Luke Vilnis
Researcher in representation learning who originated two influential region-representation methods:
- Gaussian Embedding (Vilnis & McCallum, ICLR 2015, Word Representations via Gaussian Embedding) — represents words as Gaussians, with variance encoding concept spread. See Gaussian Embedding.
- Box Lattice (Vilnis et al., 2018, Probabilistic Embedding of Knowledge Graphs with Box Lattice Measures) — the first Box Embedding, representing data as axis-aligned boxes; the foundation later refined by Smoothed Box, Gumbel Box, and Word2Box.
Key Contributions
- Gaussian Embedding
- Box Embedding — Box Lattice (the original box method)
Articles
- Express Words in a Box - Understanding Box Embedding from the Basics — covers his Gaussian Embedding and Box Lattice work
Related People
- Shib Sankar Dasgupta — extended boxes with Gumbel Box and Word2Box