Jun Zhou

Ant Group (China)

Unclaimed · AI-structured profileOpenAlex

Ant Group ML researcher spanning brain-inspired continual learning and efficient post-training for LLM agents.

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—

This axis is not scored: the snapshot contains no publication-count or output-volume data beyond the two Evolve-reviewed 2026 papers. It would measure total papers, venue spread, and yearly output volume once a fuller bibliographic record is linked.

Impact—

No citation counts, h-index, or influence metrics are available in the supplied data, so impact is unscored. This axis would measure citation accumulation, per-paper influence, and uptake of the two 2026 preprints by other researchers.

Focus0/100

The focus axis scores 0 because the catalogued output splits exactly evenly — 0.5 vs 0.5 — between two machine-learning subfields: efficient LLM post-training and NeuroAI/continual learning, with one 2026 paper in each. The raw field distribution shows no dominant single field yet, though both lines share a modular-training-and-acceleration methodological core that future output may consolidate.

Collaboration62/100

Scored 62 on collaboration. The raw record lists 22 distinct coauthors overall and 12 recurring collaborators, including Zhiyu Gui, Kexin Huang, Jia Guo, Junkang Wu, Zihao Wang, Zhiqiang Zhang, Jiancan Wu, Xiang Wang, Hongwei Yan, Kanglei Zhou, Qi Cheng and Weiyi Dong. Both catalogued 2026 papers place Jun Zhou in a middle authorship position without corresponding-author role, indicating a steady contributor within larger teams rather than a solo or lead-senior pattern.

Activity—

Activity cannot be scored: no per-year publication cadence or recent-output timestamps exist in the snapshot beyond the two 2026 arXiv entries. This axis would measure publishing frequency and the recency of ongoing research across a full career record.

Reproducibility55/100

Scored 55, computed as a mean over 2 reviewed papers, each rated medium reproducibility readiness. The agentic distillation paper is well ablated with reported 2.3–3.7× speedups, though single-seed runs temper confidence; the continual-learning paper shows strong gains in vision and robotics but informal theory. Reproducibility thus sits mid-range: solid empirical bases with acknowledged methodological gaps.

Research summary

Jun Zhou is a machine-learning researcher at Ant Group whose profile is name-resolved from public bibliographic data and the Evolve-reviewed catalog. The catalogued work spans two lines in near-equal measure — the raw field distribution splits 0.5/0.5 between efficient LLM post-training and NeuroAI/continual learning — and both entries are 2026 arXiv preprints on which Jun Zhou appears as a middle author rather than corresponding author. The first, 'Brain-Inspired Hierarchical Modularity for General Continual Learning,' draws on the Drosophila mushroom body to build a mixture-of-experts-plus-ensemble architecture that delivers large continual-learning gains in vision and robotics settings, with strong empirics though informal theory and replay baselines that remain competitive. The second, 'Know When to Stop, Where to Restart,' speeds up multi-turn agentic on-policy distillation through action-locked early stopping and prefix reuse, reporting 2.3–3.7× acceleration in a well-ablated study, tempered by single-seed runs and a telecom-domain shortfall. The collaboration record lists 22 distinct coauthors and 12 recurring collaborators, and the collaboration axis scores 62. Reproducibility averages 55 across the two reviewed papers, both rated medium readiness — a mid-range result reflecting solid empirical grounding with acknowledged gaps. Output, impact, and activity axes remain unscored because the snapshot lacks publication counts, citation data, and per-year cadence; these limitations are detailed in the caveats.

Collaboration pattern

Jun Zhou's catalogued record shows a broad but mid-weight collaboration pattern: 22 distinct coauthors overall with 12 recurring collaborators, including Zhiyu Gui, Kexin Huang, Jia Guo, Junkang Wu, Zihao Wang, Zhiqiang Zhang, Jiancan Wu, Xiang Wang, Hongwei Yan, Kanglei Zhou, Qi Cheng and Weiyi Dong. On both reviewed 2026 papers Jun Zhou sits in a middle authorship position without corresponding-author role, consistent with contributing within larger Ant Group teams rather than leading independently; the collaboration axis scores 62.

Research keywords

continual learningcatastrophic forgettingmixture-of-expertsensemble learningDrosophila mushroom bodyparameter-efficient fine-tuningon-policy distillationLLM agentsmulti-turn trainingtraining accelerationadaptive early stoppingprefix reuse

Highlights

Brain-inspired hierarchical modularity lifts general continual learning
Reviewed by Evolve as a Drosophila mushroom-body-inspired mixture-of-experts-plus-ensemble design, this 2026 arXiv preprint reports large continual-learning gains across vision and robotics, with strong empirics offset by informal theory and replay baselines that remain competitive; reproducibility readiness is rated medium.
2.3–3.7× speedup for multi-turn agentic on-policy distillation
Evolve's review finds a well-ablated acceleration recipe: action-locked early stopping decides when to stop and prefix reuse decides where to restart, yielding 2.3–3.7× training gains, though single-seed runs and a telecom-domain shortfall temper the claims; reproducibility readiness is rated medium.

Timeline

2026Co-authored (middle position) the arXiv preprint 'Brain-Inspired Hierarchical Modularity for General Continual Learning' in the NeuroAI / continual-learning line.
2026Co-authored (middle position) the arXiv preprint 'Know When to Stop, Where to Restart: Accelerating Multi-Turn Agentic On-Policy Distillation' in the efficient LLM post-training line.
2026Both 2026 preprints entered the Evolve-reviewed catalog, each rated medium reproducibility readiness with a mean score of 55.
2026The catalogued collaboration footprint reached 22 distinct coauthors, including 12 recurring collaborators.

Papers in the Evolve catalog

Frequent co-authors

Zhiyu Gui×1 Kexin Huang×1 Jia Guo×1 Junkang Wu×1 Zihao Wang×1 Zhiqiang Zhang×1 Jiancan Wu×1 Xiang Wang×1 Hongwei Yan×1 Kanglei Zhou×1 Qi Cheng×1 Weiyi Dong×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.

Name resolution: no ORCID is provided and 'Jun Zhou' is a common name, so this profile is resolved from public bibliographic data tied to the Ant Group affiliation; identity confidence is limited.
Output, impact, and activity axes are unscored because the snapshot lacks publication counts, citation metrics, and per-year cadence.
The reproducibility mean (55) rests on only 2 Evolve-reviewed papers, both 2026 arXiv preprints; no peer-reviewed venue information is in scope.
Coverage is limited to papers reviewed in the Evolve catalog; other work by this author may exist outside it.