NeTS: Medium: Black-box Optimization of White-box Networks: Online Learning for Autonomous Resource Management in NextG Wireless Networks

Synopsis

Next-generation (NextG) wireless networks are anticipated to revolutionize various applications, such as interactive real-time applications like Augmented Reality (AR), while meeting the high Quality-of-Experience (QoE) requirements expected by users. To achieve these goals, NextG networks are undergoing a transformation toward a white-box architecture, characterized by openness, intelligence, and a focus on user needs. Therefore, it is both timely and important to address autonomous resource management within the NextG paradigm. This project aims to facilitate the transition from traditional black-box network designs to a white-box network architecture, which will significantly reduce costs and enhance QoE performance at its core. The findings of this project will be integrated into the curricula of all participating institutions. Furthermore, this project is committed to growing workforce in networking through research opportunities and outreach activities at their respective institutions. Mechanisms will be established to foster research collaboration in an annual high-profile research workshop.

O-RAN is an operator-driven alliance dedicated to the advancement of radio access networks (RAN) toward an open architecture. This research focuses on harnessing the advanced capabilities of O-RAN, with a specific emphasis on edge-assisted low-latency AR as a key use case, to address autonomous resource management in the NextG paradigm. The research employs a data-driven approach across multiple time scales, using Bayesian optimization (BO) as a sample-efficient online learning and black-box optimization tool. The research develops versatile techniques and building blocks to optimize the QoE performance, structured around three interconnected thrusts: (i) developing a provably efficient multi-time-scale data-driven BO framework integrated with O-RAN, (ii) achieving collaborative BO for multi-RAN learning and optimization, and (iii) applying the developed BO frameworks to edge-assisted low-latency AR applications. The research establishes the analytical foundations and algorithmic frameworks that will be integrated with open-source full-stack O-RAN implementations. The evaluation process involves simulations based on 3GPP standards in ns-3, as well as collaborations with industry partners including AT&T, Qualcomm, and Nokia Bell Labs. Real-world trace data and production-grade O-RAN platforms will be leveraged for evaluation purposes. The outcomes of this research not only contribute to advancing knowledge in machine-learning-enabled NextG systems design but also address critical needs within the broader machine learning and networking research communities.

This project is supported by the National Science Foundation (NSF) under Grants CNS-2312833, CNS-2312834, CNS-2312835, and CNS-2312836 from 10/1/2023 to 9/30/2026.

Personnel

Principal Investigators

Collaborators

  • Nakjung Choi (Nokia Bell Labs, USA)

  • Sayak Ray Chowdhury (Microsoft Research, India)

  • Arnob Ghosh (New Jersey Institute of Technology, USA)

  • Ahan Kak (Nokia Bell Labs, USA)

  • Dimitrios Koutsonikolas (Northeastern University, USA)

  • Myungjin Lee (Cisco Research, USA)

  • Bin Li (Pennsylvania State University, USA)

  • Ramanujan Sheshadri (Nokia Bell Labs, USA)

  • Mark Squillante (IBM Research, USA)

Publications

Books and Book Chapters

  1. M. Shi, Y. Liang, N.B. Shroff, “Adversarial Online Reinforcement Learning Under Limited Defender Resource,” In: Chen, Y., Wu, J., Yu, P., Wang, X. (eds) Network Security Empowered by Artificial Intelligence. Advances in Information Security, vol 107, 2024, Springer.

Journal Articles

  1. Fengjiao Li, Xingyu Zhou, and Bo Ji, “Distributed Linear Bandits with Differential Privacy,” IEEE Transactions on Network Science and Engineering (TNSE), accepted, January 2024.

  2. Aleksandrs Slivkins, Xingyu Zhou, Karthik Abinav Sankararaman, Dylan J. Foster, “Contextual Bandits with Packing and Covering Constraints: A Modular Lagrangian Approach via Regression”, Journal of Machine Learning Research (JMLR) 2024.

  3. A. Kar, R. Singh, F. Liu, X. Liu, and N. B. Shroff, “Linear Bandits with Side Observations on Networks,” IEEE/ACM Transactions on Networking, October, 2024.

  4. S. I. Siam, H. Ahn, L. Liu, S. Alam, H. Shen, Z. Cao, N. B. Shroff, B. Krishnamachari, M. Srivastava, and M. Zhang, “Artificial Intelligence of Things: A Survey,” ACM Transactions on Sensor Networks, 2024.

Conference Papers

  1. Duo Cheng, Xingyu Zhou, and Bo Ji, “Taming Heavy-Tailed Losses in Adversarial Bandits and the Best-of-Both-Worlds Setting,” In Proceedings of NeurIPS 2024, Vancouver, Canada, December 2024.

  2. Xiaoyi Wu, Bo Ji, and Bin Li, “On the Low-Complexity of Fair Learning for Combinatorial Multi-Armed Bandit,” In Proceedings of IEEE INFOCOM 2025, London, UK, May 2025.

  3. Z. Luo, J. Liu, M. Lee, and N. B. Shroff, “Prediction-Assisted Online Distributed Deep Learning Workload Scheduling in GPU Clusters,” In Proceedings of IEEE INFOCOM, London, UK, May 2025.

  4. Y. Ahmed, A. Ghosh, C.-C. Wang, and N. B. Shroff, “Communication Efficient Asynchronous Stochastic Gradient Descent,” In Proceedings of IEEE INFOCOM, London, UK, May 2025.

  5. Y. Li, P. Ju, and N. B. Shroff, “How to Find the Exact Pareto Front for Multi-Objective MDPs?,” In Proceedings of ICLR 2025 (spotlight publication), Singapore, April 2025.

