I have been fascinated by the rise and fall of graph cut algorithms in recent years, which I described in this question: Was there something specific that caused graph cuts to lose popularity in the last 5 years?
On that question, I received an answer from an expert in the field. The main reason given for the drop in popularity of these algorithms, was the fact that graph cuts were mainly used for problems that deep neural networks are now able to solve more accurately, without the need for optimization.
But graph cuts were immensely popular during their heyday, and lead to many developments in mathematics (such as quadratization, as most applications of quadratization and QUBO involve graph cuts). I would be amazed if nothing they provided us can beat neural networks for a useful problem. 2015 was not long ago, and in that year papers on graph cuts like this one got almost 1000 citations in a single year alone! (more than some researchers get across all of their papers in a lifetime).
This makes me wonder: Is there any reason why I would rather use graph cuts than deep learning?