About

An overview course on basic neural network models and their application to image, text, and sound processing tasks.

Syllabus

• Feedforward networks. Matrix-vector differentiation. Automatic differentiation.
• Neural network optimization. DropOut. PyTorch.
• Initialization of neural networks. Types of normalization, convolutional neural networks, implementation of convolutional networks in PyTorch.
• Recurrent neural networks. Machine translation task. Attention mechanism. Transformer model.
• Object detection.
• Semantic segmentation. Instance/Panoptic segmentation.
• Reinforcement learning. Markov’s decision making process. Bellman equations.
• Policy iteration, Value iteration algorithms. Q-learning. DQN model.
• Algorithm Reinforce, A2C algorithm. Multi-armed bandits.
• Generative models. Variational autoencoder model.
• Autoregressive models: PixelCNN, PixelRNN. Reparametrization trick.
• Generative adversarial networks and their modifications. Image quality metrics.

Labworks

5 practical tasks and 1 lab work. Each task is evaluated from 10 points. The exam is evaluated from 10 points.

Grading

0.7 * semester points + 0.3 * exam points.

Prerequisites

Machine learning, statistics, optimization methods.