Research map / Architectures and efficiency / RNNs and normalization
Memory Optimization for Deep Networks
Aashaka Shah, Chao-Yuan Wu, Jayashree Mohan, Vijay Chidambaram, Philipp Kraehenbuehl
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 checkpointing and operator implementations | MONeT · ResNet-50, GoogleNet, UNet, VGG-16, MobileNet-V2 | memory at fixed compute overhead: 1.2-1.8× less memory than Checkmate; 2-3× less memory than PyTorch | Improved |
| Jointly optimize convolution algorithms with checkpointing (ablation) | MONeT · ResNet-50, GoogleNet, VGG-16 | compute overhead | Improved |
| Use output-activated operator optimization (ablation) | MONeT · ResNet-50, GoogleNet, VGG-16 | compute overhead | Improved |
| Use intermediate-activated operator optimization (ablation) | MONeT · Memory-intensive networks, including VGG-16 | memory | Improved |
| Enable all operator optimizations together (ablation) | MONeT · ResNet-50, GoogleNet, VGG-16, MobileNet-V2 | compute overhead | Improved |
| Apply convolution optimization to depthwise-separable convolutions (ablation) | MONeT · MobileNet-V2 | compute overhead | No clear effect |
| Use intermediate-activated backward optimization for hardtanh (ablation) | MONeT · MobileNet-V2 | memory | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PyTorch — PyTorch: An Imperative Style, High-Performance Deep Learning Library (NeurIPS 2019) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Optimal Gradient Checkpoint Search for Arbitrary Computation Graphs (CVPR 2021 · similarity 0.87)
- Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks (CVPR 2023 · similarity 0.85)
- Refined Coreset Selection: Towards Minimal Coreset Size under Model Performance Constraints (ICML 2024 · similarity 0.85)
- Adam-mini: Use Fewer Learning Rates To Gain More (ICLR 2025 · similarity 0.83)
- MEET: Towards Memory-Efficient Temporal Sparse Deep Neural Networks (CVPR 2025 · similarity 0.83)
- Stable Coresets via Posterior Sampling: Aligning Induced and Full Loss Landscapes (NeurIPS 2025 · similarity 0.83)
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.
