Assessing the Value of Transfer Learning Metrics for Radio Frequency Domain Adaptation - Article

Aug 2, 2026

Upload by Oliver

Transfer learning (TL) is widely adopted in computer vision and natural language processing because it leverages prior knowledge from related data distributions, enabling stronger performance with less training effort. In radio frequency machine learning (RFML) and wireless communications, TL remains underutilized and its practical guidance is limited. This study evaluates whether established transferability metrics can support model selection and accuracy forecasting for radio frequency domain adaptation. Results indicate LEEP and LogME correlate with post-transfer accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages21

Reading Time53 min

Introduction

Radio communications and RFML context

Transfer learning motivation in RFML

Transferability metrics in related modalities

Evaluated metrics and expected behavior

Experimental results and analysis

Correlation with post-transfer accuracy

machine learning & knowledge extraction Article Assessing the Value of Transfer Learning Metrics for Radio Frequency Domain Adaptation Lauren J. Wong 1,2,3, *, Braeden P. Muller 2,3, Sean McPherson 1 and Alan J. Michaels 2,3 Citation: Wong, L.J.; Muller, B.P.; McPherson, S.; Michaels, A.J. Assessing the Value of Transfer Learning Metrics for Radio Frequency Domain Adaptation. Mach. Learn. Knowl. Extr. 2024, 6, 1699–1719 . [https://doi.org/10.3390/](https://doi.org/10.3390/)[ ](https://doi.org/10.3390/)make6030084 Academic Editor: Razavi-Far Roozbeh Received: 13 June 2024 Revised: 8 July 2024 Accepted: 18 July 2024 Published: 25 July 2024 Copyright: © 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license ([https://](https://)[ ](https://)[creativecommons.org/licenses/by/](creativecommons.org/licenses/by/)[ ](creativecommons.org/licenses/by/)[4.0/](4.0/)) . 1 Intel AI Lab, Santa Clara, CA 95054, USA; [sean.mcpherson@intel.com](sean.mcpherson@intel.com) 2 National Security Institute, Virginia Tech, Blacksburg, VA 24060, USA; [braedenm@vt.edu](braedenm@vt.edu) (B.P.M.); [ajm@vt.edu](ajm@vt.edu) (A.J.M.) 3 Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, VA 24060, USA * Correspondence: [lauren.wong@intel.com](lauren.wong@intel.com) Abstract: The use of transfer learning (TL) techniques has become common practice in fields such as computer vision (CV) and natural language processing (NLP) . Leveraging prior knowledge gained from data with different distributions, TL offers higher performance and reduced training time, but has yet to be fully utilized in applications of machine learning (ML) and deep learning (DL) techniques and applications related to wireless communications, a field loosely termed radio frequency machine learning (RFML) . This work examines whether existing transferability metrics, used in other modalities, might be useful in the context of RFML. Results show that the two existing metrics tested, Log Expected Empirical Prediction (LEEP) and Logarithm of Maximum Evidence (LogME), correlate well with post-transfer accuracy and can therefore be used to select source models for radio frequency (RF) domain adaptation and to predict post-transfer accuracy. Keywords: machine learning; deep learning; transfer learning; domain adaptation; radio frequency machine learning 1. Introduction Modern day radio communications systems (Figure 1) allow users to send information across vast distances at near instantaneous speeds. The introduction of ML and DL techniques to modern radio communications systems has the potential to provide increased performance and flexibility when compared to traditional signal processing techniques. For example, cognitive radios (CRs) are capable of autonomously modifying parameters such as the modulation scheme, center frequency, bandwidth, and power in response to the external RF environment to provide continuous, high quality service to the end-user while complying with system and regulatory constraints [1] . While RFML and CR approaches inevitably overlap, RFML differs from CR in that RFML only aims to utilize autonomous feature learning from raw RF data to learn the characteristics to detect, identify, and recognize signals-of-interest [2] and is sometimes used off-board the radio itself and without the intent to re-configure the radio. In other words, RFML approaches can be seen as a component of a larger CR system. Nevertheless, both CR and RFML have broad utility in both the commercial and defense sectors [2–4] and are expected to be critical components of the upcoming 6G standard [5] . The RF system overview shown in Figure 1 identifies the parameters/variables that each component of an RF system impacts. Such components make up the domain that may differ significantly across transmitte

FAQs

Why is transfer learning useful in radio frequency domain adaptation?

Transfer learning reuses prior knowledge from a source domain to improve performance on a related target domain, addressing performance degradations when deployment conditions change.

Which transferability metrics are evaluated in this work?

The study evaluates Log Expected Empirical Prediction (LEEP) and Logarithm of Maximum Evidence (LogME) for their usefulness in RFML settings.

How do LEEP and LogME relate to post-transfer accuracy?

Results show both metrics correlate well with post-transfer accuracy, enabling source model selection for radio frequency domain adaptation and prediction of transfer outcomes.

Aug 2, 2026

Upload by Oliver