Mengwei Yang

Mengwei Yang

PhD candidate at University of California, Irvine

Biography

I am a PhD candidate in the department of Electrical Engineering at the University of California, Irvine, advised by Prof. Athina Markopoulou. I received my B.E. degree in Electrical Engineering from Northeastern University (China) in 2019, my M.Sc. degree in Electrical Engineering from University of Southern California in 2021. I was a visiting student at City University of Hong Kong in 2018, supervised by Prof. Linqi Song.

My research focuses on privacy-preserving machine learning, principled attribution, and agentic AI across edge devices and the cloud. I am a member of the ProperData center and the Engineering+Society Institute on AI.

Resume and Curriculum Vitae

Work Experience

  • Company Logo Syntiant Corp. — Engineering Intern (2022.7 – 2022.9)

  • Company Logo Ericsson Inc. — Data Science Intern (2020.6 – 2021.1)

  • Company Logo UC Irvine EECS Department — Teaching Assistant (Winter and Spring 2024; Winter and Spring 2025)

For detailed Work experience, please visit the Experience

Research Highlights

1. Agentic AI Systems for Voice and Video

  • Multilingual Video Avatar Agent

    AI agent pipeline that transforms input videos into multilingual versions while preserving speaker identity.

  • LLM-Powered Voice Assistant with Voice Cloning

    Interactive voice assistant leveraging LLMs for natural conversations with personalized synthetic voice.

2. Principled Attribution

  • DuoShapley: Adaptive and Scalable Shapley Value Approximation for Federated Learning (📄 Read Paper)

    DuoShapley framework that efficiently balances individual and collaborative user contributions in FL.

  • Maverick-Aware Shapley Valuation for Client Selection in FL (📄 Read Paper)

    Framework that quantifies client contributions when rare or underrepresented classes are present.

3. Model Pruning (Efficiency & Privacy)

  • PriPrune: Quantifying and Preserving Privacy in Pruned FL (📄 Read Paper)

    Privacy-aware pruning algorithm with personalized masks balancing privacy and model performance.

4. Privacy-Preserving Machine Learning

  • Location Leakage in Federated Signal Maps (📄 Read Paper)

    Analysis of gradient leakage attacks on federated signal mapping with defense strategies.

For detailed research experience and complete publication list, please visit the Experience and Publications pages.