Next to my interest for Cloud Computing, I recently started to focus more and more on Machine Learning and became passionate about it. To me, it is a truly fascinating field of work! However, when I talk to friends, more often than not they don’t have a clear understanding of Machine Learning: yes, it is about making computers smart. But, other than that? For example: how does it work in general and what are some use cases? Even in the Silicon Valley, when I introduce myself as a Machine Learning Engineer, people sometimes ask me to explain it in more detail. Hence, in this post I will try to briefly describe, what Machine Learning actually is, before talking about some exciting use cases like Google Translate.
What is Machine Learning?
The fundamental research question in the field of Machine Learning can be described as the following:
“How can we build computer systems that automatically improve with experience, and what are the fundamental laws that govern all learning processes?” (Tom Mitchell)
An important keyword of the research question is “experience”. I think, what is meant by experience can be best explained with the help of an example which also illustrates the general process of Machine Learning:
Suppose, you want to train a computer system to classify the category of a twitter post, e.g. sports, politics, nightlife. Then, at first you need to feed some sample tweets for all of those categories to the computer. These examples are called “labeled data” because their category is known and in our case they serve as the “experience”. In a next step, you tell the computer to use a special algorithm (e.g. logistic regression or SVM) to detect patterns in the twitter messages and to learn and save those in form of a model. Finally, when you provide a new twitter post to the computer without revealing the category (= unlabeled data), it will classify the category based on the model and will hopefully be accurate.
So in summary, the main objective of learning is to generalize from a set of few training examples to be able to answer queries on previously unseen examples.
Alright. But why is it awesome?
Most of us use products based on Machine Learning techniques several times a day. But – at least in my case – either the products enhance my user experience in such a subtle and enjoyable way that I don’t recognize the technology or I would have never guessed on the product using Machine Learning in the first place. Sure, everybody knows about the Amazon recommendation system, but how about these examples:
- Ever wondered how Facebook’s personalized “News Feed” is being created?
- How does Siri know what to answer? And how does it decode your speech into text beforehand?
- How do gaming companies like Zynga optimize the process of when to suggest costly upgrades to the player and thereby increase their revenues?
- How does Goolge’s self driving car work?
- How do people’s faces on images get detected?
- How are spam mails identified?
The answer to all those questions is: Machine Learning (sometimes in combination with other techniques).
Even Google Translate is based on Machine Learning techniques, which might sound surprising. Intuitively one might think, that language experts crafted tons of rules and translated millions of words. This work would then serve as a basis for the translation service. Far from it! In reality, Google uses publicly available data on the Internet which already exists in a translated version. For example all the European Union documents need to be translated by experts into the several languages of the European countries. Google grabs those documents and many many more and allows a Machine Learning algorithm to crunch this massive amount of data to detect patterns and learn translations. This approach makes it possible to translate texts to 58 different languages in a decent quality.
To me personally, Machine Learning is especially appealing because it originates from the domain of science and yet had a major commercial success. Often times, research in Mathematics or Computer Science does not have an immediate impact on the business world due to rather abstract concepts. For the discipline of Machine Learning it is different. Research achievements in recent years allowed for such “magical” products like Siri. And the outlook for Machine Learning appears to be bright, as the amount of stored data increases rapidly and recommendation systems gain in importance due to the rise of social media.