05/13/2021
Percona Live Online 2021
Machine Learning inside databases is becoming a hot trend. Last time at Percona Live 2020, our team presented AI Tables – an opensource solution that enables automated machine learning capabilities inside databases. The main idea of AI Tables is to allow anyone who works with databases to implement ML projects in a matter of hours without requiring data science skills.
It is as simple as using SQL queries!
In the journey of bringing AI Tables to the community, we have discovered and solved Machine Learning problems that are hard even for ML engineers but are common for data inside databases.
For example:
Forecasting inventory for all products in all stores (**GROUP BY store, product_id**), given a table that contains all inventory updates over time (**ORDER BY time**).
This problem is complex even for experienced ML engineering teams. In a traditional ML approach, you would need to train one model for each product at each store, which can mean thousands or hundreds of thousands of models, not even thinking of the logistic nightmare to bring such many models to production.
Another example of a challenge solved is creating views that do **joins between data tables and ML models**. It significantly streamlines using machine learning inside BI tools to forecast data trends. Also, it opens broader possibilities for anomaly detection and much more!
We have made significant progress in solving those problems automatically through AI-Tables, and we would like to share with you our approach and discuss some interesting insights that we have made in the process.
Agenda:
– 5 min | Advantages of ML inside a database over the traditional approach
– 15 min | Machine learning workflows inside databases
– 15 min | Automated multivariate time-series forecasting
– 15 min | Joining tables with ML models
– 10 min | Q&A
Speakers: Jorge Torres and Patricio Cerda-Mardini – MindsDB
