Green AI for 6G: Quantization of Neural Networks for Predictive Handover from Theory to Practice
Résumé
Quantization is widely regarded as a key technique for reducing inference latency and energy consumption in AI models, especially as 6G networks move toward AI-native architectures. However, its practical effectiveness varies significantly depending on the model and hardware platform. This paper investigates the impact of post-training quantization on LSTM, CNN and Transformer models used for predictive handover, a representative 6G telecom use case. We benchmark quantized and full-precision models on both CPU and GPU platforms, evaluating performance in terms of accuracy, latency and energy consumption. Our results show that while quantization can offer clear benefits, its gains are inconsistent particularly on many GPU hardware, due to limited INT8 support and suboptimal kernel execution. We identify current limitations and provide practical guidance for deploying quantized models effectively in telecom environments.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |