Sichuan University · College of Computer Science

Wendian Luo

M.Eng. candidate in Computer Technology

Download CV View Publications
Portrait of Wendian Luo

01 · Profile

About Me

Research interests in Edge Intelligence, Federated Learning, Computer Vision, Large Language Models, and Embodied Intelligence.

Master's candidate in Computer Technology at Sichuan University with research experience in edge intelligence, federated learning, computer vision, and energy-efficient computing.

First author of peer-reviewed work on semi-asynchronous federated prototype learning and low-power real-time scheduling. Experienced in PyTorch-based algorithm development, heterogeneous edge systems, DVFS-aware optimization, and large language model fine-tuning and deployment.

Research Interests

  • Edge Intelligence
  • Federated & Distributed Learning
  • Zero-Shot Industrial Anomaly Detection
  • Energy-Efficient AI & Real-Time Systems
  • LLM Evaluation, Fine-Tuning & Deployment
  • Embodied Intelligence

02 · Academic Path

Education

Sep 2024 · Jun 2027 Expected

M.Eng. in Computer Technology

Sichuan University · College of Computer Science · Chengdu, China

  • GPA: 3.65 / 4.00.
  • First-Class Graduate Academic Scholarship.
Sep 2020 · Jun 2024

B.Eng. in Computer Science and Technology

Sichuan University · College of Computer Science · Chengdu, China

  • GPA: 3.76 / 4.00; GPA ranking: 38 / 318 (top 12%); graduate recommendation ranking: 8 / 318.
  • Outstanding Graduate of Sichuan Province, 2024.

03 · Research

Research Experience

TMSC: Zero-Shot Industrial Anomaly Detection

First Author Open Source

Researcher and Developer · Project Repository ↗

2026

  • Developed a zero-shot anomaly detection pipeline using DINOv3 and CLIP, integrating multi-layer representations, dual-branch prompt adapters, cross-modal attention, and three-region semantic calibration.

FedSAE: Semi-Asynchronous Energy-Efficient Federated Prototype Learning

First Author

First Author and Researcher · Sichuan University

2025

  • Designed a semi-asynchronous end-edge-cloud federated prototype learning framework that exchanges class-level feature statistics instead of raw data or full model parameters.
  • Validated the energy-saving effectiveness of the framework's dynamic DVFS mechanism through experiments on an NVIDIA Jetson TX2 edge platform.

LPSAFS: Low-Power Scheduling with Segmented Frequency Scaling

First Author

First Author and Researcher · Sichuan University

2024

  • Designed a segmented frequency-scaling algorithm based on minimum-slack allocation for periodic real-time tasks under discrete DVFS states, deadline constraints, and reliability requirements.

04 · Scholarly Output

Publications

Journal Articles

  1. Journal Major Revision

    Wendian Luo, Tong Yu, Junnan Li, Shengxin Dai, Bing Guo, and Zhuohang Jiang, “TMSC: A Tri-Regional Mask-Guided Semantic Calibration Framework for Zero-Shot Anomaly Detection,” Pattern Recognition.

  2. Journal

    Wendian Luo, Tong Yu, Shengxin Dai, Bing Guo, Xuesen Lin, and Yanglin Pu, “Semi-Asynchronous Energy-Efficient Federated Prototype Learning for End-Edge-Cloud Architectures,” Future Generation Computer Systems, vol. 179, Art. no. 108351, 2026.

    doi:10.1016/j.future.2025.108351 ↗
  3. Journal

    Wendian Luo, Shengxin Dai, Cheng Dai, Bing Guo, Sherif Moussa, and Mubarak Alrashoud, “BDTest: A Diversity-Oriented Test Case Generation Framework for Deep Neural Networks in 6G-IoT,” IEEE Internet of Things Journal, vol. 13, no. 5, pp. 8190-8203, 2026.

    doi:10.1109/JIOT.2025.3616358 ↗
  4. Journal

    Shengxin Dai, Zheng Luo, Wendian Luo, Siyi Wang, Cheng Dai, Bing Guo, and Xiaokang Zhou, “Energy-Efficient Inference With Software-Hardware Co-Design for Sustainable Artificial Intelligence of Things,” IEEE Internet of Things Journal, vol. 11, no. 24, pp. 39170-39182, 2024.

    doi:10.1109/JIOT.2024.3482288 ↗

Conference Papers

  1. Conference

    Wendian Luo, Shengxin Dai, and Bing Guo, “A Low Power Scheduling Algorithm for Segmented Frequency Scaling,” in 2025 IEEE International Symposium on Parallel and Distributed Processing with Applications (ISPA), pp. 828-837, 2025.

    doi:10.1109/ISPA67752.2025.00112 ↗

05 · Industry

Professional Experience

Jul 2024 · Sep 2024 Internship

Large Language Model Algorithm R&D Intern

Media Intelligence Laboratory, Chengdu Sobey · Chengdu, China

  • Constructed and cleaned English instruction-tuning datasets for Story-to-Scene generation, scene keyword extraction, and shot-level visual description tasks; developed scripts for translation, provenance tracing, deduplication, and format validation.
  • Fine-tuned and evaluated Qwen-72B with LLaMA-Factory and LoRA on an eight-GPU NVIDIA A800 server.
  • Deployed inference services using vLLM, Flask, and HTTP APIs; completed capability integration, API documentation, and project handover.

