If we look at any person from near we can also recognize them by getting some steps far from them, This means our visual system having seen any object or face of a person, recognize it despite of object position or scale, for example, we know that state-of-the-art classifiers, such as vanilla deep networks, will fail this simple test.
To recognize a particular face under a range of transformations, Neural Network needs to be trained with many examples of the face in different conditions. In other case if only one image is there, they cannot achieve invariance through memorization. Thus, understanding how human vision can pull off this remarkable feat is relevant for engineers aiming to improve their existing classifiers. In particular, it is possible that the invariance with one-shot learning exhibited by biological vision requires a rather different computational strategy than that of deep networks.
“Humans can learn from examples, unlike deep networks. There is a huge difference with vast implications for the engineering of vision systems and for understanding how human’s vision works”. The reason behind this difference is the relative invariance of the visual system to scale, shift and other transformation. Strangely, this has been very unconcern in the AI Community because the psychophysical data were so far less than clear-cut.
To differentiate invariance growing from built-in computation with that from experience and memorization, the new study measured the scope of invariance in exclusive learning. The exclusive learning was performed by introducing the Korean letter stimuli to human subjects who were unexplored with this language. These words were initially introduced a one time under one particular condition and checked at different scales and positions than the original conditions. This Structure is different from commonly used neural network models where an image is processed under specific resolution with the same shared filters.
Check our Best Articles: