Research map / Architectures and efficiency / Neural architecture search
MnasNet: Platform-Aware Neural Architecture Search for Mobile
Check 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 MnasNet-A1 instead of MobileNetV2 (1.4x) | ImageNet mobile classification · ImageNet validation | Top-1 accuracy: 74.7 -> 75.2 | Helped |
| use MnasNet-A1 instead of MobileNetV2 (1.4x) | ImageNet mobile classification · ImageNet validation | Inference latency: 143ms -> 78ms; 1.8× faster | Helped |
| use MnasNet-A1 instead of NASNet-A | ImageNet mobile classification · ImageNet validation | Top-1 accuracy: 74.0 -> 75.2 | Helped |
| use MnasNet-A1 instead of NASNet-A | ImageNet mobile classification · ImageNet validation | Inference latency: 183ms -> 78ms; 2.3× faster | Helped |
| use MnasNet-A1 + SSDLite instead of MobileNetV2 + SSDLite | SSDLite object detection · COCO test-dev2017 | mAP: 22.1 -> 23.0 | Helped |
| use MnasNet-A1 + SSDLite instead of SSD300 | COCO object detection · COCO test-dev2017 | mAP: 23.2 -> 23.0 | No clear effect |
| remove SE from the search space (ablation) | MnasNet search · ImageNet validation | Top-1 accuracy: 75.2% -> 74.5% | Hurt |
| use a single repeated layer type instead of diverse layers (ablation) | MnasNet-A1 · ImageNet validation | Top-1 accuracy: 75.2% vs. 71.8%, 72.5%, 74.9%, 75.6% | Mixed |
| use a single repeated layer type instead of diverse layers (ablation) | MnasNet-A1 · ImageNet validation | Inference latency: 78ms vs. 63ms, 79ms, 116ms, 146ms | Mixed |
| add the multi-objective latency reward to the cell-based search (ablation) | NASNet-style cell-based search · ImageNet validation | Inference latency: 183ms -> 100ms | Helped |
| add the multi-objective latency reward to the cell-based search (ablation) | NASNet-style cell-based search · ImageNet validation | Top-1 accuracy: 74.0% -> 72.0% | Hurt |
| replace the cell-based search space with the factorized hierarchical MnasNet search space (ablation) | Multi-objective architecture search · ImageNet validation | Top-1 accuracy: 72.0% -> 75.2% | Helped |
Closest papers by meaning
- MobileOne: An Improved One Millisecond Mobile Backbone (CVPR 2023 · similarity 0.90)
- MnasFPN: Learning Latency-Aware Pyramid Architecture for Object Detection on Mobile Devices (CVPR 2020 · similarity 0.89)
- IFORMER: INTEGRATING CONVNET AND TRANSFORMER FOR MOBILE APPLICATION (ICLR 2025 · similarity 0.87)
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.87)
- SnapGen: Taming High-Resolution Text-to-Image Models for Mobile Devices with Efficient Architectures and Training (CVPR 2025 · similarity 0.87)
- MobileMamba: Lightweight Multi-Receptive Visual Mamba Network (CVPR 2025 · 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.