 I'm just going to do a quick demo of just one of the machine learning models that we came up with in our Ready for Research Weeks workshop. So the task we set was to use these little Raspberry Pi that have quite restricted capabilities with their camera to automatically detect something. Students could decide what that was going to be and we were going to use retraining techniques rather than training from scratch and we were going to use run detection code on the Raspberry Pi and Python and just from a research point of view what we were researching is how accurate are these retrained models, how specific to the training circumstances are they because they're built from very general models but as we see they end up being very specific models and ultimately we want to know whether these can be used in practice and relied upon. So the students pick their own tasks and I'll make some videos for the other ones. This video is demonstrating the car versus person detector which we set up outside the office so I'll get this running and ultimately this is all we're going to show off the students work so you can see I'm panning down they use the scroll hat there to display whether there was a person or a car in the frame and there's a camera hanging down there and because it's a busy time of day on campus there's always a person but in fact if there's no one in the frame it'll show you an X to show you there's no one in the frame and this one ended up being really really accurate now this is the location that it was trained at but it was only trained with three or four people it was only trained with a dozen or so positive images and a dozen or so negative images and it manages to pick every time a person walks past and every time a car goes past so you can see me struggling with the focus on my camera but eventually a car will go past and we'll see it detected. That's quite a fun demo but the most important thing from my point of view is that it's shown that these devices the power in these devices is enough to run custom detectors that are detecting things straight from the camera there we go the car has been detected taking things straight from the camera reliably so there's all kinds of used cases for that we got much more research to do in terms of how reliable are not circumstances are they reliable there's a million other variables here but this is a really interesting starting point and to show that this is not something magical that we're looking at here here's the same model running in the office or the workspace we were using so this is our hate of lovelace room where we were using came down after the workshop was finished and took a recording and it thinks that the room is a car and that's because this model was trained for cars people are nothing and it was trained outside and it has to make a guess about what it's looking at but I really enjoyed doing the workshop and I really enjoyed watching people come to understand much more about machine learning and the capabilities in machine learning are exciting but the restrictions on what it's able to do are substantial and the students discovered all of that and create a great demo