Fixed-Point Quantization for Neural Networks

This project studies the robustness of neural networks to fixed-point quantization for deployment on embedded hardware.

Research Goal

Understand how quantization affects model performance and develop training methods that improve robustness.

Initial Experiments

• Synthetic regression datasets
• Linear regression with MSE loss
• Quantization of trained weights to different precision levels

Project Structure

src/ – core implementation
experiments/ – experiment scripts
notebooks/ – exploratory experiments
results/ – plots and experiment outputs

Authors

Project laboratory research.