What is Machine Learning?

While machine learning is technically a subcategory of artificial intelligence, machine learning differs from AI in both its goal and purpose. While AI is intended to mimic human intelligence, machine learning does not. Machine learning simply aims to teach a machine how to perform tasks by identifying patterns. Machine learning can be tested and studied in order to improve the perception, cognition, and action of a computer system, and allow it to come to accurate conclusions regarding the data that has been input.

Machine learning is what allows artificial intelligence to progress, as it is how a computer system is able to develop intelligence. Machine learning uses algorithms trained on data in order to produce models that are able to complete complex tasks. These tasks are more widespread than one may even know. They include email sites filtering spam mail, websites making personalized recommendations, and even banking software detecting unusual transactions.

There are four main machine learning techniques: supervised learning, unsupervised learning, semi-supervised learning, and re-inforcement learning. Supervised learning is when a machine has input and output data with correct labels, which teaches it to make predictions about output values. For example a machine learning algorithm being trained with images labelled with what they are allows the machine to more accurately identify and sort them in the future. Unsupervised learning is when a machine is trained using input or labels, but no samples. It may sort the input into different categories, but may not be completely accurate.

Semi-supervised learning is, unsurprisingly, a cross between supervised and unsupervised learning. In semi-supervised learning, the machine performs tasks on datasets with both labeled and unlabeled data. Since labels can be expensive, semi-supervised learning is a good compromise between supervised and unsupervised learning, and is more accurate than unsupervised learning.

In reinforcement learning, there is no input, the machine simply explores and interacts with its environment, and is meant to learn by experience only. If the machine does a good action, it gets a positive reward, and if it does a bad action, it gets a negative reward. The rewards are feedback, and the goal is for the reinforcement learning agent to maximize the positive rewards.

Deep learning is machine learning taken one step further. Deep learning utilizes a larger network than machine learning. These networks neural networks function similarly to the human brain, using logic to come to conclusions about data. These large neural networks are used in order to allow machines to learn complex patterns and make predictions without having to rely on human input. Both machine learning and deep learning allow organizations to gain insight into data with greater speed and efficiency than could be done by a human.

Opinions expressed by the author are not necessarily those of WITI.