Udayan Kanade, CEO of Noumenon Multiphysics gave a keynote talk at the COMSOL conference in Bangalore this year. It was an insightful look into how data science methods can be combined with traditional physics-based simulations to solve complicated real-world problems.
Both data science and simulations are ways of modeling aspects of the real world. Kanade explained how data science fails when the degrees of freedom of the model increase, and in turn data requirements explode. In such cases, simulations become the only viable option, since simulations are equation based and equations are great at reducing the degrees of freedom. But where do these equations come from in the first place?
Kanade showed how the need for equations leads modelers into making unsubstantiated assumptions — giving approximate equations, and in turn, approximate simulations. He proposed a new approach where ad-hoc assumptions are avoided and only true assumptions are used. This is great since the equations are now not only exact, but also more generic i.e. universally applicable. However, the reduced assumptions still leave us with a few degrees of freedom that have to be modeled accurately. Fortunately, modeling a few degrees of freedom is exactly what data science does best!
Kanade was excited about the prospect of such a future where data science can be put to good use when combined with sound physics-based simulations. He stressed that this can only be done if we get our fundamental physics right, i.e. by developing the correct equations for representing the real world. Noumenon Multiphysics has already set the ball rolling in this direction with its ‘universal’ plasticity equation.
