Representational Bottlenecks in Deep Learning

Representational bottlenecks in deep learning

Introduction In a Sequential model, each successive representation layer is made on top of the previous one, which suggests it only has access to the knowledge contained within the activation of the previous layer. If one layer is just too small (for example, its features are too low-dimensional), then the model is going to be constrained by what proportion of information is often crammed into the activations of this layer. Description We can grasp this … Read more