FEEC/CPG - Caderno de Horário (Oferecimento de Disciplinas) - CPG
Caderno Horário:
IA376 - Tópicos em Engenharia de Computação VII
Turma: N - Período: 1/2026 - Tipo Período: 1o. período letivo - Disciplina: 4 créditos.
Tema: Deep Generative Modeling
Ementa: Generative models based on deep neural networks have achieved surprising results in the realistic synthesis of diverse signals, such as text, audio, images, and biosignals, demonstrating their potential in time series modeling, sample generation for data-scarce areas, and expanding the horizons of artificial intelligence. This advanced course provides a comprehensive overview of the principles, architectures, and applications of modern generative modeling. Topics include: a historical overview of generative models; distinctions between generative and discriminative approaches; statistical foundations of generative modeling; deep learning–based architectures such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Transformers, Diffusion Models, and Energy-Based Models; multimodal and foundation models; and the ethical, social, and economic implications of generative AI. Students will engage actively in the course by presenting seminars in English on recent state-of-the-art research papers in generative modeling. Throughout the semester, students will also be required to develop a final project that demonstrates both their theoretical understanding and practical skills in designing and applying generative AI models. Prior experience in Deep Learning is required to ensure meaningful participation and high-level discussion.
Bibliografia: Tomczak, J. M. (2024). Deep Generative Modeling. Springer International Publishing.
Conteudo Programático: 1. Foundations
Classical Synthesis Models (Rule-based vs. Data-driven)
Modeling & Evaluating Data Distributions
Generative vs. Discriminative Paradigms
2. Generative Adversarial Networks (GANs)
The Adversarial Principle
Key Architectures (DCGAN, StyleGAN)
Training Challenges & Evaluation
3. Variational Autoencoders (VAEs)
Latent Variable Modeling
The ELBO Objective
Applications in Generation & Representation
4. Autoregressive Models
Sequential Generation Principle
Transformers & Attention Mechanisms
Large Language Models (LLMs) & Foundation Models
5. Advanced Architectures
Diffusion Models (Denoising Processes)
Energy-Based Models (EBMs)
6. Broader Context
Multimodal Generative Models
Ethical & Socioeconomic Impacts
Obs.: Consultar Catálogo vigente na DAC.
Obs.: Consultar Catálogo vigente na DAC.
Forma Avaliação: Student evaluation will be based on research seminars (20%), engagement in computational activities (20%), and a final applied project (60%).
Ofertar para Graduação:
Sim Número Limite de Alunos de Graduação:
5
Aceita Estudante Especial:
Sim
Número de Alunos Total:
de 6 até 20
Mais informações:
Docente(s):
- Paula Dornhofer Paro Costa ( Matrícula: 309352 - e-mail: paula@fee.unicamp.br - Ramal: 13706 )
Horário(s):
- quarta-feira das 08:00 às 12:00
Sala(s):
- FE13
Turma(s):
- N