Esther Mondragón

City, University of London · United Kingdom

Unclaimed · AI-structured profileORCID OpenAlex

Groupoid-based reinforcement learning at City, University of London, after a long memory-and-neuroscience record

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) 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 is not yet scored: this axis would measure total publication volume and productivity over time, but the complete output record behind this profile is unavailable in the current snapshot, so no estimate is offered.

Impact—

Impact is not yet scored: this axis measures citation-based influence such as per-work citations and field-weighted impact; the snapshot carries an h-index of 13 and i10-index of 14, but fuller citation data is missing.

Focus0/100

Focus scores 0, the bottom of the scale, reflecting an unusually dispersed record: OpenAlex topics span Memory and Neural Mechanisms (33 works), Neural and Behavioral Psychology Studies (30), Reinforcement Learning in Robotics (13), Neurotransmitter Receptor Influence on Behavior (11) and Neural Networks and Applications (5), while the recent field distribution splits evenly (0.5/0.5) between reinforcement learning and machine learning (reinforcement learning).

Collaboration32/100

Collaboration scores 32, a modest value consistent with a compact working pattern: the recent window lists only 4 distinct coauthors, and the recurring partners are Ben Opperman and Eduardo Alonso. Both catalogued 2026 papers place Mondragón in the last author position without the corresponding-author role, suggesting a senior but not lead-corresponding role within a small, stable reinforcement-learning team rather than a broad collaborative network.

Activity79/100

Activity scores 79, the strongest axis in the record: 14 works in the last three years, with the yearly curve rising from 3 recorded works in 2024 to 4 in 2025 and 7 in 2026 — the highest single-year count since the record began in 1996. This exceeds the previous peak of 6 works in 2012, indicating a clear late-period acceleration.

Reproducibility55/100

Reproducibility scores 55 (reproMean 55 across 2 papers carrying reproducibility assessments). Both catalogued papers are rated medium readiness: Evolve notes the groupoid orbit-quotient evaluation lacks baselines, seeds and statistical rigor, and the state-abstraction study's evaluation is described as thin — setups repeatable in principle, but under-controlled experiments.

Research summary

Esther Mondragón is a researcher at City, University of London whose OpenAlex record shows two distinct research programmes. The long-running strand is behavioural neuroscience: Memory and Neural Mechanisms (33 works) and Neural and Behavioral Psychology Studies (30 works) dominate the topic history, alongside Neurotransmitter Receptor Influence on Behavior (11) and Neural Networks and Applications (5). The active strand is reinforcement learning: recent field attribution splits evenly between reinforcement learning and machine learning (reinforcement learning), and Reinforcement Learning in Robotics accounts for 13 topical works. The record carries an h-index of 13 and an i10-index of 14, with 14 works in the last three years and an annual curve climbing from 3 works in 2024 to 7 in 2026 — the highest single-year count since recording began in 1996. Activity scores 79 while focus scores 0, reflecting this two-programme spread. Evolve reviewed two 2026 arXiv papers on groupoid-based reinforcement learning, in both of which Mondragón is last author and not corresponding: a categorical internalisation framework whose orbit-quotient Q-learning beats POMCP on RockSample and Tag, and a groupoid state abstraction that beats Q-learning in dense and large gridworlds. Both are judged promising symmetry-sharing ideas whose evaluation remains thin; reproducibility scores 55 across the two assessments. Collaboration is compact: 4 distinct recent coauthors, chiefly Ben Opperman and Eduardo Alonso.

Collaboration pattern

Collaboration is compact and stable: the recent window lists 4 distinct coauthors, and the recurring names are Ben Opperman and Eduardo Alonso. On both 2026 catalogued papers Mondragón sits in the last author position without being the corresponding author, consistent with a senior supervisory role inside a small reinforcement-learning team at City, University of London. The collaboration dimension scores 32, and no large cross-institution network is visible in the available data.

Research keywords

reinforcement learningPOMDPgroupoidcategory theorylocal symmetrystate abstractioncanonicalisationQ-learninggeneralisationmemory and neural mechanismsneural and behavioral psychologyneural networks and applications

Highlights

Groupoid orbit-quotient Q-learning beats POMCP on RockSample and Tag
The 2026 arXiv paper 'Categorical Internalisation of Environmental Groupoids for Generalisable POMDP Solving' introduces a categorical, symmetry-sharing internalisation of environment groupoids; Evolve's review finds the orbit-quotient Q-learning agent outperforms POMCP on RockSample and Tag, but flags evaluation that lacks baselines, seeds and statistical rigor (reproReadiness: medium).
Groupoid-based state abstraction exploits local symmetries in reinforcement learning
In 'Groupoid-Based Internal State Representations for Reinforcement Learning with Local Symmetries' (2026, arXiv), a groupoid-based internal state representation exploits local symmetries as a state abstraction and beats Q-learning in dense and large gridworlds; Evolve judges the idea promising but the evaluation thin (reproReadiness: medium).
Record-wide output surge: 7 works in 2026, the highest single year since 1996
The OpenAlex counts-by-year series rises from 3 works in 2024 and 4 in 2025 to 7 in 2026, overtaking the previous peak of 6 works in 2012; the two groupoid-based reinforcement learning papers reviewed by Evolve fall within this surge, against a record-wide h-index of 13 and i10-index of 14.

Timeline

1996Earliest recorded publication year in the OpenAlex record (1 work), opening a three-decade publication history.
2004An early cluster of 4 recorded works, the highest annual output of the record's first decade.
20126 recorded works — the strongest year of the middle period, surpassed only in 2026.
2021Output resumes after a sparse stretch (1 work in 2019), with 1 recorded work beginning a renewed run that continues through 2022 and 2023.
20267 recorded works, the highest single-year count since 1996, including the two groupoid-based reinforcement learning papers reviewed by Evolve.

Papers in the Evolve catalog

Frequent co-authors

Ben Opperman×1 Eduardo Alonso×1 Ben Opperman×1 Eduardo Alonso×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 are derived from OpenAlex only; identity is name-resolved from public bibliographic data, and the record combines a neuroscience-dominated topic history with recent reinforcement-learning output, so attribution of the long-tail record to the same person should be treated with care.
Output and impact dimensions are unscored (insufficient data); the citation evidence in this snapshot is limited to h-index 13 and i10-index 14.
Evolve's qualitative verdicts cover only the two 2026 arXiv preprints, which were not peer-reviewed at review time; the reproducibility score (mean 55) rests on those 2 papers alone.
Topic counts span 1996–2026 and likely include works beyond the two catalogued reinforcement-learning papers; recent3yWorks (14) and distinctCoauthors (4) are windowed OpenAlex aggregates.