ML and OR
AA: ML, and DL specifically, has its limitations, and one way to overcome these is via the integration of ML and OR, allowing us to combine both data and domain knowledge – you’re one of the leading advocates of this.
Can you please give a brief overview of OR, its history, how to integrate it with ML, and why this combination has so much potential?
NVO: All approaches have strengths and limitations. Most of the analytical approaches are somewhat related to each other. In theory, you could consider how to transform a problem so that you could obtain the same kind of solutions by one or another approach, but practically, each approach has its pros and cons. Sometimes the differences are huge; for instance, if you try to optimize with ML or OR. Also, the approaches and goals are somewhat different. For instance, ML relies heavily on data and you hope to find patterns through automatic study (optimization of parameters and hyper-parameters), while OR is more...