University of Twente Student Theses

Login

Are classical features still relevant in the era of deep learning?

Snoijer, Jesse (2022) Are classical features still relevant in the era of deep learning?

This is the latest version of this item.

[img] PDF
711kB
Abstract:A deep-learning AI model for point clouds often requires a vast amount of training data, which is not always available or requires a time-expensive process to create. To reduce the amount of data needed, the complexity of the model can be decreased. To reduce the loss of accuracy by doing this, it is investigated if appending classical features to the data increases the accuracy for a model with low complexity. Multiple tests will be performed where data with and without classical features appended is compared. Also, models with different complexity are compared. Here we show that appending these classical features does increase the test accuracy for a deep-learning AI model with low complexity by 20%. It is also shown that decreasing the complexity of the model and appending classical features prevents the f1-score from dropping 22.4%. These findings could possibly translate to other kinds of data as well and solve the problem of not having enough data available. Because a less complex model is used, this also lowers the computational demand which speeds up training and has a lower power usage.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/96420
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page