Sirry Chen

Fudan University · China

Unclaimed · AI-structured profileOpenAlex

Fudan University ML researcher: time-series foundation models for wearable health and RL-based LLM post-training

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—

Not yet scored: there is not enough indexed record data to compute aggregate output. This axis would measure total works, venue types, and authorship roles across the career; the visible signal — 8 works in three years — is too thin.

Impact—

Not yet scored: citation-based influence cannot be computed reliably. This axis would measure citations, citation velocity, and uptake of the works. The only recorded signals — h-index 1 and i10-index 1 — come from a very early career profile.

Focus0/100

Scored 0 because the record is two-track rather than concentrated: fieldDist splits 50/50 between machine learning for time-series foundation models in wearable health and machine learning for LLM post-training. Topic data reinforces the spread — Speech Recognition and Synthesis (4), Artificial Intelligence in Healthcare and Education (4), and Speech and Dialogue Systems (4) sit alongside Reinforcement Learning in Robotics (1) and Handwritten Text Recognition (1). No single theme dominates yet.

Collaboration70/100

Scored 70 from OpenAlex co-authorship data: 33 distinct coauthors over the recent three-year window indicate a strongly team-based working style. Recurring partner names include Fangke Chen, Wei Chen, Zhongyu Wei, Shuai Dong, Yongfu Zhu, Yuqi Xu, Weichu Xie, Liuwenpu, and Ziyue Wang. Both reviewed 2026 papers list Sirry Chen in a middle author position without corresponding-author status, consistent with contributing inside larger project groups rather than leading small independent teams.

Activity64/100

Scored 64 from OpenAlex yearly counts: recent3yWorks is 8, and countsByYear shows 1 work in 2024 rising to 7 in 2026, so the current burst accounts for nearly all indexed output. Both catalog papers appeared on arXiv in 2026. With no earlier years recorded, the score reflects a high current tempo on a short track rather than sustained multi-year volume — consistent with an accelerating early-career record.

Reproducibility78/100

Scored 78 from Evolve's reproducibility assessment of 2 reviewed papers (reproMean 78). SOTER is rated high readiness, with broad, ablated zero-shot evidence for a wearable-signal foundation model; the on-policy distillation RL paper is rated medium, with strong diagnostics but single-seed runs that leave small margins uncertain. The pattern suggests a habit of ablation-heavy evidence, with statistical testing and multi-seed runs the main remaining gap.

Research summary

Sirry Chen is an early-career machine-learning researcher at Fudan University whose profile, name-resolved from public OpenAlex bibliographic data, spans two subfields in roughly equal measure: time-series foundation models for wearable health and LLM post-training. OpenAlex indexes 8 works over the last three years, with 7 appearing in 2026 alone — a high current tempo on a very short record (h-index 1, i10-index 1). Evolve's catalog contains two 2026 arXiv papers, both with middle author positions and no corresponding-author role: SOTER, a generative time-series foundation model for wearable physiological signals built on mixture-of-experts and neural CDE components, and a study arguing that on-policy distillation prepares reinforcement learning better than Base or SFT starts, with the reverse/forward KL ranking flipping after RL. Topic data adds recurring clusters in speech recognition and synthesis, speech and dialogue systems, and AI in healthcare and education. Reproducibility scoring across 2 papers averages 78 (SOTER high, the RL paper medium). With 33 distinct coauthors and a two-track portfolio, the profile reads as an emerging, fast-moving research identity rather than a settled agenda.

Collaboration pattern

Co-authorship data shows a strongly team-based pattern: 33 distinct coauthors across the recent three-year window, with recurring partners including Fangke Chen, Wei Chen, Zhongyu Wei, Shuai Dong, Yongfu Zhu, Yuqi Xu, Weichu Xie, Liuwenpu, and Ziyue Wang. Both reviewed 2026 arXiv papers place Sirry Chen in a middle author position without corresponding-author status, suggesting contribution within larger project groups spanning wearable-health modeling and LLM post-training rather than solo or small-team leadership.

Research keywords

time-series foundation modelswearable sensorsphysiological signalszero-shot forecastingmixture-of-expertsneural CDEon-policy distillationreinforcement learning with verifiable rewards (RLVR)GRPOreverse/forward KL divergencePass@kLLM post-training

Highlights

SOTER: generative time-series foundation model for wearable physiological signals
This 2026 arXiv paper proposes a generative time-series foundation model for wearable human physiological signals, built on mixture-of-experts and neural CDE components, and claims zero-shot wins over generalist time-series foundation models. Evolve finds the evidence broad and ablated, though lacking significance tests; reproducibility readiness is rated high.
On-policy distillation as a better starting point for RL
A second 2026 arXiv paper argues that RL starts before RL: on-policy distillation (OPD) preparation beats Base/SFT starting points for RLVR-style training with GRPO, and the reverse/forward KL divergence ranking flips after RL. Evolve notes strong diagnostics, with single-seed runs limiting small margins; reproducibility readiness is medium.
A deliberately two-track research portfolio
fieldDist places wearable-health time-series foundation models and LLM post-training at 50/50 each, and topic counts (speech recognition and synthesis 4, AI in healthcare and education 4, speech and dialogue systems 4) show breadth beyond the two catalog papers — an early-stage profile still searching for its center of gravity.

Timeline

2024First indexed work appears in OpenAlex countsByYear (1 work recorded for 2024).
2026Output accelerates sharply: 7 works indexed in 2026, nearly all of the recent three-year total.
2026Releases SOTER (arXiv), a generative time-series foundation model for wearable physiological signals.
2026Releases the on-policy distillation study (arXiv), arguing RL starts before RL.

Papers in the Evolve catalog

Frequent co-authors

Fangke Chen×2 Fangke Chen×1 Sirry Chen×1 Wei Chen×1 Zhongyu Wei×1 Wei Chen×1 Shuai Dong×1 Yongfu Zhu×1 Yuqi Xu×1 Weichu Xie×1 Liuwenpu×1 Ziyue Wang×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 come from OpenAlex only; no ORCID is linked and the display name is uncommon, so identity resolution carries some uncertainty.
Citation impact indicators (h-index 1, i10-index 1) reflect a very early career stage and should not be read as quality verdicts.
Evolve has reviewed only 2 of this researcher's papers; the full corpus, including any peer-reviewed venue publications, is not covered here.
countsByYear covers only 2024 and 2026; missing years may reflect OpenAlex indexing lag rather than actual inactivity.