Research map / Architectures and efficiency / Neural architecture search
$\Lambda$-DARTS: Mitigating Performance Collapse by Harmonizing Operation Selection among Cells
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 |
|---|---|---|---|
| Add Λ layer-alignment regularization to DARTS | DARTS · DARTS search space, CIFAR-100 | accuracy: 83.10 -> 83.47±0.20 | Helped |
| Add Λ layer-alignment regularization to DARTS | DARTS · DARTS search space, CIFAR-10 | accuracy: 97.51±0.06 -> 97.48±0.11; best 97.59 | Mixed |
| Use sign-gradient alignment Λ± instead of Λ (ablation) | Λ-DARTS · DARTS search space, CIFAR-100 | accuracy: 83.47±0.20 -> 83.85±0.38 | Helped |
| Use sign-gradient alignment Λ± instead of Λ (ablation) | Λ-DARTS · DARTS search space, CIFAR-10 | accuracy: 97.48±0.11 -> 97.57±0.05 | Helped |
| Lower ε₀ in the finite-difference gradient estimate (ablation) | Λ-DARTS · NAS-Bench-201, CIFAR-10 | validation accuracy | Helped |
| Reduce ε₀ to its smallest tested value (ablation) | Λ-DARTS · NAS-Bench-201, CIFAR-10 | validation accuracy | Mixed |
| Increase regularization strength λ from 0.125 to 0.5 (ablation) | Λ-DARTS · NAS-Bench-201, CIFAR-10 | mean validation accuracy: about 2% | Hurt |
| Extend search duration from the converged setting to longer epochs (ablation) | Λ-DARTS · NAS-Bench-201, CIFAR-10 | validation accuracy: about 0.4% | Hurt |
| Apply the alignment regularizer to the outer objective instead of the inner objective (ablation) | Λ-DARTS · NAS-Bench-201 | architecture performance | Hurt |
| Reduce PC-DARTS batch size from 256 to 64 (ablation) | PC-DARTS · NAS-Bench-201, CIFAR-100 | accuracy: 67.48±0.89 -> 71.11±0.00 | Helped |
| Apply layer-alignment regularization | DARTS · Reduced DARTS search spaces, 12 dataset-space experiments | test error rate | Mixed |
Closest papers by meaning
- IS-DARTS: Stabilizing DARTS through Precise Measurement on Candidate Importance (AAAI 2024 · similarity 0.92)
- DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021 · similarity 0.92)
- iDARTS: Differentiable Architecture Search with Stochastic Implicit Gradients (ICML 2021 · similarity 0.92)
- b-DARTS: Beta-Decay Regularization for Differentiable Architecture Search (CVPR 2022 · similarity 0.92)
- Interpreting Operation Selection in Differentiable Architecture Search: A Perspective from Influence-Directed Explanations (NeurIPS 2022 · similarity 0.91)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.90)
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.