I recently read that Google has developed a new AI that anyone can upload data to and it will instantly generate models, i.e. an image recognition model based on that data.
Can someone explain to me in a detailed and intuitive manner how this AI works?
Google's AutoML is really a good idea in terms of autonomous model design. You can find the details in this blog. Let me explain briefly.
We, data scientists, design new networks by following existing models, trying and failing and trying again and again by analyzing weaknesses and strengths of the created models. However, we, as humans, have limited capabilities of designing/analyzing such networks. That's why Google created an AI which analyzes the strengths and weaknesses of each node while making a prediction. This AI analyzes each node and tries to improve the results by adding/removing/modifying connections of each node/layer. I guess AutoML AI takes state-of-art network as a base and starts modifying network according to your data to create a customized model.
While doing that, two technologies are being used: Transfer learning and Reinforcement learning.
Transfer learning is being used to start training from the most accurate point possible.
Reinforcement learning is being used to modify network to achieve a better success. This is the key part of this technology.
So, for users, it is more like upload your data, let AI modifies the network for you and give you a custom model which is specific to your data.
