A Study on Encodings for Neural Architecture Search

By Colin White, Willie Neiswanger, Sam Nolen, Yash Savani. 2020

85 citations recorded by the source corpus.

Neural architecture search (NAS) has been extensively studied in the past few years. A popular approach is to represent each neural architecture in the search space as a directed acyclic graph (DAG), and then search over all DAGs by encoding the adjacency matrix and list of operations as a set of hyperparameters. Recent work has demonstrated that even small changes to the way each architecture is encoded can have a significant effect on the performance of NAS algorithms. In this work, we present the first formal study on the effect of architecture encodings for NAS, including a theoretical grounding and an empirical study. First we formally define architecture encodings and give a theoretical characterization on the scalability of the encodings we study Then we identify the main encoding-dependent subroutines which NAS algorithms employ, running experiments to show which encodings work best with each subroutine for many popular algorithms. The experiments act as an ablation study for prior work, disentangling the algorithmic and encoding-based contributions, as well as a guideline for future work. Our results demonstrate that NAS encodings are an important design decision which can have a significant impact on overall performance. Our code is available at this https URL.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Colin White
2. Yash Savani
3. Sam Nolen
4. Willie Neiswanger
5. Arber Zela
6. Julien N. Siems
7. Mi Zhang
8. Shen Yan

In Computer Vision and Pattern Recognition

1. Colin White
2. Yash Savani
3. Sam Nolen
4. Arber Zela
5. Willie Neiswanger
6. Mi Zhang
7. Julien N. Siems
8. Shen Yan

Citations in the corpus

• BANANAS: Bayesian Optimization with Neural Architectures for Neural Architecture Search (2019, methodology): “2019; Dong and Yang 2020), and there are different methods of encoding the DAGs before they are passed to the neural predictor (Ying et al. 2019; White et al. 2020), which we describe in the next section.”
• Local Search is State of the Art for NAS Benchmarks (2020, background): “Recent papers have highlighted the need for fair and reproducible NAS comparisons (Li and Talwalkar, 2019; Lindauer and Hutter, 2019; White et al., 2020), spurring the release of two NAS benchmark datasets (Ying et al.”
• NAS-Bench-301 and the Case for Surrogate Benchmarks for Neural Architecture Search (2020, background): “As recently shown by White et al. (2020a), this representation can be critical for the performance of a NAS optimizer.”
• Does Unsupervised Architecture Representation Learning Help Neural Architecture Search? (2020, background): “[52] emphasizes the importance of studying architecture encodings, and we focus on encoding adjacency matrix-based architectures into low-dimensional embeddings in the continuous space.”
• Multi-View Encoders for Performance Prediction in LLM-Based Agentic Workflows (2025, unknown): “BRP-NAS (Dudziak et al., 2020) integrates graph convolutional networks to forecast hardware-aware performance metrics, improving the practicality of NAS under resource constraints.”