Research map / Probabilistic and scientific modeling / Variational and generative sampling
Connectivity-Optimized Representation Learning via Persistent Homology
Christoph Hofer, Roland Kwitt, Marc Niethammer, Mandar Dixit
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use persistent-homology connectivity regularization in the autoencoder representation | One-class learning with 120 training samples per class · CIFAR-10 | AUC: 0.69 -> 0.76 | Improved |
| Use persistent-homology connectivity regularization in the autoencoder representation | One-class learning with 120 training samples per class · CIFAR-20 | AUC: 0.66 -> 0.72 | Improved |
| Use a connectivity-regularized representation rather than ADT | One-class learning · CIFAR-100, 120 samples/class | AUC: 0.75 -> 0.79 | Improved |
| Transfer the autoencoder representation trained on CIFAR-100 | One-class learning · Tiny-ImageNet | AUC: 0.73 vs. 0.72 | No clear effect |
| Transfer the autoencoder representation trained on CIFAR-100 | One-class learning · ImageNet | AUC: 0.72 vs. 0.72 | No clear effect |
| Add 16 separate branches to control connectivity in low-dimensional subspaces (ablation) | Connectivity-regularized autoencoder · CIFAR-10 one-class learning | AUC: 0.70 -> 0.78 | Improved |
| Use 32 branches instead of no branching (ablation) | Connectivity-regularized autoencoder · CIFAR-10 one-class learning | AUC: 0.70 -> 0.75 | Improved |
| Reduce connectivity-loss weight to lambda=1.0 (ablation) | 16-branch autoencoder · CIFAR-100 training; CIFAR-10 one-class evaluation | AUC | Hurt |
| Use connectivity-loss weight lambda>=10.0 rather than lambda=1.0 (ablation) | 16-branch autoencoder · CIFAR-100 training | connectivity loss | Improved |
| Train with smaller batches, e.g. batch size 5 (ablation) | Connectivity-loss toy mapping · Three-Gaussian toy data | range of death times | Improved |
| Increase latent dimensionality while keeping batch size fixed (ablation) | Connectivity-loss toy mapping · Three-Gaussian toy data | range of death times | Improved |
Closest papers by meaning
- Topological Autoencoders (ICML 2020 · similarity 0.90)
- Graphcode: Learning from multiparameter persistent homology using graph neural networks (NeurIPS 2024 · similarity 0.86)
- Positional Encoding meets Persistent Homology on Graphs (ICML 2025 · similarity 0.86)
- Homology Consistency Constrained Efficient Tuning for Vision-Language Models (NeurIPS 2024 · similarity 0.85)
- Persistence Homology Distillation for Semi-supervised Continual Learning (NeurIPS 2024 · similarity 0.85)
- Mapping the Multiverse of Latent Representations (ICML 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
