AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

PASHA: Efficient HPO and NAS with Progressive Resource Allocation

Ondrej Bohdal, Lukas Balles, Martin Wistuba, Beyza Ermis, Cedric Archambeau, Giovanni Zappella

ICLR 2023 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
dynamically increase maximum resources only when successive rung rankings remain inconsistentASHA-based HPO/NAS · CIFAR-10Runtime: 3.0h ± 0.6h -> 1.3h ± 0.6hImproved
dynamically increase maximum resources only when successive rung rankings remain inconsistentASHA-based HPO/NAS · CIFAR-100Accuracy (%): 71.69 ± 1.05 -> 71.84 ± 1.41No clear effect
dynamically increase maximum resources only when successive rung rankings remain inconsistentASHA-based HPO/NAS · ImageNet16-120Runtime: 8.8h ± 2.2h -> 2.9h ± 1.7hImproved
dynamically increase maximum resources only when successive rung rankings remain inconsistentASHA-based HPO · WMTRuntime: 43.7h ± 37.2h -> 2.8h ± 0.6hImproved
dynamically increase maximum resources only when successive rung rankings remain inconsistentASHA-based HPO · ImageNetAccuracy (%): 75.10 ± 2.03 -> 73.37 ± 2.71Hurt
combine progressive resource allocation with Bayesian optimization searchMOBSTER · CIFAR-100Runtime: 5.7h ± 1.4h -> 1.6h ± 0.5hImproved
replace automatically estimated soft ranking with direct ranking (ablation)PASHA · CIFAR-100Runtime: 0.9h ± 0.4h -> 2.8h ± 0.7hHurt
fix soft-ranking ε at 2.5% instead of estimating ranking noise (ablation)PASHA · CIFAR-100Accuracy (%): 71.84 ± 1.41 -> 71.41 ± 1.15No clear effect
use soft-ranking ε=2σ instead of estimating ranking noise (ablation)PASHA · CIFAR-100Runtime: 0.9h ± 0.4h -> 1.9h ± 0.7hHurt
replace soft ranking with Rank Biased Overlap (ablation)PASHA · CIFAR-100Runtime: 0.9h ± 0.4h -> 2.4h ± 0.7hHurt
replace soft ranking with Reciprocal Rank Regret (ablation)PASHA · CIFAR-100Runtime: 0.9h ± 0.4h -> 1.2h ± 0.5hImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy