This is a repository of all the projects I've worked on while doing the fastai courses. The courses are incredibly good, and I would recommend them to everyone regardless of their background.

Find the detailed explanations for the projects in the READMEs of the respective folders or on my blog

• Image classification: Pneumonia detection using X-rays [code]
• Mixed precision training on X-ray dataset [article] [code]

• Building a custom classifier using google images [code]
• SGD and discriminative learning rates [article] [code]

• Multi-label classification of satellite images [article] [code]
• Image segmentation of street data [article] [code]
• Transfer learning in NLP for sentiment analysis of movie reviews [article] [code]
• Image regression to predict the centre of a person's face [article] [code]

• Machine learning vs Deep learning for Tabular data [article] [code_ml] [code_dl]
• Collaborative filtering on song dataset [article] [code]

• Multiclass classification on seedling dataset and understanding weight decay [article] [code]
• Sound classification of urban noises using spectograms [article]
• Momentum, Adam's optimizer and more [article] [code]

• Dropout in neural networks [article]
• Data augmentation for computer vision [article] [code]
• Heatmaps and CNNs [article] [code]

• CNNs and Resnet using fastai and PyTorch. [[article]] [code]

1. Matrix multiplication (from the normal loops to PyTorch speed) [notebook][article]

2. Initializing neural networks [notebook] [article]

3. Customize your training loop using callbacks [notebook1] [notebook2] [article]

4. Convolutional neural networks and hooks [notebook] - In this notebook we use nn.Sequential to build a basic convolutional neural network. We then learn how to use PyTorch's hooks to access data during training. This can be used to figure out how the mean and standard deviation of the output of every layer varies with respect to number of epochs. Check the notebook for more information.