From Pixels to Perception: Interpretable Predictions via Instance-wise Grouped Feature Selection
This addresses the need for interpretable AI in image classification, though it appears incremental as it builds on existing sparsification approaches.
The paper tackles the problem of making machine learning model predictions interpretable by developing a method that performs instance-wise sparsification of input images using semantically meaningful pixel regions, with dynamic sparsity levels. The result is an inherently interpretable classifier that produces more meaningful, human-understandable predictions than state-of-the-art benchmarks on semi-synthetic and natural image datasets.
Understanding the decision-making process of machine learning models provides valuable insights into the task, the data, and the reasons behind a model's failures. In this work, we propose a method that performs inherently interpretable predictions through the instance-wise sparsification of input images. To align the sparsification with human perception, we learn the masking in the space of semantically meaningful pixel regions rather than on pixel-level. Additionally, we introduce an explicit way to dynamically determine the required level of sparsity for each instance. We show empirically on semi-synthetic and natural image datasets that our inherently interpretable classifier produces more meaningful, human-understandable predictions than state-of-the-art benchmarks.