Paper Review: Parameter Prediction for Unseen Deep Architectures

Just think, what if we can train a model in one single epoch, wouldn’t this be great? So, for all the AI enthusiasts out there, I have great news for you. This recent paper from Facebook research tries to do that only. It’s been long known that as we are progressing in the field towards more and more complex tasks, it is becoming tougher to train networks due to the sheer time it takes to train these huge networks. This paper can be the paradigm shift in the training of neural networks. Why not use the previous information and knowledge to build…