AI reading noteIan Goodfellow, Yoshua Bengio & Aaron Courville
Deep Learning
A foundation in the mathematics and practice of representation learning. It connects optimization, regularization, convolutional networks, sequence models, and the difficulties of training deep systems.
Deep Learning
Key ideas
- Generalization is a separate problem from fitting the training set.
- Optimization choices and data representation shape what a model can learn.
- Architecture should follow the structure of the input and the task.
How I would apply it
Compare training and validation curves before changing model size.