Bc. Tomáš Jevočin

Master's thesis

Explaining Features of LSTM Model Learned on Human Motion Data

Explaining Features of LSTM Model Learned on Human Motion Data
Abstract:
Cieľom práce je vysvetliť, aké features sa model neurónovej siete LSTM naučil na základe údajov o kostre ľudského pohybu. Na získanie usporiadania dôležitosti deep features aplikujeme DeepSHAP. Následne použijeme LRP upravené pre architektúru LSTM na získanie vstupného mapovania relevancie pre ľubovoľný fixný feature, ktorý môžeme vizualizovať. Získané usporiadania funkcií potom vyhodnotíme pomocou …more
Abstract:
The thesis aims to explain what features an LSTM neural network model learned on top of human motion skeleton data. We apply DeepSHAP to obtain deep feature importance ordering. We then use LRP adjusted for LSTM architecture to acquire input relevance mapping for any fixed feature, which we can visualize. We then evaluate the obtained feature orderings with the help of a 1-NN model trained and evaluated …more
 
 
Language used: English
Date on which the thesis was submitted / produced: 15. 12. 2023

Thesis defence

  • Date of defence: 8. 2. 2024
  • Supervisor: doc. RNDr. Jan Sedmidubský, Ph.D.
  • Reader: prof. Ing. Pavel Zezula, CSc.

Citation record

Full text of thesis

Contents of on-line thesis archive
Published in Theses:
  • světu
Other ways of accessing the text
Institution archiving the thesis and making it accessible: Masarykova univerzita, Fakulta informatiky

Masaryk University

Faculty of Informatics

Master programme / field:
Artificial intelligence and data processing / Big data

Theses on a related topic