Master thesis presentation Arash Darakhsh, MPCAS

Abstract: In this presentation we investigate estimation methods and continuous improvement algorithms for deep learning models in industrial settings. First we try to answer whether it is possible to make sample size estimations for training data, based on statistics from another data set, that is similar and belongs to the same domain of problem. Deployed machine learning models in industry will continually predict queries on new incoming data. This data can then be labeled used to improve the current model. We compare current and novel continuous improvement algorithms and how one can use these results to establish a retraining protocol.

Supervisors: Philipp Westphal & Edvard Svenmark
Examiner: Kristian Gustafsson
Opponent: Jack Sandberg