Machine Learning

What is it & How Do We Use It?

There's still a lot of confusion around the topic of machine learning. In this guide, we explain everything you need to know.

As the name would suggest, machine learning (ML) is an area of technology that allows computer systems to develop their skills, as well as their understanding of certain situations. Because artificial intelligence is generally quite a complicated concept, and the technology used in machine learning is still relatively new, there’s a lot of confusion surrounding the topic. So, we’ve put together this introduction to machine learning, which contains all of the information you’ll need to get a better understanding of the technology.

Here, we’re going to explain exactly what machine learning is and how it works. We’ll also take you through some of the most common applications, and explain how we use the technology as part of our computer vision and machine learning service here at Luminous Group. Simply read on to find out more.

What is machine learning?

The term “machine learning” was coined back in 1959 by Arthur Samuel, an American pioneer in the field of computer gaming and artificial intelligence. It’s an area of technology that provides computer systems with the ability to “learn” using statistical information and techniques. As a result, computers are able to improve their performance on certain tasks and work out how to react to specific situations without having to be explicitly programmed.

How does machine learning work?

In machine learning, computers use algorithms and statistical models to determine how they should solve a problem or react to a particular situation. They’re able to build a mathematical model using sample data — also known as training data —which allows them to make predictions or decisions. The idea is that they don’t need to be explicitly programmed to perform the tasks they do. Instead, they’re able to learn along the way. There are two main types of machine learning algorithms: supervised and unsupervised. Semi-supervised and reinforced learning techniques are also used under certain circumstances. Here’s how these four kinds of machine learning work.

TRAINING DATA UNSUPERVISEDALGORITHM Comparing data to find a pattern. SEMI-SUPERVISEDALGORITHM Using a semi-supervised learning algorithm is able to considerably improve learning accuracy. REINFORCEMENTALGORITHM Reinforcement learning can work out how to handle data through trial and error. REGRESSION Regression tasks involve predicting a continuous quantity. EXAMPLE This kind of algorithm might be able to predict how tall a child will grow up to be based on their current height. CLASSIFICATION Tasks require a machine learning algorithm to place observations into categories. EXAMPLE A computer using this kind of algorithm might be able to tell whether a particular image contains a car or not, and then assign it the correct classification.

Supervised learning algorithms

Supervised machine learning algorithms are typically used for tasks that are relatively straightforward. In supervised learning, computer systems are shown “training” examples of correct input-output pairs, and they can then apply what they learn from this to new sets of data.

One of the most common uses for supervised learning algorithms is handwriting recognition. A computer will be shown various images of handwritten digits, along with their correct labels. It can, therefore, learn the necessary patterns it needs to recognise words and numbers that have been written by hand, regardless of any differences in style.

Supervised machine learning tasks can be further categorised into two different subgroups: regression and classification. Regression tasks involve predicting a continuous quantity. For example, this kind of algorithm might be able to predict how tall a child will grow up to be based on their current height, or whether a company’s customer base is likely to grow based on past data.

As the name suggests, classification tasks require a machine learning algorithm to place observations into categories. For example, a computer using this kind of algorithm might be able to tell whether a particular image contains a car or not, and then assign it the correct classification.

Reinforcement learning algorithms

Much like humans, algorithms that use reinforcement learning can work out how to handle data through trial and error. This means they develop their processes by determining what kind of actions yield the greatest rewards.

Reinforcement learning algorithms have a range of common applications and are used by the likes of computer-based board games and self-driving cars algorithm to place observations into categories. For example, a computer using this kind of algorithm might be able to tell whether a particular image contains a car or not, and then assign it the correct classification.

Unsupervised learning algorithms

Unsupervised learning algorithms are typically used when you don’t know exactly what you want to get out of the model. For example, you could have a batch of data and suspect that there are patterns to be found in it but mightn’t know exactly what those look like yet. In this situation, you can put your data into a format that makes it easy to compare, and then feed it through the computer’s algorithm, which will look for any correlations.

Unsupervised learning tasks are most commonly used to solve problems to do with clustering (i.e. grouping observations together). This is often used in marketing, where it’s helpful to group customers with similar buying habits together.

Semi-supervised learning algorithms

As the name suggest, semi-supervised learning algorithms fall somewhere between those that are supervised and unsupervised. This is because they use both labelled and unlabelled training date — typically a small amount of the former, and a large amount of the latter. As a result, computer systems that use a semi-supervised learning algorithm are able to considerably improve their learning accuracy.

