Università Roma Tre

Course Program

Introduction to Deep Learning and the main paradigms of modern Artificial Intelligence. Fundamentals of deep neural networks: artificial neurons, activation functions, optimization methods, and training techniques. Discriminative and generative neural architectures.

Convolutional Neural Networks (CNNs) for image and signal analysis: convolution, pooling, feature extraction, and major applications in computer vision. Embeddings and distributed data representations.

Neural models for Natural Language Processing and introduction to Transformer architectures: attention mechanisms, tokenization, Large Language Models (LLMs), pre-training, and the use of generative models for text and multimodal content generation. Modern prompting techniques and interaction paradigms with generative models.

Introduction to modern generative models for images and text, with particular focus on autoregressive models and diffusion models. Applications to image generation, multimedia content generation, and intelligent conversational systems.

Recent paradigms based on intelligent agents and tool orchestration: agent-based architectures, external tool usage, memory mechanisms, and planning of complex tasks. Introduction to multi-agent systems and fundamental concepts of decision making and reinforcement learning for AI-based autonomous systems.

Applications of Deep Learning in several domains, including computer vision, speech recognition, natural language processing, machine translation, and intelligent autonomous systems.

Practical activities in Python using modern Deep Learning frameworks for the design, training, evaluation, and application of discriminative and generative neural models.

Lecture notes by the professor.