Hybrid KAN-MLP Architecture Improves Human Activity Recognition from Wearable Sensors
Researchers developed a hybrid neural network combining Kolmogorov-Arnold Networks (KANs) with traditional multi-layer perceptrons (MLPs) for human activity recognition from wearable inertial sensors. While KANs excel at learning complex functions on clean data, they struggle with noisy real-world datasets, whereas MLPs are more robust but less precise. The hybrid model achieved 5.33% average improvement over pure-MLP baselines across eight public datasets, suggesting strategic component placement matters more than wholesale replacement.
A new study on arXiv investigates how to effectively integrate Kolmogorov-Arnold Networks (KANs)—a recently developed neural architecture—into human activity recognition (HAR) systems that process data from wearable inertial measurement units (IMUs). The research identifies a fundamental trade-off: KANs demonstrate exceptional precision on clean, low-dimensional data but degrade in performance when exposed to noise and imperfect real-world sensor data, while conventional MLPs sacrifice some precision for robustness and computational efficiency. Rather than replacing all MLP components with KANs, the authors propose a hybrid architecture that strategically positions KAN modules in the input embedding layer, retains MLPs for intermediate feature processing, and uses a specialized LarctanKAN module for final classification. Testing across eight public HAR datasets, the hybrid model consistently outperformed both pure-KAN and pure-MLP baselines, with an average macro F1 score improvement of 5.33% over the MLP baseline. The findings suggest that careful architectural design combining multiple neural paradigms yields more robust and accurate models for real-world wearable sensing applications.
What's missing
The study does not discuss computational cost comparisons (inference time, memory usage) between the hybrid model and pure-MLP or pure-KAN baselines, despite mentioning computational efficiency as a motivation. Additionally, the paper does not specify which of the eight datasets are proprietary versus publicly available, limiting reproducibility assessment.
What different sources said
- arXiv cs.AICenter
KAN-MLP-Mixer: A comprehensive investigation of the usage of Kolmogorov-Arnold Networks (KANs) for improving IMU-based Human Activity Recognition
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.