KAN-MLP-Mixer: A comprehensive investigation of the usage of Kolmogorov-Arnold Networks (KANs) for improving IMU-based Human Activity Recognition

2025-04-10 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Internet of Things (IoT) & Connected Devices · Depth: Advanced, extended

Summary

The KAN-MLP-Mixer is a novel hybrid neural network architecture designed to improve IMU-based Human Activity Recognition (HAR) by combining the strengths of Kolmogorov-Arnold Networks (KANs) and Multi-Layer Perceptrons (MLPs). While KANs offer high precision on clean data, they struggle with noisy real-world sensor inputs, unlike robust MLPs. This architecture addresses this by employing an EfficientKAN module for adaptive input embedding, standard MLP layers for intermediate feature mixing, and a specialized LarctanKAN module for final activity classification. Across eight public HAR datasets, the KAN-MLP-Mixer achieved an average macro F1 score improvement of 5.33% compared to pure-MLP models, significantly outperforming standalone KAN and MLP baselines. The hybrid strategy also consistently boosted the performance of other state-of-the-art HAR architectures.

Key takeaway

For machine learning engineers developing Human Activity Recognition (HAR) systems for wearable devices, you should adopt a modular hybrid approach rather than fully replacing MLPs with KANs. Implement EfficientKAN for initial data embedding and LarctanKAN for classification, while retaining MLPs for intermediate feature mixing. This strategy, exemplified by KAN-MLP-Mixer, consistently improves accuracy and robustness on noisy IMU data, offering a practical pathway to leverage KANs' expressiveness without sacrificing MLP's efficiency.

Key insights

Strategically combining KANs for data embedding and classification with MLPs for feature mixing significantly enhances noisy IMU-based HAR.

Principles

• KANs excel at initial data embedding for smooth signals.
• MLPs provide robust, efficient intermediate feature mixing.
• LarctanKANs offer stable, discriminative classification boundaries.

Method

The KAN-MLP-Mixer architecture processes raw sensor input via an EfficientKAN embedding, uses MLP layers for feature mixing, and a LarctanKAN module for final activity classification.

In practice

• Apply EfficientKAN for raw sensor data embedding.
• Utilize MLPs for stable latent feature transformation.
• Integrate LarctanKAN for robust classification boundaries.

Topics

• Kolmogorov-Arnold Networks
• Human Activity Recognition
• Hybrid Neural Networks
• Wearable Sensors
• IMU Data
• EfficientKAN

Code references

• Blealtan/efficient-kan

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• A Practitioner's Guide to Kolmogorov-Arnold Networks — KANs offer enhanced expressivity and interpretability over MLPs by using learnable univariate basis functions on network edges.
• WISE-HAR: A Generalizable Ensemble Deep Learning Framework for WiFi-Based Human Activity Recognition — Ensemble deep learning with data augmentation significantly enhances WiFi-based HAR generalization and robustness across diverse environments.
• HiRo: A Compact Four-Directional Hierarchical Reservoir Token-Mixer for Efficient Image Classification — HiRo combines hierarchical reservoir computing and shifted windows for efficient, high-accuracy image classification with few parameters.
• Chaotic Oscillator Networks for Classification Tasks — Machine learning can approximate coupling terms in chaotic oscillator networks for scalable, efficient classification and pattern recognition.
• Improved state mixing in higher-order and block diagonal linear recurrent networks — Richer state mixing in LRNNs enhances expressivity and efficiency for long-sequence modeling.
• Selective Correlation Based Knowledge Distillation for Ground Reaction Force Estimation — SCKD distills knowledge from large models to create efficient, accurate GRF estimation models for wearable sensors.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.