Cheston Tan

Agency for Science, Technology and Research

Unclaimed · AI-structured profileORCID OpenAlex

A*STAR researcher spanning LLM multi-agent safety in trading systems and medical vision-language AI evaluation

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—

The output axis is not scored: no publication count, venue list, or productivity series is available in the current snapshot. Once full bibliographic records are linked, it would measure total publication volume, venue quality, and output distribution over time.

Impact—

The impact axis is not scored: citation counts, h-index, or altmetrics are absent from the input data. Once linked, it would measure citation-based influence across fields and uptake of the two reviewed arXiv preprints beyond this catalog.

Focus0/100

Focus scores 0/100, and the raw field distribution explains why: the portfolio splits exactly 50/50 between 'AI safety (LLM agents) / quantitative finance' and 'medical image analysis / AI evaluation methodology' — two clusters with little topical overlap. The split reads as a deliberate dual-track programme rather than drift, but the catalog does not yet demonstrate sustained depth within either subfield.

Collaboration41/100

Collaboration scores 41/100, with 7 distinct co-authors recorded in the snapshot — a small, concentrated network. Both cataloged papers are middle-author and non-corresponding, indicating team-based contribution inside larger groups rather than leading authorship. Frequent co-authors include Yew-Soon Ong, Qi Rong Sua, Junhao Dong, Nguyen Duc Thai, Yuqing Wen, Flint Xiaofeng Fan and Roger Wattenhofer, spanning both research tracks.

Activity—

The activity axis is not scored: no publication-year series underlies it, so recent publishing cadence cannot be computed. It would measure output continuity and momentum — such as papers per recent year — beyond the two 2026 preprints in this catalog.

Reproducibility78/100

Reproducibility is the strongest measured axis at 78/100 (reproMean 78 over 2 assessed papers). The medical evaluation paper (AuthEval) rates high readiness; the adversarial-trading paper rates medium, with Evolve noting a single ticker/month and no variance data, while AuthEval's evidence is entirely simulated. Despite the thin sample, the score indicates above-average artifact transparency and methodological openness.

Research summary

Cheston Tan is a researcher at the Agency for Science, Technology and Research (A*STAR), whose Evolve profile — name-resolved from public bibliographic data under ORCID 0000-0003-1248-4906 — currently centers on two 2026 arXiv preprints in deliberately different domains. The first, 'Contagion on the Trading Floor', presents what its authors describe as the first systematic study of how social-media poisoning signals spread through LLM multi-agent trading systems; Evolve rates the framework solid but flags a single-ticker, single-month setup with no variance data. The second, 'Authority-Preserving Evaluation of Medical Vision-Language Assistants', formalizes AuthEval, an evaluation methodology capturing the proposal-vs-selection 'authority gap' in medical VLM triage; Evolve finds it sound and honest while noting that all evidence is simulated. This pairing — 50/50 across 'AI safety (LLM agents)/quantitative finance' and 'medical image analysis/AI evaluation methodology' — yields a focus score of 0/100, the signature of a bifurcated portfolio. Measured axes show moderate collaboration (41/100, 7 distinct co-authors, both papers middle-author and non-corresponding) and the strongest reproducibility reading (78/100 mean across 2 assessed papers). Output, impact and activity axes remain unscored for lack of underlying data. Frequent co-authors include Yew-Soon Ong, Roger Wattenhofer, Flint Xiaofeng Fan and Yuqing Wen.

Collaboration pattern

The snapshot records 7 distinct co-authors, with both cataloged papers in middle-author, non-corresponding positions — a team-contributor pattern rather than solo leadership. Recurring names include Yew-Soon Ong, Qi Rong Sua, Junhao Dong, Nguyen Duc Thai, Yuqing Wen, Flint Xiaofeng Fan and Roger Wattenhofer. The moderate collaboration score (41/100) reflects a small, concentrated network rather than an extensive international web.

Research keywords

LLM multi-agent systemsalgorithmic tradingadversarial robustnessdata poisoningsocial media manipulationcontagion metricsevaluation methodologymedical vision-language modelsclinical decision supportteledermatology triageauthority gaphuman-AI workflow

Highlights

First systematic study of poisoning contagion in LLM trading swarms
The paper claims the first systematic study of how social-media poisoning signals spread through LLM multi-agent trading systems; Evolve credits a solid contagion framework while noting a single ticker/month and no variance data weaken the evidence.
A contagion-metrics framework for multi-agent trading
Introduces contagion metrics for tracking how manipulated signals propagate across agent interactions, framing adversarial robustness and data poisoning as first-class quantitative-finance concerns.
AuthEval: authority-preserving evaluation for medical VLMs
Formalizes the proposal-vs-selection 'authority gap' in medical vision-language triage, including teledermatology workflows; Evolve finds it sound and honest, while noting all evidence is simulated and proposer accuracy is weak.
High reproducibility readiness anchoring a 78/100 mean
The medical evaluation paper is rated high reproducibility readiness, anchoring the researcher's 78/100 mean across 2 assessed papers; the trading paper is rated medium, giving reproducibility its strongest measured dimension.

Timeline

2026arXiv preprint 'Contagion on the Trading Floor' studies adversarial signal contagion in LLM multi-agent trading systems (middle author, non-corresponding).
2026arXiv preprint 'Authority-Preserving Evaluation of Medical Vision-Language Assistants' formalizes the AuthEval evaluation methodology (middle author, non-corresponding).
2026Evolve catalog snapshot: 2 reviewed papers across two field clusters, 7 distinct co-authors, reproducibility mean 78/100.

Papers in the Evolve catalog

Frequent co-authors

Yew-Soon Ong×2 Qi Rong Sua×1 Junhao Dong×1 Nguyen Duc Thai×1 Yuqing Wen×1 Flint Xiaofeng Fan×1 Roger Wattenhofer×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 scores derive from an OpenAlex-based snapshot; output, impact and activity axes lack underlying data and are unscored.
Only two papers, both 2026 arXiv preprints, have been reviewed by Evolve; no peer-reviewed venue or citation data is captured.
Reproducibility (78/100) is computed from just 2 papers (reproPapers = 2), so the estimate is statistically thin.
Profile name-resolved from public bibliographic data via ORCID 0000-0003-1248-4906; country and detailed affiliation beyond the institution name are unverified.