A few weeks ago, Engineering.com wrote an article that nicely summarized Eureqa’s impact in the manufacturing space, slashing the cost and boosting the performance of metals. But while we lovingly call our product the “Robotic Data Scientist” for its ability to automatically inform engineers (or any analyst/data scientist) how to optimize processes, one thing Eureqa will never do is eliminate the need for good engineers.
Eureqa extends engineers’ capabilities so they can get better results, faster. Manufacturing complex widgets and running efficient supply chains involves a significant amount of predictive and analytical modeling to problem-solve: If we feed these inputs into the assembly line and manufacture them in this way, what’s likely to happen? Based on how we currently run this process and all the inputs at our disposal, what can we change to generate a more durable, lower-cost widget or an entirely new material?
Typically, the modeling process is extremely time-consuming. Most engineers do not come from statistical or computer science backgrounds. While brilliant, their comparative advantage is not creating analytical models to decipher which inputs and processes lead to the best results. Engineers are smart enough to figure it out, but it’s asking them to put extraneous time into something that’s not their specialty, and something that can be accomplished faster and more accurately with computers than with any possible team of humans.
Engineers are domain experts. They understand their project, and they understand all of the various factors at play. They are Eureqa’s coaches. They can apply Eureqa to a problem (why is this process failing 15% of the time?), feed it all of their data, and let it churn through to determine the best equations – from the very simple to the very complex – that characterize the problem. Eureqa can do in minutes what it takes even a well-trained team to do in weeks or months (in some cases, if ever). Under the hood, Eureqa leverages free form modeling, a treasure chest of techniques developed by the world’s top data scientists, to automatically search the infinite space of equations that could explain the data. Once Eureqa converges on completion, it gives the engineer a few models to choose from and iterate on.
This is where the engineer’s expertise is not only a nice to have, but a need to have. There may be certain features Eureqa finds and recommends changing, for example, which can’t be touched due to regulatory compliance measures. Other attributes, like process conditions, may be prohibitively expensive to adjust. Engineers can recognize this and omit such variables from the data for more actionable, telling models. They can put thresholds on variables, stray from models too simple or complex, and easily interpret what the models actually say. They don’t need to work for weeks manually testing hypotheses. They can build accurate predictive and analytical models, that are easily understandable in plain English, orders of magnitude faster than they could with legacy tools and techniques.
One of Nutonian’s goals is to enable any company – any engineer – to accelerate the speed and scale of any data science initiative. We’re teaching engineers to fish, not snatching the fish from their hands.