Posted by The Lad on Sunday, March 11, 2018 at 04:04:37A 3D systems for learning physics has been developed by researchers at the University of Manchester.
This new method can be used to teach 3D simulations of physical processes and to generate computer programs that are capable of reproducing these experiments.
The study of a 3D object is the key to understanding its properties, explains lead researcher Dr Jonathan Wilson, who also holds the chair of physics at the university.
He explains that it is the ability to understand a 3d object’s properties that allows you to understand the behaviour of its environment and to create computer programs to simulate its behaviour.
“It is really important to know the properties of a system, because then we can predict what will happen when it is interacting with its environment, and we can then build computer programs for those interactions,” he says.
The new method, which he calls a learning system, uses an existing technique for teaching 3D physics.
The method involves the use of a set of simple 3D objects, such as a cup of coffee, a bowl of soup, a table of chess pieces, and so on, and then a system of reinforcement learning, a technique that relies on reinforcement learning theory.
“The first thing you do is to learn how to recognise the properties that these 3D things have,” explains Wilson.
“So, for example, a cup is really a single object, but it’s a very complex object, and it’s got a lot of properties that are hard to predict.”
Then you can take the properties and make some algorithms to predict what the properties are of those objects.””
That’s the problem with 3D: it doesn’t tell you what the model is, and that’s the first thing that gets lost.””
But what if you go and create a model for the table and you see that the properties aren’t the same?
That’s the problem with 3D: it doesn’t tell you what the model is, and that’s the first thing that gets lost.”
Wilson explains that the learning system works by giving the learner a series of objects to learn about.
The objects are then randomly selected from a list of objects that the learter has been taught to recognise.
“When you get a particular object that the system has learnt to recognise, then the system makes a prediction about how it will behave in the future,” explains the professor.
“So, the system then makes a decision about what the future is going to be like.
So, for instance, if it’s learning to predict the behaviour when it sees a bowl that’s a little bit larger than the bowl it was taught to predict, then it will be a little more aggressive in predicting the behaviour that way.”
In this case, the learmer recognises that there’s a bowl on the floor that is a little bigger than the bowls it was trained to predict so that it has an opportunity to pick up on the future behaviour of the bowls.
“For this reason, it doesn ‘t like to predict anything that’s larger than it is, but when it does see something that is larger, it just doesn’t like to use the prediction system, and the behaviour is more aggressive.”
The learner then has to make a choice between making a prediction and learning to do the opposite, Wilson explains.
If it has learnt that the bowl is a bit bigger than it was predicted to be, then, in the first iteration, it will make a prediction that the object will behave like it predicted, but in the second iteration, if the learcher doesn’t pick up this prediction, it’s going to pick the bowl up and drop it in a bucket.
“What we do in this system is that we teach it to use this learning system to make the predictions, and by the time we’ve done that, it has learned that it’s not going to make that prediction,” says Wilson.
“Now, when it gets a bowl, we then use the system to predict that it will move to the bowl and drop the bowl in the bucket.
So you’ve trained it to predict when it’s picking up a bowl but not picking it up, and when it has picked up a cup, it makes a very aggressive prediction that it won’t make that mistake.”
This is a very powerful learning system.
It’s very powerful because it’s able to predict future behaviour very quickly, and to make predictions about the future that are very accurate.
“The team has now developed a system that can be easily integrated into existing 3D-based simulations and can be run in a variety of different environments.
It can be developed for use in educational settings, where students are taught to use computer games to explore the physical world, and for use on the real world, where 3D models are used to explore complex physical processes.”
We’ve used the system for a few different applications,” says Prof Wilson.