SMORE: Similarity-based Hyperdimensional Domain Adaptation for Multi-Sensor Time Series Classification

Junyao Wang,Mohammad Abdullah Al Faruque

2024-02-21

Abstract:Many real-world applications of the Internet of Things (IoT) employ machine learning (ML) algorithms to analyze time series information collected by interconnected sensors. However, distribution shift, a fundamental challenge in data-driven ML, arises when a model is deployed on a data distribution different from the training data and can substantially degrade model performance. Additionally, increasingly sophisticated deep neural networks (DNNs) are required to capture intricate spatial and temporal dependencies in multi-sensor time series data, often exceeding the capabilities of today's edge devices. In this paper, we propose SMORE, a novel resource-efficient domain adaptation (DA) algorithm for multi-sensor time series classification, leveraging the efficient and parallel operations of hyperdimensional computing. SMORE dynamically customizes test-time models with explicit consideration of the domain context of each sample to mitigate the negative impacts of domain shifts. Our evaluation on a variety of multi-sensor time series classification tasks shows that SMORE achieves on average 1.98% higher accuracy than state-of-the-art (SOTA) DNN-based DA algorithms with 18.81x faster training and 4.63x faster inference.

Machine Learning

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address the issue of **distribution shift** in multi-sensor time series data classification for Internet of Things (IoT) applications. Specifically:

1. **Distribution Shift Challenge**: When machine learning (ML) models are deployed on data distributions different from the training data, the model performance significantly degrades. For example, a human activity recognition model may systematically fail when tested on populations with different age groups or demographic characteristics.

2. **Efficient Solutions under Resource Constraints**: Existing deep learning methods perform poorly in handling multi-sensor time series data with complex spatial and temporal dependencies and require complex architectures and substantial computational resources. Therefore, these methods are not suitable for resource-constrained edge devices.

To address these issues, the authors propose a novel domain adaptation algorithm based on Hyperdimensional Computing (HDC) called SMORE. SMORE leverages the efficient parallel operations of HDC to dynamically customize the model at test time, explicitly considering the domain context of each sample to mitigate the negative impact of domain shift. Experimental results show that SMORE achieves an average accuracy improvement of 1.98% over state-of-the-art deep learning domain adaptation algorithms on various multi-sensor time series classification tasks, with a training speed increase of 18.81 times and an inference speed increase of 4.63 times. Additionally, SMORE demonstrates lower inference latency and energy consumption on resource-constrained hardware devices such as Raspberry Pi and NVIDIA Jetson Nano.