With our tool you can predict incidents and control your manufacturing
We find the reason of incidents and show it in human readable format to engineers
Using explainable features of our tool you can the reasons of crashes and prevent it effectively!
We use ensembles of neural networks and gradient boostings to detect incidents. With recent paper on uncertainty estimation published at NIPS conference we get uncertainty of model prediction, so we can see if it is confident.
For each prediction we find features that differ the most from usual behaviour and show them to the engineer to allow him to understand what is wrong.
With uncertainty estimation we understand if the model is confident in current prediction. For 83% of data the model has high confidence and 0% errors. This 83% of data can be processed fully automatically. And remaining 17% go to the engineer to check them.
We use ensembles of neural networks and gradient boostings to detect incidents. With recent paper on uncertainty estimation published at NIPS conference we get uncertainty of model prediction, so we can see if it is confident.
For each prediction we find features that differ the most from usual behaviour and show them to the engineer to allow him to understand what is wrong.