Polaris : Coupled Orbital Polar Embeddings for Hierarchical Concept Learning
Sahil Mishra, Srinitish Srinivasan, Sourish Dasgupta, Tanmoy Chakraborty · 2026
Read Polaris : Coupled Orbital Polar Embeddings for Hierarchical Concept Learning by Sahil Mishra, Srinitish Srinivasan, Sourish Dasgupta, Tanmoy Chakraborty (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-00265.txt.
Description
Sahil Mishra and 3 co-authors present Polaris, a method for embedding hierarchical knowledge, such as product taxonomies, medical ontologies, and label trees. Hierarchies are lopsided and their labels are noisy, which makes them hard to represent in the flat vector spaces models usually learn. Polaris places concepts on a sphere and splits the job in two.
Direction on the sphere carries meaning, while distance from the center carries rank in the hierarchy, so the two signals do not interfere. Training mixes local constraints, a global term that stops the geometry from collapsing, and objectives that respect the one-way nature of parent and child links. At lookup time, structure guides a narrowing of candidate parents. Across tree, graph, and multimodal hierarchies, Polaris improved retrieval over fourteen strong baselines.
Catalog record
| CREATOR | Sahil Mishra, Srinitish Srinivasan, Sourish Dasgupta, Tanmoy Chakraborty |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 175 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.00265