Neurons connected to each other in circuits can perform specific computations. Looking at these circuits gives us clues about what they are doing, and understanding these computations helps us understand why we see (or not see) the way that we do.
We introduce each neuron's "receptive field" as a way neuroscientists describe what patterns of photons that neuron responds to: what image makes a visual neuron most excited? Depending on the neuron and where it is in the visual pathway, its receptive field may be a donut, a vertical stripe, a face with two eyes and a mouth, or even a movie of upwards movement. The mathematical operation corresponding to these circuit computation is known as a "convolution," which directly inspired convolutional neural networks in computer vision and AI. Beyond vision, the concept of a receptive field applies to other sensory and motor modalities as well.
Finally, we motivate the sparse coding hypothesis, which postulates that visual circuits have such receptive fields because they are the most efficient way to represent natural images.
Playlist for all videos in series: [ Ссылка ]
Professor Bing Wen Brunton
www.bingbrunton.com
