AFFS: Adaptive Fast Frequency Selection Algorithm for Deep Learning Feature Extraction
Keywords: Discrete Cosine Transform, feature extraction, frequency domain, frequency components selection.
Abstract: As deep learning(DL) advances, effective feature extraction from big data remains critical for enhancing DL model's performance. This paper proposes a method for feature extraction in the frequency domain, utilizing advantages such as concentrated signal energy and pronounced data features. However, existing frequency component selection algorithms face challenges like difficulty adapting to diverse tasks and achieving only locally optimal results with extended processing times. To address these challenges, we introduce the Adaptive Fast Frequency Selection (AFFS) algorithm, tailored for various subsequent tasks. AFFS incorporates a frequency component selection factor layer, integrating it with the subsequent DL model to select globally optimal frequency component combinations for the DL model. Additionally, we propose a fast selection algorithm to expedite the process, leveraging the experimental observation of rapid convergence of selection factor ranking. Experimental results demonstrate that AFFS achieves superior performance across three datasets and three DL models. By using AFFS to select appropriate frequency components, even though our input data size is only 10\% of the original frequency feature, the classification accuracy of the model is improved by about 1\%. Furthermore, the early stopping mechanism can shorten the selection process by approximately 80\%.
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7102
Loading