Inner layer hyperparametric meta-learning based on feature reuse

Jihong Cao; Xiangpeng Sun

doi:10.1117/12.2674729

2 May 2023 Inner layer hyperparametric meta-learning based on feature reuse

Jihong Cao, Xiangpeng Sun

Proceedings Volume 12642, Second International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2023); 126421H (2023) https://doi.org/10.1117/12.2674729
Event: Second International Conference on Electronic Information Engineering, Big Data and Computer Technology (EIBDCT 2023), 2023, Xishuangbanna, China

Abstract

Few-shot learning requires fast learning and adaptation during the learning process, which has been seen as a very challenging problem because of its stringent training requirements. With the development of deep learning, meta-learning is increasingly applied to solve few-shot problems. Model-agnostic meta-learning (MAML) is applied to model-agnostic tasks by finding common initialization parameters applicable to the model through both inner and outer loops. However, the generalization performance of MAML is not strong, and its inner loop is not fully functional. Therefore, we add learnable hyperparameters to the inner loop to make the model training results more relevant to the current task and reduce the number of layers of gradient updates without unnecessary computation. Experimental results show that the inner loop hyperparameter learning based on feature reuse (HLFR), finds better initialization parameters compared to MAML.

Citation Download Citation

Jihong Cao and Xiangpeng Sun "Inner layer hyperparametric meta-learning based on feature reuse", Proc. SPIE 12642, Second International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2023), 126421H (2 May 2023); https://doi.org/10.1117/12.2674729

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available