Essentially, these computer systems are taught using the labelled data, and are then able to apply what they know to the information that hasn’t yet been labelled. Semi-supervised learning is used for a range of tasks, such as speech recognition and webpage classification.

Common machine learning applications

While many aspects of artificial intelligence seem so futuristic that you would be forgiven for thinking they only exist in sci-fi films, it’s likely that you come into contact with examples of machine learning on a daily basis. Whenever you check your e-mails, talk to your smart speaker, or shop online, you’re feeding information to algorithms that are designed to make your life easier. Here, we’re going to talk you through some of the most common examples we all benefit from.

Personal-Assistant

1. Virtual personal assistants

Do you have an Amazon Alexa or Google Home device? These smart speakers are designed to make your life easier by providing you with the answers to any questions you have. For example, you might enquire about the weather if you aren’t sure whether you’ll need an umbrella, or you could be curious about the age of an actor in the film you’re watching. They’re also able to process commands by setting alarms, putting important dates in your virtual diary, and playing songs of your choice.

However, what you mightn’t be aware of is that, every time you speak to your assistant, they use this information to refine the way they work. This means that, over time, they can learn how best to help you, and tailor the way they work to your preferences.

Video-Surveillance

2. Video surveillance

Many modern video surveillance systems are equipped with artificial intelligence, which means they can detect the possibility of crime before it even happens. This is because the technology is able to recognise unusual behaviour — for example, if someone’s standing motionless in a suspicious spot for a particularly long time.

These camera systems can then alert security guards or police who can step in and prevent anything from going further. If a problem is reported and it turns out to be a genuine threat to security, this can be logged, helping the system to improve its surveillance abilities.

Social-Media

3. Social media

There are a number of ways in which machine learning is used by social media companies. Firstly, platforms such as Instagram and Facebook now provide users with curated feeds, which are assembled based on what kinds of posts you ‘like’ and generally engage with the most.

You will also find that social media platforms often present you with lists of people you might want to follow or add as a friend. These recommendations are based on a whole host of different details, including who you’re already friends with or following, what kinds of content you seem to like, the area you live in, and even where you work.

And, finally, one of the most noticeable ways that social media platform use artificial intelligence is by using the information you provide them with to target you with ads. For example, if you engage with a lot of eco-friendly accounts, they might start to place adverts for sustainable products in your newsfeed. Or, if you follow a lot of fashion bloggers, it’s likely you’ll see a lot of promotions from high street clothing stores. Social media platforms’ algorithms can tell a lot about your interests through the posts you look at, like, and comment on, and they’re able to target your accordingly.

Email-Filtering

4. E-mail filtering

If you continuously ignore e-mails from a particular sender, you’ll find that they soon start disappearing into your junk folder. This is because your e-mail provider is able to learn from your behaviour and filter your mail with this in mind.

Not only can this be very helpful when it comes to keeping your inbox free of spam, but it can also help to increase your security. Many machine learning algorithms that deal with filtering emails are able to detect most fraud emails, as well as communications containing malware that could damage your computer. So, while you might be inconvenienced every once in a while when an important message is accidentally sent straight to your junk folder, machine learning is still incredibly valuable when it comes to filtering your online mail. And, the algorithm should take note and learn from its mistakes whenever you move an email back into your inbox.

Customer-Support

5. Online customer support

While shopping online, you’ll find that a lot of e-commerce sites now offer you the option of live chatting with a customer support representative. But, many of these companies don’t actually have workers on hand to respond to your queries — instead, some choose to use chatbots that can answer basic questions, send you the information you might be looking for, and point you in the direct of a human representative if they’re unable to help you properly.

Of course, thanks to machine learning, these chatbots learn how best to deal with certain situations as time goes on, which means they’re developing all of the time. So, if shoppers find that they have one bad experience with this kind of technology, it doesn’t mean they shouldn’t try again in the future. Due to its nature, it’s improving as time goes on.

Search-Engine

6. Refining of search engine results

Every time you submit a query using a search engine like Google or Bing, their algorithms will take note of how you respond to the results you’re given. For example, if you click on one of the first results and stay on that page for a while, they should infer that they’ve provided you with the correct information.

However, if you have to scroll past the first few results and still don’t find what you’re looking for, they’ll guess that the results you were presented with didn’t suit your search well enough. This is happening constantly and, as a result, search engines are always refining their results pages accordingly.

Product-Recommendation

7. Recommended products

When you view or buy products online, you’ll often find that you start to receive emails or online ads recommending items in a similar style or from the same brand. For example, you might place an online clothing order only to find that, when you revisit the site, they’re suddenly promoting an array of pieces that wouldn’t look out of place in your wardrobe. Or, you might purchase a camera, which results in the retailer advertising a variety of compatible accessories to you.

