Back to blog

How to use machine learning for our safety?

Any technology user today has benefitted from machine learning

July 19, 2022

8 minutes read

What is machine learning? According to Arthur Samuel, who popularized the term, ML-based software enables computers to learn from data improving themselves without being explicitly programmed. Machine learning (ML) is a category of algorithms that allows software applications to receive input data and statisticaly analyse it to update outputs each time new data becomes available. ML is a subfield of artificial intelligence (AI). Any technology user today has benefitted from machine learning. 

Machine learning is used anywhere from offering intelligent insights to automating monotonous daily tasks. Different industries want to use it for their benefit. You may already be using a device that utilizes it. For example, a wearable fitness tracker or an intelligent home assistant. But there are much more examples of usage ML in prediction systems, image recognition, speech recognition, medical diagnoses, financial industry, and others. 

Related: ML in medical imaging: feature extraction in diagnostics

Facial recognition technology allows social media platforms to help users tag and share photos of friends. Optical character recognition technology converts scanned images of handwritten, typed, or printed text into machine-encoded text. ML-based recommendation algorithms help choose television shows or movies to watch using individual preferences. ML-based programs in self-driving cars may soon be a reality. Machine learning is continuously developing, but some considerations have to be kept in mind while working with ML or analyzing the impact of machine learning processes.

A short history of machine learning

ENIAC (Electronic Numerical Integrator and Computer) was the first manually operated computer system invented in the 1940s. The idea was to emulate human thinking and learn with the computer’s help. In the 1950s, the first computer game helped check players improve their skills by beating them. At that time, it was a real breakthrough. 

Machine learning became very famous in the 1990s with data-driven approaches. Scientists started to use large-scale data while building intelligent systems to analyze and learn from large amounts of data. As a highlight, IBM’s Deep Blue design beat the world champion of chess, the grand-master Garry Kasparov.

ML in practical use

ML classifies tasks into different categories based on how feedback on the learning is given to the system developed or how learning is received. The most common ML methods are the following: supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Of course, there are other approaches, and sometimes more than one is used by the same machine learning system. For example topic modeling, dimensionality reduction, meta-learning, deep learning, and others. Let’s explore some of these methods in more detail.

Supervised learning

Supervised learning uses data that humans label for training algorithms based on example, input, and output. Labeled data means a set of training examples, where each model is a pair consisting of information and the desired output value. The algorithm that uses supervised learning can “learn” by comparing the “taught” outputs with its actual work to look for errors and improve the model. For recognizing label values on additional unlabeled data, supervised learning uses patterns.

For instance, imagine that an algorithm uses supervised learning and is fed with the following data: oceans images are labeled as water, and sharks images are labeled as fish. The supervised learning algorithm trained on this data can later identify unlabeled ocean images as water and unlabeled shark images as fish.

Supervised learning is mostly used to predict statistically likely future events by analyzing historical data. For example, filter out spam emails or anticipate upcoming fluctuations in the stock market based on analysis of historical data. Supervised learning can use tagged dog photos as input data to classify untagged photos of dogs.

Unsupervised learning

Unsupervised learning provides algorithms with no labeled data to find structure within the input data. ML methods that facilitate unsupervised learning are precious because unlabeled data are more abundant than labeled data.

Unsupervised learning can discover hidden patterns within a dataset, but it may also have a feature learning goal, which automatically allows classifying information.

Unsupervised learning algorithms are often used to classify transactional data. For example, in a large dataset containing customer purchases, a human eye will not find similarities between different customer profiles and their most common assets. An unsupervised learning algorithm can use this data to determine a specific age range of women who often buy unscented soaps, which can mean that they are pregnant. Therefore, baby and pregnancy products can be offered to this audience in the market campaign to increase the number of appropriate purchases.

Not knowing a “correct” answer, unsupervised learning algorithms can analyze complex data and help organize it in a meaningful way. Unsupervised learning methods are very effective for the detection of anomalies. For example, fraudulent credit card purchases. Recommender systems use such algorithms to advise products to be bought next.

If skilled human experts have to label data, it raises costs. You can cut costs by using the model building semi-supervised algorithms without labels. These methods exploit the idea that even though the group memberships of the unlabeled data are unknown, this data carries essential information about the group parameters.

Reinforcement learning 

Reinforcement learning allows machines and software agents to continuously learn from the environment and automatically determine the ideal behavior within a specific context to maximize its performance. A computer program can drive a car or play a game against an opponent. As it navigates its problem space, the program provides feedback analogous to rewards, which it tries to maximize. Reinforcement learning doesn’t need labeled data set.

Topic modeling

Topic modeling can discover the abstract “topics” in the documents. This statistical model is frequently used as a text-mining tool to find hidden semantic structures in a text. In each text about a specific topic, some words appear more frequently than others: “bone” and “dog” will appear more often in texts about dogs, “meow” and “cat” will appear in texts about cats, and “is” or “the” may appear equally in both texts.

Each text concerns different topics. For example, a text that is 90% about dogs and 10% about cats would probably consist of about nine times fewer cat words than dog words. Topic modeling techniques produce “topics” organized in clusters of similar words by capturing them in a mathematical framework, which allows examining a set of documents and discovering, based on the statistics of the words in each, what the topics might be and what each document’s balance of topics iss.

How to use ML for our safety?

Face recognition tools can not only be used on social media for tagging photos. They are often developed to protect people. For instance, the ML-based software can analyze video streams from CCTV cameras and control the proper usage of personal protective equipment in the workplace or other institutions. 

The algorithm cheques workers on the accuracy of wearing protective items (coat, glasses, glove, mask, helmet). When somebody doesn’t wear the PPE, the system can identify it and send an automatic notification to a safety engineer. The system analyses the CCTV stream and saves pictures documenting safety rules violations. After that, you can review saved photographs and see the time and place noted on the image. It improves the workplace’s safety and decreases the injury rate. The solution can be used for different locations such as building sites, fabrication lines, steel, oil & gas enterprises, and other industrial environments where safety rules should be strictly followed. 

Why do we need an automated solution for PPE monitoring? Of course, all this can also be performed by a safety engineer manually, but it takes a lot of time and effort. Additionally, human observation is less stable and accurate because people can not focus their attention for a long time. They get tired and can be injured easier. When monitoring is not good enough, it can often provide injuries.

Monitoring in real-time allows researchers to each case and to see all details to avoid misinterpretation in case litigation or court. Real-time monitoring and automated alerts save time for controllers as well.

More articles

prev
next

More articles

Leave us a message

You’re in a good company: