Deep learning has made a large contribution to the field of artificial intelligence. But the concern is that it requires a large amount of data to operate. This is the issue on which the researcher and critics will be agreed on. However, all the researchers have minimized the reliance on deep learning.
A prominent Computer scientist, Yann LeCun, during AAAI conference, talked about the limitations of deep learning in artificial intelligence. Well, He gave a new technique of self-supervising to solve the problem of high-quality data.
As you remember, LeCun is a godfather of deep learning, and he invented CNN (Convolutional Neural Networks), which brought the upheaval in the past decade.
Self-supervised learning is a better edition of deep learning, which help to create a data-efficient intelligence network. Although, at this juncture, we cannot predict which technique will succeed in the future.
The data scientist, with the help of supervised learning, can perform nicely on complex tasks such as image samples. But these models will succeed when the only large database is obtainable. However, high-quality data is not easily available, which creates hurdles in the technique.
The self-supervised learning can figure out the problem through derive supervision signal from data itself without the interference of human. According to some well-known researchers, self provide learning has the ability to improve network robustness and minimize the cost of model training.
According to LeCun, the self- supervised learning is the future which helps our artificial intelligence system to move at the next level. However, He also informed us that it also provides a large amount of information developed by AI.
‘In self-supervised learning, we can get some knowledge about the world with a fewer database, unlike deep learning.’ says Luen.
You can also say that if artificial intelligence is a cake, then self-supervised learning is a bulk of cake. However, the upcoming revolution in AI will not be supervised and like before, but it is self-supervised.