AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

U-Boost NAS: Utilization-Boosted Differentiable Neural Architecture Search

Ahmet Caner Yüzügüler, Nikolaos Dimitriadis, Pascal Frossard

ECCV 2022 · Architectures and efficiency › Neural architecture search · approach: Efficient NAS search strategies · 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
add utilization term to the NAS loss (U-Boost vs utilization-agnostic baselines)differentiable hardware-aware NAS · CIFAR10hypervolume: 2.68, 1.86, 1.47 -> 0.39Improved
use utilization-aware U-Boost instead of FLOPS/Roofline/Blackbox latency modelsdifferentiable NAS · CIFAR10runtime speedup: 2.8-4x speedupImproved
use U-Boost instead of Blackbox latency modeldifferentiable NAS · ImageNet100hardware utilization (%): 69.3 -> 91.1Improved
use U-Boost instead of Blackbox latency modeldifferentiable NAS · ImageNet100top-1 accuracy (%): 87.8 -> 87.9Improved
use U-Boost instead of Blackbox latency modeldifferentiable NAS · ImageNet100runtime (ms): 4.8 -> 2.2Improved
use U-Boost instead of FLOPS latency modeldifferentiable NAS · ImageNet100hypervolume: 108 -> 12.7Improved
use U-Boost instead of FLOPS model at lambda=1.0differentiable NAS · ImageNet100top-1 accuracy (%): 78.4 -> 87.9Improved
use U-Boost instead of Roofline model at lambda=1.0differentiable NAS · ImageNet100top-1 accuracy (%): 84.0 -> 87.9Improved
use U-Boost instead of Blackbox at lambda=0.1 on CIFAR10differentiable NAS · CIFAR10top-1 accuracy (%): 91.4 -> 90.9No clear effect
use U-Boost instead of Blackbox at lambda=5.0 on CIFAR10differentiable NAS · CIFAR10runtime (us): 122 -> 30Improved
use the proposed utilization model instead of the roofline model for estimating utilizationCloud TPU v2 layers · TPU v2 measurementestimation error: up to 40% -> up to 2%Improved
replace discrete lookup-table (Blackbox) search by gradient-based utilization modelchannel dimension search · CIFAR10selected channel dimensions: concentrated on 128 or 256Improved

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