Frequently Asked Questions (FAQ)#

Q: What is Neural Architecture Search (NAS)?

A: NAS is a machine learning technique that aims to automatically discover effective neural network architectures for a given task, without the need for human intervention or trial-and-error experimentation. By using algorithms to search the space of possible architectures, NAS can help identify novel and effective designs that may not have been discovered through traditional manual design methods.

Q: What is Archai and how does it relate to NAS?

A: Archai is an open-source project that aims to accelerate NAS research and serve as a turnkey NAS framework for Microsoft and the broader machine learning community. Archai has made significant progress in both the research and engineering aspects of NAS, and is a generic platform that can be applied to a wide range of tasks and hardware targets.

Q: What are some of the open problems in NAS research?

A: There are many open questions and challenges in the field of NAS, including:

  • Search space design: How much inductive bias should be included in the search space, and how can the search space be designed to balance the need for complexity and expressiveness with the need for efficiency and scalability?

  • Algorithmic scalability: How can NAS algorithms be scaled to handle the vast number of potential architectures, and how can they be made more efficient and effective in their search for the best architectures?

  • Robustness and uncertainty: How can NAS algorithms be made more robust to noise and uncertainty in the data, and how can they incorporate uncertainty estimates in their search and evaluation processes?

  • Learning-based search: How can NAS algorithms incorporate learned priors or meta-learning techniques to improve their search and evaluation, and how can they be integrated with other learning-based approaches to NAS?

  • Novel search methods: What are the next-generation search algorithms and techniques that will push the boundaries of NAS and enable the discovery of even more effective and efficient architectures?

Q: How can researchers get involved with Archai and NAS?

A: Researchers interested in exploring the open problems in NAS and contributing to the Archai project are encouraged to join the community on GitHub. There are many opportunities for collaboration and research within the Archai community, including working on existing open problems, proposing new directions for research, and contributing to the development of the platform.