Machine learning theorist at Harvard proving guarantees for optimization and learning under symmetry and invariance
3
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
5
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—
Not yet scored: the underlying data on this researcher's total publication output are insufficient in the current record. This axis would measure cumulative and per-year output volume, venue placement, and authorship roles across the full publication list.
Impact—
Not yet scored: citation-based evidence for this record is insufficient, so no impact value is reported. This axis would measure citation counts, h-index trajectory, and uptake of the work by other researchers over time.
Focus0/100
Scores 0 on focus, the lowest concentration band. The raw field distribution splits evenly (0.5 and 0.5) between two machine learning theory subfields — optimization under symmetry and geometric/invariant learning — and OpenAlex topics spread across five clusters, from stochastic gradient optimization techniques (11) to cryptography and data security (3). The work is deep but not yet concentrated in a single niche.
Collaboration39/100
Scores 39 on collaboration. The record lists 6 distinct coauthors over the last three years against 13 works in the same window, pointing to a small, stable circle rather than a broad network. Recurring partners include Melanie Weber, Parsa Moradi, Mohammad Ali Maddah-Ali, Ashkan Soleymani, Patrick Jaillet, and Stefanie Jegelka, spanning the current invariant-learning agenda and an earlier coded-computing line.
Activity77/100
Scores 77 on activity, the strongest axis in this profile. OpenAlex records 13 works in the last three years, and counts by year climb from 2 in 2023 and 2 in 2024 to 6 in 2025 and 5 already in 2026, including three 2026 arXiv preprints in the Evolve catalog — a clear recent acceleration in theoretical output.
Reproducibility38/100
Scores 38 on reproducibility, computed from 2 assessed papers with a mean readiness of 38. Both are theory-first: the sparse augmentation paper carries medium readiness but toy-only experiments, while the exact-invariance paper is rated low and reports no experiments at all. The guarantees are precise, yet empirical artifacts for independent verification are largely absent.
Research summary
Behrooz Tahmasebi is a machine learning theorist at Harvard University whose recorded work centers on provable guarantees for learning and optimization under symmetry. Bibliometric data from OpenAlex sketches an early-career, theory-first profile: h-index 5, i10-index 1, and 13 works in the last three years, with the field distribution split evenly (0.5/0.5) between optimization under symmetry and geometric or invariant learning. Topic counts are led by stochastic gradient optimization techniques (11) and machine learning and algorithms (8), with smaller threads in graph algorithms, metaheuristics, and cryptography. The Evolve catalog holds three 2026 arXiv papers. As first author, he proves that one-shot sparse data augmentation with O(log|G|/ε²) sampled transformations reaches ε-stationarity of the group-averaged loss, though the experiments are toy-scale. A second paper establishes polynomial-time exact-invariant regression for infinite groups together with provable symmetry discovery, using kernel regression, spectral averaging, and random Cayley and expander graphs — purely theoretical, without experiments. A third lays a learning-theoretic foundation for general coded computing in the straggler setting. Citation counts remain modest and the output and impact axes cannot yet be scored, consistent with a preprint-stage agenda that targets structural guarantees rather than benchmark performance.
Collaboration pattern
The record shows a compact collaboration footprint: 6 distinct coauthors over the last three years against 13 works in the same window, with recurring partners Melanie Weber, Parsa Moradi, Mohammad Ali Maddah-Ali, Ashkan Soleymani, Patrick Jaillet, and Stefanie Jegelka. In the 2026 catalog papers he is first author once and a middle author twice, suggesting he leads some projects while contributing to team efforts.
Research keywords
Highlights
One-shot sparse data augmentation with O(log|G|/ε²) stationarity guarantee
Polynomial-time exact-invariant learning for infinite groups
Learning-theoretic foundations for coded computing under stragglers
A third 2026 arXiv paper, with a middle authorship position, develops a learning-theoretic treatment of general coded computing in the straggler setting. It is not yet reviewed in the Evolve catalog, so no verdict, keywords, or reproducibility rating are available.
Timeline
- 2026Posts a first-author arXiv paper on sparse data augmentation for optimization with provable guarantees.
- 2026Co-authors an arXiv paper proving efficient learning and symmetry discovery under exact invariances.
- 2026Co-authors a learning-theoretic foundation for general coded computing in the straggler setting.
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, i10-index, counts by year, topics, coauthor counts) come from a single OpenAlex record and may not capture every publication or name variant.
- Output and impact dimensions are unscored because the underlying data are insufficient; citation counts and venue tiers are not quoted in this profile.
- All three catalog papers are 2026 arXiv preprints, so peer-review status and final publication venues are not reflected in this data.
- Profile is name-resolved from public bibliographic data (OpenAlex, ORCID 0000-0001-8544-6623); no biographical information beyond the provided record was used.