I am working on making generative AI efficient enough to run on real devices – LLM/VLM/LVM model quantization, MoE inference, code generation with efficient test-time-scaling, ML compiler optimization, and neural scheduling – so state-of-the-art AI can run faster, smaller, and cheaper at the edge.
I’m with Qualcomm AI Research, San Diego, CA, where I’m now a Sr Staff System Engineer/Manager. I received my Ph.D. degree from the Department of Electronic Engineering, Tsinghua University, Beijing, China, in 2012 and my B.S. degree in electronic engineering from the University of Electronic Science and Technology of China (UESTC), Chengdu, China, in 2007, both with the highest honor. I was a Research Scholar (2009-2011) and a Post-Doctoral Fellow (2013-2014) at Missouri University of Science and Technology (formerly University of Missouri, Rolla, MO).
I was honored to receive the prestigious Qualcomm IP Achievement Award (AI domain) in 2025, multiple QualStar Awards from Qualcomm, the Excellent Doctoral Student Scholarship from the Ministry of Education of China, the Distinguished Honor Graduate award from the Beijing Municipal Commission of Education, and the First-Class Scholarship for graduate students from Tsinghua University. I’m a frequent author and reviewer for top-tier journals/conferences, TPC Co-chair for IEEE Globecom'17, and TPC member for various flagship conferences.
Research Interests:
Calculating the MI for MIMO channels with finite-alphabet inputs can be difficult – the embedded multiple integrals are hard to evaluate accurately, and Monte-Carlo estimation is slow. This code derives a lower bound with no integrals and low computational effort, approximating the MI (with a constant shift) across settings.
Details in: W. Zeng, C. Xiao, and J. Lu, “A Low-Complexity Design of Linear Precoding for MIMO Channels with Finite-Alphabet Inputs,” IEEE Wireless Commun. Letters, vol. 1, no. 1, pp. 38-41, Feb. 2012.
Calculating the AMI under statistical CSI is even harder than MI, since it also averages over the fading channel. This code develops the lower bound and approximation for doubly-correlated MIMO channels.
Details in: W. Zeng, C. Xiao, M. Wang, and J. Lu, “Linear Precoding for Finite-Alphabet Inputs Over MIMO Fading Channels With Statistical CSI,” IEEE Trans. Signal Process., vol. 60, no. 6, pp. 3134-3148, Jun. 2012.
Feel free to reach out by email, or scan the QR code below to add my contact info to your phone.
