Table 5. Summary of neural architecture search paradigms for DL.
| Study | Model | Strengths and limitations |
|---|---|---|
| Baker et al. (2016) | MetaQNN | The neural architecture selection process is modelled as a Markov decision process. |
| Liu et al. (2021) | Large-scale evolution | Evolutionary algorithms are used to learn the best neural architecture automatically. |
| Liu et al. (2021) | GeNet | Agent-based neural architecture search |
| Xie & Yuille (2017) | NAS-RL | Reinforcement learning-derived neural architectures |
| Zoph & Le (2016) | NASNet | Using the concept of artificial neural architecture design proposes a modular search space |
| Zoph et al. (2018) | GDAS-NSAS | Weight sharing is implemented to train a new neural architecture sequentially. |
| Zhao et al. (2021) | NAS-RUL | A gradient-based neural architecture search method |
| Alyasseri et al. (2022) | NAS-Bench-NLP | Search based on natural language processing (NLP) |
| Shi et al. (2022) | Genetic-GNN | Evolutionary algorithms are used to learn the best neural architecture automatically. |
| Ding et al. (2022) | NAP | Neural architecture search with pruning |
| Mun, Ha & Lee (2022) | DE-DARTS | Neural architecture search with dynamic exploration |