Differentiable Clustering with Perturbed Spanning Forests
This enables clustering to be integrated into neural networks for supervised and semi-supervised tasks, though it appears incremental as it builds on existing spanning forest techniques.
The paper tackles the problem of making clustering differentiable for end-to-end trainable pipelines by introducing a method based on stochastic perturbations of minimum-weight spanning forests, achieving good performance on datasets with high noise and challenging geometries.
We introduce a differentiable clustering method based on stochastic perturbations of minimum-weight spanning forests. This allows us to include clustering in end-to-end trainable pipelines, with efficient gradients. We show that our method performs well even in difficult settings, such as data sets with high noise and challenging geometries. We also formulate an ad hoc loss to efficiently learn from partial clustering data using this operation. We demonstrate its performance on several data sets for supervised and semi-supervised tasks.