I work on trustworthy AI, with a focus on privacy, safety, and decisions made from incomplete information.
I’m a Postdoctoral Research Associate at Northeastern University, working with Mahdi Imani. I’m also Chief Scientist at AgentMark.ai, where I build AI agents for advertising.
What I’m working on
Bayesian inference connects three parts of my work: learning across different datasets, understanding human state, and helping agents make decisions together.
Learning across devices with different data, personalizing models, and examining what shared information can reveal.
Modeling human state and checking safety properties, from cybersickness to engagement under cognitive attack.
Combining evidence and coordinating exploration while accounting for what other agents already know.
At AgentMark.ai, I build AI agents for ads—connecting campaign data, analysis, content creation, and automation.
I designed and implemented an automated advertising system for managing strategies and data. My work on LLM agents brings conversational interfaces into those workflows, helping connect analysis, advertising management, and content creation.
I’ve also applied deep and federated learning to satellite-navigation interference detection and WiFi indoor localization.
One example: learning to recognize jamming signals across different receivers, without pooling their raw observations.
Robot teams share new evidence and account for teammates’ plans, addressing repeated information and redundant exploration.
Peng Wu, Mohsen Imani, Amidu Kamara, Md Tamzeed Islam, Seyede Fatemeh Ghoreishi, Mahdi Imani
arXiv:2609.17384 · 2026
A Bayesian approach to grouping clients and learning personalized models from heterogeneous data.
Peng Wu, Tales Imbiriba, Pau Closas
IEEE Transactions on Pattern Analysis and Machine Intelligence · 48(3), 3471–3481 · 2026
Bayesian networks connect human-state observations to cybersickness, with probabilistic checks of risk properties within the model.
Peng Wu, Nasim Ahmed, Abhiram Sarma, Kaiming Huang, Rifatul Islam, Bin Li, Tian Lan, Gang Tan, Mahdi Imani
IEEE International Symposium on Mixed and Augmented Reality · 782–792 · 2025
Agents share probabilistic knowledge of their environment to support collaborative learning and decision-making.
Yuxi Wang, Peng Wu, Mahdi Imani
Learning for Dynamics & Control · PMLR 283, 817–829 · 2025
Combining local Bayesian models while accounting for shared prior information, so common evidence is not counted twice.
Peng Wu, Tales Imbiriba, Víctor Elvira, Pau Closas
IEEE Transactions on Signal Processing · 72, 275–288 · 2024
I started in physics and optical engineering before moving into electrical engineering and machine learning. I completed my Ph.D. at Northeastern University in 2024, working with Pau Closas on Bayesian data fusion for distributed learning.
Today, my research at Northeastern includes modeling for a DARPA-funded project on cognitive security in mixed reality, alongside Bayesian methods for multi-agent learning.
The question running through this work: how can distributed systems learn from incomplete information and make decisions safely?
More about my background