Jeongmin Bae, Ph.D

Enabling Large AI for Real-Time Network Control
  Large AI models, including large language models, offer strong representation and generalization capabilities across diverse network environments. Yet their inference latency, computational cost, and energy demands make direct use impractical for control tasks that operate at millisecond timescales, particularly wireless resource management.
  My research explores system architectures that reconcile the broad capabilities of large AI with the stringent timing and efficiency requirements of real-time network control. The goal is to bring high-capacity AI support to frequent control loops without placing the full model on the latency-critical path.

Runtime Learning for Robust Network Control
  Learning-based network control can capture complex relationships between network conditions and control decisions that are difficult to encode in handcrafted rules. Yet its performance often degrades outside the training range, while covering the full range of possible operating conditions through offline training is rarely practical.
  My research investigates runtime learning mechanisms that expand a controller’s operating range as new conditions arise, enabling reliable performance without repeated manual data collection, retraining, and redeployment. The central challenges are to determine when additional learning is needed and to obtain useful training experience without significantly degrading ongoing network performance. I study these challenges in fundamental and technically demanding network-control problems, including wireless resource management and congestion control.

In-Network Adaptation for Timely Multimedia Delivery
  Computationally intensive AI services often offload multimedia processing to remote edge or cloud servers. In applications such as visual analytics, immersive media, and interactive rendering, content must be transmitted, processed, and returned within a service deadline. Meeting this deadline requires balancing content quality against transmission delay through an appropriate encoding level.
  However, this choice is typically made at the sender, which has limited visibility into downstream network conditions. Its information may be incomplete or outdated, and path conditions may change while the content is already in transit. My research explores in-network adaptation mechanisms that allow selected in-path nodes to dynamically apply additional compression to already encoded content based on current downstream conditions. This enables the quality–delay tradeoff to be refined during delivery, supporting timely multimedia services under dynamic network conditions.

Network Architectures for AI Service Delivery
  Traditional network architectures have limited awareness of the end-to-end requirements and execution characteristics of the services they support. This limitation becomes increasingly important for AI services, whose performance often depends jointly on communication and distributed computation. While networks can observe conditions such as channel quality, queueing, and traffic load, they typically have limited visibility into service-level information such as response-time requirements, computation demands, and processing workflows.
  My research explores service-aware network architectures that bring such information directly into network operation. The goal is to translate changing service requirements into timely network-control actions and coordinate communication and computing resources across the RAN, edge, and cloud, enabling responsive and predictable delivery of AI services.

2024
Gold (1st) Prize, Samsung Humantech Paper Award (CLINE: A Learning-Based Congestion Control that Continually Learns Unseen Network Environments)

2023
Best Presentation Award, A3 Foresight Program Workshop on AI-Based Future IoT Technologies and Services (A Learning-Based Congestion Control that Continually Learns Unseen Network Environments)

2019
Bronze (3rd) Prize, Samsung Humantech Paper Award (Learning to Schedule Network Resources Throughput and Delay Optimally Using Q⁺-Learning)

2015–2017
National Science and Engineering Scholarship (Full Tuition), Korea Student Aid Foundation, Ministry of Education (국가우수장학금 - 이공계, 한국장학재단, 교육부) 

2014
Academic Excellence Award, College of Engineering, Sogang University (Awarded to Top 1% Students of the College of Engineering)

Jeongmin Bae, Harim Kang, Taehun Kim, Taegun An, Seunghyun Lee, Junhao Cai, Changhee Joo, and Kyunghan Lee, “RXC: Runtime Extra Compression for Time-Critical Content Delivery,” IEEE Communications Magazine, under minor revision (IF: 8.3). 

Jeongmin Bae, Junseon Kim, Carlee Joe-Wong⁺, and Kyunghan Lee, “Redefining the Role of Large AI Models in Real-Time RAN Control,” IEEE Communications Magazine, accepted for publication, 2026 (IF: 8.3).

Junseon Kim, Jeongmin Bae, and Kyunghan Lee, “ATC: An Active Cellular Network Architecture for Application Performance Guarantee in 6G,” IEEE Communications Magazine, 2026 (IF: 8.3) (Corresponding author).

Jongyun Lee, Sanghyun Han, Jeongmin Bae, Sangtae Ha⁺, and Kyunghan Lee, “DECENTO: A New Scalable Interactive Live Streaming System via Control Plane Decentralization,” IEEE INFOCOM, 2026 (BK21 Top Conference, 한국정보과학회 최우수학술대회) (Co-corresponding author).

Jeongmin Bae, Joohyun Lee, and Song Chong, “Learning to Schedule Network Resources Throughput and Delay Optimally Using Q⁺-Learning,” IEEE/ACM Transactions on Networking, vol. 29, no. 2, pp. 750–763, 2021 (IF: 5.3).

Jeongmin Bae, Joohyun Lee, and Song Chong, “Beyond Max-Weight Scheduling: A Reinforcement Learning-Based Approach,” IEEE WiOPT, 2019 (KIISE Rank-A Conference, 한국정보과학회 우수학술대회).

“Neural Resource Allocation with Application-Level Performance Guarantee for O-RAN and AI-RAN,” Korean Patent Application No. 10-2025-0066601, filed May 22, 2025 (applied for).

“Lightweight Policy Derivation Method from a Large Model for Real-Time Network Scheduling,” Korean Patent Application No. 10-2025-0005867, filed January 15, 2025 (applied for).

2024–2025
[Principal Investigator] Development of a Continual Learning-Based Real-Time Network Control Framework Adaptable to Changing Environments, NRF (National Research Foundation of Korea), MSIT (Ministry of Science and ICT). 

2024–2028
[Lead Researcher] Next-Generation Communications Cloud Leadership Program, IITP (Institute of Information & Communications Technology Planning & Evaluation), MSIT (Ministry of Science and ICT). 

2020
[Lead Researcher] Adaptive Autonomous Multi-Resource Management Using Meta-Reinforcement Learning, ETRI (Electronics and Telecommunications Research Institute), MSIT (Ministry of Science and ICT). 

2019
[Lead Researcher] Hierarchical Distributed Learning Methods for Data-Centric Distributed Autonomous Networking, ETRI (Electronics and Telecommunications Research Institute), MSIT (Ministry of Science and ICT). 

2018
[Lead Researcher] Learning-Based Network Data Collection Architecture and Resource Management Algorithms, ETRI (Electronics and Telecommunications Research Institute), MSIT (Ministry of Science and ICT). 

2017
[Lead Researcher] Predictive Modeling for Autonomous Operation, ETRI (Electronics and Telecommunications Research Institute), MSIT (Ministry of Science and ICT). 

Technical Program Committee
International Conference on ICT Convergence (ICTC), 2025

Reviewer
IEEE/ACM Transactions on Networking, 2026 - 2023
IEEE Transactions on Mobile Computing, 2026 - 2023
IEEE Transactions on Network Science and Engineering, 2024
Computer Networks, 2026 - 2023

Session Chair
KRnet 2024
KICS Fall Conference 2024