Roberto Battiti is full professor of Computer Science at Università degli Studi di Trento and Director of the LION lab (machine Learning and Intelligent OptimizatioN) for prescriptive analytics.
Mauro Brunato is assistant professor and member of machine Learning and Intelligent OptimizatioN (LION) research group at Università degli Studi di Trento.
Learning and Intelligent OptimizatioN (LION) is the combination of learning from data and optimization applied to solve complex and dynamic problems. The LION way is about increasing the automation level and connecting data directly to decisions and actions. More power is directly in the hands of decision makers in a self-service manner, without resorting to intermediate layers of data scientists. LION is a complex array of mechanisms, like the engine in an automobile, but the user (driver) does not need to know the inner workings of the engine in order to realize its tremendous benefits. LION's adoption will create a prairie fire of innovation which will reach most businesses in the next decades. Businesses, like plants in wildfire-prone ecosystems, will survive and prosper by adapting and embracing LION techniques, or they risk being transformed from giant trees to ashes by the spreading competition.
Optimization fuels automated creativity and innovation. It looks like a manifest contradiction, because creativity is usually not related to automation. This is why the message you will find in this book is disruptive, far from trivial, even irritating and provoking for people believing that machines are only for shallow mechanical and repetitive tasks. Starting from Galileo Galilei (1564-1642), to change the world with science, not only to interpret it with philosophy, one needs measurements and experiments. ”Measure what is measurable, and make measurable what is not so.” Measurements start shy and humble but permit a gradual and pragmatic conquering of the world as far as production means and quality of life are concerned. Almost all business problems can be formulated as finding an optimal decision x by maximizing a measure goodness(x). For a concrete mental image, think of x as a collective variable x = (x1, . . . , xn) describing the settings of one or more knobs to be rotated, choices to be made, parameters to be fixed. In marketing, x can be a vector of values specifying the budget allocation to different campaigns (TV, newspaper, web, social), and goodness(x) can be a count of the new customers generated by the campaign. In website optimization, x can be related to using images, links, topics, text of different size, and goodness(x) can be the conversion rate from a casual visitor to a customer. In engineering, x can be the set of all design choices of a car motor, goodness(x) can be the miles per gallon traveled. Formulating a problem as “optimize a goodness function” also encourages decision makers to use quantitative goals, to understand intents in a measurable manner, to focus on policies more than on implementation details. Getting stuck in implementations, to the point of forgetting goals, is a plague infecting businesses and limiting their speed of movement when external conditions change. Automation is the key: after formulating the problem, deliver the goodness model to a computer which will create and search for one or more optimal choices. And when conditions or priorities change, just revise the goals quantified by the goodness measure, restart the optimization process, et voila`. To be sure, CPU time is an issue and globally-optimal solutions are not always guaranteed, but for sure the speed and latitude of the search by computers surpass human capabilities by a huge and growing factor.
But the awesome power of optimization is still largely stifled in most real-world contexts. The main reason blocking its widespread adoption is that standard mathematical optimization assumes the existence of a function to be maximized, in other words, an explicitly defined model goodness(x) associating a result to each input configuration x. Now, in most real-world business contexts this function does not exist or is extremely difficult and costly to build by hand. Try asking a CEO “Can you please tell me the mathematical formula that your business is optimizing?”, probably this is not the best way to start a conversation for a consultancy job. For sure, a manager has some ideas about objectives and tradeoffs, but these objectives are not specified as a mathematical model, they are dynamic, changing.