P2W: From Power Traces to Weights Matrix -- An Unconventional Transfer Learning Approach

TL;DR

This work tackles data-scarce transfer learning for embedded ML by introducing P2W, which retrieves an approximated weights matrix from a target model running on an SoC through power-trace analysis. An EDNN maps a single power trace to a weight-like representation, enabling initialization of a new model that is subsequently fine-tuned on a small dataset, yielding up to approximately a 3x accuracy gain over training from scratch. The approach demonstrates notable transfer of knowledge (over 60%) in multiple tasks (ECG/Diabetes/Sleep) and shows robustness to imbalanced data, though it relies on a clone SoC and knowledge of the target topology. Overall, P2W expands transfer-learning applicability to scenarios where direct access to model parameters is unavailable, with practical implications for privacy-conscious and data-limited embedded deployments.

Abstract

The rapid growth of deploying machine learning (ML) models within embedded systems on a chip (SoCs) has led to transformative shifts in fields like healthcare and autonomous vehicles. One of the primary challenges for training such embedded ML models is the lack of publicly available high-quality training data. Transfer learning approaches address this challenge by utilizing the knowledge encapsulated in an existing ML model as a starting point for training a new ML model. However, existing transfer learning approaches require direct access to the existing model which is not always feasible, especially for ML models deployed on embedded SoCs. Therefore, in this paper, we introduce a novel unconventional transfer learning approach to train a new ML model by extracting and using weights from an existing ML model running on an embedded SoC without having access to the model within the SoC. Our approach captures power consumption measurements from the SoC while it is executing the ML model and translates them to an approximated weights matrix used to initialize the new ML model. This improves the learning efficiency and predictive performance of the new model, especially in scenarios with limited data available to train the model. Our novel approach can effectively increase the accuracy of the new ML model up to 3 times compared to classical training methods using the same amount of limited training data.

Figures (5)

• Figure 1: High-level overview of the phases in the P2W transfer learning approach, including EDNN Model Construction, EDNN-based Power Analysis, and Initializing and Fine-tuning of the new ML model.
• Figure 2: An example of a weights matrix construction.
• Figure 3: Accuracy of new ML models trained using only $D_\text{small}$ vs. new ML models obtained using the P2W approach.
• Figure 4: Accuracy of new ML models and target ML models performing EEG, Diabetes, and Sleep classification tasks.
• Figure 5: Accuracy and F1-score of ML models obtained with and without P2W using balanced or imbalanced $D_\text{small}$.