Popis předmětu - BECM33DPL
| BECM33DPL | Deep Learning Essentials | ||
|---|---|---|---|
| Role: | P, V, PV | Rozsah výuky: | 2P+2C |
| Katedra: | 13133 | Jazyk výuky: | EN |
| Garanti: | Zimmermann K. | Zakončení: | Z,ZK |
| Přednášející: | Neumann L. | Kreditů: | 6 |
| Cvičící: | Neumann L., Raikwar P., Škvrna J. | Semestr: | Z |
Webová stránka:
https://cw.fel.cvut.cz/wiki/courses/becm33dpl/startAnotace:
The course presents an engineering view of fundamental AI concepts.. The course starts by the basics - the two-class and multi-class linear classifier on RGB images; maximum likelihood estimation and KL divergence as the origin of loss functions and the reason overfitting arises; and then follows with neural networks - perceptron, MLP, and backpropagation expressed through the vector-Jacobian product and autograd - extended to convolutional layers. The remainder covers the components that make deep networks train and generalize (activation functions, BatchNorm, Dropout, weight decay; optimizers from SGD and momentum through RMSProp to Adam) and the architectures that put them to use: backbones such as ResNet, EfficientNet, and Transformers, then task-specific models for object detection, pose estimation, and generation, closing with reinforcement learning (RL) and implicit layers. This course is part of inter-university program prg.ai Minor. This program includes the best of AI education in Prague with the goal to provide the students deeper and broader view of artificial intelligence. More info on the program web https://prg.ai/minor.Cíle studia:
The course teaches deep learning methods on practical applications from computer vision and robotics. The overall goal is timeless, universal knowledge rather than listing all known deep learning architectures. Students are assumed to have basic working prior knowledge of mathematics (gradient, linear equations). In labs, students will build on existing templates in order to build real AI applications and models including RL, vision transformers and generative networks.Obsah:
https://cw.fel.cvut.cz/wiki/courses/becm33dpl/startOsnovy přednášek:
| 1. | Machine learning 101: engineering view on models, loss, learning, learning issues, regression, classification | |
| 2. | Linear classifier: two-class and multi-class linear classifier on RGB images | |
| 3. | Maximum Likelihood Estimation (MLE), KL Divergence: Where does loss function come from, why overfitting exists? | |
| 4. | Neural Networks: Perceptron, MLP, Backpropagation, Vector-Jacobian product, Autograd | |
| 5. | The story of the cat's brain surgery: cortex + convolutional layer and its Vector-Jacobian-Product (VJP), fun with backpropagation | |
| 6. | Activation, Normalization and Regularization: activation functions, BatchNorm, Dropout, weight decay | |
| 7. | Optimization: SGD, momentum, RMSProp, Adam | |
| 8. | Backbone architectures: ResNet, EfficientNet, Transformers | |
| 9. | Task-specific architectures: Object detection, pose estimation, generative networks | |
| 10. | Reinforcement learning | |
| 11. | Implicit layers |
Osnovy cvičení:
Literatura:
Ian Goodfellow and Yoshua Bengio and Aaron Courville, Deep learning, MIT press, 2016, http://www.deeplearningbook.org| F. | Fleuret. The Little Book of Deep Learning. lulu.com, 2023. |
Požadavky:
Klíčová slova:
Deep learning, neural networks, AIPředmět je zahrnut do těchto studijních plánů:
| Stránka vytvořena 16.11.2026 09:54:38, semestry: Z/2027-8, L/2025-6, L/2029-30, L/2027-8, Z/2026-7, L/2028-9, L/2026-7, Z/2028-9, připomínky k informační náplni zasílejte správci studijních plánů | Návrh a realizace: I. Halaška (K336), J. Novák (K336) |