Imagine a world where we don’t have to program our computers anymore, but where the computers program themselves; this is where we are heading towards with machine learning. But how does it work? And more importantly, how can we apply it to improve our lives? At MOBGEN Lab we have started to explore the possibilities by creating a boxing app that gives instant feedback on your performance by using a sensor embedded in the boxing glove.
By using sensors, we are able to learn new things about ourselves and the world around us that we can’t see with our bare eyes. But often just looking at data does not result in an epiphany; data can be messy, and interpreting it can be a challenge. Writing a computer program that interprets data for you, which can filter irregularities and recognise events, is tricky, and oftentimes consuming. At least, it used to be…
Machine learning offers a solution as it interprets the data for us. The way it works is that we set up a computer model in which we specify the input data, for example, acceleration data – and the output we expect in return, for example, the punch type. After that, we train the model by feeding it new data, and telling it what output we expect it to give. The model looks for similarities in the datasets, and writes a program that allows it to recognise the same events in new datasets. The model is then tested with new datasets to see how it performs. The more data is used to train it, the more accurate it will be.
For the boxing app, we use machine learning to recognise different types of punches such as jabs, hooks, and uppercuts to be precise. Knowing the type of punch gives us a good starting point to interpret not only the quality of the hits, but also to give advice on how to improve.
To demonstrate this function, we are creating an app that, in real time, visualises the impact you make while boxing with Bob, our office boxing dummy as seen in the image above. A small sensor (accelerometer and gyroscope) is embedded in the glove, and connects to the phone through Bluetooth. Within the app, the data is processed, classified, and translated into a visualisation as you can see in the image below.
Even though we are now applying machine learning to boxing skills, the possibilities are endless and definitely not limited to the sports domain. It could be applied to app within the medical world, or in therapy, think of interactive art installations, face and speech recognition, robots, and even fraud detection. Basically, any type of detection you can think of. With the growing pool of data that we collect through IoT devices, we are entering a world where we can predict people’s needs and desires based on how they behave; a bit like J.A.R.V.I.S. in Iron Man. The future is now: be aware and be inspired!