• EN
  • DE
  • NL

Home / Our blog / 5 inspiring examples of Machine Learning

2018 is the year of Data Science, Machine Learning and Artificial Intelligence. Three terms almost everybody is familiar with nowadays – we’ve also written about it earlier. But what about actual (real) (live) use cases? We’ve listed some of the most interesting examples we built and found online.

Identifying important signs

“AI beats doctors at visual diagnosis lung cancer”, this promising title stood out from the crowd. Scientists trained a model with over 2000 slides exhibiting a kind of cancer to identify the different traits that contribute to a diagnosis. While humans only identify a few hundred signifiers, the model identified 10.000. Not that strange, since Machine Learning models are far better in recognizing patterns than humans, it shows promising possibilities for the future of our health care though.

Source: Extremetech.com (Click for the original article)

Reading lips

Are you able to read lips? Lipreading is quite complicated and equally challenging to train an AI to do it. Imagine the possibilities however for automatic subtitling or supporting the visually and hearing impaired. Students at Oxford University succeeded in training a model that reads the correct words with an accuracy of 93.4%: Lipnet. (Humans could only read lips successfully in 52,3% of the cases). However, this model was tested by volunteers who spoke specific sentences facing the camera. Another experiment of both Google and BBC succeed in lipreading in more natural situations. Scientists trained a model with more than 5000 hours of BBC material. This generated 17 500 unique words, which the AI can predict with an accuracy of 46,8%. (Lipnet only had 51 unique words). Impressive though!

Farm work

As farmer the amount and quality of your milk depends on the health of your cows. Ida, an application created with Tensorflow, supports farmers by tracking the cow activity. It’s a wearable sensor that detects for example when a cow is eating or resting and what her temperature is. By tracking this, disorders or signs of lameness are diagnosed far earlier and farmers can be helped with recommendations. Since acting can be done sooner, the efficiency and health of their farms and cows improves.

When there's smoke, there’s Machine Learning

What about using image processing as modern fire alarm? Normal smoke detectors are made to be used indoor. When you work at for example an outside storage area, it’s difficult to detect fire in a early stage. Our own Fire and Smoke Application analyses CCTV footage and can determine the amount of fire and smoke in the images. This way fire in outdoor areas can be recognised quite quickly as well. Even when there are other elements that provide heat.

Click for the original case

Predicting Crimes

Is it possible to predict the next burglary? The Dutch police force uses a predictive police system to do exactly that. The system has been filled with a lot of data of past criminal records and reports and combines these with common neighbourhood statistics such as age, gender and household composition. By combining this amount of information it predicts where there is a higher chance on something to happen. This way the police can be more efficient in crime prevention, by for example issuing extra surveillance's or campaigns for better locks in a specific area…

Source: Erik van Gameren, NRC (Click for the original article, dutch-only) Translation image: The coloured circles are crimes. The red blocks the predicted spot)

Try it yourself

Are you inspired? We didn’t even mention self driving cars, recognising emotion or speech analysis. The good news is, it isn’t difficult to try ML yourself. Just teach the model what it has to do when you do a certain movement. And you can create your own model with your webcam.

A bit more advanced, but still very simple is the new Cloud AutoML. It’s just for image processing (more to come!), but if you feed your pictures to this program, you can teach it to for example recognise cats, shapes of clouds or whatever you want. The only thing you need is a bunch of pictures and different categories ( and some pictures to test your model with of course!) You don’t need any code to organise your cat pictures automatically.


Similar Stories