RMIT University
Browse

One-Shot Neural Architecture Search: Maximising Diversity to Overcome Catastrophic Forgetting

journal contribution
posted on 2024-11-02, 18:22 authored by Miao Zhang, Huiqi Li, Shirui Pan, Xiaojun ChangXiaojun Chang, Chuan Zhou, Zongyuan Ge, Steven Su
One-shot neural architecture search (NAS) has recently become mainstream in the NAS community because it significantly improves computational efficiency through weight sharing. However, the supernet training paradigm in one-shot NAS introduces catastrophic forgetting, where each step of the training can deteriorate the performance of other architectures that contain partially-shared weights with current architecture. To overcome this problem of catastrophic forgetting, we formulate supernet training for one-shot NAS as a constrained continual learning optimization problem such that learning the current architecture does not degrade the validation accuracy of previous architectures. The key to solving this constrained optimization problem is a novelty search based architecture selection (NSAS) loss function that regularizes the supernet training by using a greedy novelty search method to find the most representative subset. We applied the NSAS loss function to two one-shot NAS baselines and extensively tested them on both a common search space and a NAS benchmark dataset. We further derive three variants based on the NSAS loss function, the NSAS with depth constrain (NSAS-C) to improve the transferability, and NSAS-G and NSAS-LG to handle the situation with a limited number of constraints. The experiments on the common NAS search space demonstrate that NSAS and it variants improve the predictive ability of supernet training in one-shot NAS with remarkable and efficient performance on the CIFAR-10, CIFAR-100, and ImageNet datasets. The results with the NAS benchmark dataset also confirm the significant improvements these one-shot NAS baselines can make.

Funding

Towards data-efficient future action prediction in the wild

Australian Research Council

Find out more...

History

Related Materials

  1. 1.
    DOI - Is published in 10.1109/TPAMI.2020.3035351
  2. 2.
    ISSN - Is published in 01628828

Journal

IEEE Transactions on Pattern Analysis and Machine Intelligence

Volume

43

Number

9247292

Issue

9

Start page

2921

End page

2935

Total pages

15

Publisher

IEEE

Place published

United States

Language

English

Copyright

© 2020 IEEE

Former Identifier

2006109295

Esploro creation date

2021-08-28

Usage metrics

    Scholarly Works

    Categories

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC