Fastai Notebooks

Deep Learning from the Foundations

These notebooks roughly follow the Fastai course Part 2: Deep Learning from the Foundations .

Basics:

• Tensors and Matrix Multiplication
• Weight Initialization
• Fully Connected Layers
• Convolutions

Training Loop:

• Mini-Batch Training
• DataBunch, Learner, and Runner
• Recorder and Param Scheduling
• Runner v2

Inside the Network:

• CUDA, CNN with Hooks
• BatchNorm and Others

Further Flexibility and Customizing:

• DataBlock API
• Optimizers
• Learner v2

Data Augmentation:

• Data Augmentation
• MixUp Label Smoothing
• fp 16

NLP with Transfer Learning

• Preprocess Text
• Transfer Learning
• AWD LSTM
• Language Model Pretain
• Universal Language Model Fine Tuning (ULMFiT)

Extras:

• Exports
• Audio
• Cycle Gan

Papers with Code:

• LSUV
• Bag of Tricks