Neural network

• Deep learning

  • Basics of neural networks
    • intro: Introduction to neural networks along with some history.
    • Using the neural networks: Regression vs Classification.
    • Learning: it covers how the model training is done in general, and how backpropagation works.
    • Cost function
    • Optimization
  • Optimization and hyper parameters
    • Bias x Variance tradeoff: How to choose the right hyper parameters for a given dataset.
    • Overfitting vs Regularization techniques
    • Machine learning strategy: discusses the process of training a model and the decisions taken through it, like hyperparameter tuning process, the idea of orthogonalization.

• Tensorflow

  • Sequential model - notebook: it covers the most basic use of Tensorflow, through the Sequential model.
  • Model storage: this module covers how to store a trained model and how to rebuild a model from its stored version.
  • Functional API 1 - notebook: it demonstrates the use of Tensorflow functional API, which can be used to build more complex models, it works on energy + effciency dataset.
  • Functional API 2 - notebook: it demonstrates the use of Tensorflow functional API, which can be used to build more complex models.
  • Customization: custom layer - notebook: it demonstrates how to create a custom layer in Tensorflow.
  • Customization: custom loss - notebook: it demonstrates how to create a custom loss in Tensorflow.
  • Metrics: it explains what metrics are, how to use built-in metrics, and how to build your own.
  • Customization: custom model: it demonstrates how to create a custom model in Tensorflow, using Resnet18 and VGG16 as examples.
  • Callbacks: it explains what callbacks are, how to use built-in callbacks, and how to build your own.
  • Customization: custom optimizer - notebook
  • What is a tensor?: covers what tensor is, how to create it and use it.
  • Gradient Tape: it explains how to use the gradient tape to calculate gradients, as a step to customize the leasrning process.
  • custom training: it demonstrates how to create a custom training loop in Tensorflow, you may want to use that if you want to do training in an inconventional way, like using more than the first derivate.
  • Graph mode - notebook

• CNN

  • Convolutional network
  • ResNet
    • Residual network architecture
    • The results obtained from the ResNet paper
    • Tensorflow Resnet
  • Inception network
  • Object detection: YOLO and the gang
  • Summary