Back

A Bayesian Approach to Non-Metric Hyperbolic Multi-Dimensional Scaling

Jolis, M.; Praturu, A.; Sharpee, T. O.

2023-12-11 neuroscience
10.1101/2023.12.08.570871 bioRxiv
Show abstract

This paper explores the intersection of hyperbolic geometry, non-metric techniques, and Bayesian frameworks to extend the capabilities of Bayesian Hyperbolic Multi-Dimensional Scaling (HMDS). While hyperbolic geometry is gaining attention for its ability to represent hierarchical relationships, traditional metrics impose constraints on distances. Non-metric techniques offer flexibility in capturing complex structures, making them suitable for scenarios where metric distances are less meaningful. The paper introduces a novel extension of Bayesian HMDS, incorporating non-metric techniques, enabling the embedding of Euclidean data within a hyperbolic space. The approach simultaneously fits for curvature and coordinates, leveraging the scaling properties of hyperbolic space. The non-metric Bayesian Hyperbolic MDS is expected to unveil new insights into hierarchical structures within complex datasets, providing a versa-tile tool for analyzing high-dimensional data flexibly and accurately. The efficacy of the proposed method is demonstrated through synthetic data experiments, showcasing its ability to capture non-linear transformations and accurately predict underlying curvature, with an emphasis on its ro-bustness to hyperparameter choices.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.