Jin Zhao

Professor
Fudan University
Research

1. Space-Air-Ground Integrated Networks

My research in Space-Air-Ground Integrated Networks focuses on designing resilient and resource-efficient architectures that seamlessly bridge satellite, aerial, and terrestrial domains. As network topologies become increasingly heterogeneous and dynamic, a primary challenge is maintaining service continuity and optimizing resource allocation under strict latency and bandwidth constraints. To address this, I investigate advanced orchestration mechanisms, including service function chain (SFC) placement and migration strategies tailored for LEO satellite environments. I also explore distributed intelligence over satellite networks through split learning and federated learning frameworks, as well as synergistic satellite-ground systems for earth observation and AI applications.

Typical Works:
  1. S-Leon: An Efficient Split Learning Framework Over Heterogeneous LEO Satellite Networks, IEEE TPDS 2026
  2. SatFed: A Resource-Efficient LEO-Satellite-Assisted Heterogeneous Federated Learning Framework, Engineering 2025
  3. Towards Fast and Robust Split Federated Learning over Satellite-Based Computing Networks, DAC 2026
  4. A Satellite-Ground Synergistic Large Vision-Language Model System for Earth Observation, ACM Multimedia 2025
  5. HRA-SFCP: Hybrid Resource-Aware SFC Placement in LEO Satellite Networks, ICCC 2025
  6. Adaptive Service Function Chain Migration in Satellite-Terrestrial Integrated Networks, ICCC 2025

2. Edge AI

My research in Edge AI focuses on making large-scale AI models viable on resource-constrained edge devices by fundamentally rethinking how neural networks consume memory and compute. The "memory wall" remains a critical bottleneck for deploying Large Language Models (LLMs) at the edge, where hardware limitations often prevent real-time inference. To break this barrier, I pioneer techniques that exploit inherent model redundancies without sacrificing accuracy. Beyond inference optimization, I also focus on efficient training and fine-tuning in distributed edge environments, particularly in federated learning settings, developing methods to drastically reduce computational and memory demands while preserving model performance.

Typical Works:
  1. Breaking Memory Wall for Fast Edge LLM Inference Using Contextual Sparsity, IEEE TMC 2026
  2. SpBatch: Reclaiming Lost Sparsity for Batched Sparse FFN Execution, IEEE TC 2026
  3. Selective Tensor Freezing for Efficient Fine-Tuning in Resource-Constrained Federated Learning, IEEE JSAC 2026
  4. MemFerry: A Fast and Memory Efficient Offload Training Framework with Hybrid GPU Computation, IEEE INFOCOM 2025
  5. Amoeba: Runtime Tensor Parallel Transformation for LLM Inference Services, DAC 2026
  6. LCFed: An Efficient Clustered Federated Learning Framework for Heterogeneous Data, ICASSP 2025
  7. Gradient Free Personalized Federated Learning, ICPP 2024

3. Network Traffic Analytics

My research in Network Traffic Analytics centers on pushing intelligence directly into the data plane to achieve line-rate processing and unprecedented network agility. Traditional software-based traffic analysis struggles to keep pace with modern high-speed networks, creating a bottleneck for real-time security and telemetry. To overcome this, I leverage programmable switches (P4) to implement custom packet processing logic directly in hardware. My work covers deep packet inspection, real-time traffic classification, in-network caching, and service function chain offloading, as well as resilient control plane mechanisms for software-defined networks.

Typical Works:
  1. Hyperflex: A SIMD-Based DFA Model for Deep Packet Inspection, IEEE TNSM 2026
  2. ReWeave: Traffic Engineering with Robust Path Weaving for Localized Link Failure Recovery, IEEE ICNP 2025
  3. Harry: A Scalable SIMD-based Multi-literal Pattern Matching Engine for Deep Packet Inspection, IEEE INFOCOM 2023
  4. RET-Net: A CNN Framework for Real-Time Traffic Classification Using Key-Byte Mechanism, IEEE TNSM 2026
  5. Flexible Offloading of Service Function Chains to Programmable Switches, IEEE TSC 2023
  6. P4Neighbor: Efficient Link Failure Recovery with Programmable Switches, IEEE TNSM 2021