GAC-KAN: An Ultra-Lightweight GNSS Interference Classifier for GenAI-Powered Consumer Edge Devices

2026-02-13 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Internet of Things (IoT) & Connected Devices, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

The GAC-KAN framework introduces an ultra-lightweight Global Navigation Satellite System (GNSS) interference classifier designed for resource-constrained consumer edge devices, particularly those running Generative AI (GenAI) applications. It addresses the dual challenges of data scarcity and computational efficiency by first employing a physics-guided simulation to synthesize a large-scale jamming dataset. Second, it utilizes a Multi-Scale Ghost-ACB-Coordinate (MS-GAC) backbone for efficient feature extraction, combining Asymmetric Convolution Blocks (ACB) and Ghost modules. Finally, it replaces traditional Multi-Layer Perceptron (MLP) decision heads with a Kolmogorov-Arnold Network (KAN) for superior non-linear mapping with fewer parameters. GAC-KAN achieves 98.0% overall accuracy with only 0.13 million parameters, approximately 660 times fewer than Vision Transformer (ViT) baselines, and requires 0.19 G FLOPs, making it suitable for "always-on" background operation.

Key takeaway

For AI scientists and embedded systems engineers developing GenAI-powered consumer electronics, GAC-KAN demonstrates a viable path to integrate critical security functions without compromising primary application performance. Your teams should consider adopting physics-guided data synthesis and KAN-based architectures to achieve high accuracy in GNSS interference classification with minimal computational and memory footprints, ensuring robust "always-on" security on resource-limited edge devices.

Key insights

GAC-KAN offers an ultra-lightweight, high-accuracy GNSS interference classifier for GenAI edge devices via synthetic data and KANs.

Principles

• Synthesize data to overcome real-world scarcity.
• Combine Ghost modules and ACBs for efficient feature extraction.
• KANs offer superior non-linear mapping with fewer parameters.

Method

The GAC-KAN method involves physics-guided simulation for data generation, a Multi-Scale Ghost-ACB-Coordinate (MS-GAC) backbone for feature extraction, and a Kolmogorov-Arnold Network (KAN) as the classification head.

In practice

• Use physics-guided simulation for scarce real-world data.
• Integrate Ghost modules and ACBs for model compression.
• Employ KANs for parameter-efficient non-linear classification.

Topics

• GNSS Interference Classification
• Kolmogorov-Arnold Networks
• Edge AI
• Lightweight Neural Networks
• Generative AI

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

Related on AIssential

• Kolmogorov Arnold networks (KAN) for aerodynamic prediction: a comparison with MLPs and GNNs — KANs show promise for aerodynamic prediction but are marginally inferior to MLPs and less stable than GNNs.
• GenAI for Energy-Efficient and Interference-Aware Compressed Sensing of GNSS Signals on a Google Edge TPU — GenAI on Edge TPUs enables real-time, energy-efficient GNSS jamming classification and data compression at the receiver.
• GEAK-Triton v2 Family of AI Agents: Kernel Optimization for AMD Instinct GPUs — AI agents can automate and significantly optimize GPU kernel generation and refinement for AMD Instinct GPUs.
• GeMoE: Gating Entropy is All You Need for Uncertainty-aware Adaptive Routing in MoE-based Large Vision-Language Models — GeMoE adaptively routes tokens in MoE models by using gating entropy to manage complexity and uncertainty.
• KAN-MLP-Mixer: A comprehensive investigation of the usage of Kolmogorov-Arnold Networks (KANs) for improving IMU-based Human Activity Recognition — Strategically combining KANs for data embedding and classification with MLPs for feature mixing significantly enhances noisy IMU-based HAR.
• Structural Kolmogorov-Arnold Convolutions: Learnable Function on the Values or the Filter Shape as Parameter-Efficient Alternative to Per-Edge Convolutional KANs — Structural KANs achieve parameter efficiency and strong performance by applying learnable functions to convolution structure or filter shape.

Open in AIssential →

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