What is machine learning?

April 23rd, 2018
Machine learning
Let's read the text

Required reading: 2 minutes

Machine learning is a subarea of the Artificial intelligence (AI), which combines data science and computer science. Computers learn from input values and continuously improve their task solving skills. The aim is to extract knowledge from data.

This technology offers numerous applications: from automatic spam filters for e-mails to personalized user recommendations at Netflix, Spotify and Amazon to self-propelled cars.

Further definitions of the term machine learning

In 1959, the AI pioneer Arthur Samuel first described the concept of machine learning as

"Research area designed to enable computers to learn without being explicitly programmed to do so."

In 1997, Tom M. Mitchell provided a more modern and formal definition of machine learning:

"A computer program shall learn from experience E in relation to a class of tasks T and performance measure P if its performance in tasks T, as measured by P, improves with experience E."

The idea behind it is that a general program enables computers to independently find solutions to a problem. Instead of meticulously giving precise instructions (such as rule-based) on how to solve this problem. Further development through experience plays a fundamental role in machine learning. Similar to humans, computers learn by failure or repetition.

Machine learning allowed the computer program "Watson" to win the Jeopardy quiz show. Already in 1996 "Deep Blue" defeated the then world champion Garry Kasparow in chess. Both programs were developed by the IT company IBM and were based on the experiences from countless games.

Tasks of machine learning

Machine learning distinguishes different types of learning tasks. Two basic approaches are supervised learning (learning with a teacher) and unsupervised learning (learning without a teacher). Both methods use learning algorithms.

In supervised learning, algorithms are trained on the basis of given examples that already contain correct answers. This learning task is used in regression analyses (goal: describing relationships and predicting values) and in the area of classification (goal: differentiating data in classes). Forecasting property prices based on the size of houses is an example of a regression problem. If, however, the goal is to predict, for example, whether an e-mail is a spam e-mail depending on sender and content, there is a classification problem.

The second basic learning approach is unsupervised learning. Algorithms search for trends, structures, and patterns in data without a goal being set for them. Algorithms find similarities in this way and can group them into segments (clustering). In practice, the process is suitable for customer and market segmentation, for example.

If you want to be inspired by one of the pioneers of AI, Andrew Ng, I recommend the video recordings of his Machine Learning course at Standford University in 2008.

My conclusion: Compared to the countless definitions of artificial intelligence, the different definitions of machine learning are quite similar. This makes the term much more tangible. It is important to understand the meaning of statistical words such as regression or classification. You can hide the statistical calculations confidently.




Introduction to Machine Learning with Python, Andreas C. Müller & Sarah Guido, 2017

These contributions should please you

No comment

Comment now