Yiming Zhang

Tianjin University of Technology

Unclaimed · AI-structured profileOpenAlex

Early-career researcher spanning LLM-agent reliability, embodied RL environments, and neural-field reconstruction.

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—

Not yet scored: the catalog holds only three 2026 arXiv entries, with no full publication count, venue history, or productivity trend. This axis would measure output volume, venue tier, and year-over-year publication trajectory.

Impact—

Not yet scored: no citation counts, h-index, or downstream usage data exist in the input. This axis would measure citations, influence, and uptake of the three reviewed works by follow-up research.

Focus0/100

Scored 0/100, reflecting maximal dispersion rather than weakness: the field distribution splits exactly into thirds — NLP on LLM agents and RAG, machine learning on LLM/VLM agents and reinforcement learning, and computer vision on 3D reconstruction and novel view synthesis. Across only three 2026 papers, this even three-way spread leaves no dominant field yet; concentration should be re-scored once the publication sample grows.

Collaboration58/100

Scored 58/100 from 17 distinct coauthors across the three 2026 papers. The pattern is team-based arXiv collaboration: one first-authored NLP paper plus middle-author roles on the RL-environment and NeRF projects, with recurring partners such as Jinghong Zhang, Haoran Zhao, and Chunlei Zhao pointing to small working groups rather than one large lab network.

Activity—

Not yet scored: the data covers only 2026 preprints, so publication cadence, submission rhythm, and multi-year trends cannot be computed. This axis would measure annual output rate and sustained research tempo over time.

Reproducibility43/100

Scored 43/100 as the mean repro-readiness across the three reviewed papers: the memory decision controller and DeliveryGym rate medium, while WS-NeRF rates low — Evolve flags roughly 0.1 dB gains without error bars and table anomalies needing checks. The zero-parameter design and verifiable rewards are positives; missing statistical rigor and artifacts hold the score down.

Research summary

Yiming Zhang, affiliated with Tianjin University of Technology, is an early-stage researcher whose 2026 arXiv output spans three areas in near-equal measure: natural language processing on LLM agents and retrieval-augmented generation, machine learning on LLM/VLM agents and reinforcement learning, and computer vision on 3D reconstruction and novel view synthesis — the field distribution sits at exactly one third each, which drives the focus score to 0/100. The clearest signal of independent leadership is a first-authored interpretable memory decision controller for LLM agents: a zero-parameter post-retrieval trust gate that decouples confidence and consistency to curb RAG memory-conflict hallucination; Evolve credits its efficiency and candor while noting that wins ride on risk-triggered abstention. As a middle author, Zhang contributed to DeliveryGym, a courier-shift RL environment whose GRPO training lifts Qwen3-VL-4B test income by 54.3% (adaptive curriculum +18% versus uniform, one model trained), and to WS-NeRF, a Mamba-gated deblurring NeRF whose reported ~0.1 dB gains over DeepDeblurRF lack error bars and include table anomalies flagged for checking. Reproducibility readiness averages 43/100 across the three reviewed papers, with 17 distinct coauthors supporting a collaboration score of 58/100.

Collaboration pattern

All three 2026 papers are team-based arXiv efforts: Zhang is first author on the memory decision controller and a middle author on DeliveryGym and WS-NeRF, with no corresponding-author role recorded. Seventeen distinct coauthors underpin a collaboration score of 58/100; recurring partners include Jinghong Zhang, Haoran Zhao, Chunlei Zhao, Haoqiang Kang, Yiyang Guo, and Lianhui Qin. The coauthor list also contains another "Yiming Zhang", so name disambiguation within the collaboration graph remains uncertain.

Research keywords

retrieval-augmented generation (RAG)LLM agentsmemory systemshallucination mitigationknowledge conflictreinforcement learning environmentslong-horizon planningadaptive curriculumneural radiance fieldsimage deblurringnovel view synthesisstate-space models (Mamba)

Highlights

Zero-parameter trust gate against RAG memory conflict
First-authored interpretable memory decision controller uses three-signal complementarity to decouple confidence and consistency, curbing memory-conflict hallucination in LLM agents; Evolve credits its efficiency and candor while noting the wins ride on risk-triggered abstention.
RL environment with verifiable rewards lifts embodied planning
Co-authored DeliveryGym, a courier-shift RL environment with verifiable rewards for long-horizon agent planning; GRPO training raises Qwen3-VL-4B test income by 54.3%, and adaptive curriculum adds 18% over uniform scheduling from a single trained model.
Mamba-gated deblurring NeRF with open reproducibility questions
Co-authored WS-NeRF, a world-state-aware adaptive deblurring neural radiance field driven by Mamba gating and mixture-of-experts; reported gains over DeepDeblurRF are about 0.1 dB with no error bars, and Evolve flags table anomalies requiring verification.

Timeline

2026Co-authored DeliveryGym, a long-horizon embodied-agent RL environment with verifiable rewards and adaptive curriculum; GRPO lifts Qwen3-VL-4B test income 54.3%.
2026Co-authored WS-NeRF, a Mamba-driven world-state-aware adaptive deblurring neural radiance field for novel view synthesis.
2026First-authored an interpretable memory decision controller for LLM agents that decouples confidence and consistency to curb RAG memory-conflict hallucination.

Papers in the Evolve catalog

Frequent co-authors

Jinghong Zhang×1 Haoran Zhao×1 Yiren Ma×1 Chunlei Zhao×1 Haoqiang Kang×1 Yiyang Guo×1 Chuying Li×1 Jianzhi Shen×1 Tianruo Rose Xu×1 Xiaokang Ye×1 Lianhui Qin×1 Yiming Zhang×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.

Bibliometrics are drawn only from the provided snapshot: no citation counts, h-index, venue history, or ORCID were available for output, impact, and activity scoring.
"Yiming Zhang" is a common name; this profile is name-resolved from public bibliographic data and the three reviewed arXiv entries, without an ORCID to confirm identity.
All three catalog papers are 2026 arXiv preprints; peer-reviewed venue status and publication outcomes are unknown.
The frequent-coauthor list itself contains a second "Yiming Zhang", so collaboration-graph edges may include same-name noise.