MNIST ("Modified National Institute of Standards and Technology") is the de facto “hello world” dataset of computer vision. Since its release in 1999, this classic dataset of handwritten images has served as the basis for benchmarking classification algorithms. As new machine learning techniques emerge, MNIST remains a reliable resource for researchers and learners alike.
Goal is to correctly identify digits from a dataset of tens of thousands of handwritten images.
Creating LeNet5 nn class module
conv2d => relu => maxpooling => conv2d => relu => maxpooling => fully connected layer(fc)1 => fc2 => softmax output
pytorch
numpy
pandas
matplotlib
conda install pytorch torchvision -c pytorch
- Activate the fastai environment:
source activate pytorch
- Run the notebook:
jupyter notebook