Neural networks (IMAGE)
Caption
MIT researchers have found that neural networks can be designed so they minimize the probability of misclassifying a data input. They show that, 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.
Credit
Jose-Luis Olivares, MIT, with figures from iStock
Usage Restrictions
Credit must be given to the creator. No derivatives or adaptations of the work are permitted.
License
CC BY-ND