Our Technology

A self-consistent, iterative algorithm that imputes sparse data.

• Unlike most machine learning approaches, which require a critical mass of complete inputs and outputs for training, Alchemite™ works for datasets with significant gaps in the data.
• It does this through techniques to estimate missing data and then iterate rapidly to improve its model until it converges on a solution.

Reliable non-parametric uncertainty quantification.

• Alchemite™ tells you, clearly and instantly, how much reliance you can place on every estimated data point.
• You can quickly see where more experimental data will improve your model, or identify the most reliable candidate for a new product.

Simultaneous prediction of multiple properties.

• Unlike traditional methods, which model one output at a time, Alchemite™ predicts all targets and their uncertainties simultaneously.
• Alchemite™ can tackle complex problems that are intractacble for other algorithms; it was originally validated on a project to discover new aerospace alloys that simultaneously satisfied eleven physical criteria.

~1000 times less computational resource than alternatives.

• Alchemite™ algorithms are highly efficient compared to other leading machine learning methodologies.
• It can return exceptionally fast results and handle large databases.

A probabilistic, goal-driven methodology.

• Based on your objectives, Alchemite™ mathematical methods enable efficient design of experiments.
• Focus time and expense based on rational assessment of potential performance improvements and the likelihood of success.