POET: Training Neural Networks on Tiny Devices with Integrated Rematerialization and Paging
This work addresses the challenge of privacy-preserving personalization on edge devices for users with limited memory and battery, representing a strong specific gain in edge training capabilities.
The paper tackles the problem of training large neural networks on memory-constrained edge devices by introducing POET, an algorithm that jointly optimizes rematerialization and paging to reduce memory usage and energy consumption, enabling fine-tuning of models like ResNet-18 and BERT on embedded devices while outperforming existing methods in energy efficiency.
Fine-tuning models on edge devices like mobile phones would enable privacy-preserving personalization over sensitive data. However, edge training has historically been limited to relatively small models with simple architectures because training is both memory and energy intensive. We present POET, an algorithm to enable training large neural networks on memory-scarce battery-operated edge devices. POET jointly optimizes the integrated search search spaces of rematerialization and paging, two algorithms to reduce the memory consumption of backpropagation. Given a memory budget and a run-time constraint, we formulate a mixed-integer linear program (MILP) for energy-optimal training. Our approach enables training significantly larger models on embedded devices while reducing energy consumption while not modifying mathematical correctness of backpropagation. We demonstrate that it is possible to fine-tune both ResNet-18 and BERT within the memory constraints of a Cortex-M class embedded device while outperforming current edge training methods in energy efficiency. POET is an open-source project available at https://github.com/ShishirPatil/poet