Scientio is getting towards the end of a successful collaboration with a medical devices start up. Basically the product works, and barring some tinkering and approvals the initial development phase is over. I’m not going to talk about this company; there’ll be a separate splash when they are ready to publicise things, but, obviously, having built up this expertise we’d like to reuse it.
My feeling is that there are other enterprises like this, who may not be aware of what we do, or that what we do can be done. I don’t intend to break any confidences so I’m going to talk about our experiences in general terms in this post.
This company had a unique way of interpreting and conditioning a kind of sensor that is frequently used. They also had a set of tests built around this and other sensors, and an expert who could detect a range of conditions using this set up. Obviously with only one expert and only so many hours in a day the earning potential of this idea was limited, so how could they automate and reproduce this idea, so it would be available across America?
Scientio’s interest in this was the automation of the expert’s diagnostic knowledge, and the provision of this as a central cloud based diagnosis engine. The result is that this diagnostic method is now leveraged so that thousands of tests can be handled in the time required for one manual test. This previous post talks about the architecture we used.
We’ve discovered through this process that Scientio’s engine is ideal for such tasks.
First of all, in an environment where approvals and compliance are important, The rules, though stored as XML, are easily displayed as English language if…then text, so the function of the system can be easily verified.
The rules are testable, either as a complete functional block or individually, and we supply software in our Lacuna product that can find any unintentional gaps in the rule sets, i.e. combinations of inputs that ought to produce a valid result but don’t.
When you add a new fact to a conventional expert system you have no idea how long it will process before stable results are generated.
XmlMiner uses pre-processing to format the rules for straight through processing. The run time is defined and exceedingly speedy.
The power of fuzzy logic also makes it easier to transfer the expert’s knowledge to a set of rules. Scientio’s fuzzy logic inference engine is entirely capable of handling competing solutions and handling them in a rational way. Fuzzy logic makes for very expressive rules: we were amazed how small the set of rules used in the final product were.
Smaller rule sets mean lower maintenance costs and easier approvals.
XmlMiner can tell you when a set of input data is outside of the circumstances the rule set was created to handle. This means it’s easy to flag exceptional circumstances for human supervision or monitoring.
So, if you are trying to make that jump from a human expert based process to an automatic, semi automatic or human supervised process contact Scientio, we’d be happy to hear from you.