Machine-learning researcher applying LLM agents to autonomous research and pre-training data curation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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: this axis measures publication volume and productivity over time. Only two reviewed 2026 arXiv papers are available, too small a base to estimate an output profile.
Impact—
Not yet scored: this axis measures citation counts, uptake, and scholarly influence of the work. No citation or readership data exists in the current snapshot, so impact cannot be estimated.
Focus0/100
Focus scores 0 because the reviewed work splits evenly — 50/50 in the field distribution — between autonomous research agents and LLM pre-training data curation. One line (AIDE², recursive self-improvement) builds agents that improve agents; the other (AutoData) uses agentic search to select pre-training data. Both are LLM-agent-centric, reading more as one method applied to two problems than scattered interests, but the even balance still yields the lowest focus score.
Collaboration41/100
Collaboration scores 41, based on 7 distinct coauthors in the snapshot. Recurring names include Dhruv Srikanth, Yuxiang Wu, Zhengyao Jiang, Dixing Xu, and Yan Meng. On both reviewed papers Zhao is a middle-position author and not corresponding, consistent with contributing inside larger teams rather than leading them. The moderate score and small coauthor pool suggest an early-stage, team-embedded collaboration pattern built around a stable core group.
Activity—
Not yet scored: this axis measures publication cadence and sustained research engagement. Only two 2026 papers are in scope, with no multi-year timeline to assess activity trends.
Reproducibility38/100
Reproducibility scores 38, a mean computed across the 2 papers with reproducibility assessments. 'Recursive self-improvement of AI research agents' is rated low — Evolve flags few runs and noisy grade comparisons behind the 0.703→0.778 gain — while 'AutoData' is rated medium, its val-bpb improvements holding across scales even though CORE gains lack statistical support. Overall, results are promising but rest on limited statistical grounding.
Research summary
Bingchen Zhao is a machine learning researcher whose reviewed work spans two tightly related threads: autonomous research agents built on large language models, and pre-training data curation for LLMs. Evolve's snapshot resolves two 2026 arXiv papers, on both of which Zhao is a middle-position author and not the corresponding author — a contributor profile within larger collaborative teams. The field distribution splits evenly, 50/50, between artificial intelligence (autonomous research agents / LLM agent systems) and machine learning / NLP (LLM pre-training data curation). 'Recursive self-improvement of AI research agents' reports an AIDE² self-improvement loop that, over 8 days, lifts agent grade from 0.703 to 0.778 with held-out transfer; Evolve finds it compelling but notes it rests on few runs and noisy grade comparisons. 'AutoData: Agentic Search for Pre-training Data Selection' shows LLM-agent search over executable data-selection recipes beating curated-pipeline baselines on val-bpb across scales, though CORE gains lack statistical support. Radar scores are collaboration 41 (7 distinct coauthors), reproducibility 38 (mean across 2 papers; readiness low and medium), and focus 0, reflecting the even two-field split. Output, impact, and activity dimensions lack sufficient underlying data and are not scored. No institution, country, or ORCID is provided; the profile is name-resolved from public bibliographic data.
Collaboration pattern
The collaboration pattern is team-embedded: 7 distinct coauthors appear in the snapshot, with Dhruv Srikanth, Yuxiang Wu, Zhengyao Jiang, Dixing Xu, and Yan Meng recurring most often. On both reviewed 2026 papers Zhao is a middle-position, non-corresponding author, consistent with contributing specific components within larger groups rather than leading them. The collaboration radar score of 41 reflects this moderate-size but stable network.
Research keywords
Highlights
AIDE²: an 8-day recursive self-improvement loop for research agents
AutoData: agentic search replaces hand-built pre-training data pipelines
One agentic-search playbook applied to two problems
Timeline
- 2026Publishes 'Recursive self-improvement of AI research agents' on arXiv as a middle author, reporting AIDE²'s 8-day loop lifting agent grade from 0.703 to 0.778 with held-out transfer.
- 2026Publishes 'AutoData: Agentic Search for Pre-training Data Selection' on arXiv as a middle author, showing LLM-agent search over executable data-selection recipes beating curated pipelines on val-bpb across scales.
- 2026Both papers enter Evolve's review catalog, establishing a 50/50 field split between autonomous research agents and LLM pre-training data curation.
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.
- Bibliometrics come from Evolve's OpenAlex-based snapshot only; no citation counts, h-index, or venue history are available.
- No institution, country, or ORCID is provided; the profile is name-resolved from public bibliographic data and may conflate researchers sharing this name.
- Scope is limited to two reviewed 2026 arXiv preprints; output, impact, and activity dimensions are unscored for insufficient data.
- The reproducibility score (38) is a mean over just 2 papers with differing readiness (low and medium), so it should be read as preliminary.