Research map / Architectures and efficiency / Pruning and efficient tuning
POET: Training Neural Networks on Tiny Devices with Integrated Rematerialization and Paging
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| jointly optimize rematerialization and paging instead of using POFO's schedule | ResNet-18 training on Jetson TX2 · ResNet-18, Jetson TX2 | peak memory: lowers peak memory consumption by 8.3% | Helped |
| jointly optimize rematerialization and paging instead of using POFO's schedule | ResNet-18 training on Jetson TX2 · ResNet-18, Jetson TX2 | throughput: improves throughput by 13% | Helped |
| integrate rematerialization with paging rather than using Capuchin's paging-first heuristic | ResNet-18 training on A72 · ResNet-18, A72 | energy overhead: 73% to 141% more energy | Helped |
| integrate paging and rematerialization instead of using either strategy alone (ablation) | POET · training across evaluated models and devices | energy consumption: up to ${40\%}$ lower energy | Helped |
| use rematerialization only rather than integrated paging and rematerialization (ablation) | POET · training under a lower runtime budget ($0.5$ ms) | energy consumption | Hurt |
| use paging only rather than integrated paging and rematerialization (ablation) | POET · training at higher runtime budgets ($0.6$ - $0.9$ ms) | energy consumption | Hurt |
| rematerialize cheap operations earlier while also paging expensive operations (ablation) | POET · VGG training on CIFAR-10 | energy consumption | Helped |
| allow the MILP solve budget to exceed 10 min (ablation) | POET on BERT, Cortex M4 and TX2 · BERT training | solution optimality: <30 min | Helped |
Closest papers by meaning
- TinyTrain: Resource-Aware Task-Adaptive Sparse Training of DNNs at the Data-Scarce Edge (ICML 2024 · similarity 0.90)
- AtomNet: Designing Tiny Models from Operators Under Extreme MCU Constraints (AAAI 2025 · similarity 0.86)
- Stepping Forward on the Last Mile (NeurIPS 2024 · similarity 0.85)
- TinyTTA: Efficient Test-time Adaptation via Early-exit Ensembles on Edge Devices (NeurIPS 2024 · similarity 0.84)
- Beyond Low-rank Decomposition: A Shortcut Approach for Efficient On-Device Learning (ICML 2025 · similarity 0.84)
- An Empirical Study of Distributed Deep Learning Training on Edge (Student Abstract) (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.