06 · Selected Work

Selected Projects

OpenClaw-Enabled Mobile Robot Control & Vision-Guided Grasping

System Integration

Project Developer · ROSMASTER M3 PRO

May 2026 · Jul 2026

  • Deployed and configured an OpenClaw-based natural-language robot-control platform integrating MCP/FastMCP with ROS 2, enabling tool-based access to the omnidirectional chassis, 6-DoF arm, and perception modules.
  • Integrated and validated a closed-loop RGB-D tracking-and-grasping workflow spanning visual grounding, mobile-base alignment, 3D grasp-point estimation, inverse-kinematics control, and collision-aware retreat; tuned grasp offsets and diagnosed visual/depth localization errors.

PaddlePaddle-Based 3D Medical Data Analysis Platform

National Second Prize

Primary Developer · 12th China Software Cup

Jun 2023 · Aug 2023

  • Developed the medical platform, 3D imaging workspace, and UI/UX using Vue.js, Vuetify, and VTK.js; contributed to the Python/Django backend and Tencent Cloud COS integration.
  • Integrated PaddlePaddle MedicalSeg into an end-to-end workflow covering medical-image upload, AI-assisted segmentation, 3D visualization, and patient management.

Tingyin: Deep Learning-Based Acoustic Life Detection for Earthquake Rescue

National Innovation Program

Core Team Member · Sichuan University

Oct 2022 · Sep 2023

  • Developed the mobile application, conducted hardware feasibility analysis, and contributed to algorithm deployment and end-to-end system integration.
  • Integrated TSTNN-based speech enhancement and Whisper-based speech recognition; completed the project ahead of schedule and received First Prize in the Sichuan Division of the 2023 Chinese College Students Computer Design Competition.

06.1 · Active Reproductions

Recent Active Reproductions

Hands-on reproduction and deployment attempts across VLA, world models, and large-model inference.

TurboVLA

VLA

Reproduction attempt · Repository ↗

2026 · Jetson Thor 128GB · LIBERO goal / object / spatial

  • Real-time vision-language-action model evaluation at 32 Hz on an RTX 4090 with less than 1 GB VRAM.
  • MuJoCo video visualization is supported.

Isaac Lab on Jetson Thor

VLA / RL

Reproduction attempt · Isaac Lab Repository ↗

2026 · Jetson Thor 128GB · LIBERO goal / object / spatial

  • Isaac Lab is being evaluated as an open-source robotics research framework for reinforcement learning, imitation learning, and motion planning.
  • Isaac Sim cannot load 3D resources on Jetson Thor because of hardware capability limits; the original RL training process remains feasible, with inference visualization available through Newton/Web.

ABot-World-0

World Models

Reproduction attempt · Repository ↗

2026 · Jetson Thor 128GB

  • Interactive generative scenes are running at under 10 FPS in the current test.
  • Jetson Thor does not support SageAttention; FlashAttention 4 works after a small FlashAttention 2 to 4 code adaptation.

MiniMax-H3 on ComfyUI

Large Models

Deployment attempt · MiniMax-H3 ↗ · ComfyUI ↗

2026 · Jetson Thor 128GB · Open-source video generation

  • MiniMax-H3 is being evaluated as an open-source large-model workflow deployed through ComfyUI.
  • Generated video: Bilibili ↗

07 · Toolbox

Technical Skills

Programming

Python C LaTeX

Machine Learning

PyTorch CLIP DINOv3 Federated Learning Zero-Shot Anomaly Detection

Edge AI

NVIDIA Jetson Thor / AGX / Orin DVFS Model Deployment vLLM Ollama

AI Development

Cursor Codex Trae OpenClaw

Languages

Chinese · native English · CET-4 & CET-6 Preparing for IELTS

08 · Recognition

Honors & Awards

  • Outstanding Graduate of Sichuan Province

    Sichuan Province

    2024

  • First-Class Graduate Academic Scholarship

    Sichuan University

    2024 - 27

  • National Second Prize

    12th China Software Cup

    2023

  • First Prize, Sichuan Division

    Chinese College Students Computer Design Competition

    2023

  • Merit and Academic Scholarships

    Sichuan University

    2020 - 24