Ágnes Vathy-Fogarassy, PhD; Tibor Dulai, PhD
Prerequirements:
-
Topics covered:
Students will learn about the following topics, including their individual training plans and interests:
T1. The transformer architecture and its variants:
- description of Generative Pre-trained Transformer, XLNet, and Text-to-Text Transfer Transformer architectures and their properties
- detailed explanation of transformer architectures (multi-head self-attention mechanism, positional encoding, their training process)
- transfer learning and fine-tuning for applying pre-trained transformers in new domains
T2. Generative Adversarial Networks (GANs) and the transformer-based generative models:
- the structure and training of GANs (the tasks of the generator and discriminator)
- the structure and operation of transformer-based generative models
T3. Embeddings and vector databases. The application of generative models and its ethical implications:
- the concept of embedding, possible formats of input and their mapping
- specialties and characteristics of vector databases
- possible applications of generative models with their ethical implications (e.g. content generation, style transfer, data correction)
- responsible AI
Literature:
1. David Foster: Generative Deep Learning: Teaching Machines to Paint, Write, Compose, and Play. O’Reilly Media, 2019.
2. Ian Goodfellow, Yoshua Bengio, Aaron Courville: Deep learning. MIT Press, 2016.