Research map / Architectures and efficiency / RNNs and normalization
Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks
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 |
|---|---|---|---|
| Use FasterNet-T0 instead of MobileViT-XXS | ImageNet-1k classification · ImageNet-1k val | top-1 accuracy: 69.0 -> 71.9 | Helped |
| Use FasterNet-T0 instead of MobileViT-XXS | ImageNet-1k classification · ImageNet-1k val | GPU throughput: 2393 -> 6807 fps | Helped |
| Use FasterNet-T0 instead of MobileViT-XXS | ImageNet-1k classification · ImageNet-1k val | CPU latency: 30.8 -> 9.2 ms | Helped |
| Use FasterNet-L instead of Swin-B | ImageNet-1k classification · ImageNet-1k val | GPU throughput: 237 -> 323 fps | Helped |
| Use FasterNet-L instead of Swin-B | ImageNet-1k classification · ImageNet-1k val | top-1 accuracy: 83.5 -> 83.5 | No clear effect |
| Replace ResNet50 backbone with FasterNet-S | Mask R-CNN on COCO · COCO object detection | box AP: +1.9 | Helped |
| Replace ResNet50 backbone with FasterNet-S | Mask R-CNN on COCO · COCO instance segmentation | mask AP: +2.4 | Helped |
| Replace PVT-Large backbone with FasterNet-L | Mask R-CNN on COCO · COCO object detection | box AP: +1.1 | Helped |
| Replace PVT-Large backbone with FasterNet-L | Mask R-CNN on COCO · COCO instance segmentation | mask AP: +0.4 | Helped |
| Use PConv + PWConv instead of DWConv + PWConv | ResNet50 feature transformation · ImageNet feature transformation, average over stages | test loss: 0.0148 -> 0.0123 | Helped |
| Increase partial ratio from 1/4 to 1/2 (ablation) | FasterNet-T0 · ImageNet-1k val | top-1 accuracy: 71.9 -> 71.7 | Hurt |
| Reduce partial ratio from 1/4 to 1/8 (ablation) | FasterNet-T0 · ImageNet-1k val | top-1 accuracy: 71.9 -> 71.3 | Hurt |
Closest papers by meaning
- Fast Sparse ConvNets (CVPR 2020 · similarity 0.89)
- Fast and Accurate Model Scaling (CVPR 2021 · similarity 0.87)
- High Performance Zero-Memory Overhead Direct Convolutions (ICML 2018 · similarity 0.87)
- MEET: Towards Memory-Efficient Temporal Sparse Deep Neural Networks (CVPR 2025 · similarity 0.87)
- FlashRNN: I/O-Aware Optimization of Traditional RNNs on modern hardware (ICLR 2025 · similarity 0.86)
- Harnessing Manycore Processors with Distributed Memory for Accelerated Training of Sparse and Recurrent Models (AAAI 2024 · similarity 0.86)
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.