Xubin Wang's Site

Welcome to my personal website, which serves as a platform for self-presentation and communication with others. Here, you can learn about my research progress in Intelligent Computing and other areas of interest. If you have any suggestions, please feel free to let me know.

About Me Research Works

About me

Dr. Wang's Cute Cat Avatar

Research Focus and Role-Relevant Direction

I work on foundation models, agent systems, and data-centric optimization, with a practical goal: convert model capability into reliable product value under real-world constraints (quality, latency, memory, privacy, cost). My work studies how better selection mechanisms across data, prompts, and deployment paths can improve both model reasoning and system efficiency.

Technically, I combine reinforcement learning, evolutionary optimization, and robust representation learning to build an end-to-end stack from data curation to LLM/agent deployment. This includes: (1) demonstration selection for in-context learning (RDES, ICML 2025), (2) lightweight meta-selection for scalable prompting (Meta-Sel, 2026), (3) robust data-centric pipelines for imbalanced and noisy structured data (MEL/EODE/HSNOE/GMR), and (4) system-level optimization for on-device and edge intelligence (On-Device AI, CSUR 2025; Cognitive Edge Computing, 2026).

For frontier foundation-model and agent roles, I position this foundation into four capability directions: data engine quality (selection/filtering for reliable training and evaluation sets), post-training and reasoning efficiency (prompt/demo optimization), agent reliability (tool-use orchestration and controllable autonomy), and evaluation and deployment (joint measurement of quality, efficiency, and safety trade-offs). Publications appear in ICML, TKDE, TCBB, CSUR, KBS and related venues.

Contact: xubin.wang [at] kindlab.site · 欢迎合作 / Collaboration welcome.

Selection-Centric AI: Framework, Evolution, and Capability Translation (2022-2026)

This framework organizes all studies (2022-2026) into one systematic line: build resource-constrained intelligence through selection. The trajectory evolves from data/feature optimization to LLM and agent intelligence, and translates into role-relevant capabilities: data engine quality, reasoning efficiency, agent orchestration, and reliable evaluation/deployment.

Unified Framework

Systematic research framework map from 2022 to 2026

Evolution and Value Trajectory

Research evolution map labeled by venue and year

Systematic Progression (Method to Capability to Value)

Stage I · 2022
Data and Feature Selection

FWPSO and SaWDE establish search-space reduction, supporting efficient data filtering and curation.

Stage II · 2024
Robust Data-Centric Learning

MEL/EODE/HSNOE improve transferability and robustness, enabling higher-quality training/eval data pipelines.

Stage III · 2025
LLM Prompt and Context Optimization

RDES and Meta-Sel target low-cost, high-quality reasoning through better demonstration selection.

Stage IV · 2026
Agentic and Deployable Intelligence

On-device and Cognitive Edge studies connect model quality, system constraints, and reliable deployment.

Capability Mapping for Frontier Model Research and Product Roles

Capability A
Data Engine and Quality Control

Selection, denoising, and imbalance-aware data processing for robust training and evaluation sets.

Capability B
Post-Training Reasoning Efficiency

Prompt/demo optimization to improve reasoning quality under strict latency and token budgets.

Capability C
Agent Reliability and Tool Use

From single-model inference to controllable multi-agent workflows for practical task completion.

Capability D
Evaluation and Deployment Trade-offs

Joint optimization of quality, latency, memory, privacy, and operational safety in real-world systems.

Research Topics

The research program is organized into two core directions with clear translational value for frontier model and agent product work: Large Language Models, Agents and Their Applications and Data-Centric AI and Its Applications.

  • Large Language Models, Agents and Their Applications

    LLM and agent research topic map with methods and applications

    Core Focus: Build efficient and deployable intelligence for LLM-based reasoning and autonomous agent collaboration under real-world constraints.
    Role-Relevant Translation: inference-time optimization, post-training style selection, controllable tool-use, and robust agent execution.
    Representative Works: [RDES, ICML 2025] [Cognitive Edge Computing, 2026] [On-Device AI, CSUR 2025] [Meta-Sel, arXiv 2026]
    Application Value: intelligent text annotation, resource-aware in-context reasoning, and edge-side assistants/agents with better cost-latency-privacy trade-offs.

  • Data-Centric AI and Its Applications

    Data-centric AI research topic map with robust learning and science applications

    Core Focus: Improve learning reliability by selecting and refining informative features and samples from high-dimensional, noisy, and imbalanced data.
    Role-Relevant Translation: data quality engineering, benchmark robustness, and stable generalization for model development and evaluation.
    Representative Works: [SaWDE, KBS 2022] [FWPSO, BIBM 2022] [MEL, TKDE 2024] [EODE, TCBB 2024] [HSNOE, ESWA 2024] [GMR, arXiv 2026]
    Application Value: robust biomedical data processing, biomarker/pathway discovery, and stable decision support for imbalanced structured learning tasks.

Selected Publications

    Full list of publications can be found at:
  • 1
    Demonstration Selection for In-Context Learning via Reinforcement Learning
    Xubin Wang, Jianfei Wu, Yichen Yuan, Deyu Cai, Mingzhe Li, Weijia Jia
    Forty-Second International Conference on Machine Learning (ICML), 2025 CCF A
  • 2
    Empowering Edge Intelligence: A Comprehensive Survey on On-Device AI Models
    Xubin Wang, Zhiqing Tang, Jianxiong Guo, Tianhui Meng, Chenhao Wang, Tian Wang, Weijia Jia
    ACM Computing Surveys, 57(9):1-39, 2025 SCI I
  • 3
    MEL: Efficient Multi-Task Evolutionary Learning for High-Dimensional Feature Selection
    Xubin Wang, Haojiong Shangguan, Fengyi Huang, Shangrui Wu, Weijia Jia
    IEEE Transactions on Knowledge and Data Engineering, 36(08):4020-4033, 2024 CCF A
  • 4
    Exhaustive Exploitation of Nature-inspired Computation for Cancer Screening in an Ensemble Manner
    Xubin Wang, Yunhe Wang, Zhiqiang Ma, Ka-Chun Wong, Xiangtao Li
    IEEE/ACM Transactions on Computational Biology and Bioinformatics, 21(5):1366-1379, 2024 CCF B
  • 5
    Evolving Pathway Activation from Cancer Gene Expression Data using Nature-inspired Ensemble Optimization
    Xubin Wang, Yunhe Wang, Zhiqiang Ma, Ka-Chun Wong, Xiangtao Li
    Expert Systems with Applications, 248:123469, 2024 SCI I
  • 6
    A self-adaptive weighted differential evolution approach for large-scale feature selection
    Xubin Wang, Yunhe Wang, Ka-Chun Wong, Xiangtao Li
    Knowledge-Based Systems, 235:107633, 2022 SCI I
  • 7
    A Feature Weighting Particle Swarm Optimization Method to Identify Biomarker Genes
    Xubin Wang, Weijia Jia
    2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 830-834, 2022 CCF B

Contact

Email xubin.wang [at] kindlab.site