Research map / Architectures and efficiency / Neural architecture search
U-Boost NAS: Utilization-Boosted Differentiable Neural Architecture Search
Ahmet Caner Yüzügüler, Nikolaos Dimitriadis, Pascal Frossard
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 utilization term to the NAS loss (U-Boost vs utilization-agnostic baselines) | differentiable hardware-aware NAS · CIFAR10 | hypervolume: 2.68, 1.86, 1.47 -> 0.39 | Improved |
| use utilization-aware U-Boost instead of FLOPS/Roofline/Blackbox latency models | differentiable NAS · CIFAR10 | runtime speedup: 2.8-4x speedup | Improved |
| use U-Boost instead of Blackbox latency model | differentiable NAS · ImageNet100 | hardware utilization (%): 69.3 -> 91.1 | Improved |
| use U-Boost instead of Blackbox latency model | differentiable NAS · ImageNet100 | top-1 accuracy (%): 87.8 -> 87.9 | Improved |
| use U-Boost instead of Blackbox latency model | differentiable NAS · ImageNet100 | runtime (ms): 4.8 -> 2.2 | Improved |
| use U-Boost instead of FLOPS latency model | differentiable NAS · ImageNet100 | hypervolume: 108 -> 12.7 | Improved |
| use U-Boost instead of FLOPS model at lambda=1.0 | differentiable NAS · ImageNet100 | top-1 accuracy (%): 78.4 -> 87.9 | Improved |
| use U-Boost instead of Roofline model at lambda=1.0 | differentiable NAS · ImageNet100 | top-1 accuracy (%): 84.0 -> 87.9 | Improved |
| use U-Boost instead of Blackbox at lambda=0.1 on CIFAR10 | differentiable NAS · CIFAR10 | top-1 accuracy (%): 91.4 -> 90.9 | No clear effect |
| use U-Boost instead of Blackbox at lambda=5.0 on CIFAR10 | differentiable NAS · CIFAR10 | runtime (us): 122 -> 30 | Improved |
| use the proposed utilization model instead of the roofline model for estimating utilization | Cloud TPU v2 layers · TPU v2 measurement | estimation error: up to 40% -> up to 2% | Improved |
| replace discrete lookup-table (Blackbox) search by gradient-based utilization model | channel dimension search · CIFAR10 | selected channel dimensions: concentrated on 128 or 256 | Improved |
Closest papers by meaning
- HW-NAS-Bench: Hardware-Aware Neural Architecture Search Benchmark (ICLR 2021 · similarity 0.88)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.88)
- NAS-Bench-360: Benchmarking Neural Architecture Search on Diverse Tasks (NeurIPS 2022 · similarity 0.88)
- DA-NAS: Data Adapted Pruning for Efficient Neural Architecture Search (ECCV 2020 · similarity 0.88)
- FP-NAS: Fast Probabilistic Neural Architecture Search (CVPR 2021 · similarity 0.88)
- MOTE-NAS: Multi-Objective Training-based Estimate for Efficient Neural Architecture Search (NeurIPS 2024 · similarity 0.86)
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.
