The Alchemite™ method provides a unique combination
Value from sparse, noisy data
Traditional machine learning (ML) models struggle with sparse data (data with gaps) or data that is noisy. Alchemite™ uses a self-consistent, iterative algorithm to impute sparse data, filling in the gaps. This enables Alchemite™ to train machine learning (ML) models using real-world experimental and process data so that it can solve problems where most ML methods fail.
Accurate uncertainty quantification
Truly understanding the accuracy of predictions from ML models is vital to their effectiveness. It enables sound decisions and lets you get started even with small amounts of data. But ML solutions often use fairly crude approximations to calculate uncertainty. Alchemite™ offers advanced uncertainty quantification based on nonparametric probability distributions for many properties simultaneously.
Optimise against multiple targets
Alchemite™ enables you to solve high-dimensional problems with multiple targets (for example, a formulation design) that are intractable without an ML approach. And it does this even when the training data is sparse and noisy.
Start fast
Alchemite™ generates a model from the available data that you can then refine and apply. There is minimal need for data cleaning and model setup, removing the high dependence on starting assumptions that is typical of many methods. With accurate uncertainty calculations and DOE tools to advise on the next experiment, you can iterate quickly towards your objectives.
For large or small datasets
Alchemite™ computations have a light memory and CPU footprint, so not only is Alchemite™ lightning fast, it can handle huge databases (where needed). But its Design of Experiments (DOE) tools also make it ideal for problems where you have very little data. By providing guidance on which experiments to do, it can get you to your targets with 50-80% fewer experiments than conventional DOE methods.
A global view
Alchemite™ works across all available information to extract useful knowledge. This supports a global view (e.g., considering both ingredients and processing for a formulation), finding relationships that humans may miss.
Intellegens development
Our Cambridge-based team of scientists and software developers continually develops the Alchemite™ software, delivering both algorithmic improvements and usability enhancements in regular releases of the Alchemite™ Suite. We’re have also developed complementary technology to support enterprise application of machine learning. For example, Ichnite™ is an application framework which helps you to combine methods, models, and data sources to maximise machine learning insights
