Researchers at the Massachusetts Institute of Technology have accelerated the construction of transport routes within a large group of cities using machine learning.

Researchers have focused on improving the efficiency of routes such as last mile logistics, where a merchant needs to deliver goods from a warehouse to multiple cities. Such problems are solved using heuristic algorithms. Usually a heuristic algorithm works by starting with the simplest, but not the best solution, gradually finding the optimal one. However, for such a large task as building routes between, for example, 2000 cities, this approach will take too long.

“In the world of logistics, time is money, and you can't keep all warehouse operations waiting for a slow algorithm to build routes. For an algorithm to be practical, it must be ultra-fast, ”explains Mark Kuo, founder of Routific, an intelligent logistics platform for optimizing delivery routes.

To solve a problem, the algorithm breaks it down into smaller ones, and then solves them in a certain order or randomly. Katie Wu, assistant professor of civil engineering and environmental engineering at Gilbert W. Winslow, and her students complemented this process with a new machine learning algorithm that picks the most useful subproblems instead of solving all of them to improve performance with less computation.

The MIT researchers ran a set of subtasks through a neural network they created, and found that this technique accelerated the process of solving subtasks by 1.5-2 times. At the same time, it is not known by what principle the neural network, which selects the best subtasks, works.

“We don't know why these sub-tasks are better than others. This will be the direction of our future work. These ideas can lead to the development of even better algorithms, ”concludes Wu.

The researchers say their approach, which they call “learning delegation,” can be used in a variety of areas, including route planning for warehouse robots.

bbabo.Net