Research map / Architectures and efficiency / Neural architecture search
Rethinking Neural Operations for Diverse Tasks
Nicholas Carl Roberts, Mikhail Khodak, Tri Dao, Liam Li, Christopher Re, Ameet Talwalkar
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 |
|---|---|---|---|
| replace convolutions with XD-operations (ablation) | LeNet · Permuted CIFAR-10 | accuracy (%): 43.7\pm0.5 -> 63.0\pm1.0 | Improved |
| replace convolutions with XD-operations (ablation) | ResNet-20 · Permuted CIFAR-10 | accuracy (%): 58.6\pm0.7 -> 73.5\pm1.6 | Improved |
| replace convolutions with XD-operations (ablation) | WRN-40-4 · Permuted CIFAR-10 | accuracy (%): 64.7\pm0.9 -> 72.9\pm0.8 | Improved |
| replace convolutions with XD-operations (ablation) | ResNet-20 · CIFAR-10 | accuracy (%): 91.7\pm0.2 -> 92.4\pm0.2 | No clear effect |
| replace convolutions with XD-operations (ablation) | WRN-40-4 · CIFAR-10 | accuracy (%): 95.2\pm0.1 -> 95.0\pm0.1 | No clear effect |
| use relaxed DARTS operations instead of XD-operations (ablation) | ResNet-20 · Permuted CIFAR-10 | accuracy (%): 73.5\pm1.6 -> 58.0\pm1.0 | Hurt |
| use relaxed DARTS operations instead of XD-operations (ablation) | WRN-40-4 · Permuted CIFAR-10 | accuracy (%): 72.9\pm0.8 -> 61.3\pm1.3 | Hurt |
| replace convolutions with XD-operations (ablation) | CNN backbone · 1d Burgers' equation s=256 | test relative error: 0.0518 -> 0.0141 | Improved |
| replace convolutions with XD-operations (ablation) | CNN backbone · 1d Burgers' equation s=8192 | test relative error: 0.2970 -> 0.0123 | Improved |
| replace convolutions with XD-operations (ablation) | CNN backbone · 2d Darcy Flow s=421 | test relative error: 0.1150 -> 0.0066 | Improved |
| XD-operations versus custom FNO-2d (reproduced) (ablation) | CNN XD · 2d Darcy Flow s=85 | test relative error: 0.0096 -> 0.0065 | Improved |
| XD-operations versus reported FNO-1d | CNN XD · 1d Burgers' equation s=4096 | test relative error | No clear effect |
Closest papers by meaning
- AutoML-Zero: Evolving Machine Learning Algorithms From Scratch (ICML 2020 · similarity 0.86)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.85)
- Auto-Pytorch: Multi-Fidelity MetaLearning for Efficient and Robust AutoDL (TPAMI 2021 · similarity 0.85)
- Building Optimal Neural Architectures using Interpretable Knowledge (CVPR 2024 · similarity 0.84)
- AutoAL: Automated Active Learning with Differentiable Query Strategy Search (ICML 2025 · similarity 0.83)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.82)
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.
