Docente
|
RODOLA' EMANUELE
(programma)
- Fundamentals of ML review
- Linear regression
- Classification
- Energy minimization
- Maximum likelihood estimation
- Optimization
- Quasi-newton methods
- Stochastic gradient descent
- Automatic differentiation
- Supervised, unsupervised and self-supervised learning
- Representation, geometry, stability, variability
- The curse of dimensionality in ML
- Neural networks
- Perceptron, multi-layer perceptron
- Backpropagation
- Properties of learnt representations
- Training neural networks
- Regularization
- Activation functions
- Weight initialization
- Batch normalization
- Hyperparameter optimization
- Parameter updates
- Dropout
- Convolutional neural networks
- Shift-invariance, co-variance and contra-variance
- Weight sharing
- Common architectures
- Residual networks
- Theory of deep learning
- Convergence
- Gradient flow
- Open problems
- Visualization, understanding and interpretability
- Frameworks and libraries (language: Python)
- Overview of DL frameworks (Keras, Tensorflow)
- PyTorch
- Transfer learning and domain adaptation
- Recurrent networks, long-short time memory
- Generative models
- Autoencoders
- Variational autoencoders
- Generative adversarial networks
- Geometric deep learning on non-Euclidean domains:
- Graphs
- Riemannian manifolds
- Point clouds
- Adversarial and universal attacks
- Applications
- Computer vision and graphics
- Network and graph analysis (fake news detection, Netflix problem)
- Audio synthesis
Data la natura altamente dinamica dell'area coperta da questo corso avanzato, non è previsto un testo unico di riferimento. Durante il corso verranno indicate e fornite di volta in volta le fonti sotto forma di articoli scientifici e capitoli di libri.
Come riferimento generale, i seguenti libri possono rivelarsi utili:
Deep Learning
Ian Goodfellow, Yoshua Bengio, Aaron Courville
MIT Press, 2016
Deep Learning with PyTorch
Vishnu Subramanian
Packt, 2018
|