Bin Mohd Azman, Muhammad Amir (2023) Disentangling Uncertainty in Regression for Out-of-Distribution Data: A Comparative Study of Methods and Applications to Real-world Scenarios. Bachelor's Thesis, Artificial Intelligence.
|
Text
Disentangling_Uncertainty_s4013948.pdf Download (24MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (130kB) |
Abstract
The objective of this study is to compare and evaluate previously studied uncertainty quantification methods in out-of-distribution (OOD) data using real-world scenario regression tasks, focusing on their ability to distinguish between aleatoric and epistemic uncertainties. A comprehensive analysis is conducted using four uncertainty methods: MC-Dropout, MC-DropConnect, Flipout and Ensembles, in both numerical and visual datasets to determine their effectiveness in disentangling uncertainty. The findings from the numerical dataset indicates that variations of OOD inputs on individual features produces different relationships between ID and OOD data when using the same uncertainty method. Uncertainty methods such as Dropout and Dropconnect show positive signs of increasing epistemic uncertainty in both types of datasets. However, it is inconclusive to title them as the best general disentangling method when applied on the respective datasets. Yet, methods such as Flipout show to be unreliable.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Valdenegro Toro, M.A. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 16 Aug 2023 09:28 |
| Last Modified: | 16 Aug 2023 09:28 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/31189 |
Actions (login required)
 |
View Item |