Introduction, and why would you choose PolyChord rather than Neural Networks?
At PolyChord we often collaborate with science teams from all kinds of different industry sectors. Unlike “black-box” technology providers, at PolyChord we work in close collaboration with commercial partners – bringing together our specialised data science expertise with your deep knowledge of your own domain to produce a truly effectively approach. Because our technology is complex and challenging and can more effectively explore difficult data landscapes, it’s not the kind of tool we’d simply hand over to you to use.
Many people ask us; “What’s the difference between PolyChord and a conventional Neural Network method?” There is quite a pronounced difference, Neural Networks are good at some things and not good at others. Despite this scientists and engineers often try to tackle complex and challenging data problems with NN based tools which are simply not suited to solving them. Many of these problems are suitable for the PolyChord approach. Here’s a one-pager that goes deeper into that:
Just to make things more interesting; PolyChord have been developing our own take on a better trained neural network that gives more explainable answers – this is complex, and information on it can be found on the “Technical Information for Data Scientists” page.
Why is PolyChord unusually effective at Managing Trade Offs in complex processes - such as chemicals manufacturing, or water utilities?
Essentially, managing trade-offs or “parameter optimisation” as data scientists call it, can be thought of as follows; if you turn up the volume on factors “B” and “C” in your process, how might it affect outputs “K” and “L” – whilst considering all the other letters in between? Here is another simple one-pager explaining how PolyChord performs parameter optimisation in problems where there is a combination of challenges: high dimensionality ; multiple variables and complex constraints – such as having to aim for “cheapest”, “quickest” or even “least carbon intensive”. We’re using “optimising satellite orbits” as an example here, but PolyChord can handle all kinds of variables, trade-offs and constraints which are not straightforwardly mathematical – e.g., “is it blue or red”, or “the process can’t go above 21 degrees Centigrade.” PolyChord has handled these kinds of problems in a dataset 10 billion years large, so it is powerful enough to handle your problem. Other data science technologies which attempt to do optimisation or managing trade-offs just fail where there are complex constraints plus multiple variables to manage. PolyChord does this extremely effectively and is, if you like, a next generation optimisation technology.
Optimising Sensors in a Network - a Use Case
Suppose you wanted to make the absolute best choices possible for transmitters and receivers in a network where buildings were getting in the way and acting as constraints for the passage of the signal, preventing its best possible spread through an urban environment – i.e. to maximise best coverage whilst using the least amount of expensive transmitters and receivers. This single one-pager shows you how we do that in a way which previously just did not exist. PolyChord can help your team of engineers deploy a network in the most efficient way using their knowledge and our highly advanced data science.
Leveraging Value from data you already have from sensors
Many businesses have sensors which collect data, but they only use it for simple procedures – such as alerting if there is a fault. This data contains (hidden within it) far more useful insights, and you could be leveraging the value from that data right now with PolyChord. Groups of sensors in a complex system create exactly the kind of complex data landscape that PolyChord can more deeply explore to produce actionable information. In this use case, we look at how the PolyChord technology is being used to generate predictive maintenance decision-making tools from data collected by an array of simple sensors in trams and rail.