Entity Embeddings package real world knowledge for AI algorithms

A Retail Store Embedding Matrix acts as an Input to the DNN for Store Sales Prediction

One of my eighty plus year young Dad’s favourite saying is that scientists have not yet invented a machine to read the mind.

Maybe, AI will change that in the future.

In the information systems mapping the real world, all of us deal with entities.

What are entities ?

Entities are nothing but nouns.

Some simple examples.

Customer, Supplier, Product, Store are all entities.

AI systems made predictions as regards real world entities and events based on the data we feed them.

Since the past few weeks, I have been a part of a FastAI Study group setup by Twimlai on the course titled ‘Deep Learning for Coders — Part 1’. We have weekly online meetings.

As part of the above group, I recently gave a half hour presentation on ‘Entity Embeddings’ .

‘Entity Embeddings’ is a recent AI development that can help smart organisations & individuals capture domain knowledge mathematically for better predictions.

Entity Embeddings can help explicitly package real world domain knowledge in a structured mathematical form which can then be fed into various AI algorithms.

It can help speed up organisational learnings in a systematic structured manner.

It helps give a structured mathematical shape to domain knowledge lying inside human brains — Whether vertical industry knowledge or horizontal functional knowledge or geography specific knowledge or any other knowledge.

Organizational learning becomes much more automated & dynamic leading to better AI predictions.

So, an important concept to learn not just for technology folks but also all professionals.

In any vertical, function or geography.

AI advances such as these have the potential to make industry vertical & horizontal function specific domain knowledge & experience irrelevant.

Earlier, if entry level jobs were at risk due to AI, with Entity Embeddings, it is possible that even tasks in jobs which need years of domain expertise & experience can possibly be done by machines better, faster & cheaper in the future.

This is the 14th July 2018 Youtube presentation link.

And the presentation is below.

Mahesh Khatri 14th July 2018 Presentation on Entity Embeddings for TWIMLAI FastAI Study Group

PS : My Dad had scored 108 out of 100 in Maths in his public tenth grade examination. So it will be a delicious irony if AI using Maths can indeed prove him wrong in the future.

Thanks to Sam Charrington of TWiML & AI and Jeremy Howard of FastAI.