Search papers, labs, and topics across Lattice.
This paper introduces a persistence-based data augmentation framework that leverages the Morse-Smale complex to encode local gradient flow regions and their hierarchical evolution for integration with neural networks. By retaining spatially localized topological information across multiple scales, the method aims to improve deep learning performance while preserving local geometric structure. Experiments on histopathology image classification and 3D porous material regression demonstrate that this augmentation outperforms baselines and global TDA descriptors, with an efficient $O(n \log n)$ computational complexity.
Injecting local topological structure into neural networks via persistence-based data augmentation unlocks improved performance and interpretability in histopathology image classification and 3D material regression.
Topological Data Analysis (TDA) provides tools to describe the shape of data, but integrating topological features into deep learning pipelines remains challenging, especially when preserving local geometric structure rather than summarizing it globally. We propose a persistence-based data augmentation framework that encodes local gradient flow regions and their hierarchical evolution using the Morse-Smale complex. This representation, compatible with both convolutional and graph neural networks, retains spatially localized topological information across multiple scales. Importantly, the augmentation procedure itself is efficient, with computational complexity $O(n \log n)$, making it practical for large datasets. We evaluate our method on histopathology image classification and 3D porous material regression, where it consistently outperforms baselines and global TDA descriptors such as persistence images and landscapes. We also show that pruning the base level of the hierarchy reduces memory usage while maintaining competitive performance. These results highlight the potential of local, structured topological augmentation for scalable and interpretable learning across data modalities.
