Machine learning in lildog with CEML Niro

lilbit's Chief Embedded Machine Learning (CEML): – The more you use our product the smarter it gets

Countless films and TV series portray gloomy dystopian futures where artificial intelligence (AI) has caused the end of days for humans. There are no wonder moviemakers like to present such scenarios because, undoubtedly, AI has gained an incredible potential to cause radical changes for the better or worse.

"lilbit can't predict the future; however, we can affect it by using the latest technology on the market to change the future for the better."

By using machine learning, a subtype of the greater AI, we can make it easier for people to care for the ones they love.

Can observe changes invisible to the human eye

Niro is the CEML at lilbit, which means he is responsible for the company's machine learning team. He has been passionate about technology for as long as he can remember; however, it was the Sci-Fi thriller TV series «Person of Interest» that first caught his fascination for AI, and more particularly, Machine learning. The series plots a computer with the ability to predict crimes before they happen. The machine collects data from mobile phones, web cameras, and surveillance systems and then makes conclusions based on the gathered data to prevent crimes. Even though «Person of Interest» is a Sci-Fi series, it gave Niro a greater understanding of how great a potential AI has.

The most fascinating aspect for Niro was how they could teach algorithms to find information invisible to the human eye. Today he is convinced that we can use AI to help us take care of those around us.

But how is it possible– for example, how could lilbit's technology help humans take care of their dogs?

Translates dog behaviour

This article refers to one of lilbit's products: lildog. A digital dog sitter detecting typical signs of illnesses and injuries among dogs. The unit monitors the dog's behaviour through movements and living patterns and translates it into a language dog owners can understand. In other words, the product creates communication between the dog and owner in a way that has never been done before. Machine learning stands central in the accuracy of the product.

Niro explains:

- To begin with, the unit knows nothing. For example, it definitely doesn't know what a dog is. However, thanks to the gyroscope and accelerometer the unit can register the dogs' movements.

With this ability, the product is already of great value to a dog owner.

- But without any input from humans, the data collected doesn't mean much to the unit, Niro says.

This is where machine learning becomes relevant. We must tell the machine what the different movements mean. For example, when the dog sits, it makes almost the same movement every time. Therefore, we tell the machine that this type of movement means that the dog is sitting down– the machine then registers this– and from there on, it can recognise whenever the dog sits. The more times the dog sits, the easier it will become for the machine to recognise the movement.

By feeding lildog's algorithms information about what the different movements mean, the preciseness will increase each time it happens.

- The unit can learn to interpret the data we give it by itself. It will learn whether the dog sits, stands or runs. It can also automatically fit the data into different categories. For each time it registers a movement for something it isn't, we can simply correct the algorithm. The more data the algorithm collects from the unit, the cleverer it will get, Niro says.

And that is what we mean by machine learning. By providing data to the algorithm, it will become smarter on its own. As a result, lilbit can provide a product that continuously improves how it recognises early signs of illnesses and health issues through registering movement. If normal living patterns are breached, it can mean that your dog's health is changing.

Supervised vs unsupervised

Today lilbit's technology is mainly based on 'supervised learning', a type of machine learning where algorithms know what to observe and look for because we already told it.

- We already have endless information about dog behaviour. We know what to look for to recognise typical signs of illnesses and health issues. Because we have a definition, we can give the algorithms something to work with. It knows what to look for; it just needs to identify the movements and place them in the correct category. Each time the unit misplaces a movement, we correct it, which makes it smarter.

In the future, lilbit can also use unsupervised machine learning.

- Unsupervised machine learning means the machine doesn't know what to look for. Nor do we know what it should look for or what it will find. The only information we have is the data it collects. We will then look for patterns in the data we gather. In advance, it is impossible to know if the data will be valuable or useless, Niro explains.

By categorising and organising the data correctly, we could find something useful. If we find something, we will tell the algorithm to keep looking for the same information repeatedly.

- lilbit's products are unique because they can provide pet owners with information about health issues that they haven't yet detected in their pet. The earlier it is detected, the sooner they can treat the issue. Machine learning is essential to provide the owners and us with the information we need to understand our pets better.

Read more: Producing hardware for lilbit's products with Product Owner Erik