Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
September 14, 2026IoTOpen Access

Meta-Learning-Driven Adaptive Control for Multi-Exit DNN Splitting at the Edge

View Full Paper
Ask AI
Bookmark
Share

Authors

LWLuyao WangJXJiahao XieHHHao Hao

Discussion

Loading...

Member takes

Overview

Simulation study demonstrates adaptive multi-exit neural network splitting in edge computing networks, indicating improved disturbance recovery and reduced latency-energy costs.

Key Points

  • To develop a meta-learning-driven adaptive control framework that stabilizes split-inference policies and early-exit routing in deep neural networks facing edge resource disruptions.
  • Formulated joint MobileViT backbone splitting and early-exit routing as a constrained Markov decision process.
  • Developed splitting-aware multi-dimensional adaptive proximal policy optimization (SMAPPO) incorporating nonlinear QoS penalties, topology-aware action masking, and cross-environment meta-initialization.
  • Simulated stationary and disturbed edge computing environments across ten seeds and compared SMAPPO against six baseline methods across varying latency-energy preference settings.
  • SMAPPO reached the highest performance-index plateau in a 500-episode stationary trace and achieved the lowest normalized total cost across three latency–energy preference settings.
  • Across ten seeds and nine scenarios, online SMAPPO attained 77.20% measured accuracy and consumed 22.40 mJ of system energy.
  • With an adaptation horizon of K=14, the framework demonstrated a post-disturbance mean latency of 37.68 ms, a QoS-violation rate of 2.24%, and an on-time completion rate of 98.69%.

Cite This Study

Wang et al. (2026) studied this question.

synapsesocial.com/papers/6aa7b2e10926e14a848b16bchttps://doi.org/10.3390/iot7030080
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Automatic regulation and performance optimization method of computer network experimental environment based on deep reinforcement learning2026
  2. 2Robust Offline Multi-Agent Reinforcement Learning for Latency-Aware SDN Path Control in 6G-Oriented Network Softwarization2026
  3. 3Joint Resource Allocation and Computation Offloading for DNN Inference with Model Partition and Early Exit in MEC Networks2026
  4. 4Learning to Partition: Dynamic Deep Neural Network Model Partitioning for Edge-Assisted Low-Latency Video Analytics2025 · 2 citations
  5. 5EdgeAutoOffloadAI a deep reinforcement learning framework for adaptive task offloading and resource management in mobile edge computing2026