Viaduct’s TSI Prevails for Anomaly Detection

After successful deployments at multiple global OEMs, Viaduct’s proprietary AI for detecting production issues in connected vehicle data is already a proven value generator (see our previous whitepaper).  However, people continue to ask how Temporal Structural Inference (TSI) is different and whether it's actually better than what other people have tried or might be tempted to try.  We decided it was time to quantify just how much better TSI performs against likely alternatives and we’ve published a white paper to explain our methods and findings.

Feel free to check out the details, but all you really need to know can be summarized in a few words here.  We tested TSI against 6 other algorithmic approaches in two scenarios to see how it stacks up.  The first scenario is what we call the “static” scenario.  This scenario mimics the type of test done in most competitive bake offs that rely purely on historical data that doesn’t change during the course of the test. In this scenario, TSI was shown to be more than 2x more precise than alternatives, achieving greater than 60% precision where the competition is less than 25%.  

In the 2nd scenario, called the “dynamic scenario”, TSI is able to maintain greater than 60% precision while all the competitive approaches drop to near 0%.  In other words, these alternative approaches just don’t work at all. The key here is that the “dynamic scenario” is the one that resembles a production operation, where production methods, products and failure modes are changing all the time. The world is not static.  Only TSI is able to perform in this kind of real-world environment.   

‍