Aske Plaat

—

Unclaimed · AI-structured profileOpenAlex

Reinforcement learning evaluation and RAG reliability researcher: metrics, benchmarks, and hallucination repair

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

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 snapshot lacks publication counts and output history, so total volume, arXiv cadence, and authorship-position trends over time cannot be computed. This axis would measure aggregate paper output and its trajectory across years and venues.

Impact—

Not yet scored: citation counts, h-index, and influence indicators are absent from the current bibliometric snapshot. This axis would measure uptake of the proposed safe reinforcement learning evaluation metrics and citations across the retrieval-augmented generation work.

Focus100/100

The focus score of 100 reflects a highly concentrated record: the field distribution currently contains exactly one classified field — safe reinforcement learning as machine learning evaluation methodology — drawn from the 2026 arXiv evaluation-metrics paper. The companion paper on hallucination detection and repair in retrieval-augmented generation suggests a consistent evaluation-methodology theme rather than topic scatter, though only one field has been classified so far.

Collaboration32/100

Collaboration scores 32, based on 4 distinct coauthors in the snapshot. The named frequent collaborators are Sai Krishna Reddy Mulakkayala, Niki van Stein, Lindsay Spoor, and Thomas Moerland. Authorship placement varies across the two 2026 arXiv papers — middle position on the safe reinforcement learning evaluation study and last position on the retrieval-augmented generation hallucination paper — indicating participation both as a team member and as the senior-listed author within a small, stable coauthor circle.

Activity—

Not yet scored: the cataloged papers all date from 2026, so multi-year publishing rhythm and continuity cannot be computed. This axis would measure annual output counts and sustained research activity across recent years.

Reproducibility100/100

Reproducibility scores 100, computed as a reproducibility mean of 100 over 1 reviewed paper. The 2026 safe reinforcement learning evaluation study carries a high reproducibility-readiness rating, consistent with Evolve's verdict highlighting a 1,440-run experimental scale and transparent reporting conventions — violation rates, normalized costs, tier definitions, and cumulative-distribution reporting. The second cataloged paper has not yet been assessed, so this score rests on a single-paper evidence base.

Research summary

Aske Plaat is a machine learning researcher whose cataloged work centers on rigorous evaluation methodology. The Evolve catalog contains two 2026 arXiv preprints. The first, 'Evaluation Metrics for Safe Reinforcement Learning' (middle authorship), proposes violation-rate and normalized-cost metrics, safety tiers, and cumulative-distribution reporting for evaluating safe reinforcement learning, backed by a 1,440-run empirical study; Evolve rates its reproducibility readiness as high, noting strong scale but narrow scope and brittle tier thresholds. The second, 'Detecting and Repairing Hallucinations in Retrieval-Augmented Generation' (last authorship), extends the evaluation-first orientation to the reliability of retrieval-augmented language models and has not yet received a full Evolve review. Bibliometric dimensions from the snapshot score focus at 100 — the field distribution currently holds a single classified field, safe reinforcement learning as machine learning evaluation methodology — and reproducibility at 100, from a reproducibility mean of 100 over one reviewed paper. Collaboration scores 32, based on 4 distinct coauthors, including frequent collaborators Sai Krishna Reddy Mulakkayala, Niki van Stein, Lindsay Spoor, and Thomas Moerland. Output, impact, and activity dimensions are not yet scored because citation counts, publication totals, and multi-year history are missing from this snapshot.

Collaboration pattern

The collaboration dimension scores 32, based on 4 distinct coauthors in the snapshot. Recurring names include Sai Krishna Reddy Mulakkayala, Niki van Stein, Lindsay Spoor, and Thomas Moerland. Across the two 2026 arXiv preprints, authorship shifts from a middle position on the safe reinforcement learning evaluation study to the last position on the retrieval-augmented generation hallucination paper, suggesting a small stable circle in which the researcher contributes both as a core team member and as the senior-listed author.

Research keywords

safe reinforcement learningconstrained Markov decision processevaluation metricsviolation ratebenchmarkingsafety tiersmachine learning evaluation methodologyretrieval-augmented generationhallucination detectionhallucination repair

Highlights

A metrics framework for safe reinforcement learning
The 2026 arXiv study proposes violation-rate and normalized-cost metrics, safety tiers, and cumulative-distribution reporting for evaluating safe reinforcement learning. Evolve notes a strong 1,440-run empirical base, while flagging narrow scope and brittle tier thresholds as limitations.
Large-scale empirical evidence with high reproducibility readiness
Evolve rates the safe reinforcement learning evaluation paper's reproducibility readiness as high; the profile's reproducibility dimension scores 100, based on a reproducibility mean of 100 across one reviewed paper.
Extending evaluation rigor to retrieval-augmented generation
The second 2026 arXiv preprint, written in the last author position, addresses detecting and repairing hallucinations in retrieval-augmented generation; it has not yet received a full Evolve review, so no verdict or keywords are available.

Timeline

2026Publishes 'Evaluation Metrics for Safe Reinforcement Learning' on arXiv in a middle authorship position, proposing violation-rate and normalized-cost metrics alongside safety tiers and cumulative-distribution reporting.
2026The safe reinforcement learning evaluation study receives a high reproducibility-readiness rating from Evolve, anchoring the profile's reproducibility dimension score of 100 (reproMean 100 over 1 paper).
2026Authors 'Detecting and Repairing Hallucinations in Retrieval-Augmented Generation' in the last position on arXiv, extending the evaluation focus to RAG reliability; a full Evolve review is pending.

Papers in the Evolve catalog

2026-09-14
Evaluation Metrics for Safe Reinforcement Learning

Frequent co-authors

Sai Krishna Reddy Mulakkayala×1 Niki van Stein×1 Lindsay Spoor×1 Thomas Moerland×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.

Bibliometric dimensions are derived from public bibliographic data (OpenAlex-based); no institutional affiliation, country, or ORCID was available in this snapshot.
This profile is name-resolved from public bibliographic records; without an ORCID link, identity attribution carries some uncertainty.
Output, impact, and activity dimensions are unscored: the snapshot lacks citation counts, publication totals, and multi-year history.
Only two 2026 arXiv preprints are cataloged; the retrieval-augmented generation paper has no Evolve verdict, keywords, field label, or slug yet.