MIT theorist working on no-regret learning in games, nonsmooth H∞ policy optimization, and provable symmetry discovery
4
papers here
4
reviewed by Evolve
4
first-author
0
corresponding
—
works · OpenAlex
—
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—
Insufficient data to score. The snapshot contains no paper counts, venue list, or authorship-share statistics, so output volume cannot be computed; this axis would measure total publications, venue tiers, and first-author roles once the record covers them.
Impact—
Insufficient data to score. No citation counts, h-index, or venue-level influence metrics are present, so scholarly impact cannot be evaluated; this axis would measure citations and uptake of the four 2026 preprints as they accumulate.
Focus0/100
Focus scores 0/100 because the field distribution is perfectly flat: algorithmic game theory / online learning theory, robust control and nonsmooth optimization, online learning and algorithmic game theory, and machine learning theory (geometric / invariant learning) each carry 0.25 weight. Four equally weighted theory clusters rather than one concentrated lane, so the raw metric reads as maximal breadth of theoretical coverage.
Collaboration36/100
Collaboration scores 36/100, built on 5 distinct coauthors in the record — a small-team footprint typical of proof-driven theory work. All four catalog papers are first-authored and none is corresponding. The named frequent coauthors are Patrick Jaillet, Georgios Piliouras, Gabriele Farina, Behrooz Tahmasebi, and Stefanie Jegelka, indicating repeated partnerships across game-theoretic learning, self-play dynamics, and invariant-learning strands rather than a single large collaboration.
Activity—
Insufficient data to score. Beyond the four 2026 arXiv entries, the snapshot provides no per-year publication series or submission cadence, so research activity over time cannot be computed; this axis would measure recent output rate and its consistency.
Reproducibility58/100
Reproducibility scores 58/100, the mean over 4 reviewed papers (reproMean 58, reproPapers 4). Evolve readiness ratings run high, medium, low, and medium across the four preprints. Since every paper is purely theoretical with no experimental section, this score reflects proof-checkability and specification clarity — verdicts note missing experiments and, in one case, large constants — rather than released code or benchmark artifacts.
Research summary
Ashkan Soleymani is a Massachusetts Institute of Technology–affiliated researcher working at the intersection of learning theory, algorithmic game theory, and optimization. The reviewed catalog contains four 2026 arXiv preprints, all first-authored and purely theoretical. In game-theoretic learning, the MORM preprint shows that one-step optimism yields constant O(√n log d) regret in general-sum self-play, targeting coarse correlated equilibria. A companion paper develops exact-form regret analysis, proving that gradient descent, mirror descent, and follow-the-regularized-leader remain no-regret against exact-form deviations, with a linear-regret obstruction from circulation. Beyond games, a third preprint proves polynomial-time exact-invariant regression for infinite groups together with provable symmetry discovery, using spectral averaging and random Cayley and expander graphs. A fourth establishes algorithmic optimality guarantees for nonsmooth H∞ output-feedback policy search, showing that ε-stationarity implies O(ε) suboptimality via Clarke subdifferential and Goldstein stationarity arguments. Radar scores place collaboration at 36/100 (5 distinct coauthors), reproducibility at 58/100 averaged over 4 papers, and focus at 0/100 given a perfectly even 0.25 split across four theory clusters; output, impact, and activity axes lack sufficient data to score. Frequent coauthors include Patrick Jaillet, Georgios Piliouras, Gabriele Farina, Behrooz Tahmasebi, and Stefanie Jegelka. This is an early-stage, name-resolved profile built from public bibliographic data and Evolve's paper reviews; no citation or venue statistics are available yet.
Collaboration pattern
Collaboration is compact: 5 distinct coauthors and a 36/100 collaboration score fit a small-team, proof-driven theory pattern, with all four catalog papers first-authored and none corresponding. Recurring named partners are Patrick Jaillet, Georgios Piliouras, Gabriele Farina, Behrooz Tahmasebi, and Stefanie Jegelka, suggesting parallel strands in game-theoretic learning, self-play dynamics, and invariant learning.
Research keywords
Highlights
Constant regret in general-sum games via one-step optimism
Exact-form regret for gradient descent, mirror descent and FTRL
Efficient learning and symmetry discovery under exact invariances
Algorithmic optimality guarantees for nonsmooth H∞ output-feedback policy search
Timeline
- 2026Releases MORM, giving constant O(√n log d) regret in general-sum self-play with one-step optimism.
- 2026Publishes exact-form regret analysis showing OGD, mirror descent, and FTRL resist exact-form deviations.
- 2026Publishes polynomial-time exact-invariant regression and provable symmetry discovery for infinite groups.
- 2026Publishes optimality guarantees for nonsmooth H∞ output-feedback policy search, linking ε-stationarity to O(ε) suboptimality.
Papers in the Evolve catalog
- first author
Multiplicative Optimism for Constant Regret in Games
- first author
Exact-Form Regret for Gradient Descent, Mirror Descent and Follow-the-Regularized-Leader
- first author
Efficient Learning and Symmetry Discovery under Exact Invariances
- first author
Algorithmic Optimality Guarantees for Nonsmooth $H_\infty$ Output-Feedback Policy Search
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.
- Dimension scores come from the platform's bibliometric snapshot only; output, impact, and activity axes are unscored due to insufficient data.
- The reviewed catalog covers four 2026 arXiv preprints; no peer-reviewed venue, citation count, or publication-volume data is included.
- All four papers are purely theoretical without experimental sections; reproducibility reflects Evolve review readiness, not released code or benchmarks.
- This profile is name-resolved from public bibliographic data (ORCID 0009-0006-7278-9014); the name may be shared by multiple researchers.