Why learning?

Original article was published by Larawehbe on Deep Learning on Medium


Meta Learning

Simple put, meta learning is a wide umbrella that covers a lot of interesting parts. For the sake of distinguishing between transfer learning and the last mentioned, we will briefly say that meta learning is the ability to learn, also known as ‘learning to learn’.

When we were kids and our math teacher gave us 1+1=2, it wasn’t only for the sake of ‘1+1=2’, yet, in fact, it was about learning how to the math behind it, in order to do math as we go and as we need.

Starting from basic principles, anyone can learn anything by knowing how to learn and how to study the case. For this reason, instead of having a large amount of data and pretrained model, we can simply let the machine learn by itself. Psychologically speaking, psychologists say that meta learning is the skill of 2020, the major reason behind success.

Learning to learn how to learn, what, when and why !

Conclusion

Learning will never get to an end, and each conclusion is a new beginning for a far horizon of thinking and innovation.

Do you have a lack of data ? Simply, use transfer learning for a field similar to yours to get benefit of pre-trained models ( make it rational, trucks and cars may pass, but cows and eye glasses wont ! ).

Do you have a lack of data and not enough pre-trained models ? Teach your model to outperform the available models with less effort! Think about the least model that can be close to yours, and teach your model the process of learning, the basics and the principles. Trust your model, it will surprise you!