Notes taken from Google Machine Learning Course provided to public for practice & correction.

Google Machine Learning Course Notes

Wiki

• Framing
  • In this section we learn the basics of Machine Learning Terminology
• Descending into Machine Learning
  • In this section we work with linear regression, learn about MSE, loss caculation and the basics of how training a model works
• Reducing Loss
  • In this section we explore loss reduction methods by explaining gradient descent, batches, iterative learning and other effective learning methods
• First Steps With Tensorflow
  • In this section we learn the basics of TensorFlow and Pandas. Through practices linked we develop our own linear regression code
• Generalization
  • In this article we discuss the problem of overfitting, learn the difference between a good and a bad model, learn about subsets used in model training & generalization
• Training and test sets
  • In this section we learn about data splitting, dangers of training on test data & test data characteristics
• Validation
  • In this section we cover the importance of validation, a 3rd partition in a dataset
• Representation
  • In this section we discuss qualities of features, learn about feature engineering and mapping values to useful features
• Feature Crosses
  • In this section we look into feature crosses, a synthetic feature used to improve model’s learning & encode non-linear data into useful features
• Regularization: Simplicity
  • In this section we look into ways of penalizing the model for being too complex using L2 regularization
• Logistic Regression
  • In this section we look into Logistic Regression to calculate probabilty, and dive deeper into it’s loss function
• Classification
  • In this section we dive into evaluation precision and recall of logistic regression, as well as ROC & AUCs curves
• Regularization: Sparsity
  • In this section we learn the differences between L1 & L2 and how they bring uninformative weights to 0 or close to 0
• Neural Networks
  • In this section we learn how to solve non-linear problems with Neural Networks. We dive into basics of Neural Networks structure & how it all works
• Training Neural Networks
  • In this article we dive into backpropagation, an algorithm used to traing Neural Networks
• Multi Class Neural Networks
  • In this article we look into multi class neural networks which are the closest to real world example of machine learning usage such as recognizing cars, faces, poses etc.
• Embeddings
  • Learn about embeddings & how they are used to translate large sparse-vectors to a lower dimensional space