Learn the foundations of machine learning.
The following concepts were studied:
- linear regression: training set, features, target variable, hypothesis, learning algorithm, parameters, cost function, optimisation problem, gradient descent, learning rate, batch gradient descent
- multivariate linear regression: feature scaling, mean normalisation, choosing learning rate, normal equation
- logistic regression: classification, logistic function, sigmoid function, decision boundary, nonlinear decision boundaries, cost function, optimisation algorithms, multiclass classification, one-vs-all
- regularisation: overfitting, regularisation parameter, regularised linear regression, regularised logistic regression
- neural networks: computer vision, sigmoid activation function, layers, bias, forward propagation, nonlinear classification, back propagation algorithm, random initialisation
- model selection: training / validation / test sets, diagnosing bias and variance, cross validation error, regularisation, learning curves, high bias, high variance, error analysis, precision, recall, F1 score
- support vector machines: SVM hypotesis, large margin classifier, kernels, similarity, Gaussian kernel, linear kernel, polynomial kernel,
- unsupervised learning: clustering, K-means
- principal component analysis: dimensionality reduction, data compression, data visualisation, covariance matrix sigma, eigenvectors of sigma, reconstruction from compressed representation, number pf principal components, learning speedup
- k-means clustering: cluster index, cluster centroid, random initialisation, elbow method
- anomaly detection: density estimation, normal vs anomalous, fraud detection, manufacturing, monitoring working parameters, Gaussian distribution, features, error analysis
- recommender systems: content based recommendations, collaborative filtering, low rank matrix factorisation, mean normalisation,
- large scale machine learning: stochastic gradient descent, mini-batch gradient descent, online learning, map-reduce and data parallelism
The tool of choice for this course was Matlab / Octave; each machine learning algorithm was coded in a project.
As a proof of accomplishment, the following certificate was issued: https://www.coursera.org/account/accomplishments/certificate/WLHZZ6TPVVM2.