We had mentioned that the architecture of CNNs is motivated by the visual system of mammals. In this segment, we will discuss an influential paper named ‘Receptive field for single neurons in the cat’s striate cortex’ published by Hubel and Wiesel.
This was basically a bunch of experiments conducted to understand the visual system of a cat. In the experiments, spots of light (of various shapes and size) were made to fall on the retina of a cat and, using an appropriate mechanism, the response of the neurons in the cat’s retina was recorded. This provided a way to observe which types of spots make some particular neurons ‘fire’, how groups of neurons respond to spots of certain shapes, etc.
Let’s look at some of the statements made in this paper.
Some of the important observations made in the study were:
- Each neuron in the retina focuses on one part of the image and that part of the image is called the receptive field of that neuron.
- There are excitatory and inhibitory regions in the receptive field. The neurons only ‘fire’ when there is a contrast between the excitatory and the inhibitory regions. If we splash light over the excitatory and inhibitory regions together, because of no contrast between them, the neurons don’t ‘fire’ (respond). If we splash light just over the excitatory region, neurons respond because of the contrast.
The figure below shows a certain region of the receptive field of a cat. The excitatory region (denoted by the triangular marks) is at the centre surrounded by the inhibitory region marked by the crosses.
- The strength of the response is proportionalto the summation over only the excitatory region (not the inhibitory region). Later, you will study the pooling layer in CNNs which corresponds to this observation.
In the next segment, we will study some more observations from this study that influenced the CNN architecture.
Report an error