Open Internet by MindsNet

Mitigating Accuracy Loss in Model Quantization

Quantizing a trained model to improve inference speed on edge devices like the Raspberry Pi 4 results in significant accuracy loss. The author experienced a drop from 99% to 73% accuracy when switching from floating-point to integer quantization. The cause of this loss is unclear and may be related to the dataset, hyperparameters, or inherent model limitations.

Computing & Technology, Computer Science, Machine Learning