Out-of-Distribution Aware Classification for Tabular Data

What is it about?

Traditional classification models work well for test samples that are similar to their training data but often struggle with samples that significantly deviate from the training data, known as out-of-distribution (OOD) samples. To address this issue, prior studies have developed OOD-aware classification models that can detect OOD samples and prevent them from being misclassified into existing classes. Our paper introduces a novel method to OOD-aware classification for tabular data. Our approach also integrates smoothly into downstream tasks, such as counterfactual explanations, enhancing these tasks through OOD-awareness.

Featured Image

Photo by Google DeepMind on Unsplash

Photo by Google DeepMind on Unsplash

Why is it important?

Most previous studies on OOD-aware classification (OOD detection) focus on the image domain, but these methods cannot effectively handle tabular data. Our method is specifically designed to address the unique challenges of tabular data. Additionally, our method integrates easily into downstream tasks that rely on pre-trained classification models. Enhancing downstream tasks through OOD-aware classification is one of the major contributions of our work.