Abstract
Few-shot meta learning aims to obtain a prior from previous experiences, which is well used for new tasks during meta-test phase. The model-agnostic meta-learning (MAML) algorithm in [3] achieves this goal by finding a proper initial parameter at meta-training phase and this initialization could be quickly adapted to a new task by using the gradient descent at meta-test phase. However, it suffers from the problem of high computational complexity. In this paper, we propose a sparse model-agnostic meta-learning (SMAML) algorithm to alleviate the complexity problem and enhance the efficiency, where $L_{1}$ regularization is considered as an extra constraint. Experimental results on classification show that our well-placed modification indeed improves the computational efficiency in comparison to the original MAML algorithm.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.