Research map / Architectures and efficiency / Neural architecture search
Saliency-Aware Neural Architecture Search
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 |
|---|---|---|---|
| replace multiplicative reweighting with addition of saliency to pixels (ablation) | Ours+darts · CIFAR-100 | test error (%): 16.4±0.09 -> 20.3±0.27 | Hurt |
| use absolute saliency values instead of signed (ablation) | Ours+darts · CIFAR-100 | test error (%): 16.4±0.09 -> 16.8±0.11 | No clear effect |
| concatenate saliency map with input instead of product (ablation) | Ours+darts · CIFAR-100 | test error (%): 16.4±0.09 -> 17.9±0.11 | Hurt |
| remove saliency reweighting entirely (ablation) | Ours+darts · CIFAR-100 | test error (%): 16.4±0.09 -> 20.6±0.44 | Hurt |
| remove saliency reweighting entirely (ablation) | Ours+pdarts · CIFAR-100 | test error (%): 15.2±0.09 -> 17.5±0.06 | Hurt |
| remove saliency reweighting entirely (ablation) | Ours+darts · CIFAR-10 | test error (%): 2.54±0.05 -> 2.76±0.09 | Hurt |
| replace adversarial-attack saliency with integrated gradients (ablation) | Ours+darts · CIFAR-100 | test error (%): 16.42±0.09 -> 16.92±0.08 | Hurt |
| replace adversarial-attack saliency with SmoothGrad (ablation) | Ours+pdarts · CIFAR-100 | test error (%): 15.16±0.09 -> 15.81±0.05 | Hurt |
| add IG saliency to DARTS2nd framework vs vanilla (ablation) | Darts2nd · CIFAR-100 | test error (%): 20.58±0.44 -> 16.92±0.08 | Helped |
| run the four stages separately instead of end-to-end (ablation) | Ours+darts · CIFAR-100 | test error (%): 16.42±0.09 -> 18.05±0.27 | Hurt |
| optimize stages I-III as weighted multi-task sum (ablation) | Ours+pdarts · CIFAR-100 | test error (%): 15.16±0.09 -> 16.83±0.10 | Hurt |
| run the four stages separately (ablation) | Ours+pdarts · CIFAR-10 | test error (%): 2.45±0.03 -> 2.51±0.03 | Hurt |
Closest papers by meaning
- NAS-Bench-360: Benchmarking Neural Architecture Search on Diverse Tasks (NeurIPS 2022 · similarity 0.87)
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation (CVPR 2026 · similarity 0.87)
- NAS-Bench-101: Towards Reproducible Neural Architecture Search (ICML 2019 · similarity 0.87)
- Geometry-Aware Gradient Algorithms for Neural Architecture Search (ICLR 2021 · similarity 0.87)
- There and Back Again: Revisiting Backpropagation Saliency Methods (CVPR 2020 · similarity 0.87)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 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.