Shunya Nagashima

—

Unclaimed · AI-structured profilename-matched identity

ML researcher linking time-series forecasting, biomedical signal reconstruction, and LLM-agent benchmarking

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-01) 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 data is not yet available: no total publication count, venue record, or yearly volume can be computed from the current snapshot, which covers only two arXiv preprints. This axis would measure the volume and consistency of published work.

Impact—

Impact metrics are unavailable: the snapshot carries no citation counts, h-index, or downstream-usage signals for this name. This axis would measure how widely the two catalog papers are cited, reused, or built upon by other work.

Focus0/100

The focus score of 0 reflects an even 50/50 split across two distinct subfields in the raw field distribution: machine learning for LLM agents and time-series forecasting on one side, machine learning for time-series and biomedical signal processing on the other. The two catalog papers share no single field tag; time-series forecasting is the connective method, while benchmark design and PPG-based vital-sign reconstruction pull the portfolio in different directions.

Collaboration14/100

A collaboration score of 14 matches a very small footprint: exactly one distinct coauthor, Ragamayi Puli, appears across the catalog snapshot. Both papers place Nagashima in a senior slot — first author on the benchmark paper and last author on the ROOSTER paper — suggesting a stable two-person working pattern rather than large-team or consortium science. The sample is small, so the low score should be read as 'small-team so far', not as an established solo signal.

Activity—

Activity data is insufficient: only the 2026 arXiv year is represented, so recent-publication cadence and multi-year trends cannot be scored. This axis would measure publishing rhythm and sustained engagement over time.

Reproducibility78/100

Reproducibility scores a 78, the mean across the two papers rated by Evolve. The ROOSTER paper reaches 'high' readiness, while Forecast Workflow Bench sits at 'medium'. Both verdicts attach the same class of warning: the benchmark's headline GPT-6 Astra win at a 2.5% budget is not statistically significant on repeat, and ROOSTER's gains rest on single-seed, cross-paper evidence. Strong methodological openness, with statistical robustness as the open gap.

Research summary

Shunya Nagashima is an early-stage machine learning researcher whose visible work attacks time-series forecasting from two directions: statistical and biomedical signal modeling, and language-model-driven decision systems. The Evolve catalog holds two 2026 arXiv papers, splitting the field distribution evenly (50/50) between machine learning for LLM agents and time-series forecasting, and machine learning for time-series and biomedical signal processing. As first author, Nagashima introduced Forecast Workflow Bench, evaluating language-model decisions with budgeted forecast tools; Evolve credits the budgeted-agent framing while noting that GPT-6 Astra's win at a 2.5% budget lacks significance on repeat and that every local model trails a free heuristic. As last author on 'Learning Where to Look' (ROOSTER), the work learns a comb relative-position bias replacing identity-skip and seasonal-lag conditioning, serving both time-series forecasting and PPG-to-vital-sign reconstruction; Evolve rates its reproducibility readiness 'high' while flagging single-seed, cross-paper evidence behind headline wins. The collaboration footprint is a compact two-person pattern with one distinct coauthor, Ragamayi Puli (collaboration score 14). Reproducibility is the strongest measured axis at a mean of 78 across both papers. Output, impact, and activity remain unscored because the snapshot covers too little; with no institution or ORCID on record, this profile is name-resolved from public bibliographic data.

Collaboration pattern

The measured footprint is a two-person pattern: exactly one distinct coauthor, Ragamayi Puli, appears in the snapshot, yielding a low collaboration score of 14. Nagashima takes senior authorship slots — first author on the benchmark paper and last author on the ROOSTER paper — consistent with a small, stable duo rather than large-team science. With only two papers observed, this should be read as a thin-sample signal rather than an established collaboration style.

Research keywords

photoplethysmographyvital-sign reconstructionPPG-to-ECGtime-series forecastingrelative position biascross-attentionLLM agentstime-series foundation modelsbenchmarkdecision-makingcost-constrained tool usecapacity planning

Highlights

ROOSTER: a shared relative-alignment module for forecasting and PPG vital-sign reconstruction
As last author, Nagashima's 2026 arXiv paper learns a comb relative-position bias that replaces identity-skip and seasonal-lag conditioning, serving both time-series forecasting and PPG-to-vital-sign reconstruction. Evolve rates reproducibility readiness 'high' but cautions that headline wins rest on single-seed, cross-paper evidence.
Forecast Workflow Bench: budgeted evaluation of LLM decisions with forecast tools
First-author 2026 arXiv benchmark evaluates language-model decision-making under explicit tool budgets. GPT-6 Astra leads at a 2.5% budget, but Evolve notes all local models trail a free heuristic and Astra's repeat run lacks statistical significance; reproducibility readiness is 'medium'.
A two-track time-series portfolio with reproducibility as the strongest measured axis
The two 2026 papers split evenly across LLM-agent benchmarking and biomedical signal processing, tied together by time-series forecasting; the mean reproducibility score across both papers is 78, with one 'high' and one 'medium' readiness rating.

Timeline

2026Publishes first-author preprint 'Forecast Workflow Bench: Evaluating Language-Model Decisions with Budgeted Forecast Tools' on arXiv.
2026Publishes last-author preprint 'Learning Where to Look' (ROOSTER), introducing a shared relative-alignment module for time-series forecasting and PPG-to-vital-sign reconstruction.
2026Both catalog papers appear in the same year, splitting the portfolio evenly between LLM-agent benchmarking and biomedical signal processing.

Papers in the Evolve catalog

Frequent co-authors

Ragamayi Puli×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.

Dimension scores come from a thin public bibliometric snapshot of only two 2026 arXiv papers; output, impact, and activity lack sufficient data and are left unscored.
No institution, country, or ORCID is on record; the profile is name-resolved from public bibliographic data, and 'Shunya Nagashima' may be shared by multiple researchers.
The reproducibility mean of 78 rests on two Evolve-reviewed papers; verdicts flag single-seed evidence and one non-significant repeat run.
Both papers are arXiv preprints; no peer-reviewed venue status is available in the snapshot.