Explainable Machine Learning via Concept Bottleneck
Improving the explainability of Computer Vision models based on Deep Learning has recently become a compelling problem, providing reliable predictions for the end user and enabling more fine-grained classifications. Recently, Concept Bottleneck models have been proposed for image classification, dividing the problem into two stages and thereby defining a hierarchy of concepts. In this lecture, we will present our contribution to the topic. We developed a conceptual bottleneck model for image segmentation, objects fine classification, and object tracking, which was trained and tested on a dataset consisting of images of refrigerators filled with various objects. The proposed model makes full use of the hierarchy in concepts, exploiting the relationships between different categories at the same hierarchical level and relying on a new method for handling multi-label classifications. We will discuss the findings of our research, especially the explainability of the proposed approach, and elaborate on the possibility of future improvements.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
- University Ss. Cyril and Methodius, Faculty of Computer Science and Engineering
- Rudzer Boshkovikj 16
- Skopje, Macedonia
- Macedonia 1000
- Room Number: AMF-FCSE
- Contact Event Hosts
- Co-sponsored by FEEIT, FCSE, Explainable AI Project, EuroCC Project, ICT-ACT
- Starts 23 November 2024 10:00 AM UTC
- Ends 27 November 2024 12:00 PM UTC
- No Admission Charge
Speakers
Vesna Dimitrievska, PhD of Silicon Austria Labs, Villach, Embedded Systems Division
Explainable Machine Learning via Concept Bottleneck
Improving the explainability of Computer Vision models based on Deep Learning has recently become a compelling problem, providing reliable predictions for the end user and enabling more fine-grained classifications. Recently, Concept Bottleneck models have been proposed for image classification, dividing the problem into two stages and thereby defining a hierarchy of concepts. In this lecture, we will present our contribution to the topic. We developed a conceptual bottleneck model for image segmentation, objects fine classification, and object tracking, which was trained and tested on a dataset consisting of images of refrigerators filled with various objects. The proposed model makes full use of the hierarchy in concepts, exploiting the relationships between different categories at the same hierarchical level and relying on a new method for handling multi-label classifications. We will discuss the findings of our research, especially the explainability of the proposed approach, and elaborate on the possibility of future improvements.
Biography:
Vesna Dimitrievska has been working as a scientist at Silicon Austria Labs, Villach, since 2019 and is currently a scientist in the Embedded Systems Division. She holds a PhD in Natural Sciences and Mathematics, specialization in Computer Science, from St. Cyril and Methodius University, Skopje, with the dissertation “Modeling of emotions in human-robot interaction”. She holds a Master's degree from the same university in Intelligent Systems Engineering – Robotics. From 2012 to 2019, Vesna worked at the University of Skopje as a laboratory demonstrator and teaching assistant at the Institute of Intelligent Systems. Her main research topics are machine learning, data analysis, intelligent systems, and robotics. Especially in the field of machine learning, she has practical experience with deep neural network models, recurrent neural networks, cluster models, reinforcement learning models, etc. She is the co-author of more than 25 international publications and has participated in more than 10 research projects.