BOFFINS FROM NVIDIA have created a way to train artificial intelligence (AI) systems and robots for industrial use in the same fashion humans learn.
Team Green's deep learning AI tech aims to learn by watching what a human does and then tries to replicate it, in much the same way as one human will learn how to do things from another more experienced person.
"For robots to perform useful tasks in real-world settings, it must be easy to communicate the task to the robot; this includes both the desired result and any hints as to the best means to achieve that result," the researchers stated. "With demonstrations, a user can communicate a task to the robot and provide clues as to how to best perform the task."
Using Nvidia's Titan X GPUs, the deep learning system carries out duties relating to perception and program generation and execution. But taking this approach is an learn how do complete a task from watching a single human demonstration.
Once the demonstration is done, the AI-powered robot creates a description, which is readable by humans, to detail the steps it'll need to take to complete the task it witnessed, thereby providing human overseers with an insight into what the system has taken from its observations.
The idea behind the system is that it'll learn through synthetic data - information it has essentially created itself. Normally, deep learning neural networks require vast amounts of labeled data to be thrown at them until they figure out patterns in the information in order to complete certain tasks.
But Nvidia's smart folks reckon using synthetic data is less intensive and training data can be produced with very little effort.
Furthermore, because the system takes into account the state of the real environment it's working in, it's able to recover from any mistakes it makes in real-time, rather than needing to process a heap of extra data to work out where it went wrong.
While Nvidia's AI advancement may be unique to it, it's not alone in trying to create smart systems that learn like humans; Facebook is doing the same with tech that mimics human learning as best it can, while Intel's Loihi AI chip also wants to make smart machines think like humans µ.
The week in Google in brief
Sega hedgehogging its bets
And not a purple duck in sight