How to use a webcam to make a viral video of yourself, according to John Hopkins Hopkins
The brain is very good at categorising images into categories, but what if it can recognise pictures that are just random?
That’s what a team of researchers at Johns Hopkins University in Maryland have shown by training a neural network to recognise the shapes of faces in a video of a random woman.
They’re calling their discovery the “happiest face” network.
The new study is published in the journal Science Advances.
It uses artificial intelligence techniques to train a new neural network, called the “happy face network”.
When the network is trained on images of random people, it recognises the happy face of the woman.
This is what the scientists call a “happy-face” network because the network recognises all of the faces that the computer can’t see.
This means the network can be used to generate funny videos, or videos that make you smile.
The researchers also found that the network was better at recognising faces that had a strong resemblance to human faces.
“Our study shows how well the network could discriminate between faces that look like us and those that are a bit more bizarre, such as a monkey or a rat,” said lead researcher Jonathan Withers, PhD, a post-doctoral fellow in neuroscience at Johns’ Hopkins School of Medicine.
“These results provide the first concrete evidence that the brain is capable of distinguishing between face features that are similar to ourselves, such that we can use it to make videos and share them online.”
Dr Witherspos first learnt about the network’s capabilities when he was working on a machine learning project at his university.
He was surprised that the neural network could recognise human faces and the faces of random strangers.
“What was surprising was how well it could recognise a monkey,” he said.
“You wouldn’t expect that the human brain would have that ability, but it is remarkable.”
Dr John Hopkins and his colleagues are currently studying the neural networks that were trained to recognize the happy-face network.
“The networks are based on a number of features that we’re very interested in, such image similarity, but we don’t have the ability to directly measure it,” Dr Wethers said.
To measure how well these networks could recognise faces that were different to themselves, the researchers also tested the network on images that were a little more human than the face of random individuals.
They then compared the networks’ performance against the faces in the videos.
“To our knowledge, this is the first time we have shown that the happy brain can distinguish between a face that looks like us versus a face which is a bit bizarre,” Dr John said.