Yale Astronomy Virtual Colloquium - Kevin Schawinski

Abstract: The past few years have shown that machine learning approaches can be useful in a wide range of scientific application. However, many implementations of ML systems in research tend to be over-engineered and sub-optimal in performance as scientists try using existing ML frameworks. These problems are largely caused by the fact that these ML frameworks are designed to be at the wrong level of abstraction. I go over some consequences of this problem, and discuss how approaches from automated machine learning can help alleviate some of them.