Brains that don’t see in greyscale first over-rely on colors: Project Prakash study

SUB: Science and tech

SEC: Human health

Context:

In May, a team of Indian and U.S. researchers reported in the journal Science that this delay in developing color vision is actually important for overall vision development.

More on news:

Project Prakash treats and rehabilitates blind children in India. These children helped the researchers shed light on how the brain learns to see.

Importance of color vision:

Humans don’t need color vision to recognise objects but colors can provide adaptation and survival advantages.
Children often described objects around them with their color.
Their reliance on colors is a little more than what normal children have.
This observation gave the researchers an idea about how to show them some things without color.
The children could recognise color images and discs quite well — even those who were barely two days out of eye surgery. But they had a tough time recognising black and white images.
Children without any visual impairment had trouble neither with color nor grayscale images, on the other hand.

Mimicking visual development:

Normally, a child first understands the world in grayscale.
The first time the children at Project Prakash experienced normal vision, their eyes had developed enough to see colors as well, so they skipped the grayscale phase.
Their brain processed black and white images differently as a result.
To understand the effects of this issue, the researchers needed a proxy to the brain that they could tweak to learn in response to different visual stimuli.
They set up a deep convolutional neural network (CNN) — a computer program that processes information the way neurons in the brain’s visual cortex do.
They trained four CNNs, one each on color and grayscale images in a particular order:
- grey-grey,
- color-color.
- color-grey.
- grey-color.
They found the grey-CNN recognised both greyscale and color images better than any of the other models.
The color-color model, which most mimicked visual development among Project Prakash’s children — fared worse at identifying greyscale images.
The researchers attributed this to the color-color model’s overreliance on color cues when examining images because its training data was composed solely of color images.
The grey-color model had learnt enough cues from the greyscale images and was thus better able to recognise color images.

Optimizing visual development:

It’s fascinating that the brain develops object recognition and color perception at different times.
For example, children could also be made to experience a room deprived of color, simulating a black and white or a greyscale environment, for a few hours at a time.

What is a convolutional neural network?

A convolutional neural network is a regularized type of feed-forward neural network that learns features by itself via filter optimization.
Vanishing gradients and exploding gradients, seen during backpropagation in earlier neural networks, are prevented by using regularized weights over fewer connections.