TensorFlow/Pytorch implementation of computer vision/character recognition.
Conclusion While there are many more combinations for these hyperparameters, and there are other CNN functionalities that were not a part of this exploration, we gained valuable insight into how to manage and improve a convolutional neural net for handwriting recognition. We successfully identified combinations of data that best helps the model learn different classes of images, layers to improve the ability to recognize detailed differences, kernel size to get a broader scope of analysis and pooling to counteract overfitting and introduce translation invariance. While improvements can still be made to this model, a 92% accuracy score from only 4 hyperparameters, shows what a significant impact they have on the model’s capabilities.