  6. M. Shi, Y. Liang, and N. B. Shroff, “Designing Near-Optimal Partially Observable Reinforcement Learning,” In Proceedings of IEEE MILCOM, Oct. 2024.

  7. Z. Jonny Kong, Nathan Hu, Y. Charlie Hu, Jiayi Meng, and Yaron Koral, “High-Fidelity Cellular Network Control-Plane Traffic Generation without Domain Knowledge,” In Proceedings of ACM IMC, November 4-6, 2024.

  8. Imran Khan, Moinak Ghoshal, Joana Angjo, Sigrid Dimce, Mushahid Hussain, Paniz Parastar, Yenchia Yu, Claudio Fiandrino, Charalampos Orfanidis, Shivang Aggarwal, Ana C. Aguiar, Ozgu Alay, Carla-Fabiana Chiasserini, Falko Dressler, Y. Charlie Hu, Steven Y. Ko, Dimitrios Koutsonikolas, and Joerg Widmer. “How Mature is 5G Deployment? A Cross-Sectional, Year-Long Study of 5G Uplink Performance,” In Proceedings of IFIP Networking 2024: 276-284.

  9. Xingyu Zhou and Wei Zhang, “Locally Private and Robust Multi-Armed Bandits”, In Proceedings of NeurIPS 2024.

  10. Z. Jonny Kong, Qiang Xu, and Y. Charlie Hu, “ARISE: High-Capacity AR Offloading Inference Serving via Proactive Scheduling,” In Proceedings of ACM MobiSys 2024, Tokyo, Japan, June 2024.

  11. Omar Basit, Phuc Dinh, Imran Khan, Z. Jonny Kong, Y. Charlie Hu, Dimitrios Koutsonikolas, Myungjin Lee, and Chaoyue Liu, “On the Predictability of Fine-grained Cellular Network Throughput using Machine Learning Models,” In Proceedings of IEEE MASS 2024, Seoul, South Korea, September 2024.

  12. Xingyu Zhou and Sayak Ray Chowdhury, “On Differentially Private Federated Linear Contextual Bandits,” In Proceedings of ICLR 2024, Vienna, Austria, May 2024.

  13. Arnob Ghosh, Xingyu Zhou, and Ness B. Shroff, “Towards Achieving Sub-linear Regret and Hard Constraint Violation in Model-free RL,” In Proceedings of AISTATS 2024, Valencia, Spain, May 2024.

Broader Impacts

  • PI Shroff and PI Ji, along with several other colleagues, co-organized the 2025 Artificial Intelligence Modeling, Analysis, and Control of Complex Systems (AIMACCS) Workshop, which consists of 15 keynotes, a discussion panel on career advice for students and postdocs, and a student poster session, where students working on this project presented their research results and progress with AI Edge Institute members and other AIMACCS workshop attendees and received useful feedback.

  • PI Ji and PI Zhou continued a close collaboration with the Mobile Network Systems department at Nokia Bell Labs. The main purpose of the collaboration is to explore novel online learning approaches for adaptive end-to-end network slicing.

  • PI Ji served as TPC Co-Chair of the 3rd Workshop on Next-generation Open and Programmable Radio Access Networks (NG-OPERA), co-located with IEEE INFOCOM 2025.

  • PI Ji served as General Co-Chair of Workshop on Modeling and Optimization for Semantic Communications (MOSC), co-located with IEEE/IFIP WiOpt 2024.

  • PI Ji gave invited talks at the 2025 NSF Workshop on Networking and Systems Challenges in Immersive Computing and the sixth Buffalo Day for 5G and Wireless Internet of Things at the University at Buffalo.

  • PI Ji has supervised four undergraduate students to conduct research on AR/VR security and privacy through the Broadening Undergraduate Research Groups (BURGs) in the Department of Computer Science at Virginia Tech. Their research was recognized with the First Prize Research Award at the 2025 VTURCS Research Symposium and the Best Poster Award at the 2025 CCI Student Researcher Showcase.

  • PI Shroff has given numerous keynote and distinguished talks on the use of ML for designing wireless networks.

  • PI Shroff has also served as a panelist at various conferences like ACM Sigmetrics 2025 and IEEE INFOCOM 2025.

  • PI Hu chaired the 2025 ACM SIGMOBILE Test-of-Time award committee.

  • PI Hu supervised an undergraduate student to conduct research on efficiency-efficient serving of Large language model on GPU servers.

  • PI Zhou continued the AI/ML summer camp at WSU for local high school students.

  • PI Zhou has served as a web (co)-chair for MobiHoc 2024/2025.

  • PI Zhou has given a tutorial on recent theoretical advances in RL at Sigmetrics 2025.

  • PI Shroff and PI Ji, along with several other colleagues, co-organized the 2024 Artificial Intelligence Modeling, Analysis, and Control of Complex Systems (AIMACCS) Workshop, which includes a student poster session, where students working on this project presented their research results and progress with AI Edge Institute members and other AIMACCS workshop attendees and received useful feedback.

  • PI Ji and PI Zhou have formed a close collaboration with the Mobile Network Systems department at Nokia Bell Labs. The main purpose of the collaboration is to explore novel online learning approaches for adaptive end-to-end network slicing.

  • PI Hu has collaborated with Cisco Research, to study cellular network throughput prediction at fine time scale needed to support dynamic adaptation of offloading strategies of edge-assisted AR design.

  • PI Ji has supervised three undergraduate students to conduct research through the Undergraduate Research to PhD (UR2PhD) program organized by Computing Research Association (CRA). Their research was recognized with People’s Choice Award at the 2024 VTURCS Research Symposium.

  • PI Zhou has initiated a new AI/ML summer camp at WSU for local high school students.