I have a very simple question: Does all deeplearning algorithms start learning from the smallest possible piece of something? (For example single pixels in computer vision)
I’m asking because that is not us humans perceive things, we look at things more wholistic, or directly start looking at parts rather than the smallest possible piece.
In Deep learning algorithm, we always look at the holistic side but try to connect one holistic side to other through deep learning where you learn about neural network and other factors affecting algorithm. So if you see Even Machine learning and Deep learning algorithm is trying to learn what human is naturally versed in identifying things.
Regards
DP
Thanks for the reading piece, made me think further. What I am trying to say is yes there is light ray/molecule reflecting from a certain point at a certain angle coming to our eyes and hence triggering the process shown in link, then signalled into the brain going through an algorithm we don’t know. Hypothesising about this algorithm I am asking that are we really putting these singular inputs together like how we put in a matrix and calculate a resolution or do we instantly recognize say that its a cat we are looking at and then looking for parts of the object to verify our ouput.
I realise maybe as we are in further steps of our evolution process, and all these algorithms are already well engraved in our brains, I’m just assuming we are not concerned with such micro inputs.
Just an itch there is something missing nothing more…
I am not a neuroscientist, but to the best of my understanding there is a fusion of optical signal with other acquired knowledge (eg language) that results in a complex, multi-factor categorization of the visual field and the elements within it. It seems wholistic and instantaneous, but I don’t think it is either. Instead, it is a complex collection of small pieces of information that help you quickly reference and integrate the appearance and behavior of the real cat approaching you with all your previous experiences of the concept.
Neuroscientists have found clear evidence that the infra temporal cortex is indeed required for object recognition; they also found that subsets of this region are responsible for distinguishing different objects.
In addition to its hypothesized role in object recognition, the infra temporal cortex also contains “patches” of neurons that respond preferentially to faces. Beginning in the 1960s, neuroscientists discovered that damage to the infra temporal cortex could produce impairments in recognizing non-face objects, but it has been difficult to determine precisely how important the infra temporal cortex is for this task.
Despite its peripheral location, the retina or neural portion of the eye, is actually part of the central nervous system.
So if one understands this part, even human brain is passing through neutrons to generate an image classification based on inputs of visual cortex, which is the primary cortical region of the brain that receives, integrates, and processes visual information relayed from the retinas. It is in the occipital lobe of the primary cerebral cortex, which is in the most posterior region of the brain
once a sensory input is captured by the retina(eye), it travels through lateral geniculate nucleus in the thalamus and then reaches the visual cortex. The area or part of visual cortex which receives this sensory input is called primary visual cortex which is Brodmann area 17 which basically detects the intensity, shape, size and location of objects in visual field.
occipital visual areas such as Brodmann areas 18 and 19 provide input on spatial vectors to the middle temporal and medial superior temporal cortexes . These areas basically help in generating smooth conjugate pursuit eye movements. Brodmann areas 19 is part of the extrastriate visual cortex that surrounds the primary visual cortex, and which processes visual information.
So now you understand how each part of human brain helps in image recognition and as the deep learning tries to find or one should say make as close as possible algorithm to identify image recognition based on how neural network works.
When you do DLS specialisation, Andrew Ng mentions this part of what deep learning is trying to achieve.
Hope it helps you understand!!!
Keep Learning!!!
Regards
DP
