Two convergent algorithms improve spatial regularisation in neural networks
DeepFlow: Two Convergent Differentiable Approaches for Pairwise Conditional Random Fields
Machine Learning
Summary
When computers try to recognize images or sounds, they often use complicated math models to improve accuracy. This paper shows two new algorithms that help make these complex models work better with deep learning. The authors proved the algorithms reliably find solutions and calculated how they perform differently. These methods help computers understand spatial patterns in data without needing lots more examples.
What this means in practice
- •For computer vision engineers: Integrate the proposed algorithms to improve image classification models with better spatial pattern handling during training.
- •For signal processing teams: Use the convergence-guaranteed methods to enhance signal classification systems by including complex probabilistic spatial dependencies.
Authors
John S. H. Baxter
Abstract
Basic probabilistic models form the basis of modern deep learning largely by defining the final activation layer of a network and allowing the networks outputs to be interpreted through a probabilistic lens. By increasing the complexity of these final activation layers to include features such as spatial regularisation, allow for these models to improve in performance without requiring significantly more data. The challenge however is to ensure that these more complex probabilistic graphical models adapt well to being used in tandem with deep learning, that is, ensuring they converge and have meaningful derivatives. The goal of this paper is to show two algorithms with provable convergence properties to allow this type of regularisation for image and signal classification models. Through both rigorous mathematical analysis and in silico experiments, we can show the differing performance between these solution types.