Peng Wu

Peng Wu

A sci-fi fan, a researcher.

About Me

I am currently a postdoctoral researcher at Northeastern University working with Professor Mahdi Imani. Current research focuses on Mixed Reality attacks, Multi-Agent/Robot collaboration and more, with an emphasis on critical themes such as security, privacy, and trustworthiness. This work intertwines concepts from machine learning, reinforcement learning and federated learning to address complex challenges in these domains.

Prior to this role, I completed my Ph.D. in Electrical Engineering at Northeastern University, where I conducted research under the guidance of Professor Pau Closas. My doctoral work concentrated on advanced deep learning and Bayesian methods within Federated Learning, particularly applied to indoor positioning, Global Navigation Satellite Systems (GNSS), and image processing. Key areas of focus included Bayesian data fusion in Federated Learning, developing WiFi fingerprint-based solutions for indoor positioning and tracking, and enhancing GNSS signal processing to mitigate the risks of spoofing and jamming.

I also co-founded a startup as Chief Scientist, Cactivate, an AI-driven platform that optimizes and automates online advertising, specifically tailored for agencies and e-commerce businesses.

Recent and past Research

In the current AI environment, where the technology’s pervasiveness is undeniable, the vital issues of data privacy, security, and equitable access require immediate attention. My current and future research efforts will focus on these concerns head-on through private, robust, equalable and interpretable Distributed Learning. These algorithms are carefully developed not only to meet the rigorous requirements of real-world applications but also to establish the foundation for Responsible AI—a paradigm in which AI systems are designed with accountability, transparency, and fairness at their core, ultimately resulting in more secure, inclusive, and trustworthy AI implementations.

Bayesian Federated Learning Framework: Compared to FL, Bayesian FL enhances FL by leveraging the benefits of Bayesian inference. By integrating prior knowledge and inferring parameter distributions, this approach effectively captures the intrinsic statistical heterogeneity of FL models, which facilitates the quantification of uncertainties and model dynamics. Consequently, it fosters the development of more robust and interpretable federated models. However, papers combining Bayesian FL with data association mechanisms are notably absent, especially in clustered FL. By introducing this framework, I have laid the groundwork for a myriad of further investigations. The idea centralizes optimizing client-cluster associations within Federated Learning to gracefully handle non-independent and identically distributed (non-IID) data, a commonplace in practical applications. My work delineates a methodical approach for balancing performance with computational complexity, setting the stage for enhanced models and innovative client knowledge-sharing methodologies. One of my papers delving into this matter is currently undergoing review for In Conference on Machine Learning (ICML) 2024, and continuous efforts are underway to develop this concept further, such as incorporating a random number of clusters for optimization.

Robust Bayesian Data Fusion Framework: The designed framework optimizes fusion accuracy while honoring privacy constraints and offers profound insights into distributed learning. Presenting a meticulous analysis of shared priors impact within Bayesian data fusion settings, published in Transactions on Signal Processing extends our understanding of distributed frameworks under a Bayesian lens, providing interpretability for Distributed Learning and Federated Learning systems.

Data Driven Machine Learning: Throughout my academic journey, I have undertaken several machine learning projects that focus on indoor positioning, integrating machine learning methodologies and tracking algorithms to refine WiFi fingerprinting and tracking systems. This work has culminated in three peer-reviewed publications. In parallel, my endeavors in GNSS applications have led to groundbreaking data-driven methodologies that eclipse traditional strategies, particularly evident in GNSS signal acquisition, deep learning-based spoofing detection, and intricate modeling of multipath effects on GNSS correlation outputs. These collaborative efforts have resulted in over six published papers.

Application to Real-world Problems: (1) Jammer Signal Classification: Addressing threats to GNSS operations, this application of Federated Learning for jamming signal classification safeguards data privacy while training robust, accurate classifiers. This groundbreaking work, which claimed the best paper award at ION PLANS 2023, takes significant strides in privacy-preserving collaborative learning. (2) Indoor Localization: I have explored personalized Federated Learning and model reliability with dropout to tackle the challenges of non-IID data for indoor localization, with outcomes indicating significant improvements over traditional techniques, charting a path closer to centralized learning effectiveness.

