Explaining Model Predictions in VQA

Jain, Jayant (2022) Explaining Model Predictions in VQA. Masters thesis, Indian Institute of Science Education and Research Kolkata.

[img] Text (MS dissertation of Jayant Jain (17MS075))
17MS075_Thesis_file.pdf - Submitted Version
Restricted to Repository staff only

Download (4MB)
Official URL: https://www.iiserkol.ac.in

Abstract

Recent state-of-the-art VQA systems have shown great accuracy results. Many recent works suggest that many VQA models do not necessarily rely on image content while making predictions and can resort to learning superficial correlations from the structure of natural language. This makes the model biased. Moreover, they do not necessarily attend to image features that we humans consider necessary for answering a question. This raises the need to be able to explain such models’ predictions. This work explores the model given by Wu and Mooney [32] and attempts to explain this model’s predictions.

Item Type: Thesis (Masters)
Additional Information: Supervisor: Prof. Utpal Garain, Computer Vision Pattern Recognition (CVPR) Unit, Indian Statistical Institute, Kolkata; IISER Kolkata: Dr. Anirvan Chakraborty
Uncontrolled Keywords: Counterfactual Analysis; Explainable Models; Language Priors; Visual Question Answering
Subjects: Q Science > QA Mathematics
Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions: Department of Mathematics and Statistics
Depositing User: IISER Kolkata Librarian
Date Deposited: 07 Jun 2023 10:40
Last Modified: 07 Jun 2023 10:40
URI: http://eprints.iiserkol.ac.in/id/eprint/1300

Actions (login required)

View Item View Item