Notes and implementations of various topics relating to Machine Learning and Artificial Intelligence
Content spanning from experience in practice, graduate studies, and extracurricular studies
- Loss
- Similarity
- Problem Frames
Focuses on leveraging the properties of hyperbolic space - a space with negative curvature - to model data more efficently than traditional Euclidian methods. Growing popularity in areas like: graph embeddings, taxonomies, hierarchical clustering, and knowledge representation
Hyperbolic space grows exponentially with distance from a point, which makes it ideal for representing hierarchical relationships. Traditional Euclidian embeddings struggle to capture these relationships effectively, whereas hyperbolic embeddings preserve the structure with minimal distortion, even for very deep hierarchies. This makes hyperbolic geometry valuable for domains like graph rep learning, recommendation systems, and biological networks
- https://hyperbolic-representation-learning.readthedocs.io/en/latest/index.html
- https://arxiv.org/pdf/2306.09118
- dude fuck this is sick someonse thesis from 2021: https://purl.stanford.edu/bj040rx3340
checking editing from app