Research map / Architectures and efficiency / Neural architecture search
Learning Latent Architectural Distribution in Differentiable Neural Architecture Search via Variational Information Maximization
Yaoming Wang, Yuchen Liu, Wenrui Dai, Chenglin Li, Junni Zou, Hongkai Xiong
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 the architectural neural network (replace vanilla 14x8 parameters) (ablation) | VIM-NAS · CIFAR-10 (DARTS space) | top-1 test error (%): 2.65 -> 2.45 | Improved |
| use a medium architectural network instead of a single ConvReLUBN module (ablation) | VIM-NAS · CIFAR-10 (DARTS space) | top-1 test error (%): 2.51 -> 2.45 | Improved |
| use the default architectural network instead of five modules (ablation) | VIM-NAS · CIFAR-10 (DARTS space) | top-1 test error (%): 2.58 -> 2.45 | Improved |
| use probabilistic posterior instead of the variational-dropout reformulation (ablation) | VIM-NAS · CIFAR-10 (DARTS space) | top-1 test error (%): 2.58 -> 2.45 | Improved |
| use probabilistic distribution instead of point estimation (ablation) | VIM-NAS · DARTS space | top-1 test error (%) | Improved |
| VIM-NAS instead of DARTS-V2 | DARTS · CIFAR-10 | top-1 test error (%): 2.76 -> 2.45 | Improved |
| VIM-NAS instead of DARTS-V2 | DARTS · CIFAR-100 | top-1 test error (%): 17.54 -> 15.80 | Improved |
| VIM-NAS instead of DARTS-V1 (search cost, GPU-days) | DARTS · CIFAR-10 | search cost (GPU-days): 0.4 -> 0.007 | Improved |
| VIM-NAS instead of DARTS- | DARTS- · NAS-Bench-201 CIFAR-100 test | accuracy (%): 71.53 -> 73.07 | Improved |
| VIM-NAS instead of DARTS- | DARTS- · NAS-Bench-201 ImageNet-16-120 test | accuracy (%): 45.12 -> 46.27 | Improved |
| search directly on ImageNet instead of transferring from CIFAR-10 | VIM-NAS · ImageNet | top-1 test error (%): 24.0 -> 23.8 | Improved |
| VIM-NAS instead of PC-DARTS | PC-DARTS · ImageNet | top-1 test error (%): 25.1 -> 24.0 | 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-V1 — DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021) · better on 8 of 8 shared results
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
- BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule (CVPR 2022 · similarity 0.91)
- DrNAS: Dirichlet Neural Architecture Search (ICLR 2021 · similarity 0.90)
- Dependency-aware Differentiable Neural Architecture Search (ECCV 2024 · similarity 0.89)
- iDARTS: Differentiable Architecture Search with Stochastic Implicit Gradients (ICML 2021 · similarity 0.88)
- SNAS: stochastic neural architecture search (ICLR 2019 · similarity 0.88)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.88)
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.