The shopping sites or apps that you use are equipped with machine learning algorithms that are able to keep track of your past purchases, as well as any items you’ve liked or added to your cart. This is how they know what you might be interested in, and what kinds of products they should recommend in the hopes of making another sale.

Face-Recognition

8. Image recognition

The latest image recognition technology is able to identify objects and even faces, but this is only possible thanks to machine learning algorithms.

Because image recognition has so many helpful and important real-world applications, it’s one of the areas in which machine learning is developing the quickest. For example, it can be used to analyse medical imagery, help autonomous cars to drive themselves safely, and allow security systems to identify criminals and people of interest.

You’ll often come across examples of image recognition when using the internet, too. For example, Facebook is able to learn what you look like from the images you publish or are tagged in. As a result, once the platform’s algorithm has enough information, it’s able to register and alert you when someone else uploads a picture that you’re featured in.

Fraud-Detection

9. Fraud detection

Machine learning can also be used by businesses that handle monetary transactions, in order to detect fraudulent activity quickly and effectively. For example, most banks will quietly monitor your spending and, if they recognise any activity that is out of the ordinary, you might be contacted, or your card could even be blocked automatically. Thanks to machine learning, this can all happen very quickly, protecting people’s finances.

PayPal is also known to use machine learning to detect money laundering. Its algorithms are able to look at and compare millions of transactions, distinguishing between those that are legitimate and illegitimate. The platform is then able to register if activity is potentially fraudulent, and this can be followed up.

This list certainly isn’t exhaustive, but it does cover some of the most common uses of machine learning, as well as ways in which you might come into contact with the technology on a daily basis.

Using AI to solve business problems

How are we using machine learning here at Luminous Group?

Here at Luminous Group, we use specific applications of machine learning in order to enhance our clients’ procedures. We primarily combine machine learning with computer vision to achieve advanced object recognition, and also use the technology for predictive maintenance.

This relatively new technology can massively benefit a range of industries, including the automotive, food processing, and manufacturing sectors.

MACHINE LEARNING input unstructured data sources, reports and files COMPUTERVISION Combine with real world intelligence and depth sensors ADVANCEDOBJECTRECOGNITION Identify, categories and sort PREDICTIVEMAINTENANCE gain actionable insight and analytics INDUSTRIES AUTOMOTIVE FOOD PROCESSING MANUFACTURING INDUSTRY 4 INDUSTRY 5

Machine learning for predictive maintenance

The majority of businesses rely on corrective maintenance, in that they will arrange for parts and equipment to be replaced once they fail. However, this can cost these companies in downtime, labour, and unscheduled maintenance requirements.

Some businesses practice preventative maintenance by predicting the useful lifespan of parts and equipment, so these can be maintained or replaced before they actually fail. But, while this can help to prevent unexpected or dangerous equipment failures, the high costs of downtime, under-utilisation of certain components, and labour are still an issue.

The predictive maintenance service that we provide using machine learning aims to find the right balance between corrective and preventative maintenance by enabling parts to be replaced or fixed at just the right time. This means components are only replaced once they’re very close to failure, extending their lifespans when compared to preventative maintenance, and reducing the cost of unscheduled maintenance and labour costs when compared to corrective maintenance.

Machine learning and computer vision

Here at Luminous Group, we also offer a machine learning and computer vision service, which allows us to build exciting new solutions for a range of industrial applications.

Using this technology, computer systems can be trained to look for specific objects and details. This can be helpful when it comes to identifying defects and improving quality control of products, recognising staff for security access, and even identifying whether workers are wearing the correct PPE for a specific job. We can also help to provide businesses with specialist equipment, such as drones that can identify cracks in pipelines, or determine whether wind turbine blades need any specific maintenance.

One key area that we’re particularly interested in researching is the use of computer vision and machine learning in the enhancement of 3D mapping. Our RIVO app allows for fast 3D data capture, but we’re currently developing algorithms that are able to search out and recognise 3D shapes, so items can be mapped out semi-autonomously.

Machine learning techniques are still relatively new, but new applications are constantly being found. And, due to the nature of this technology, it’s improving all of the time. Here at Luminous Group, we’re always looking for ways in which we can implement new and innovative methods to help our clients get the most out of their work, and machine learning is an area we’re very excited to be exploring more and more.

Clients

Contact Us

Drop us a line

If you need one of our 3D services or just want to pick our brains then please get in touch below.

Company