Wei Wang

University of California, Los Angeles · Hong Kong SAR China

Unclaimed · AI-structured profileOpenAlex

UCLA researcher on LLM agent post-training and long-term memory evaluation, spanning continual learning and diagnostics

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 cannot be scored yet: the radar marks this axis missing, and h-index and i10-index are recorded as 0 in the snapshot, consistent with an unpopulated record. This axis would measure total publication and preprint volume over time.

Impact—

Impact cannot be scored yet: no citation counts, h-index, or i10-index are available beyond zeros in an unpopulated OpenAlex record, and both catalog papers are 2026 preprints too recent for citation accumulation. This axis would measure citations, adoption, and influence.

Focus0/100

The focus axis scores 0, the lowest reading in the radar, and the raw data explains why: fieldDist splits evenly (0.5/0.5) between LLM-agent long-term-memory evaluation and LLM post-training, and the OpenAlex topic list scatters across five topics — Topic Modeling (2 papers) plus four singletons spanning information retrieval, neural computing, and even pulsars and gravitational waves. Part of this spread is genuine dual-track work; part likely reflects name-ambiguity noise in the record.

Collaboration65/100

Collaboration scores 65, backed by 25 distinct coauthors in the OpenAlex record and a stable inner circle of 12 frequent coauthors, including Zeyu Liu, Jian Zhong, Rongduo Han, and Xiaokang Xue. Both Evolve-catalog papers are multi-author arXiv submissions on which Wei Wang appears in the last author position without being corresponding author, a pattern consistent with senior-supervision collaboration on LLM-agent systems projects.

Activity47/100

Activity scores 47 on a record whose observable output is concentrated in a single year: countsByYear lists 4 works in 2026, and recent3yWorks is likewise 4, meaning no earlier works appear in the snapshot. Within that window the pace is brisk — two Evolve-reviewed arXiv preprints (ACLArena, EvalMem) plus two further indexed works — but the short horizon makes sustained productivity impossible to confirm yet.

Reproducibility55/100

Reproducibility scores 55, computed as reproMean 55 over reproPapers 2 — exactly the two Evolve-reviewed papers, each rated medium reproReadiness. Both are arXiv preprints describing benchmark-style frameworks (ACLArena's continual-learning arena, EvalMem's three-Examiner diagnostic protocol), which typically aids reimplementation, but neither carries a peer-reviewed venue or a verified released codebase in the data Evolve holds, keeping the score mid-range rather than high.

Research summary

Wei Wang is an LLM-systems researcher at the University of California, Los Angeles whose work runs along two complementary tracks: post-training of language-model agents and evaluation of long-term memory in those agents. The Evolve catalog holds two 2026 arXiv papers, ACLArena and EvalMem, on both of which Wei Wang is the last author but not the corresponding author — a position consistent with senior or supervisory involvement. ACLArena treats agent continual learning as a multi-stage post-training problem, benchmarking catastrophic forgetting and mitigation paradigms such as on-policy distillation, self-distilled fine-tuning, and model merging; Evolve credits the diagnosis and paradigm comparison but notes the setting relies on task-ID routing and is validated at 8B scale only. EvalMem is an operation-level diagnostic framework whose three Examiners attribute memory failures to encoding, retrieval, or generation; Evolve finds the retrieval-deficiency result convincing while noting limited gains. Bibliometrically, the OpenAlex-resolved record lists 25 distinct coauthors, an even 50/50 field split between the two tracks, four works in 2026, and medium reproducibility readiness on both reviewed papers. Because Wei Wang is an extremely common name, this profile is name-resolved from public bibliographic data and should be read with that limitation in mind.

Collaboration pattern

Collaboration is the strongest radar axis at 65: the OpenAlex record lists 25 distinct coauthors, and 12 recur frequently — including Zeyu Liu, Jian Zhong, Rongduo Han, Ziyang Wu, Shunye Tang, Chenghao He, Yaxuan Yang, Yihang Qiu, Ailing Wang, Xiao Liang, Guohuan Xie, and Xiaokang Xue. Both catalog papers are team-written arXiv submissions with Wei Wang as last author but not corresponding author, suggesting a supervisory role within a rotating group of LLM-agent collaborators.

Research keywords

agent continual learningmulti-stage post-trainingcatastrophic forgettingon-policy distillationself-distilled fine-tuningmodel merginglong-term memoryLLM agentsfailure attributionevaluation benchmarkretrieval-augmented generationagentic RAG

Highlights

ACLArena: diagnosing catastrophic forgetting in multi-stage agent post-training
This 2026 arXiv paper benchmarks agent continual learning across post-training stages, comparing mitigation paradigms including on-policy distillation, self-distilled fine-tuning, and model merging. Evolve's verdict: strong failure diagnosis and paradigm comparison, with the caveat that the setting depends on task-ID routing and validation is limited to 8B-scale models; reproducibility is rated medium.
EvalMem: operation-level attribution of long-term memory failures
EvalMem supplies a diagnostic framework that routes each long-term-memory failure to its responsible operation — encoding, retrieval, or generation — via three Examiners. Evolve finds the validation thorough and the retrieval-deficiency finding convincing, while observing that the measured gains remain limited; reproducibility is rated medium.

Timeline

2026EvalMem, an operation-level diagnostic framework for long-term memory systems, appears on arXiv with Wei Wang as last author.
2026ACLArena, on agent continual learning in multi-stage post-training, appears on arXiv with Wei Wang as last author.
2026The OpenAlex snapshot records 4 works for this author in 2026, alongside 25 distinct coauthors overall.

Papers in the Evolve catalog

Frequent co-authors

Zeyu Liu×1 Jian Zhong×1 Rongduo Han×1 Ziyang Wu×1 Shunye Tang×1 Chenghao He×1 Yaxuan Yang×1 Yihang Qiu×1 Ailing Wang×1 Xiao Liang×1 Guohuan Xie×1 Xiaokang Xue×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.

'Wei Wang' is among the most common researcher names; this profile is name-resolved from public bibliographic data (OpenAlex identity) and may merge or split distinct individuals.
Bibliometrics come from OpenAlex only; no ORCID is linked, and the output and impact axes are marked missing because citation indices are recorded as unpopulated zeros.
The OpenAlex topic spread — including pulsars and gravitational waves alongside LLM topics — is inconsistent with the two catalog papers and likely reflects name-ambiguity noise.
Both catalog entries are 2026 arXiv preprints without peer-reviewed venue data; affiliation is listed as UCLA with country recorded as HK, as given in the source record.