Initial Investigation of Kolmogorov-Arnold Networks (KANs) as Feature Extractors for IMU Based Human Activity Recognition

TL;DR

This work investigates Kolmogorov-Arnold Networks (KANs) as low-level feature extractors for IMU-based human activity recognition, addressing the mismatch between sensor-encoded information and image-centric neural architectures. The authors implement several KAN-based FE variants that compute edge-wise nonlinear functions, represented as B-splines, and sum them at each node, learning spline parameters rather than weights. Across four public HAR datasets, KAN-based extractors achieve higher or comparable accuracy with significantly fewer parameters than CNN baselines, with RL-KAN FE typically performing best. The study demonstrates the potential of KANs for compact, effective low-level feature extraction in sensor-based HAR and provides a foundation for future end-to-end systems and pre-training strategies that leverage KAN representations.

Abstract

In this work, we explore the use of a novel neural network architecture, the Kolmogorov-Arnold Networks (KANs) as feature extractors for sensor-based (specifically IMU) Human Activity Recognition (HAR). Where conventional networks perform a parameterized weighted sum of the inputs at each node and then feed the result into a statically defined nonlinearity, KANs perform non-linear computations represented by B-SPLINES on the edges leading to each node and then just sum up the inputs at the node. Instead of learning weights, the system learns the spline parameters. In the original work, such networks have been shown to be able to more efficiently and exactly learn sophisticated real valued functions e.g. in regression or PDE solution. We hypothesize that such an ability is also advantageous for computing low-level features for IMU-based HAR. To this end, we have implemented KAN as the feature extraction architecture for IMU-based human activity recognition tasks, including four architecture variations. We present an initial performance investigation of the KAN feature extractor on four public HAR datasets. It shows that the KAN-based feature extractor outperforms CNN-based extractors on all datasets while being more parameter efficient.

Figures (11)

• Figure 1: Proposed KAN-based Feature Extractor Models (FE: Feature Extractor)
• Figure 2: Architecture of KAN as proposed by liu2024kan
• Figure 3: KAN-based Feature Extractor Block (Channel-wise)
• Figure 4: KAN-based Feature Extractor Block (Cross-channel-wise)
• Figure 5: Architecture of CNN-based Feature Extractor