What is machine learning?Posted on: May 24, 2022
by David Diaz
Machine learning is the branch of artificial intelligence – or AI – that imitates human learning. Through the use of complex algorithms and data, computer systems can effectively learn the way a person does, and iteratively improve their understanding in order to perform tasks, solve problems, and make decisions.
Is machine learning artificial intelligence?
Machine learning is sometimes used synonymously with artificial intelligence, but while they are intrinsically linked, they aren’t the same thing.
Artificial intelligence is the technology that enables machines to simulate human behaviours. Learning is one such human behaviour, and it’s the one that machine learning seeks to emulate. So machine learning is a subset of artificial intelligence, one that enables a machine to continually learn and evolve based on the data it accesses and analyses.
Other subsets of artificial intelligence include:
- Natural language processing (NLP). Natural language processing technology emulates human speech and writing. Applications include text generation, question answering, context extraction, classification, and machine translation.
- Expert systems. Expert systems simulate the judgement and problem-solving capabilities of a human being with expert knowledge and experience within a particular field. They’re designed to solve complex problems using if/then rules.
- Robotics. Robotics are machines that are designed to perform human actions. They’re used in a diverse range of industries, including healthcare, agriculture, and automotive manufacturing, as well as for warehousing, logistics, and supply chain tasks.
- Machine vision. Machine vision simulates human sight, with applications such as image and video recognition. Machine vision, however, isn’t limited by the same barriers to human eyesight; for example, it can use other data sources, such as radio waves, to see through walls.
- Speech recognition. Speech and voice recognition allows machines to understand and even translate the spoken word. It is used for both speech-to-text and text-to-speech applications, and is commonly used in virtual assistant software, such as Siri from Apple, Cortana from Microsoft, or Alexa from Amazon.
There are also subsets of machine learning, including:
- Deep learning. Deep learning is an important and powerful subset of machine learning. It more closely mirrors the structure of the human brain, using a complex structure of algorithms to learn from raw and unstructured data, whereas traditional machine learning relies more on linear algorithms. Deep learning is frequently used in predictive modelling, data science, and statistics work, because it can collect, analyse and interpret huge amounts of data. An open-source example of deep learning is Keras, a high-level neural network API.
- Neural networks. Artificial neural networks (ANNs) or simulated neural networks (SNNs), are inspired by the human brain’s neural networks, neurons, and ability to recognise patterns and solve problems. They can accurately classify and categorise information within huge swathes of data, and there are even deep neural networks. For example, TensorFlow is an open-source platform with a focus on deep neural networks, training, and tools such as its TensorFlow Profiler, which can help optimise machine learning algorithms. It also provides TensorFlow certifications for developers.
- Computer vision. Computer vision systems learn from digital images and videos. For example, there are open-source computer vision applications, such as OpenCV, that offer real-time, optimised libraries and tools to perform tasks such as image processing and face detection, with developer languages including python and C++.
How does machine learning work?
Machine learning systems are developed using a variety of machine learning algorithms. As a machine explores datasets and identifies patterns, its algorithms enable the machine to learn – without human intervention or programming – from the data and iteratively improve itself and optimise its understanding of new data points. This enables the system to make predictions, solve problems, produce accurate outputs, and so on.
Machine learning models
Machine learning can be divided into a few different categories.
Supervised learning, or supervised machine learning, is used to classify data and predict outcomes. Its algorithms are effectively trained through the use of labelled training data, and there may be human validation of its outputs. Supervised learning typically has two purposes or metrics:
- Classification. Data is categorised into classes. Common algorithms include linear classifiers, support vector machines, and decision trees.
- Regression. Data is analysed and used to make predictions. Common algorithms include linear regression, logistic regression, and gradient descent.
Common uses for supervised learning models include image recognition and objective recognition, predictive analytics, sentiment analysis, and spam detection.
Reinforcement learning is a feedback-based technique used to improve an AI system or agent’s performance. This allows the machine to learn through trial and error.
There are two types of learning methods used:
- Positive reinforcement learning. An AI agent is rewarded with good feedback for performing a task correctly.
- Negative reinforcement learning. An AI agent is given negative feedback for performing a task incorrectly.
Reinforcement learning is used for technologies such as self-driving cars and chatbots, as well as in gaming and robotics.
Unsupervised learning is useful for unwieldy datasets where subject matter experts are unsure of common properties. Algorithms are trained with unlabelled, unclassified data, and outputs are analysed and clustered. It typically has one of three purposes:
- Clustering. Data is grouped based on similarities or differences.
- Association. Data is sorted into relationships between variables.
- Dimensionality reduction. Data is reduced – while retaining its integrity – to assist with visualisation and make it more manageable to process and analyse.
Common uses include data analysis, anomaly detection, customer segmentation, image recognition, and recommendation engines.
Semi-supervised learning is a mix of supervised and unsupervised learning, with learning occurring through a mix of labelled and unlabelled data.
Where is machine learning used?
There are a rapidly growing number of applications and use cases for machine learning. These include:
- Big data. Data scientists frequently use machine learning systems to help organise and analyse large datasets.
- Process automation. An increasing number of businesses are using machine learning to automate their processes to help save time and reduce errors.
- Recommender systems. Businesses from Amazon to Netflix use machine learning to recommend products and content based on user interests, demographics, and so on.
- Time series forecasting. Organisations use machine learning analysis to understand customer and user patterns over time, or even to predict future trends using historical data.
Take machine learning to the next level
Gain key knowledge and skills in machine learning while studying the 100% online MSc Computer Science at the University of Sunderland. You’ll develop an understanding of the tools, trends, and current developments in the field of artificial intelligence, as well as its professional, ethical, social, and legal considerations.
This flexible master’s degree is taught online full-time or part-time and is suitable for individuals with or without a computer science background.