A method for designing neural networks optimally suited for certain tasks
MIT researchers find neural networks can be designed so they minimize the probability of misclassifying data input. To create a neural network that can achieve optimal performance on any dataset, one must use a specific building block, known as an activation function, in the network’s architecture.
United States Caroline Uhler Mikhail Belkin Adityanarayanan Radhakrishnan Department Of Electrical Engineering Proceedings Of The National Academy Sciences
Source: mit.edu