Qi Cheng

McGovern Institute for Brain Research

Unclaimed · AI-structured profileOpenAlex

Machine-learning researcher at the McGovern Institute working on brain-inspired continual learning

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) 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 — cumulative publication count, venues, and authorship positions over time — is not yet present in the bibliometric snapshot, so this axis is unscored. It would measure how much Cheng publishes, where, and in which author roles.

Impact—

No citation-based evidence (citation counts, h-index, or referencing literature) is available yet, so impact is unscored. This axis would measure how widely Cheng's papers are cited and taken up by the continual-learning community.

Focus0/100

The radar scores focus at 0. The underlying field distribution splits evenly, 50/50, between two machine-learning labels: 'continual learning and representation learning' and 'NeuroAI / continual learning'. In practice both labels point at the same general-continual-learning niche shared by both cataloged papers, so the zero likely reflects label dispersion rather than genuine topic-switching; still, the score is computed from this recorded distribution.

Collaboration60/100

Collaboration scores 60 out of 100, computed from a network of 19 distinct coauthors. Recurring partners include Guanglong Sun, Kanglei Zhou, Hongwei Yan, Liyuan Wang, Shuang Cui, Hang Su, Jun Zhu, Yi Zhong, Weiyi Dong, Chunyan Lan, and Jun Zhou. On both cataloged papers Cheng sits in a middle author position rather than leading, consistent with a mid-scale, team-embedded collaboration pattern rather than a solo or corresponding-author profile.

Activity—

Recent-activity signals — publication cadence by year and career-stage trajectory — are insufficient in the current snapshot, so this axis is unscored. It would track whether research output is accelerating, holding steady, or winding down over recent years.

Reproducibility78/100

Reproducibility is the strongest measured axis at 78, the mean readiness score across 2 reviewed papers. MePo++ is rated high readiness, while the brain-inspired hierarchical-modularity paper is rated medium, reflecting heavier infrastructure needs for its mixture-of-experts and robotics experiments. Both are arXiv preprints with detailed empirical claims, giving a solid but small two-paper base for this score.

Research summary

Qi Cheng is a machine-learning researcher affiliated with the McGovern Institute for Brain Research whose cataloged work centers on general continual learning: training models that keep acquiring new tasks without catastrophically forgetting old ones. Evolve's catalog holds two 2026 arXiv preprints with Cheng as a middle author and no corresponding role. The first, 'Brain-Inspired Hierarchical Modularity for General Continual Learning,' borrows organization principles from the Drosophila mushroom body to build a mixture-of-experts plus ensemble architecture, reporting large gains in vision and robotics with strong empirics but informal theory. The second, 'MePo++,' combines meta-refinement with online alignment to refine and reconcile representations of pre-trained models under continual learning, reporting broad consistent gains despite very small representation-level effect sizes. Evolve's field distribution splits evenly, 50/50, between continual-learning/representation-learning and NeuroAI-flavored continual learning — a tightly focused niche. Collaboration scores 60 over 19 distinct coauthors, with Liyuan Wang, Hang Su, Jun Zhu, and Yi Zhong among recurring partners. Reproducibility is the strongest measured axis at 78 (mean across 2 papers). Output, impact, and activity metrics are not yet available.

Collaboration pattern

Cheng works as a middle author inside a sizable team: 19 distinct coauthors underpin a collaboration score of 60. The recurring roster spans Guanglong Sun, Kanglei Zhou, Hongwei Yan, Liyuan Wang, Shuang Cui, Hang Su, Jun Zhu, Yi Zhong, Weiyi Dong, Chunyan Lan, and Jun Zhou, and both 2026 papers are non-corresponding, middle-position contributions — a pattern of contributing expertise to group-led continual-learning projects rather than leading them.

Research keywords

continual learninggeneral continual learningcatastrophic forgettingmixture-of-expertsensemble learningDrosophila mushroom bodyparameter-efficient fine-tuningpretrained modelsmeta-learningrepresentation learningpost-training

Highlights

Drosophila-inspired modular architecture for continual learning
In 'Brain-Inspired Hierarchical Modularity for General Continual Learning' (2026, arXiv; middle author), mixture-of-experts plus ensemble modularity modeled on the Drosophila mushroom body yields large continual-learning gains in vision and robotics. Evolve notes strong empirics, informal theory, and replay-competitive results, and rates reproducibility readiness medium.
MePo++ unifies representation refinement and reconciliation
In 'MePo++' (2026, arXiv; middle author), meta-refinement plus online alignment targets general continual learning with pre-trained models, reporting broad consistent gains. Evolve flags that representation-level effect sizes are very small, while rating reproducibility readiness high.
A tightly focused general-continual-learning agenda
Both 2026 cataloged papers attack catastrophic forgetting in general continual learning from complementary angles — brain-inspired architectural modularity on one side, representation-level refinement for pre-trained models on the other — producing an even 50/50 split across the two machine-learning field labels in Evolve's data.

Timeline

2026Co-authored 'Brain-Inspired Hierarchical Modularity for General Continual Learning' (arXiv), bringing Drosophila mushroom-body-inspired mixture-of-experts modularity to continual learning in vision and robotics.
2026Co-authored 'MePo++: Unifying Representation Refinement and Reconciliation for General Continual Learning' (arXiv), a meta-refinement plus online-alignment method for pre-trained-model continual learning.
2026Both Evolve-reviewed preprints appeared in the same year, anchoring a profile centered on general continual learning, with 19 distinct coauthors and a mean reproducibility readiness of 78 across the two papers.

Papers in the Evolve catalog

Frequent co-authors

Guanglong Sun×2 Kanglei Zhou×2 Hongwei Yan×2 Liyuan Wang×1 Shuang Cui×1 Hang Su×1 Jun Zhu×1 Yi Zhong×1 Hongwei Yan×1 Weiyi Dong×1 Chunyan Lan×1 Jun Zhou×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.

'Qi Cheng' is a relatively common name; this profile is name-resolved from public bibliographic data tied to the McGovern Institute for Brain Research affiliation, without ORCID verification.
Radar scores come from Evolve's OpenAlex-based bibliometric pipeline; output, impact, and activity axes currently lack sufficient data and are unscored.
Only two 2026 arXiv preprints are reviewed in this catalog; peer-reviewed venue status and citation counts are not yet recorded.
The reproducibility score averages just 2 papers, so treat it as an early-stage signal rather than a stable estimate.