Notes and Introduction to Extreme Classification

Notes on. 
Extreme Classification   By Manik Varma  (Abstract)
Communications of the ACM, November 2019, Vol. 62 No. 11, Pages 44-45
10.1145/3355628

What would you do if you had the super-power to accurately answer, in a few milliseconds, a multiple-choice question with a billion choices? Would you design the next generation of Web search engines, which could predict which of the billions of documents might be relevant to a given query? Would you build the next generation of retail recommender systems that have things delivered to your doorstep just as you need them? Or would you try and predict the next word about to be uttered by U.S. President Donald Trump?

The objective in extreme classification, a new research area in machine learning, is to develop algorithms with such capabilities. The difficulty of the task can be judged from the fact that, even if it were to take you just a second to read out a choice, it would take you more than 30 years to go through a billion choices. In 2012, state-of-the-art multi-label classification algorithms were struggling to pick the correct subset of options in questions involving thousands of choices. Then, in 2013, a team from Microsoft Research India and IIT Delhi developed a classifier1 that could scale to 10 million choices, thereby laying the foundations of the area. The approach was based on the realization that only a handful of choices would be relevant for any given question on average. The trick was therefore to quickly eliminate the millions of irrelevant choices. The classifier could then accurately and efficiently choose from the remaining hundred or so options. ..... "

Based on:
Extreme Classification - New Paradigm for Ranking and Recommendation

A good mostly non technical video presentation from 2018:   https://www.youtube.com/watch?v=n4jM5YnUQjE 

A Method for Ranking, Recommendation, Advertising applications,  Search, NLP, Vision...  Seen as replacing recommenders based on matrix factoring

Current methods started with Entropy based methods and lead to tree-splitting algorithms.