Understanding Visual Cortex: for Deep Learning
Working in deep learning, one would often encounter the term ‘visual cortex’. Especially, while studying ConvNets. We are not the biologists, so we need not to go deeper into this. Yet, it is very essential to know, as what we use today in AI, is inspired from nature.
According to a Stanford lecture I took online, in humans, almost 50% of neurons in our cortex is involved in visual processing. Let’s pick the word ‘cortex’ from this and learn a bit more about it. So, what she meant by this is ‘cerebral cortex’. This is one of the largest and essential region of cerebrum in brain.
It is the layer of grey matter in vertebrates that covers the cerebral hemisphere and is composed of folds of neurons and axons. It is responsible for higher level functions of the nervous system, such as muscle activity, learning, language and memory.
Visual cortex is the portion that is located on the back portion of the brain. It contains neurons connected to one another. So, this is the portion where our visual processing is done. In the past, researchers have observed that a cat’s visual cortex contains nucleus, whose job is to detect features like edges from the visual information. Higher level nucleus are tend to detect more complex features.
How it looks in ConvNets ?
In ConvNet, we try to achieve something similar. Our model first extracts features from an image and then classifies. Hence, it is important for us to know little about this biological perspective of deep learning.