계층적 공간에서 FaDE를 사용한 효율적인 NAS
FaDE는 계층적 NAS 공간의 유한 영역에서 상대적 성능 예측을 사용하여 효율적인 신경망 구조 탐색을 가능하게 한다.
Accel-NASBench: A Sustainable Benchmark for Accelerator-Aware Neural Architecture Search
The authors present a technique to efficiently construct a realistic neural architecture search (NAS) benchmark for the large-scale ImageNet2012 dataset, combined with performance metrics for various hardware accelerators including GPUs, TPUs, and FPGAs.
Evaluating Neural Architecture Search Methods on Diverse and Unseen Datasets
Neural Architecture Search (NAS) methods should be able to find optimal neural network architectures for diverse datasets, not just common benchmarks. This work introduces eight new datasets to challenge NAS approaches and evaluate their generalization capabilities.
TG-NAS: A Universal Zero-Cost Proxy for Efficient Neural Architecture Search
TG-NAS proposes a universally applicable, data-independent performance predictor model that can handle unseen operators in new search spaces without retraining, acting as a zero-cost proxy to guide efficient neural architecture search.
Assembling Zero-Cost Proxies for Efficient Network Architecture Search
Assembling multiple zero-cost proxies that capture distinct network characteristics to efficiently predict the performance of candidate architectures without training.
SWAP-NAS: Sample-Wise Activation Patterns for Ultra-Fast NAS at ICLR 2024
샘플별 활성화 패턴을 활용한 SWAP-Score는 효율적인 NAS를 위한 혁신적인 메트릭스입니다.
FlatNAS: Optimizing Neural Architecture for Out-of-Distribution Robustness
Neural Architecture Search (NAS) explores flat regions in loss landscapes to optimize NN architectures for out-of-distribution (OOD) robustness.
Building Optimal Neural Architectures using Interpretable Knowledge
AutoBuild proposes a method to construct high-quality neural architectures by assigning importance scores to architecture modules, reducing the need for exhaustive search.
Pareto-wise Ranking Classifier for Multi-objective Evolutionary Neural Architecture Search
複数目的の進化ニューラルアーキテクチャ検索におけるパレート順位付け分類器の重要性
Pareto-wise Ranking Classifier for Multi-objective Evolutionary Neural Architecture Search
The author proposes a Pareto-wise end-to-end ranking classifier to simplify the architecture search process in multi-objective NAS, addressing the rank disorder issue and outperforming other methods.
© 2024 by Linnk AI