Computer Science master thesis

Latex code for my computer science master thesis, "A comparison of frequentist methods and Bayesian approximations in the implementation of Convolutional Neural Networks in an Active Learning setting".

Supervised by Alfredo Garbuno Iñigo.

Code can be found here.

And the final PDF file can be found here.

To compile document locally

Must have latex and biber installed. Follow the following instructions:

1. Clone repository.

2. cd to latex directory.

3. Folder has a Makefile, so all you have to do is execute the command make all

To compile document on the cloud

Go to https://v2.overleaf.com/read/dscncbsfvjvf and compile. This version may not reflect most recent changes because I have to manually pull changes from the repository and I don't do this with every single commit.

Description of R scripts

• File gradient_descent_example.R implements gradient descent for logistic regression in simulated data. Generates file GD_plots.pdf.

• File mini_batch_gradient_descent_example.R implements mini-batch gradient descent for logistic regression in simulated data. Uses mini_batch_gd_log_reg.cpp file to compile C++ code via Rcpp package. Generates file Mini-batch_GD_plots.png.

• File BBVI_logistic_regression.R implements gradient ascent for Mean-field Variational Approximation of posterior distribution of logistic regression in simulated data. Generates file BBVI_plots.pdf.

• File plot_ANN.R creates ANN diagrams. Generates files plot_ANN_01.pdf, plot_ANN_02.pdf and plot_ANN_03.pdf.

• File plot_KL_example.R shows the difference between forward and reverse KL-divergence in a Gaussian mixture. Creates files KL_example_1.pdf, KL_example_2.pdf and KL_example_3.pdf.

• File logistic_regression_AL_example.R implements active learning in two logistic regression models with simulated data. Creates files log_reg_AL_decision_boundary_plot.pdf and log_reg_AL_accuracies_plot.pdf.