20255 works recorded, keeping the publishing cadence steady into the mid-2020s.
- 2026Two new arXiv preprints enter the Evolve catalog — Safe Meta-Reinforcement Learning via Information Space Reachability and Newton Matching — both with Azizan in the last author position.
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 12, i10-index 15, yearly counts, topics, coauthor counts) come from an OpenAlex-indexed snapshot and may lag or omit some venues.
- The output and impact radar axes are unscored because their underlying metrics were absent from the provided data.
- Qualitative verdicts cover only the 2 papers Evolve has reviewed; all other works are represented by bibliometric metadata alone.
- Identity is name-resolved from public bibliographic data and anchored to ORCID 0000-0002-4299-2963; homonyms cannot be fully excluded.