What is Machine Learning? Exploring its Impact and Role at Devant

In today’s digital age, machine learning (ML) has become a cornerstone of technological innovation. A key branch of artificial intelligence (AI), ML empowers computers to learn from data while adapting and making decisions or predictions without being explicitly programmed for every task. Rather than coding specific instructions for each scenario, ML algorithms are trained with data, enabling them to recognise patterns and make generalisations. This flexibility allows them to handle new, unseen data with remarkable accuracy.

In recent years, computer vision—powered by ML—has matured from an academic specialty to a practical tool, reshaping entire industries like healthcare, automotive and retail. From in-cabin monitoring systems and facial recognition to breakthroughs in medical imaging and retail automation, computer vision applications are revolutionising how we interact with technology. At the heart of these advancements lies machine learning which is uniquely positioned to help computers interpret and analyse visual data.

This year, the societal impact of ML was underscored by the Nobel Prize in Physics being awarded to Geoffrey Hinton and John Hopfield for their "foundational discoveries and inventions that enable machine learning with artificial neural networks." Hinton’s groundbreaking work and lectures on Convolutional Neural Networks are applied in our own ML models today.

At Devant, we recognize the vast potential of ML in driving business and technological advancement. In 2024, we significantly enhanced our in-house ML capabilities by welcoming Tomas Björklund as Head of Machine Learning. Tomas brings a wealth of experience from both academic and commercial sectors, equipping him with the insight and skills to expand Devant’s ML resources effectively. His work focuses on identifying and implementing cutting-edge tools and features that not only enhance Devant’s capabilities in ML training on our in-house produced synthetic data but also benchmarking it in order to deliver the most value to our clients.

Benchmarking is fundamental to Devant’s commitment to excellence in ML and computer vision. By regularly benchmarking results on our data, we ensure our image datasets are optimised for training, validation and testing, in order to deliver high performance in computer vision applications. This rigorous evaluation process ensures our solutions are reliable, robust, and prepared to meet the diverse needs of our clients.

ML is more than just technology at Devant; it’s a commitment to continuous improvement, innovation, and delivering powerful, data-driven solutions that make a tangible impact. As we continue to explore the possibilities of ML, our focus remains on using it as a tool for practical, transformative change across industries.

Our preliminary benchmarking efforts on standard ML tasks demonstrate that training exclusively on our synthetic data achieves results which are comparable to training on real-world image data. As shown in the example images below, the landmarks and segmentation outputs from a model trained entirely on 100% in-house synthetic data exhibit high accuracy and deliver excellent results when applied to real images.

Left: Prediction of 68 facial landmark locations in the regions described in the legend. A black dot indicates that the landmark itself is not visible in the picture.

Right: Pixel by pixel semantic segmentation of the same photo, into 13 color coded classes as indicated by the legend.

We will continue to advance our ML efforts by identifying, benchmarking, and implementing new features, further strengthening Devant's value proposition for both current and future customers.

If you want to learn more how Devant can support your business - contact us directly