Research map / Architectures and efficiency / Neural architecture search
MixPath: A Unified Approach for One-shot Neural Architecture Search
Xiangxiang Chu, Shun Lu, Xudong Li, Bo Zhang
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 |
|---|---|---|---|
| add shadow batch normalization (SBN) to multi-path supernet training (ablation) | MixPath supernet (m=4) · NAS-Bench-101 subset | Kendall tau (no calibration): 0.117 -> 0.393 | Improved |
| add BN post-calibration on top of SBN (ablation) | MixPath supernet (m=4) · NAS-Bench-101 subset | Kendall tau: 0.393 -> 0.597 | Improved |
| use linear (m) SBNs instead of exponential SBNs with calibration (ablation) | MixPath supernet (m=3) · NAS-Bench-101 subset | Kendall tau (with calibration): 0.34 -> 0.59 | Improved |
| use exponential SBNs instead of linear when no calibration is applied (ablation) | MixPath supernet (m=3) · NAS-Bench-101 subset | Kendall tau (no calibration): 0.32 -> 0.37 | Improved |
| replace simple feature magnitude scaling with SBNs (ablation) | MixPath supernet (m=4) · NAS-Bench-101 subset | Kendall tau (with calibration): 0.472 -> 0.597 | Improved |
| train with SBN in the full search pipeline (ablation) | MixPath on CIFAR-10 S1 · CIFAR-10, under 500M FLOPS | top-1 accuracy (%): 97.12 -> 97.44 | Improved |
| use MixPath multi-path supernet instead of SPOS for branch composition | NAS-Bench-101 m=4 branch composition · NAS-Bench-101 | Kendall tau: -0.225 -> 0.504 | Improved |
| use MixPath instead of DARTS in reduced NAS-Bench-101 | NAS-Bench-101 S4 · Reduced NAS-Bench-101 | top-1 accuracy (%): 79.42 -> 94.22 | Improved |
| use NSGA-II search instead of random search (ablation) | MixPath supernet S3 · ImageNet Pareto front | accuracy and FLOPs of elite models | Improved |
| swap MixNet-M for MixPath-A backbone | RetinaNet · COCO val2017 | AP (%): 31.3 -> 31.5 | Improved |
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.
- DARTS — DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021) · better on 1 of 1 shared result
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
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.88)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.87)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.87)
- Revisiting Neural Networks for Few-Shot Learning: A Zero-Cost NAS Perspective (ICML 2025 · similarity 0.86)
- Neural Fine-Tuning Search for Few-Shot Learning (ICLR 2024 · similarity 0.86)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 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.