Publications

  • Borhani-Darian, Parisa, Li, Haoqing, Wu, Peng, and Closas, Pau. (2024). Detecting GNSS spoofing using deep learning. EURASIP Journal on Advances in Signal Processing, 2024(1), 14. Springer International Publishing Cham.

  • Wu, Peng, Imbiriba, Tales, and Closas, Pau. (2023). A Bayesian Framework for Clustered Federated Learning.

  • Li, Haoqing, Tang, Shuo, Wu, Peng, and Closas, Pau. (2023). Robust Interference Mitigation Techniques for Direct Position Estimation. IEEE Transactions on Aerospace and Electronic Systems. IEEE.

  • Wu, Peng, Calatrava, Helena, Imbiriba, Tales, and Closas, Pau. (2023). Jammer classification with federated learning. 2023 IEEE/ION Position, Location and Navigation Symposium (PLANS), 228–234. IEEE.

  • Darian, Parisa Borhani, Li, Haoqing, Wu, Peng, and Closas, Pau. (2023). Detecting GNSS spoofing using deep learning.

  • Borhani-Darian, Parisa, Li, Haoqing, Wu, Peng, and Closas, Pau. (2023). Deep learning of GNSS acquisition. Sensors, 23(3), 1566. MDPI.

  • Wu, Peng, Imbiriba, Tales, Elvira, Víctor, and Closas, Pau. (2023). Bayesian data fusion with shared priors. IEEE Transactions on Signal Processing. IEEE.

  • Park, Junha, Moon, Jiseon, Kim, Taekyoon, Wu, Peng, Imbiriba, Tales, Closas, Pau, and Kim, Sunwoo. (2022). Federated learning for indoor localization via model reliability with dropout. IEEE Communications Letters, 26(7), 1553–1557. IEEE.

  • Li, Haoqing, Borhani-Darian, Parisa, Wu, Peng, and Closas, Pau. (2022). Deep neural network correlators for GNSS multipath mitigation. IEEE Transactions on Aerospace and Electronic Systems, 59(2), 1249–1259. IEEE.

  • Wu, Peng, Imbiriba, Tales, Park, Junha, Kim, Sunwoo, and Closas, Pau. (2021). Personalized federated learning over non-iid data for indoor localization. 2021 IEEE 22nd International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), 421–425. IEEE.

  • Li, Haoqing, Borhani-Darian, Parisa, Wu, Peng, and Closas, Pau. (2020). Deep learning of GNSS signal correlation. Proceedings of the 33rd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2020), 2836–2847.

  • Borhani-Darian, Parisa, Li, Haoqing, Wu, Peng, and Closas, Pau. (2020). Deep neural network approach to detect GNSS spoofing attacks. Proceedings of the 33rd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2020), 3241–3252.

  • Imbiriba, Tales, Wu, Peng, LaMountain, Gerald, Erdoğmuş, Deniz, and Closas, Pau. (2020). Recursive Gaussian processes and fingerprinting for indoor navigation. 2020 IEEE/ION Position, Location and Navigation Symposium (PLANS), 933–940. IEEE.

  • Imbiriba, Tales, LaMountain, Gerald, Wu, Peng, Erdoğmuş, Deniz, and Closas, Pau. (2019). Change detection and Gaussian process inference in piecewise stationary environments under noisy inputs. 2019 IEEE 8th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP), 530–534. IEEE.

  • Wu, Peng, Imbiriba, Tales, LaMountain, Gerald, Vilá-Valls, Jordi, and Closas, Pau. (2019). Wifi fingerprinting and tracking using neural networks. Proceedings of the 32nd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2019), 2314–2324.