Idea
A scalable AutoML platform for enterprises to analyze large multi-table databases with explainable Bayesian models.
Research Paper
Core Innovation
This paper presents Khiops, which uniquely integrates Bayesian variable selection, naive Bayesian classifiers with weight learning, and automatic aggregate construction for multi-table databases. It scales efficiently to very large datasets with millions of individuals and hundreds of millions of records. The approach combines AutoML and explainability in a single open source tool accessible via Python and UI.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: growing demand for scalable AutoML and explainable AI in enterprise data analytics.
Potential Customers & Pain Points
- Enterprises with Large Multi-Table Databases Needing Scalable ML
- Data Scientists Seeking Automated Variable Selection and Explainability
- Businesses Requiring Efficient Multi-Table Aggregation and Classification
- Organizations Handling Millions of Records with Limited ML Resources
Business Model
Open source core with enterprise subscription for advanced features, support, and cloud deployment options.
Competitive Landscape
- DataRobot
- H2O.ai
- Google AutoML
Implementation Challenges
- Integration with diverse database systems
- Competition from established AutoML platforms
- User adoption in complex enterprise environments
Validation Strategy
- Pilot deployments with large enterprises managing multi-table databases
- Benchmarking against leading AutoML tools on scalability and explainability
- User feedback cycles to refine UI and Python API integration
Research Paper Overview
Khiops: An End-to-End, Frugal AutoML and XAI Machine Learning Solution for Large, Multi-Table Databases
Summary
Khiops is an open source machine learning tool designed for mining large multi-table databases using a Bayesian approach. It offers variable selection, classification, decision trees, and co-clustering with predictive variable importance measures. It supports naive Bayesian classifiers with variable selection and weight learning, and automatically constructs aggregates for multi-table databases. It scales to millions of individuals, tens of thousands of variables, and hundreds of millions of records, accessible via Python and UI.