STAI CDT student Munkhtulga Battogtokh presented the paper “Simple Framework for Interpretable Fine-grained Text Classification” at the 26th European Conference on Artificial Intelligence (ECAI 2023) in Kraków, Poland.
Munkhtulga explains the paper here; “As machine learning (ML) models have become better, people have used them for harder and harder tasks. For example, whereas an ML model was used to classify between two opposite categories before (e.g., positive and negative sentiment), now in practice they are often used to classify between several dozens of highly similar classes. In this context, explaining why a model chooses one class over another has become more challenging due to the classes being more similar. Our paper proposes an approach for dealing with this challenge, which is to make the fine-grained aspects that characterize each class explicit during model prediction. This allows us to precisely explain for a given class what aspects the model needs to look for, whether it finds those in a data input or not (either of which is equally important), and what are the aspects that make the class different from other ones. This framework is relevant for a wide range of difficult ML tasks and application domains where interpretability is indispensible (e.g., medicine, banking, etc.)”.
Munkhtulga enjoyed the conference, particularly the opportunity to get feedback on the paper. He said, “For me, the highlight was having conversations with other researchers after my presentation. It is rewarding to hear that they connect to the problems that I have addressed and find my proposed approach valuable. These conversations are essential for getting the “feel” for what problems are actually troubling people, where my research sits in relation to them, and what I can do more in the future. Great food too!”
The full paper can be read here: Simple Framework for Interpretable Fine-grained Text Classification. / Battogtokh, Munkhtulga; Luck, Michael; Davidescu, Cosmin et al.
ECAI 2023: 3rd International Workshop on Explainable and Interpretable Machine Learning (XI-ML). 2023.