Curriculum Vitae

Teng Li

Ph.D. Candidate in Electrical and Computer Engineering

University of Illinois Chicago · Chicago, Illinois

Download PDF LinkedIn Demo & GitHub

Profile

Professional Summary

Ph.D. candidate in Edge AI and distributed learning and inference for resilient, communication-efficient ML systems. Research focuses on wireless and decentralized network optimization and resource scheduling. Deployed vision models and LLMs with Python/PyTorch on NVIDIA Jetson, developed RF signal detection systems, and optimized GPU inference latency and throughput with ONNX Runtime / TensorRT. Wireless systems experience includes O-RAN, OpenAirInterface (OAI), and SDR (USRP).

Quick Profile

At a Glance

Education

University of Illinois Chicago · Ph.D. Candidate, ECE

2019–Expected December 2026 · GPA 4.0/4.0

Research Interests: Distributed Learning · Edge AI · Network Optimization · Resilient Systems · LLMs on Edge Devices

University of Illinois Chicago · M.S., ECE

2017–2019 · GPA 4.0/4.0

Concentration: Computer Communication Networks · Deep Learning · Wireless Communications · Computer Architecture

Northeastern University (China) · B.Eng., Automation

2013–2017 · GPA 80.8/100

Concentration: Automation Control Systems · Intelligent Optimization · Embedded Systems

Skills

Programming
Python, C/C++, MATLAB, Java, Bash
ML & Deployment
PyTorch, TensorFlow, Hugging Face Transformers, ONNX / ONNX Runtime, TensorRT
Platforms & Tools
NVIDIA Jetson SDK, Linux, Docker, Git, LaTeX, sockets, HTTP, Wi-Fi, O-RAN, OpenAirInterface (OAI), SDR (USRP)
Languages
Mandarin Chinese (Native), English (Proficient)

Research Projects

PA-MDI · Distributed Inference

Partitions and allocates models by task priority for faster inference with resilience to node mobility and dynamic networks. Tested on Jetson and Colosseum.

IEEE ICMLCN 2025

REDIT · Distributed Speech Inference

Uses text summarization as redundancy to reduce distributed text-to-speech delay when devices are slow. Tested on Jetson Nano.

IEEE Globecom 2022

Communication-Aware MDI · Activation Compression

Dynamic activation compression meets throughput targets missed by uncompressed inference, improves accuracy over baseline compression policies, and reduces delay on Jetson. I implemented compressors and T5 and ResNet inference on NVIDIA Jetson, with testing and evaluation across datasets.

ACM MobiHoc 2026 · Accepted

CEDAR-YOLO / CEDAR-MDI · Wireless Signal Detection

Detects radar/LTE signals in spectrograms with a compact YOLO detector and distributed inference for higher throughput than local execution while preserving overlap-detection quality. Tested on Jetson Orin Nano. Separately explored local GPU deployment with ONNX Runtime / TensorRT workflows for multiple model configurations, comparing latency, throughput, and detection accuracy.

under submission

Demo & GitHub Research Details ↓

Academic Background

Education

Sept 2019 –
Expected December 2026Chicago, Illinois

University of Illinois Chicago

Ph.D. Candidate in Electrical and Computer Engineering

  • Advisor: Prof. Hulya Seferoglu
  • GPA: 4.0 / 4.0
  • Research Interests: Distributed Learning, Edge AI, Network Optimization, Resilient Systems, LLMs on Edge Devices.
Aug 2017 – Apr 2019Chicago, Illinois

University of Illinois Chicago

M.S. in Electrical and Computer Engineering

  • GPA: 4.0 / 4.0
  • Concentration: Computer Communication Networks, Deep Learning, Wireless Communications, Computer Architecture.
  • Key Coursework: Advanced Microprocessor Design, Neural Networks, Digital Image Processing, Pattern Recognition, Computer Network, Robotics Algorithms & Control, Wireless Communication.
Aug 2013 – May 2017Shenyang, China

Northeastern University (China)

Bachelor of Engineering in Automation · GPA: 80.8/100

  • Honors: The College-level Scholarship (2017)
  • Concentration: Automation Control Systems, Intelligent Optimization, Embedded Systems.

Experience

Research Experience

First Author · IEEE ICMLCN 2025Networking Research Lab, UIC

Priority-Aware Model-Distributed Inference at Edge Networks

Demo
  • Formulated and solved an optimization problem for distributed inference in multi-source edge networks, allocating models according to task priorities.
  • Designed the Priority-Aware Model-Distributed Inference (PA-MDI) algorithm to partition and allocate models across devices. Incorporated task failure probabilities into model allocation for resilience to node mobility and dynamic network changes.
  • Evaluated ResNet-50 and ResNet-56 on NVIDIA Jetson Xavier and Nano edge testbeds and GPT-2 on the Colosseum testbed. PA-MDI optimized model partitioning and allocation based on task priorities and reduced inference time versus baselines.
First Author · IEEE Globecom 2022Networking Research Lab, UIC

REDIT: Resilient Distributed Text-to-Speech at Edge Networks

  • Designed and implemented REDIT (REsilient DIstributed TTS) for deep learning based text-to-speech (TTS), using text summarization as redundancy for straggler resilience.
  • Distributed computationally intensive TTS tasks across multiple resource-constrained devices in edge networks.
  • Implemented and evaluated the framework on an NVIDIA Jetson Nano Developer Kit testbed, demonstrating reduced task completion delay compared with distributed TTS without resiliency.
Co-author · ACM MobiHoc 2026 (Accepted)Networking Research Lab, UIC

