This is the second part in our series answering the question: “With regards to what you do - scheduling, route and fleet optimization - what is artificial intelligence, and how does it compare to machine learning, or other technologies?”
You can read the first part in the series, a look at artificial intelligence in general, here.
What is Machine Learning?
The general definition from Wikipedia is succinct:
Machine learning is a field of computer science that uses statistical techniques to give computer systems the ability to “learn” (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed.
Machine learning is a sub-domain of artificial intelligence which focuses on learning by example. Machine learning has been one of the central sources of success in the field of artificial intelligence in recent years, which is why AI and machine learning are sometimes used interchangeably.
The Harvard Business Review gives a good example of how it works:
Artificial intelligence and machine learning come in many flavors, but most of the successes in recent years have been in one category: supervised learning systems, in which the machine is given lots of examples of the correct answer to a particular problem. This process almost always involves mapping from a set of inputs, X, to a set of outputs, Y. For instance, the inputs might be pictures of various animals, and the correct outputs might be labels for those animals: dog, cat, horse. The inputs could also be waveforms from a sound recording and the outputs could be words: “yes,” “no,” “hello,” “good-bye.”
Many of the applications of machine learning today use an approach called deep learning, which we won’t discuss in detail here, but can make use of much larger datasets than was previously possible.
How Machine Learning Works
Above, we mentioned that in general, machine learning gives computers the ability to “learn” with data.
In general, to “train” a machine learning model, one gives the model a large number of inputs, for which the correct output is already known. As an example, an image recognition algorithm would be fed a lot of images (input X), and told the subject of the image (output Y). This way it learns the correct answers, and eventually, the hope is that it will be able to predict Y when an input X is given on it’s own.
There are many different algorithms and techniques one can apply to do this, but he end goal is always the same - develop a model that can successfully predict output Y from input X.
Applications of Machine Learning
Again, machine learning is typically used in “Any situation in which you have a lot of data on behavior and are trying to predict an outcome”.
Last time we mentioned a few examples of AI, many of which would be valid applications of machine learning, and relevant to our own domain in scheduling and logistics:
Predict charter demand by analyzing historical schedules.
Input: historical charter schedules.
Output: new schedule for next two weeks.
Predict the impact of a particular weather event by analyzing historical weather events and their impacts.
Input: historic weather disruptions.
Output: more robust schedule for next two weeks taking into account predicted weather disruptions.
Predict the frequency and impact of unplanned maintenance requirements on a transportation/aviation fleet through a given week/month/year.
Input: historic unplanned maintenance disruptions.
Output: more robust schedule for next two weeks taking into account predicted disruptions.
Some other common applications of machine learning are:
Speech recognition from large numbers of inputs of voice recordings with their corresponding transcripts.
Trading bots using historical market data and current market indicators to execute stock trades.
Purchase histories and responses to ads of customers to predict success of future ads and placements.
Netflix movie recommendations are based on a model that includes machine learning.
And of course, the example we mentioned in our last post:
Predicting where and when a baseball player will hit the ball during a game based on his historical career data.
In the next post in this series, we’re going to talk about mathematical modeling, and then we’ll discuss the differences between the two techniques.
Read our last post about artificial intelligence here.