InDeed: Interpretable image deep decomposition with guaranteed generalizability

TL;DR

This paper introduces InDeed, a novel interpretable deep learning framework for image decomposition that combines hierarchical Bayesian modeling with deep neural networks, enhancing interpretability and generalizability for tasks like denoising and anomaly detection.

Contribution

The work presents a new modular deep neural network architecture grounded in hierarchical Bayesian modeling, with a theoretical link to generalization error and a test-time adaptation method for out-of-distribution data.

Findings

Improved generalizability in image denoising and anomaly detection

Enhanced interpretability of deep image decomposition methods

Theoretical connection between loss function and generalization error

Abstract

Image decomposition aims to analyze an image into elementary components, which is essential for numerous downstream tasks and also by nature provides certain interpretability to the analysis. Deep learning can be powerful for such tasks, but surprisingly their combination with a focus on interpretability and generalizability is rarely explored. In this work, we introduce a novel framework for interpretable deep image decomposition, combining hierarchical Bayesian modeling and deep learning to create an architecture-modularized and model-generalizable deep neural network (DNN). The proposed framework includes three steps: (1) hierarchical Bayesian modeling of image decomposition, (2) transforming the inference problem into optimization tasks, and (3) deep inference via a modularized Bayesian DNN. We further establish a theoretical connection between the loss function and the generalization error bound, which inspires a new test-time adaptation approach for out-of-distribution scenarios. We instantiated the application using two downstream tasks, \textit{i.e.}, image denoising, and unsupervised anomaly detection, and the results demonstrated improved generalizability as well as interpretability of our methods. The source code will be released upon the acceptance of this paper.