Visão Geral
Este curso apresenta os fundamentos do algoritmo de Backpropagation, o principal método de aprendizado utilizado em redes neurais artificiais e Deep Learning. O curso explora a intuição, a base matemática e o fluxo completo de treinamento de redes neurais, explicando como os erros são calculados e propagados para ajuste dos pesos. Ao final, o aluno compreenderá claramente como redes neurais aprendem e estará preparado para arquiteturas mais avançadas como CNNs, RNNs e Transformers.
Conteúdo Programatico
Module 1: Introduction to Backpropagation
- What is learning in neural networks
- Role of backpropagation in Deep Learning
- Supervised learning overview
Module 2: Neural Network Training Flow
- Forward pass
- Prediction and output generation
- Error calculation
Module 3: Loss Functions
- Mean Squared Error
- Cross-Entropy Loss
- Choosing the right loss function
Module 4: Gradients and Derivatives
- Derivatives intuition
- Gradients in neural networks
- Sensitivity analysis
Module 5: Chain Rule Fundamentals
- Chain rule concept
- Applying chain rule step by step
- Backward flow of errors
Module 6: Backward Pass
- Error propagation
- Gradient calculation for weights
- Gradient flow across layers
Module 7: Weight Update Mechanism
- Gradient Descent
- Learning rate impact
- Convergence behavior
Module 8: Common Problems in Backpropagation
- Vanishing gradients
- Exploding gradients
- Training instability
Module 9: Practical Understanding
- Simple neural network example
- Visualizing loss reduction
- Preparing for advanced architectures