Communication-Aware Model-Distributed Inference via Latent Representation Compression

GitHub Repo
  • I explored and implemented activation compressors and implemented, tested, and evaluated T5 (Flan-T5) and ResNet inference on NVIDIA Jetson across datasets.
  • The paper presents a communication-aware model-distributed inference framework for ResNet, Flan-T5, Gemma, and Llama under dynamic edge network conditions.
  • The framework dynamically adjusts activation compression using Top-k sparsification, quantization, and LLM.int8 to reduce communication overhead and optimize the accuracy-throughput trade-off.
  • The paper evaluates the framework on Raspberry Pi, NVIDIA Jetson, and GPU-based edge testbeds. Experiments met throughput targets missed by uncompressed inference and achieved higher accuracy than baseline compression policies; Jetson tests also reduced inference delay at comparable accuracy.
First Author · under submission

YOLO-Based CBRS Detection and Distributed Edge Inference (under submission)

  • Built a CBRS detection pipeline using spectrogram-based object detection to identify and localize radar and LTE signals from raw radio data.
  • Developed CEDAR-YOLO, an improved YOLO detector that simplifies the model architecture and reduces model size and activation-transfer volume, enabling flexible partitioning and faster edge inference while preserving detection quality.
  • Designed CEDAR-MDI, a model-distributed inference framework that accelerates detection across edge devices by optimizing model partitioning, node assignment, and activation compression.
  • Implemented and evaluated the system on a multi-node NVIDIA Jetson Orin Nano testbed, demonstrating higher inference throughput than local execution while maintaining radar/LTE overlap-detection quality.

Separately explored local GPU inference deployment using the CBRS models, developing ONNX Runtime and TensorRT workflows for desktop NVIDIA GPUs and Jetson. Built a configurable benchmark pipeline for multiple model configurations, evaluating FP32/FP16 inference and calibrated post-training INT8 quantization against PyTorch baselines on latency, throughput, and detection accuracy.

Research Output

Publications

  1. T. Li, H. Seferoglu. “Priority-Aware Model-Distributed Inference at Edge Networks.” IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), Barcelona, Spain, May 2025.
  2. T. Li, H. Seferoglu, E. Koyuncu. “REDIT: Resilient Distributed Text-to-Speech at Edge Networks.” IEEE Global Communications Conference (Globecom), Rio de Janeiro, Brazil, Dec. 2022.
  3. P. Gholami, T.-T. Davarakis, T. Li, et al. “Communication-Aware Model Distributed Inference via Latent Representation Compression.” ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc), 2026. (Accepted)

Instruction

Teaching Experience

Teaching AssistantChicago, Illinois

Department of ECE, UIC

  • ECE 568 Advanced Microprocessor Architecture: Instruct on advanced hardware acceleration, like GPU, TPU, trusted processor, and SMT design.
  • ECE 466 Computer Architecture: Cover instruction set architecture, pipelining, instruction-level parallelism, and memory hierarchy.
  • ECE 417 Digital Signal Processing II: Assist with digital filter design, FFT algorithms, and multirate signal processing.
  • ECE 265 Introduction to Logic Design: Lead labs on Boolean algebra, logic gates, and sequential circuit synthesis.

Additional Work

Selected Projects

Northeastern UniversityMar 2026 · Boston, Massachusetts

Open6G+AI Hackathon: 5G Multi-Service Scheduler

  • Co-designed and implemented a 5G MAC scheduler via Lua in a six-member team on a physical O-RAN testbed, balancing high eMBB throughput with low latency and packet loss for URLLC UEs.
  • Deployed and tested an end-to-end OTA 5G network on Kubernetes (OpenShift) with OAI gNB, commercial O-RAN RU/5G UEs, and Open5GS core.
NASA Space Apps Challenge 2022Oct 2022 · Chicago, Illinois

DASPA (Delta Socioeconomic Predictor Application)

Developed a Python/PyTorch web application using Neural Networks to predict socioeconomic indicators from NASA Delta-X datasets.

Course ProjectDec 2017 – Mar 2018 · Chicago, Illinois

Music Selector Based on Image Recognition Features

Led a team to implement a deep learning-based music recommendation system that maps user facial features to song selections.

Graduation ProjectDec 2016 – May 2017 · Shenyang, China

Research on Containment Control for Linear Multi-Agent Systems

Designed and simulated a state feedback control algorithm for linear multi-agent systems, proving stability via Lyapunov principles.

Service

Academic Service

  • Peer Reviewer, IEEE Military Communications Conference (MILCOM), 2025, 2026.
  • Peer Reviewer, IEEE Global Communications Conference (Globecom), 2025, 2026.
  • Peer Reviewer, IEEE Transactions on Mobile Computing (TMC), 2024.

Industry

Professional Experience

June 2018 – July 2018Zhengzhou, China

Henan Xin Xin Lan Network Technology Co., Ltd.

Java Back-end Developer

Contributed to Java back-end development for information system upgrades and performed maintenance on Linux servers.

Toolkit

Skills

Programming

Python, C/C++, MATLAB, Java, Bash

ML & Deployment

PyTorch, TensorFlow, Hugging Face Transformers, ONNX / ONNX Runtime, TensorRT

Platforms & Tools

NVIDIA Jetson SDK, Linux, Docker, Git, LaTeX, sockets, HTTP, Wi-Fi, O-RAN, OpenAirInterface (OAI), SDR (USRP)

Languages

Mandarin Chinese (Native), English (Proficient)