SIAM News Blog

Letter to the Editor: Deep Learning Effects in Detroit Auto and Beyond

I'm a SIAM member and a Ph.D. computational multibody dynamicist working at Detroit Auto and truck transportation. I also worked for a company that made a modeling and simulation software called Adams, fueled and solved by former SIAM President C. William Gear’s stiff differential equation integration algorithm.

I was happy to read Michael Elad’s review of deep learning and the growing application of success versus governing equation-controlled results. In the auto world, I call this “business” engineering versus “science” engineering. Harvard MBA CEOs tend to support the former over the latter, and costs are the controlling direction. Will this driven intention of making huge wealth for a small amount of people result in overturning scientific rigor and precise process? And do humans learn anything here, or does that matter anymore, given the advent of robots and autonomous “vehicles of knowledge”? I know a lot of low academic types who favor the business engineering approach and would gladly accept the answer from a neural network algorithm. I remember this methodology using feedback control devices, and I’m sure the concepts are being arranged and addressed for future autonomous vehicle developments.

Yet, I’m sad to see how science is being battered, if not ignored, in the matter. But it’s also like the mathematics of finite element analysis, where too many degrees of freedom are being reduced using a variety of projection methods to remove all the computational noise, getting to the essential mechanics equations that yield the critical solutions with great accuracy. In this case, the large computational problem is being slimmed down. Will this happen in deep learning, with its excessive data collection and analysis approach?

Very good article provided here by asking more about the philosophy of human value in what transpires on the planet. I mean at what point do humans, even the very rich ones, become inconsequential debris in the way of much deeper thinkers that are not human? Deep learning could be writing a new science fiction to fact-scripted plot. Al Kovacs, South Lyon, MI

Read Michael Elad’s article, “Deep, Deep Trouble,” in the May issue of SIAM News.

blog comments powered by Disqus