Visão Geral

Este curso de treinamento prático e ao vivo de Deep Learning com Python se baseia em nosso Curso Data Science with Python e ensina aos participantes os fundamentos do Deep Learning e como implementar aplicativos de rede neural artificial (ANN), usando Keras e TensorFlow.

Objetivo

Após realizar este Curso Deep Learning with Python Advanced você será capaz de:

• Aprenda a teoria fundamental por trás das redes neurais
• Modele uma função arbitrária usando uma rede neural artificial (RNA)
• Pratique a interpretação de métricas de perda e condições de convergência
• Aplicar uma rede neural a um problema de regressão
• Entenda a regularização no contexto das RNAs
• Implementar dropout e LASSO como estratégias de regularização de rede
• Aplicar Deep Learning a um problema de classificação
• Implemente métodos de processamento de imagem em Python e Keras
• Estenda arquiteturas de rede feed-forward para camadas convolucionais
• Construir arquiteturas de classificação de imagens convolucionais 2D
• Execute uma classificação multiclasse
• Aplicar Deep Learning a dados sequenciais usando arquiteturas recorrentes (RNNS, LSTMs e GRUs)
• Aplique Deep Learning a aplicativos de previsão de séries temporais
• Automatize a seleção de arquitetura de RNA usando Autokeras
• Compreender o conceito de Representações Semânticas Latentes e incorporações de palavras

Pre-Requisitos

• Todos os participantes devem ter concluído o curso Data Science with Python  ou ter experiência equivalente.

Materiais

Conteúdo Programatico

Introduction to Artificial Neural Networks (ANNs) and Deep Learning

1. Why artificial neural networks? Advantages of ANNs
2. Understanding the essential concepts
3. Activation functions, optimizers, back-propagation
4. Components and architectures of artificial neural networks
5. Evaluate the performance of neural networks on a known function
6. Define and monitor convergence of a neural network
7. Model selection
8. Scoring new datasets with a model

Constructing Deep Learning Models

1. Preprocessing structured datasets for Deep Learning workflows
2. Model validation strategies
3. Architectural modifications to manage generalization error
4. Regularization strategies
5. Deep Learning: regression models
6. Deep Learning: classification models

Introduction to Image Processing with Python and Keras

1. Management and preparation of image data for Deep Learning models
2. The dimensionality of image data
3. Handling image metadata
4. Conversion of images to NumPy arrays
5. Python Image Library (PIL) and skimage
6. Keras' load_img() function
7. Image standardization and resampling
8. Augmentation strategies for image data

Deep Learning for Image Classification with Convolutional Architectures

1. Image data is multidimensional
2. Overview of convolutional architectures
3. Convolution layers act as filters
4. Pooling layers reduce computation
5. Data augmentation through image transformation for smaller datasets
6. Image transformation using the pillow library
7. Applying a model to a multi class labeled dataset
8. Evaluating a confusion matrix for multiple classes

Time Series Forecasting with Deep Recurrent Architectures

1. Identify limitations of feed-forward ANN architectures for sequential data
2. Modify model architecture to include recurrent (RNN) components
3. Preprocessing time series data for ingestion into RNN models
4. Examine improvements to RNNs: The LSTM and GRU networks
5. Time series forecasting with recurrent architectures
6. Time series forecasting with 1D convolutional architectures

Deep Learning and Natural Language Processing (NLP)

1. Text manipulation with TensorFlow
2. Categorical representations and word embeddings
3. Text embeddings as layers in an ANN
4. Word2vec
5. Exploiting pre-trained word embedding models
6. Visualizing semantic relationships between words using t-SNE

Transfer Learning

1. Exploiting pre-trained models (VGG16) for image classification
2. Selecting layers to unlock for specific applications
3. Transfer learning and fine tuning

Variational Autoencoders

1. What is an autoencoder?
2. Building a simple autoencoder from a fully connected layer
3. Sparse autoencoders
4. Deep convolutional autoencoders
5. Applications of autoencoders to image denoising
6. Sequential autoencoders
7. Variational autoencoders

Generative Adversarial Networks (GANs)

1. Adversarial examples
2. Generational and discriminative networks
3. Building a simple generative adversarial network
4. Generating images with a GAN

Transformer Architectures

1. The problems with recurrent architectures for sequential data
2. Attention-based architectures
3. Positional encoding
4. The Transformer: attention is all you need
5. Time series classification using transformers
6. GPT-3 and the future of natural language generation
7. Open AI Codex and the future of programmatic code generation