Dimensional Embedding Techniques for Knowledge Graph Mapping and LLMs

Liubov Tupikina

Keywords: Dimensional Embedding, Knowledge Mapping, LLMs, latent space

TL;DR: We propose method to study lattent space of models which combines dimensional embedding techniques for some contexts of LLMs usage.

Abstract:

We propose a framework, which aims at construction of explainable data embedding methods. It is specifically based on the low-dimensional embedding techniques which are connecting higher-order geometric analysis, topological data analysis and natural language processing methods.
We consider the applications of our framework to the navigation of the knowledge scape is non-trivial in the everyday context, when knowledge/data growth is beyond exponential. Moreover our framework supports methods for generating knowledge graph (node) embeddings, and temporal knowledge graph embeddings.

Submission Number: 9