Testing various CNN architectures to detect grayscale circles in images with varying degrees of noise. Both DIOU (Distance-IoU) and MSE (Mean square error) loss are used.
Kaggle Fuits-360 dataset, containing 60 fruit classes and ~30k images. The CNN implemented has three convolutional layers, trains on a CPU in <30 minutes, and achieves 92% accuracy on this task.
Scripts to aggregate and format lists of English, French, and random words. This is part of a larger machine learning project to detect the language of a new string given a training set. Amazon Machine Learning and the AWS Prediction API were used.
Image classification of MNIST dataset (handwritten numbers). The processed dataset can be found here. This classic ML problem is part of a tensorflow tutorial, howevever my implementation deviates significantly. I achieve 94% accuracy.
Teaching a computer to play air hockey.