NeurIPS 2026
Zero-Violation Regret for Cooperative Markov Games with Coupled Instantaneous Hard Constraints.
View publication record →
Assistant Professor · Electrical and Computer Engineering · University at Buffalo
Ming Shi Learning and optimization for reliable networked intelligence.
Theory and algorithms for systems that must learn safely, adapt economically, coordinate across components, and remain useful under partial information and change.
My group works across reinforcement learning, online optimization, bandits, multi-agent systems, edge AI, security, and autonomous systems. We seek guarantees that connect what can be learned to what can be executed in the real world.
Recent updates
New results, student achievements, and university service.
ACM MobiHoc AIoT 2026
Fresh Enough to Decide: Age of Intelligence for Cost-Aware Model Synchronization in Dynamic IoT Systems.
Read the paper →WiOpt Best Paper Runner-Up
Jialei Liu led our work on bi-level provisioning and scheduling with switching costs and cross-level constraints.
Read the paper →ECE Tenure and Promotion Committee
Serving as a member of the University at Buffalo ECE Tenure and Promotion Committee.
View professional service →Research agenda
Full research program →Reliable intelligence requires more than accurate prediction.
It requires deciding what information to acquire, which uncertainty to trust, when to update or switch, how to coordinate across agents, and how to preserve safety and service while learning.
Central question
How should an intelligent system learn, communicate, update, and act when information is incomplete, environments change, actions interact, and every adaptation has operational consequences?
regretcompetitive ratiozero violationinformation limitsrobustness
01
Safe & verifiable learning
Instantaneous hard constraints, viability, non-convex safety, and coupled multi-agent certification.
Explore direction → 02Prediction & costly adaptation
Online convex optimization, switching and ramp costs, lookahead, bi-level control, and best-of-both-worlds guarantees.
Explore direction → 03Partial & human information
POMDPs, active probes, experience sharing, multi-objective learning, and imperfect preferences.
Explore direction → 04Distributed & multi-agent intelligence
Graph-structured coordination, cloud–edge–device learning, communication, and shared constraints.
Explore direction → 05Dynamic, secure & robust AI
Nonstationarity, adversarial learning, Age of Intelligence, cyber defense, and model robustness.
Explore direction →Selected recent work
All publications →Results that connect mathematical structure to operational consequences.
Representative publications span safe learning, partial observability, costly adaptation, edge intelligence, and student-led systems research.
From theory to practice
Four questions organize the path from a mathematical model to a reliable system.
What is observable?
State information, forecasts, probes, preferences, communication, and their physical or statistical costs.
What is admissible?
Instantaneous safety, ramp constraints, cross-component coupling, safe continuation, and resource feasibility.
What should adapt?
Actions, policies, model versions, schedules, communication patterns, and cloud–edge–device placement.
What can be guaranteed?
Zero violation, regret, competitive ratio, information limits, robustness, and scalable dependence on system structure.
Application domainsrobotics and autonomous systemswireless and edge AIdata centers and NFVcyber and LLM securityquantum networkingfoundation models
Mentoring, teaching, and outreach
Build theory deeply, explain it clearly, and connect it to systems people can see.
Students learn to move between real-world questions, mathematical abstractions, proofs, algorithms, experiments, and public communication.
Honors, talks, and leadership
Research recognition and engagement with the broader community.
Collaborate or join the group
Research at the intersection of learning theory, optimization, and intelligent systems.
I welcome conversations with prospective students, collaborators, educators, and partners interested in rigorous and deployable intelligent decision-making.