Why Optimize?

3D Optimization Problem

What is the best proportion of different crops to grow together in the same field to maximize yield at harvest time? Do you get flats when you pump your bike tyres to a higher pressure than recommended? What is the perfect amount of time to boil an egg to get that center soft but the white cooked?

All these repeating problems can be framed as data optimization problems. In our lives and work, we make many repeated decisions, both big and small, which benefit from being optimised.

Optimize combines the latest research from the University of Oxford and advanced software engineering to deliver efficient data optimization to you via a robust and user-friendly API.

What does Optimize do?

Optimize is a global data optimization tool. This means that it does not rely on any special properties of your problem to find the best configurations - it is a general purpose optimization solution.

To do this, Optimize implements Bayesian Optimization under the hood.

What is Bayesian Optimization?

Bayes was an 18th century statistician. He invented a method of reasoning about risk which incorporates both evidence and prior knowledge which we call Bayesianism.

Bayesian Optimization leverages this method of reasoning to efficiently find optimal configurations for anything that requires parameterisation.

What is Optimize best at?

Is the process you want to optimize a real and messy procedure, like planting your crop for the season or manufacturing a new component to an industry standard? Or is your problem highly complex and sensitive to its parameter settings, like a complex data science pipeline?

Bayesian Optimization is particularly efficient when it is either very expensive for you to evaluate a single set of parameters, when the relationship of performance to the parameter settings is unknown or very complicated, or when the measurement of performance is a noisy process.

Benefits of using Optimize:

Mind Foundry model health and performance statistics
  • Optimize is a core component of our platform and is constantly improved for our internal needs.
  • It is robust and simple to use, making it extremely easy to integrate and plug into your existing workflows and products.
  • Accelerate your model R&D and catalyze your creativity.
  • Optimize is constantly benchmarked on a library of 300+ of the most challenging data optimization and model calibration tasks in order to guarantee performance improvements.

Key Features:

Visualization of the surrogate model projected on a 2D space

Optimize is more than just an optimizer:

  • Multi-objective: optimize more than one metric at a time
  • Batching: Ask for multiple parameter recommendations at each iteration
  • Secure: Optimize doesn’t access your data or models to solve your tasks
  • Visualise the probabilistic models generated by the tool for each task
  • Online tutorials and documentation available to support adoption
  • Available in Python and R to suit your team's preferences

Schedule a Demo