Early-career researcher spanning AI safety evaluation and lightweight deep learning for sleep staging
2
papers here
2
reviewed by Evolve
1
first-author
1
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-19) 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 yet: publication counts, venue coverage and yearly volume are unavailable for this researcher, whose visible record here is only two 2026 arXiv preprints. This dimension would measure total and per-year paper output once a fuller bibliographic record is linked to the profile.
Impact—
Impact is unscored because no citation counts, h-index or downstream-usage data are available for these two very recent preprints. This axis would measure citations, uptake and influence of the work across fields once citation records can be reliably linked to this researcher.
Focus0/100
Focus scores 0 because the field distribution is an exactly even 50/50 split: NLP/AI safety evaluation and machine learning for biomedical signal processing each account for half of the catalog, with no dominant field yet. The two 2026 preprints sit one in each area, an early dual-track pattern rather than an established specialization.
Collaboration39/100
Collaboration scores 39, based on 6 distinct coauthors across the two catalog papers. The footprint is modest and role-mixed: the repeated-abuse study is a middle-author team effort, while LightSleepX is a first-author, corresponding-author work. Frequent collaborators include William Guey, Wei Zhang, Pierrick Bougault, Agoston Bodo, Vitor D de Moura and José O Gomes.
Activity—
Activity cannot be scored yet: the data offers no multi-year publication cadence, only two papers both dated 2026. This axis would measure publishing tempo, recency and consistency across the career timeline, and should be revisited as richer dated records accumulate for this researcher.
Reproducibility78/100
Reproducibility scores 78 as the mean over 2 scored papers. The repeated-abuse evaluation rates high for careful, transparent measurement of a bilingual multi-turn benchmark, while LightSleepX rates medium: the 0.049M-parameter efficiency results are promising, but there are no significance tests, inconsistent baselines and no released code.
Research summary
Yi Wang is an early-stage researcher whose visible record spans two distinct areas in near-equal measure: NLP/AI safety evaluation and machine learning for biomedical signal processing, each accounting for 50% of the two-paper Evolve catalog. In 2026, Wang appeared as a middle author on 'How AI Assistants Respond to Repeated Abuse', a bilingual, multi-turn benchmark study showing that refusal is not a single axis and that AI assistant configurations diverge sharply under repeated verbal abuse; Evolve judged its measurement careful and transparent but exploratory and deployment-bound, with high reproducibility readiness. In the same year, Wang led 'LightSleepX' as first and corresponding author, proposing a 0.049M-parameter Inception- and Mamba-based dual-modal network for EEG/EOG sleep staging; Evolve called the efficiency claims promising but noted missing significance tests, inconsistent baselines and no released code, giving medium reproducibility readiness. The overall reproducibility mean is 78 across 2 scored papers. Collaboration is still compact: 6 distinct coauthors, including William Guey, Wei Zhang, Pierrick Bougault, Agoston Bodo, Vitor D de Moura and José O Gomes, yielding a collaboration score of 39. Output, impact and activity metrics are not yet available. With both 2026 preprints on arXiv and no institution or ORCID on file, this is a name-resolved, early-stage profile whose dual-track focus may consolidate as the record grows.
Collaboration pattern
Collaboration remains compact (score 39), with 6 distinct coauthors across the two 2026 catalog papers. The pattern mixes roles: a middle-author team contribution on the repeated-abuse evaluation, and first-plus-corresponding authorship on LightSleepX. Recurring collaborators include William Guey, Wei Zhang, Pierrick Bougault, Agoston Bodo, Vitor D de Moura and José O Gomes, suggesting small, possibly cross-institutional teams, though no affiliations are on file.
Research keywords
Highlights
Refusal is not one axis: profiling AI assistants under repeated verbal abuse
First-and-corresponding authorship on a 0.049M-parameter sleep staging network
Timeline
- 2026Co-authored 'How AI Assistants Respond to Repeated Abuse' on arXiv as a middle author — a bilingual multi-turn safety evaluation rated high for reproducibility readiness.
- 2026Led 'LightSleepX' as first and corresponding author on arXiv, proposing a 0.049M-parameter Mamba+Inception dual-modal network for EEG/EOG sleep staging.
- 2026Published in two distinct fields in the same year — a 50/50 split between NLP/AI safety evaluation and biomedical signal processing — establishing an early dual-track profile.
Papers in the Evolve catalog
Frequent co-authors
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.
- All bibliometric scores derive from Evolve's review of just two 2026 arXiv preprints; they may not represent the researcher's complete publication record.
- 'Yi Wang' is a common name; this profile is name-resolved from public bibliographic data, and no ORCID or institutional affiliation is available to confirm identity.
- Output, impact and activity axes lack underlying data (no citation counts or multi-year publication history), so they are left unscored.
- The reproducibility score (mean 78) is averaged over only 2 papers, making it sensitive to any single new publication.