Research map / Architectures and efficiency / RNNs and normalization
Hyperbolic Associative Memory Networks
Boliang Hao, Bailing Zhang, Fangyu Wu
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 |
|---|---|---|---|
| replace Euclidean MHN retrieval with hyperbolic HAMNs retrieval | ResNet-18 + memory module · CIFAR-100 4-layer hierarchy | top accuracy: 89.39 -> 90.98 | Improved |
| replace Euclidean MHN retrieval with HAMNs | ResNet-18 + memory module · CIFAR-100 4-layer hierarchy | cophenetic correlation: 0.5902 -> 0.7184 | Improved |
| replace Euclidean MHN retrieval with HAMNs | ResNet-18 + memory module · CIFAR-100 3-layer hierarchy | cophenetic correlation: 0.7152 -> 0.7017 | No clear effect |
| use HAMNs instead of HypNN on fine-grained recognition | ResNet-18 + memory module · CIFAR-100 4-layer hierarchy | fine accuracy: 55.93 -> 53.49 | Hurt |
| use hyperbolic retrieval instead of Euclidean MHN on SBERT features | WordNet hypernym prediction · WordNet nouns | accuracy (%): 24.98 -> 47.25 | Improved |
| add ontology embedding front-end (OntEuc) to HAMNs | HAMNs · WordNet nouns | accuracy (%): 47.25 -> 96.97 | Improved |
| use HAMNs in weak/shallow hierarchy tasks versus Euclidean MHNs | HAMNs · MIL and MoleculeNet | task metrics | No clear effect |
| vary curvature c (moderate 0.7-2.0 vs extremes) (ablation) | HAMNs · CIFAR-100 hierarchy | accuracy | Mixed |
| increase number of stored patterns (ablation) | HAMNs · CIFAR-100 hierarchy | accuracy | Mixed |
| use hyperbolic instead of Euclidean operations | HAMNs · CIFAR-100 and WN18RR | runtime and peak memory | Hurt |
Closest papers by meaning
- Nested Hyperbolic Spaces for Dimensionality Reduction and Hyperbolic NN Design (CVPR 2022 · similarity 0.88)
- Hyperbolic Representation Learning: Revisiting and Advancing (ICML 2023 · similarity 0.88)
- Curvature Learning for Generalization of Hyperbolic Neural Networks (IJCV 2025 · similarity 0.88)
- Robust Hyperbolic Learning with Curvature-Aware Optimization (NeurIPS 2025 · similarity 0.87)
- Uniform Memory Retrieval with Larger Capacity for Modern Hopfield Models (ICML 2024 · similarity 0.86)
- Provably Optimal Memory Capacity for Modern Hopfield Models: Transformer-Compatible Dense Associative Memories as Spherical Codes (NeurIPS 2024 · similarity 0.86)
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.
