Experimenting with training and running ML models on a “normal computer” (ie. no GPU or cloud servers, just locally on CPU/ram)
Don’t have a background in ML, so mostly just for learning purposes
Been playing a lot with the MNIST dataset. Trying things like training only on the examples the model gets wrong, or training only on random samples of the image (ie. using only a small subset of the pixels of the image as the input), or creating one model per label to overfit the data and then merging the models for testing, or just testing performance of different architectures and frameworks on the same problem