Patrick Jaillet

Massachusetts Institute of Technology · Russia

Unclaimed · AI-structured profileORCID OpenAlex

MIT theorist working across online learning, invariant ML, and nonsmooth robust control, on a record dating to 1982

papers here

reviewed by Evolve

first-author

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—

The output axis is unscored: the pipeline does not yet hold the per-work publication counts needed to quantify this researcher's total and per-year output volume. It would measure cumulative works, annual productivity, and the output trajectory across the indexed record; no score or estimate is given here.

Impact—

The impact axis is unscored: apart from aggregate indicators, per-work citation evidence is not yet available in this dataset. It would measure citations, citing works, and the downstream influence of each publication; no score or estimate is made here.

Focus0/100

Focus scores 0/100, reflecting an exceptionally broad span rather than concentration. The recent field distribution splits exactly into thirds across robust control and nonsmooth optimization, online learning and algorithmic game theory, and geometric/invariant machine-learning theory (0.333 each), while topic assignments spread across vehicle routing (83), optimization and search (77), bandit algorithms (68), graph complexity (66), and transportation planning (50); no single dominant topic emerges.

Collaboration32/100

Collaboration scores 32/100, on the low side. The recent window shows 42 works with only 4 distinct co-authors, and the recurring set is small and stable: Ashkan Soleymani, Gabriele Farina, Behrooz Tahmasebi, and Stefanie Jegelka. The 2026 catalog entries place Jaillet in last or middle author positions without corresponding authorship, consistent with a senior-collaborator role in compact theoretical teams rather than large consortium-style projects.

Activity100/100

Activity scores a perfect 100/100. The record shows 42 works in the last three years, with 18 indexed works in 2024, 11 in 2025, and 13 already in 2026 — sustaining productivity across a span OpenAlex indexes back to 1982, and peaking historically at 30 works in 2014 and 29 in 2021. Output in the 2020s remains near the career high-water mark.

Reproducibility43/100

Reproducibility scores 43/100, averaged over 3 assessed papers. All three 2026 preprints are purely theoretical with no experiments — Evolve notes 'rigorous novel theory, no experiments' for the exact-form regret paper and 'purely theoretical, no experiments' for the invariance paper — and reproducibility readiness is rated medium, low, and medium respectively, reflecting proofs that are checkable but not empirically demonstrable.

Research summary

Patrick Jaillet is a researcher at the Massachusetts Institute of Technology whose OpenAlex-resolved record spans more than four decades, with indexed works from 1982 through 2026, an h-index of 46, and an i10-index of 139. Activity is exceptional: the radar activity score is 100/100, with 42 works in the last three years and 18, 11, and 13 indexed works in 2024, 2025, and 2026 respectively. The record's topical center of gravity is optimization: vehicle routing (83 topic-assigned works), optimization and search (77), bandit algorithms (68), graph algorithms and complexity (66), and transportation planning (50). The Evolve-reviewed recent work concentrates on three theory threads that split the recent field distribution evenly: nonsmooth optimization for H-infinity output-feedback policy search, exact-form regret for gradient descent, mirror descent, and FTRL, and polynomial-time learning and symmetry discovery under exact group invariances. All three 2026 preprints are judged rigorous but purely theoretical, with low-to-medium reproducibility readiness (reproducibility score 43 across 3 assessed papers). A focus score of 0 and an exact three-way field split mark an unusually broad span; the collaboration score of 32 reflects 4 distinct recent co-authors, with Ashkan Soleymani, Gabriele Farina, Behrooz Tahmasebi, and Stefanie Jegelka the most frequent. The platform's output and impact axes are unscored because the underlying per-work data is incomplete.

Collaboration pattern

Jaillet's recent collaboration pattern is compact: 42 works over the last three years with just 4 distinct co-authors, and a stable recurring circle of Ashkan Soleymani, Gabriele Farina, Behrooz Tahmasebi, and Stefanie Jegelka. In the three Evolve-reviewed 2026 papers Jaillet holds last or middle author positions and is corresponding author on none, consistent with a senior collaborator contributing to focused two-to-four-person theory teams rather than large consortium projects.

Research keywords

online learningregret boundsmirror descentfollow-the-regularized-leader (FTRL)bandit algorithmsgroup invariancesymmetry discoverykernel regressionnonsmooth optimizationH-infinity robust controlvehicle routing optimizationgraph algorithms and complexity

Highlights

Exact-form regret: OGD, mirror descent and FTRL without approximation
This 2026 arXiv preprint (last author) shows that classical online-learning dynamics are no-regret exactly against exact-form deviations, while circulation yields a linear-regret obstruction; Evolve rates it rigorous novel theory with no experiments, at medium reproducibility readiness.
Polynomial-time exact-invariant learning and provable symmetry discovery
A 2026 preprint (middle author) proves poly-time exact-invariant regression for infinite groups and gives provable symmetry discovery, built on kernel regression, spectral averaging, and random Cayley and expander graph constructions; Evolve calls it rigorous but purely theoretical, with low reproducibility readiness.
Algorithmic optimality guarantees for nonsmooth H∞ output-feedback policy search
The third 2026 preprint (last author) establishes that ε-stationarity implies O(ε) suboptimality in nonsmooth H-infinity output-feedback policy search, via Clarke subdifferentials and Goldstein stationarity; Evolve judges the theory sound but notes the rates need exact-slice iterates, at medium reproducibility readiness.

Timeline

1982Earliest indexed work in the OpenAlex record (1 work), marking the start of a four-decade publication history.
2013Output accelerates sharply: 27 indexed works, opening a multi-year high plateau.
2014Peak annual output in the record: 30 indexed works, the highest single-year count.
202129 indexed works, the second-highest year on record, confirming sustained late-career productivity.
2026Three arXiv preprints enter the Evolve catalog, spanning online learning regret, exact-invariant learning, and nonsmooth H∞ control; 13 works indexed so far this year.

Papers in the Evolve catalog

Frequent co-authors

Ashkan Soleymani×3 Gabriele Farina×1 Behrooz Tahmasebi×1 Stefanie Jegelka×1

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, i10-index, counts by year, topics) come from OpenAlex only; per-work citation data was unavailable, so the output and impact radar axes are unscored.
The source record codes the affiliation country as 'RU' despite an MIT affiliation — likely a data artifact; this profile is name-resolved from public bibliographic data.
All three Evolve-reviewed papers are 2026 arXiv preprints; assessments cover preprint versions, not peer-reviewed publications.
The reproducibility score (43) averages only 3 assessed papers, all purely theoretical without experiments, so it says little about empirical artifact reuse.