Machine Learning

• In my opinion one of the best videos out there to understand neural networks (highly recommend it to bootcampers) https://www.youtube.com/watch?v=aircAruvnKk

• A great resource for learning about the PyTorch framework. Chapters 1 to 3 are very useful for the bootcamp https://www.learnpytorch.io/

• https://developers.google.com/machine-learning/glossary/

• Yet another great resource on neural networks. A bit outdated but still contains a lot of relevant information. http://neuralnetworksanddeeplearning.com/

• https://towardsdatascience.com/what-are-tensors-in-machine-learning-5671814646ff

Computer Vision

• Computer Vision system developed by Istanbul Technical University for competing in the SUAS 2023 competition: Istanbul Technical University - ITUNOM UAV Team | SUAS 2023 - YouTube

Statistics

Unsupervised Data

A goal of processing large amounts of unsupervised (i.e. unlabelled) data is to be able to find relationships between and group points without knowing how they are distributed originally.

A mixture model is the grouping of points. As with all models, the resulting model can be a very good representation or be completely off.

There are several algorithms:

• k-means: Groups closest points to each other (the mean) given a known group count

• Gaussian Mixture Model: Groups points in a Gaussian (Normal) manner given a known group count

  • How it is implemented: https://brilliant.org/wiki/gaussian-mixture-model/

• Variational Bayesian Gaussian Mixture Model: Groups points but the group count is unknown

  • https://scikit-learn.org/stable/modules/mixture.html#variational-bayesian-gaussian-mixture

Belief Update

Log odds make it easier to operate with odds by making the odds symmetric on both sides: https://towardsdatascience.com/https-towardsdatascience-com-what-and-why-of-log-odds-64ba988bf704

Video encoding

H.264

The theory behind video encoding: https://sidbala.com/h-264-is-magic/

Gstreamer: https://rianadon.github.io/blog/2019/04/04/guide-to-h264-streaming-frc.html