What is machine learning? In the past year, whether it was during a conference, a seminar or an interview, a lot of people have asked me to define machine learning. There is a lot of curiosity around what it is, as human nature requires us to define something before we begin to understand what its potential impact on our lives may be, what this new thing may mean for us in the future.
Similar to other disciplines that become suddenly popular, machine learning is not new. A lot of people in the scientific community have been working with algorithms to automate repetitive activities over time for several years now. An algorithm where the parameters are fixed is called static algorithm and its output is predictable and function only of the input variables. On the other hand, when the parameters of the algorithm are dynamic and function of external factors (most frequently, the previous outputs of the same algorithm), then it is called dynamic ts output is no longer function only of the input variables and that is the founding pillar of machine learning: a set of instructions that can learn from the data generated during previous iterations to make a better output the following time.
Scientists, developers, and engineers have been dealing with fuzzy logic, neural networks, and other kinds of machine learning techniques for years, but it is only now that this discipline is becoming popular, as its applications have left the lab and are now used in marketing, sales, and finance—basically, every activity that requires the repetition of the same operation over and over again could benefit from machine learning.
The implications are easy to grasp and will have a deep impact on our society. The best way I can think of to describe what will likely happen in the next 5 to 10 years with machine learning is recalling what happened during the industrial revolution. Before the advent of the steam engine, lots of people were performing highly repetitive physical tasks, often risking their lives or their health for minimum wages; thanks to the industrial revolution, society evolved and machines took over the relevant parts of manufacturing processes, leading to improved yields, more predictable and stable outputs, improved quality of the products and new kinds of jobs, controlling the machines that were replacing physical labor. This was the first time in the history of mankind where man had delegated the responsibility for the creation of something else to a thing we had designed and invented. In the same way, machine learning will change the way data operations are performed, reducing the need of human intervention and leaving optimization to machines and algorithms. Operators will no longer have a direct control over data, but they will control algorithms that, in turn, will control data. This will allow faster execution of operations, larger datasets will be manageable by fewer people, errors will be reduced, and more stable and predictable outcomes will be guaranteed. As many things that have a deep impact on our society, it is easy to love it as it is to hate it. Lovers will praise the benefits that machine learning will drive to their lives, haters will be criticizing the fact that, in order to be effective, machine learning needs lots of iterations, hence, lots of data. Usually, the data we feed algorithms with is our own personal information.
In fact, the main applications where machine learning is taking off as a tool to improve productivity are marketing and customer support, where a deep knowledge of the customer is required to give him/her the personal service that will make the difference between a purchase or a visit or between a happy and an unhappy customer.
In marketing, for example, marketers are starting to take into consideration information, such as location, device, past purchases, what websites one has visited, weather conditions, to name just a few of the parameters that determine whether a company would decide to display its ads to a specific set of customers.
Long gone are the days of broadcasting marketing messages through untraceable media, such as TV or newspapers. Today's marketers want to know everything about who clicks and buys their products so that they can optimize creatives, spend, and allocate budget to make the best possible use of the resources at their disposal. This leads to unprecedented levels of personalization that, when exploited properly, make customers feel valued as individuals and not part of a socio-demographic group.
It is intriguing and challenging at the same time, but there is no doubt that the winners of the next decade will be those companies or individuals who can understand unstructured data and make decisions based on them in a scalable way: I see no other way than machine learning to achieve such a feat.
Andrea Isoni's book is a step into this world; reading it will be like a peek down the rabbit hole, where you'll be able to see a few applications of these techniques, mostly applied to web development, where machine learning serves to create customized websites and allow customers to see their own, optimized version of a service
If you want to futureproof your career, this is a must read; anyone dealing with data in the next decade will need to be proficient in these techniques to succeed.
Davide Cervellin, @ingdave
Head of EU Analytics at eBay