less about review of KGs more about Dr. Maximilian Nickel

Today I finished skimming through “A Review of Relational Machine Learning for Knowledge Graphs” by Maximilian Nickel et al. I encourage you to read this review to get a quick yet formal introduction to Knowledge representation, Knowledge base construction, latent feature and graph feature models, Knowledge Vault components on much more.

A little bit about Dr. Nickel

From whatever limited literature I have read till now, I have always found work of Dr. Nickel (publications) quite elegant. He formulates all the problems very mathematically, provides theoretical proofs rather than just empirical evidence, opposed to the de facto in majority of current ML community (on a side note I encourage you to watch this popular and entertaining talk “Machine Learning has become alchemy” by Ali Rahimi on receiving his test of time award in NIPS 2017), brings ideas from diverse fields that makes you want to explore these fields, may it be riemannian manifolds, graph theory, or holographic models. If I recall correctly, his first work I came across was “Poincaré Embeddings for Learning Hierarchical Representations”. In this paper they propose that hierarchical relationships can be more naturally and compactly embedded in hyperbolic space rather than in euclidean space. 5 dimensions in Poincare space outperforms 200 dimensions in Euclidean on link prediction task on WordNet.

I won’t say that I understood this paper completely as the math is quite complex and I have no understanding of manifolds, but that is what encourages me read the book “An Introduction to Manifolds” by Loring W. Tu. Hope the shubh muhurat(auspicious time in hindi) will come very soon :). Thats all for short story writing for today.