Intro to PyTorch

Lecture 1 notebook

(local file at Lecture-1/)

Homework: Improve the Sonar base model

• Add more hidden layers or change the number of neurons in each layer.
• Read what the sigmoid activation does, then try alternatives such as ReLU, GELU, Tanh, or others from the PyTorch activation layers: https://docs.pytorch.org/docs/main/nn.html
• Optional: read what learning-rate schedulers are and try one from PyTorch: https://docs.pytorch.org/docs/2.12/generated/torch.optim.lr_scheduler.LRScheduler.html

Example solution

Lecture 2 notebook

(local file at Lecture-2/)

Homework: Improve the FashionMNIST FFNN and submit it

• Train a better feed-forward neural network for FashionMNIST.
• Try different hidden sizes, number of layers, activations, learning rates, schedulers, and data augmentations.
• Save your trained model as submission.pt with model_state_dict and a small config containing hidden_units and activation.
• Run the final notebook cell to submit your .pt file to the leaderboard.

Leaderboard: https://tues-ai.github.io/PyTorch-Intro/