ReStNet: A Reusable & Stitchable Network for Dynamic Adaptation on IoT Devices
Preprint
Abstract
With the rapid development of deep learning, a growing number of pre-trained models have been publicly available. However, deploying these fixed models in real-world IoT applications is challenging because different devices possess heterogeneous computational and memory resources, making it impossible to deploy a single model across all platforms. Although traditional compression methods, such as pruning, quantization, and knowledge distillation, can improve efficiency, they become inflexible once applied and cannot adapt to changing resource constraints. To address these issues, we propose ReStNet, a Reusable and Stitchable Network that dynamically constructs a hybrid network by stitching two pre-trained models together. Implementing ReStNet requires addressing several key challenges, including how to select the optimal stitching points, determine the stitching order of the two pre-trained models, and choose an effective fine-tuning strategy. To systematically address these challenges and adapt to varying resource constraints, ReStNet determines the stitching point by calculating layer-wise similarity via Centered Kernel Alignment (CKA). It then constructs the hybrid model by retaining early layers from a larger-capacity model and appending deeper layers from a smaller one. To facilitate efficient deployment, only the stitching layer is fine-tuned. This design enables rapid adaptation to changing budgets while fully leveraging available resources. Moreover, ReStNet supports both homogeneous (CNN-CNN, Transformer-Transformer) and heterogeneous (CNN-Transformer) stitching, allowing to combine different model families flexibly. Extensive experiments on multiple benchmarks demonstrate that ReStNet achieve flexible accuracy-efficiency trade-offs at runtime while significantly reducing training cost.Keywords
All Related Versions
This publication has 15 references indexed in Scilit:
- A Generic Layer Pruning Method for Signal Modulation Recognition Deep Learning ModelsIEEE Transactions on Cognitive Communications and Networking, 2024
- Graph-Based Similarity of Deep Neural NetworksNeurocomputing, 2024
- Low-Resource Scenario Classification Through Model Pruning Toward Refined Edge IntelligenceIEEE Internet of Things Journal, 2023
- Lightweight Automatic Modulation Classification via Progressive Differentiable Architecture SearchIEEE Transactions on Cognitive Communications and Networking, 2023
- RGP: Neural Network Pruning Through Regular Graph With Edges SwappingIEEE Transactions on Neural Networks and Learning Systems, 2023
- Stitchable Neural Networks2023
- Scalable Vision Transformers with Hierarchical Pooling2021
- Complex-Valued Networks for Automatic Modulation ClassificationIEEE Transactions on Vehicular Technology, 2020
- An Improved Neural Network Pruning Technology for Automatic Modulation Classification in Edge DevicesIEEE Transactions on Vehicular Technology, 2020
- Performance-optimized hierarchical models predict neural responses in higher visual cortexProceedings of the National Academy of Sciences of the United States of America, 2014