Interpretable AI develops proprietary machine learning algorithms that achieve state-of-the-art performance while remaining completely transparent and understandable
Our algorithms are based on years of research at MIT and published in the top peer-reviewed academic journals
Industry-ready: practical and scalable
Each algorithm has a proven track record of success in large-scale experiments and industry applications.
Classical models revisited with modern optimization
We take a fresh perspective on these problems and leverage modern optimization techniques to lift the performance to the level of black-box models without sacrificing interpretability.
These algorithms form the core of the recent graduate-level textbook Machine Learning Under A Modern Optimization Lens by co-founders Bertsimas and Dunn. This book details the transformative effect modern optimization is bringing to the fields of machine learning and artificial intelligence, and is guiding teaching at leading universities like MIT. The book received the Frederick W. Lanchester Prize in 2021, awarded for the best contribution to operations research and the management sciences published in the last five years.
Interpretable AI Algorithms
Software modules spanning the entire data science lifecycle
Our technologies apply to all phases of the data science lifecycle, from data cleaning and exploration, through predictive modelling, to the end goal of data-driven decision making and driving business value.
Unlock the full power of data with missing values or quality issues
Optimal Feature Selection
Automatic selection of optimal features from the noise
Optimal Decision Trees
As powerful as black-box artificial intelligence with the interpretability of a single decision tree
Interpretable Matrix Completion
Powerful recommender system that gives detailed explanations for each suggestion
Causal Inference and Policy Learning
Learning interpretable rules while accounting for biases from observational data
More flexible than deep learning
Works with the data you actually have right now
Our algorithms can work with datasets of any size or quality, whereas significant volumes of data are required just to get started with deep learning
No specialized hardware requirements
All model training runs on standard CPUs without required specialized GPU hardware
Instant predictions and simple deployment at the edge
Our models are lightweight with near-instant prediction even on embedded IoT devices, as opposed to deep learning where specialized hardware is required for edge inference
Accessible and Intuitive
Support for major languages
Our algorithms are accessible from Julia, Python and R
Familiar and intuitive API
Our API is well-documented and very similar to Scikit-Learn, so there is very little learning curve for data scientists
Our software can run on-premises or in private cloud systems as desired