Bayesian Optimisation

Bayesian Optimisation is a technique used to try and find the optimum combination of input values to try and produce a maximum output value.

In this case, we use it to try and give you the combination of attributes (such as speed, pressure, temperature etc.) you need to try and achieve a specific metric (such as quality, colour or other feedback measurement).

To perform optimisation, you need a number of existing captures that already have values for the attributes you're asking for.

For example, if you want to calculate the best oven temperature, fan speed and conveyor speed to produce cupcakes with a specific colour, you'll need to specify some details about the cupcake (ie. the size, flavour etc.) and have several captures containing samples of those cupcakes.

The system will then try to estimate the temperature, fan and conveyor speeds you need to make a cupcake with the perfect colouring.

The analytic isn't aware of the way your various inputs might be related, so it may produce suggestions that aren't physically possible or practical.

If your data exhibits strong 'clustering' (ie. each production run appears vastly different from others), this indicates that there may be complex interrelationships - the results of your optimisation might not be usable in real-world scenarios.