Machine Learning is tightly related to optimization. Non-gradient algorithms are used when the objective function is not smooth or a closed form of the objective function is not available. There are many non-gradient algorithms (derivative-free optimization) as well such as Bayesian optimization, Cuckoo search, Genetic algorithms, etc. The Gradient Descent, one of the most popular optimization algorithms in Machine Learning, is a gradient-based (obvious from its name) optimization.
Drivers can drive only for so many hours a day. For instance, each car has a distance from potential customers. X_i=1 means the driver i is selected and will be sent to the customer. cost, loss, risk of some undesirable event, etc.) X’s are the decision variables. There is an objective function to be maximized (i.e. This type of formulation is called optimization or mathematical programming. In mathematical terms, the problem above can be written as: Maximize F(X1, X2, …, Xn) Such that it meets the constraints C1, C2, …, Cm. Operations Research in one sentence: Do things best under constraints. There also will be a cost associated with every dispatch and the routing plan should meet the constraints specific to Uber’s policy. I do not know what Uber’s objective function is, but there is something that they are trying to maximize by dispatching the drivers. And those decisions must be made while optimally using available resources. You evaluate every possible option by weighing each option’s pros and cons.įor example, in order for Uber to have a master routing plan, it has to decide which driver should be sent where, when, and how much they should charge the customers. If we have to make the best decision possible, what should we do? Operations Research in one word: Optimization.
So I thought I’d give OR, the subject that I studied at graduate school, a nice little update that it deserves. When you google “Operation s Research”, you get a very long Wikipedia article, however, the explanation is a little bit all over the place and to be honest, outdated as well.