Yang (Kody) Deng

What I do

• AI infrastructure & distributed ML — federated learning across edge networks, reinforcement learning agents that manage live network slices, and ML-driven optimization running inside optical transport systems. Currently building an SLO-aware, multi-tenant LLM inference-serving control plane (vLLM/SGLang on NVIDIA DGX Spark).
• Networked systems engineering — end-to-end 5G/O-RAN stacks (OpenAirInterface, Open5GS, near-/non-RT RIC xApps and rApps), WebRTC/RTP internals down to a modified libwebrtc, edge–cloud architectures, and low-latency delivery for interactive XR.
• Forward deployment — turning research into systems that work outside the lab: a multi-institution NSF project demonstrated live at the University of Michigan, and system demos at OFC and IEEE INFOCOM.

Highlights

• Invented an edge–cloud 3D Gaussian Splatting scene-delivery system that jointly optimizes network- and application-level performance. Approved by the Nokia Patent Board (2025); expert reviewers assessed it as the first XR streaming solution to combine network- and application-layer optimization, with direct relevance to Nokia’s MantaRay RIC and AI-RAN product lines.
• Key contributor to the NSF “Breaking Low” / XRNet program (~$3.5M, with the University of Michigan, USC, Columbia, and Duke): built one of the first open end-to-end implementations of the 3GPP Release-18 PDU Set Marking RTP extension (TS 26.522) inside a production WebRTC stack (Unity + a modified libwebrtc), showed on a hardware testbed that PDU-Set-aware traffic handling keeps interactive XR smooth under impairment that freezes conventional delivery, and demonstrated the complete system live at Michigan in June 2026.
• Built PSASlicing, an SLA-aware reinforcement-learning system for O-RAN slice management (IEEE GLOBECOM 2024), and an ML-driven spectrum defragmentation system demonstrated at OFC 2025.
• PhD in Electrical Engineering (NJIT, 4.0 GPA), dissertation on federated intelligence across edge computing and networking. Author of FedVision (federated video analytics, 45+ citations) and FLEX (blockchain marketplace for federated learning, demoed at INFOCOM 2021).

See Projects for the systems behind these, and Publications for papers, the patent, and professional service.

Background

Before Bell Labs, I completed my PhD at the New Jersey Institute of Technology advised by Tao Han, following research years at UNC Charlotte. I hold a master’s in Control Science and Engineering from Harbin Institute of Technology and a bachelor’s in Automation from South China University of Technology. I serve on technical program committees and review for IEEE/ACM venues including INFOCOM, MobiCom ImmerCom, and IEEE IoT Journal.