Episode 1: Training a classification model on MNIST with PyTorch
This video covers how to create a PyTorch classification model from scratch! It introduces all the fundamental components like architecture definition, optimizer, loss function, data loader, and Alfredo’s infamous 5 steps training! It shows you also how to train on a GPU, and how to add residual connections.
Check out our next video to find out and how to use dropout to fight overfitting, how to interactivly debug your model the Lightning way, and how to pick the best batch size for classification.
Alfredo Canziani is a Computer Science professor at NYU (check out his deep learning class -https://www.youtube.com/playlist?list=PLLHTzKZzVU9eaEyErdV26ikyolxOsz6mq)
Willam Falcon is an AI Ph.D. researcher at NYU, and creator and founder of PyTorch Lightning
01:58 Defining a model
05:34 Loss function
06:13 The training Loop
08:52 Alfredo’s 5 steps of training
16:26 Loading and preparing data
19:45 PyTorch: Recap
23:03 Logging training and validation losses
29:32 Training on GPUs
35:06 Improving the model with Residual connections