Frequency Composition for Compressed and Domain-Adaptive Neural Networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kwon, Yoojin, Suh, Hongjun, Lee, Wooseok, Gong, Taesik, Han, Songyi, Kim, Hyung-Sin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912397470466048
author Kwon, Yoojin
Suh, Hongjun
Lee, Wooseok
Gong, Taesik
Han, Songyi
Kim, Hyung-Sin
author_facet Kwon, Yoojin
Suh, Hongjun
Lee, Wooseok
Gong, Taesik
Han, Songyi
Kim, Hyung-Sin
contents Modern on-device neural network applications must operate under resource constraints while adapting to unpredictable domain shifts. However, this combined challenge-model compression and domain adaptation-remains largely unaddressed, as prior work has tackled each issue in isolation: compressed networks prioritize efficiency within a fixed domain, whereas large, capable models focus on handling domain shifts. In this work, we propose CoDA, a frequency composition-based framework that unifies compression and domain adaptation. During training, CoDA employs quantization-aware training (QAT) with low-frequency components, enabling a compressed model to selectively learn robust, generalizable features. At test time, it refines the compact model in a source-free manner (i.e., test-time adaptation, TTA), leveraging the full-frequency information from incoming data to adapt to target domains while treating high-frequency components as domain-specific cues. LFC are aligned with the trained distribution, while HFC unique to the target distribution are solely utilized for batch normalization. CoDA can be integrated synergistically into existing QAT and TTA methods. CoDA is evaluated on widely used domain-shift benchmarks, including CIFAR10-C and ImageNet-C, across various model architectures. With significant compression, it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA baseline.
format Preprint
id arxiv_https___arxiv_org_abs_2505_20890
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Frequency Composition for Compressed and Domain-Adaptive Neural Networks
Kwon, Yoojin
Suh, Hongjun
Lee, Wooseok
Gong, Taesik
Han, Songyi
Kim, Hyung-Sin
Computer Vision and Pattern Recognition
Artificial Intelligence
Modern on-device neural network applications must operate under resource constraints while adapting to unpredictable domain shifts. However, this combined challenge-model compression and domain adaptation-remains largely unaddressed, as prior work has tackled each issue in isolation: compressed networks prioritize efficiency within a fixed domain, whereas large, capable models focus on handling domain shifts. In this work, we propose CoDA, a frequency composition-based framework that unifies compression and domain adaptation. During training, CoDA employs quantization-aware training (QAT) with low-frequency components, enabling a compressed model to selectively learn robust, generalizable features. At test time, it refines the compact model in a source-free manner (i.e., test-time adaptation, TTA), leveraging the full-frequency information from incoming data to adapt to target domains while treating high-frequency components as domain-specific cues. LFC are aligned with the trained distribution, while HFC unique to the target distribution are solely utilized for batch normalization. CoDA can be integrated synergistically into existing QAT and TTA methods. CoDA is evaluated on widely used domain-shift benchmarks, including CIFAR10-C and ImageNet-C, across various model architectures. With significant compression, it achieves accuracy improvements of 7.96%p on CIFAR10-C and 5.37%p on ImageNet-C over the full-precision TTA baseline.
title Frequency Composition for Compressed and Domain-Adaptive Neural Networks
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2505.20890