Search Spaces Overview

This documentation offers insights into a variety of search spaces within NASLib. Delve into these search spaces designed to address specific tasks and domains, and tailor your architecture search experiments accordingly.

1. NAS-Bench-101

NAS-Bench-101 provides a comprehensive search space for evaluating neural architectures in the context of computer vision tasks. It is a valuable resource for benchmarking and comparison.

2. NAS-Bench-201

NAS-Bench-201 is designed to support architecture search for various tasks, including computer vision and beyond. It offers diverse search spaces to cater to different application domains.

3. NAS-Bench-301

NAS-Bench-301 extends the capabilities of our library by providing a versatile search space for exploring neural architectures in the context of computer vision and related fields.

4. NAS-Bench-ASR

NAS-Bench-ASR is tailored for the domain of Automatic Speech Recognition (ASR). It offers specific search spaces and configurations optimized for ASR tasks.

5. NAS-Bench-NLP

NAS-Bench-NLP focuses on Natural Language Processing (NLP) tasks, providing search spaces designed to create and evaluate neural architectures for NLP applications.

6. Simple Cell

Simple Cell is a straightforward and customizable search space that can be used for a variety of NAS experiments. It allows researchers to define and explore their own architectures.

7. Transbench 101

Transbench 101 is a specialized search space designed for transformer-based models, suitable for a wide range of tasks, including language modeling and machine translation.