• Test on STM32MPU - real-time inferencing
• Figure out original model from STMCubeMX-AI Analysis
• Test to see if this model predicts with consistency
• Visualization - accuracy confusion matrix for predictions (heatmap)
• Visualization for MNIST dataset
• Added quantization within TFLite

Quantization was achieved from float32 to uint8 datatypes via TensorFlow Lite. This quantized model is outputted as a .tflite file within the ./results/ directory. Additionally, quantization may be achieved within STM32CubeMX from the .h5 file, which has unquantized, float32 weights.

EMNIST - Cohen, G., Afshar, S., Tapson, J., & van Schaik, A. (2017). EMNIST: an extension of MNIST to handwritten letters. Retrieved from https://arxiv.org/abs/1702.05373