Mingyu Yang

I am a fifth year PhD student in Electrical and Computer Engineering (ECE) at University of Michigan, supervised by Prof. Hun-Seok Kim. I am working on Machine Learning and Signal Processing.

Before my PhD, I achieved my M.S degree at University of Michigan, majored in Signal & Image Processing and Machine Learning (SIPML). I did my B.S degree from Beijing University of Technology (BJUT) and University College Dublin (UCD), Ireland.

Email  /  CV  /  Google Scholar  /  Github

profile photo
Research

My research interests include machine learning, signal processing, optimization, and computer vision.

prl SAM-guided Pseudo Label Enhancement for Multi-modal 3D Semantic Segmentation
Mingyu Yang, Jitong Lu, Hun-Seok Kim
Under review

Apply the SAM model to aid domain adaptations for multi-modal 3D semantic segmentation

prl Diffusion-Aided Joint Source Channel Coding For High Realism Wireless Image Transmission
Mingyu Yang, Bowen Liu, Boyang Wang, Hun-Seok Kim
Under review
arxiv / code

Apply conditional diffusion models to enable high realism deep JSCC reconstructions

prl Efficient Computation Sharing for Multi-Task Visual Scene Understanding
Sara Shoouri, Mingyu Yang, Zichen Fan, Hun-Seok Kim
ICCV, 2023
arxiv / code

A weight and computation sharing scheme for Multi-Task ViTs that leads to SOTA performace with fewer amount of parameters and FLOPs

prl Efficient Deep Visual and Inertial Odometry with Adaptive Visual Modality Selection
Mingyu Yang, Yu Chen, Hun-Seok Kim
ECCV, 2022
arXiv / code / poster / video / Extention with NAS

An efficient deep visual inertial odometry (VIO) model that adaptively disables the visual modality according to the previous motion state and current IMU reading.

prl Deep Joint Source Channel Coding for Wireless Image Transmission with Adaptive Rate Control
Mingyu Yang, Hun-Seok Kim
ICASSP, 2022
arXiv / code

Deep JSCC with dynamic rate control conditioned on the channel and image contents. Gumbel-Softmax is adopted to make the whole system differentiable.

prl Deep Joint Source Channel Coding for Wireless Image Transmission with OFDM
Mingyu Yang, Chenghong Bian, Hun-Seok Kim
ICC, 2021 (short)
IEEE Transactions on Cognitive Communications and Networking (long)
arXiv / code / extended journal version

End-to-end optimized joint source channel coding system with OFDM for the multi-path fading channel.

prl Migrating Monarch Butterfly Localization Using Multi-Modal Sensor Fusion Neural Networks
Mingyu Yang, Roger Hsiao, Gordy Carichner, Katherine Ernst, Jaechan Lim, Delbert A. Green II, Inhee Lee, David Blaauw, Hun-Seok Kim
EUSIPCO, 2020
Mobicom, 2021 (system paper)
project page / code / paper / system paper

A multi-modal neural network that estimates the location likelihood given light intensity and temperature observations.

prl iLPS: Local Positioning System with Simultaneous Localization and Wireless Communication
Mingyu Yang, Li-Xuan Chuo, Karan Suri, Lu Liu, Hao Zheng, Hun-Seok Kim
INFOCOM, 2019
paper

An indoor localization system using reflective anchors and TDoA estimation, allowing simultaneous location of infinite number of targets with high accuracy.

Course Projects
prl Point Cloud Completion with Transformer
Mingyu Yang, Shiyu Liu, Luya Gao
Course project of EECS 598-012, W21
report

Point cloud completion using self-attention layers

prl Drift-Aware Predictive Coding for Adaptation in Changing Environments
Haozhu Wang, Zhangxing Bian, Rui Guo, Anthony Liang, Mingyu Yang
Course project of EECS 542, F20
report

Drift-aware predictive coding for continuous domain adaptation

Award

BJUT, Best 10 Graduates, 2017

BJUT, China National Scholarship, 2016

BJUT, President Scholarship (10/27000), 2016

Review Service

IEEE journals: WCL, TWC, TCOM, TCCN, TGCN, JSAC, TMC, TSP

Conferences: NIPS'23'24, ICLR'24'25, ICML'24, ICASSP'24'25, CVPR'24, ICRA'24‘25, AAAI'25


The source code is from Jon Barron.