Lecture/ Practice | Date | Topic | Lecturer | Materials | Notes | Feedback / discuss |
---|---|---|---|---|---|---|
1. | 19.2 | Recap: linear classifiers, linear regression, logistic regression, loss function, empirical risk minimisation, regularisation | BF | slides | thread | |
20.2 | (no lab) | - | ||||
2. | 26.2 | Artificial neurons, activation functions, network architectures; sidestep: stochastic neurons; sidestep: biological neurons | BF | slides | thread | |
27.2 | Lab1: generative vs. discriminative learning I | BF | task model | due 9.04 | ||
3. | 4.3 | Neural networks as classifiers, empirical risk minimisation, loss functions, model complexity and generalisation bounds; neural networks as nonlinear regression models | BF | slides | thread | |
5.3 | Seminar | BF | problems | |||
Reading: generalization bounds, VC dimension, large margin: (J. Mount), (V. V. V’yugin, Theorem 6.8) | ||||||
4. | 25.3 | Backpropagation | AS | slides bbb mp4 | thread | |
26.3 | Lab2: generative vs. discriminative learning II | BF | assignment template record 9:00 | due 7.05 | thread | |
5. | 1.4 | Stochastic Gradient Descent | AS | slides bbb mp4 | ||
2.4 | Seminar | AS | problems record | |||
Reading: Practical Recommendations for Gradient-Based Training Y. Bengio | ||||||
6. | 8.4 | Convolutional Neural Networks | AS | slides bbb mp4 | ||
9.4 | No meeting – Friday schedule | - | ||||
7. | 15.4 | Training neural networks 0: project pipeline, data collection, training/validation/test set, model selection (architecture), overfitting, early stopping | BF | slides bbb mp4 | thread | |
16.4 | Lab3: PyTorch, training pipeline | AS | assignment template | due 21.05 | ||
8. | 22.4 | Training neural networks 1: Data augmentation, Weight initialisation, Batch normalisation | BF | slides bbb mp4 | ||
23.4 | Seminar | AS | problems | thread | ||
9. | 29.4 | Training neural networks 2: Adaptive SGD methods | AS | slides bbb mp4 | ||
30.4 | Lab4: Pretrained CNN Finetuning | AS | task | due 28.05 | ||
Reading: Overview of GD Optimization, Mirror Descent | ||||||
10. | 6.5 | Training neural networks 3: L2 weight regularization, dropout, and a bit beyond | AS | slides bbb mp4 | ||
7.5 | Seminar | AS | problems notes | |||
11. | 13.5 | Training neural networks 4: adversarial patterns, robust learning approaches | BF | slides bbb mp4 | ||
14.5 | Lab5: Network visualisation, adversarial patterns | BF | assignment template record | due 11.06 | ||
12. | 20.5 | Generative models: VAE, GANs (introductory level) | BF | slides bbb mp4 | ||
21.5 | Lab6: Generative networks (optional/bonus lab) | BF | assignment bbb | submit before 18.6. | ||
13 | 27.5 | Supervised representation and similarity learning | Giorgos Tolias | slides bbb mp4 | ||
28.5 | Seminar | BF | problems | |||
14. | 3.6 | Recurrent neural networks: recurrent back-propagation, RNN, GRU, LSTM | BF | slides | ||
4.6 | (no lab) | |||||