Can ML and Operations Research Lift Each Other Up?
Opinions on deep learning's actual capability differ. Geoffrey Hinton, awarded for pioneering deep understanding (DL), is not entirely unbiased. Still, others, including Hinton's DL collaborator Yoshua Bengio, are seeking to infuse DL with domain factors under the radar: operations research. In 2020, Bengio and his partners surveyed recent attempts from the ML and operations research communities to use ML tech to solve combinatorial optimization issues. They support pushing further the merging of ML and combinatorial optimization and approach detail.
Ways to Merge ML & Operations Research (OR)
- ML and OR are used in conjunction to help each other. ML can be leveraged to enhance OR algorithms and vice versa.
- Nikolaj van Omme isn't out to strike ML. What he is supporting is a method that merges ML and OR. Van Omme said; that you first use ML to get some estimates and then leverage those as input for OR algorithm to optimize.
- Construct new algorithms. If you understand the weaknesses and strengths of ML and OR, there are methods to merge both so that one's weaknesses are balanced by the other's strengths. Van Omme explained neural graph networks as an example of this method.