Machine learning explained through OCCO’s digital platform

OCCO’s Head of Machine Learning Edward Erelt explained how computers can see and why even manual labourers should not be afraid of the development of artificial intelligence and technology.

Machine learning and computer vision are two keywords that are increasingly becoming more and more talked about and are conquering even the most unexpected sectors. Machine learning plays an important role in the work of the DesignTech start-up OCCO, for example. OCCO’s mission is to simplify and speed up the work of architects by automating the process of finding design and financial alternatives to interior design elements. The company’s Head of Machine Learning explained what are the future promises in this area and how computer vision is important to OCCO.

With regard to the future, what are the promises of machine learning – what is its potential?

If we talk about machine learning in general, not just computer vision, then I see machine learning as a technology that can make a person’s life much better. Many jobs today are very monotonous – people do the same thing all day, whether it’s sending emails or making some simple decisions. If you collect enough data and train an algorithm to make decisions on the basis of that data, you can automate the process.

The most interesting part for me personally at the moment is to combine different types of data, such as text and images. In other words, we tell the computer in words to change the image in some way, for example, add make-up to a photo of me and it will, just as I tell it to. Or when it comes to the world of design, for example, it’s exciting to combine CAD (computer-aided design) and images so that we can make different CAD models using machines.

Are there any dangers?

There has been a lot of talk about all the dangers of artificial intelligence and how it will replace humans. Personally, I think this is a phenomenon that accompanies every technology. When the first steam engines arrived, people also said that they would replace people soon, but in fact, people’s lives have improved steadily thanks to automation and technology.

Even people who do simple jobs today should not be afraid of machine learning, as this will allow them to automate their work, leaving them a more interesting and comprehensive role, as they will be able to deal with situations that artificial intelligence cannot solve.

Of course, another, more sci-fi threat may be that computers will become even smarter than humans at some point, but it will probably take hundreds of years. Unfortunately or fortunately, the human brain is extremely complex and smart compared to the computers that we currently have.

What are the biggest challenges in machine learning?

Currently, the main challenges in machine learning are either a lack of data or problems dealing with unsuitable data. Companies often have a lot of data, but whether that data is untagged, that is, it is just raw data and can’t really train artificial intelligence, or it looks good at first glance, but when you start using it, some problems will emerge.

Let me give you an example: we had a lot of images with a nice product on a white background, and we trained a very good categorizer based on that – our product could provide categories very easily, very well, very accurately. But when we wanted to use the same model to show people recommendations and help them find new products based on what they could add to the product (OCCO’s selection), it turned out that people were sending (to OCCO’s search) completely different pictures.

They sent images where there is little light, where the product is only partially visible, where the product is at the wrong angle – we did not have images at that angle in our database. When it came to light, we collected new data, trained a new model, and solved the problem, so now people can take whatever picture and search any way.

How many of these categories and products does a computer (for OCCO) have to chew through, so to speak, and how long does it take?

Right now, we actually have a little over 100 product categories. They are being added every day, so I don’t know the exact number at the moment. When it comes to how long it takes for the computer to review all of these products, it’s actually essentially none. While a person can take hours or days to review these products, a computer can easily review hundreds of thousands or even millions of images in a matter of twenty milliseconds and decide what to display to the person and what not.

How common is such a machine learning system and how much is it used for example in the field of design?

There are some other companies in the field of design, but to our knowledge, no one is doing exactly the same thing as us. At the same time, machine learning is a growing field in the world of design. And this is possible due to the fact that more and more data is being generated and it is increasingly better.

Design requires a lot of data compared to another avenues. This is such a wide field that in order to make any decisions, you need to see essentially the whole field, millions of pictures. But thanks to social media, for example, it is possible to find billions of pictures of furniture items.

What is the next big step for OCCO that you are currently working on?

Expanding the business is certainly a key step for OCCO at the moment. We are expanding very steadily into different countries across Europe. We are increasing our turnover very fast, and our team is also growing, so we need a lot of new people.

OCCO’s goal in general is to become a household name that interior designers mention on a daily basis in their offices, that is, what they talk about and what they praise – that’s our main goal.

What is your role in the company as a machine learning expert?

As a Head of Machine Learning I am responsible for the development of machine learning. This means that I communicate with both the product representatives and the end customer and decide which functionalities should be included. Also, I talk to other developers and decide how the new functionalities would fit into the product elements they are developing.

Once the initial functionality is in place, I’ll develop it. Of course, we also need to collect data for this – in some cases we already have the data, other times we have to go out and start collecting the data ourselves.

Once the data has been collected, I train the neural network or its mass learning algorithm and develop a service based on it. The service then goes up in the cloud and other developers add their own software, after which it reaches the user. After the first users, we get some kind of feedback, that is, if something needs to be changed – whether more data needs to be collected, it needs to be made more functional, it needs to be made more user-friendly or more understandable to people. And then the service starts to get better and better.

The interview was conducted at the Digit IT conference in Tartu in April 2022 by foundme.io. Author: Johanna Adojaan