Question

A 2013 paper by Ian Goodfellow et al. introduced a technique named analogously to dropout in which layers that perform this operation are added to a neural network. The LogSumExp, or LSE, function essentially performs a smooth version of this operation. This is the most common operation used for pooling in convolutional (15[1])neural networks. The rectifier function, which is by far the most popular activation function in neural networks today, performs this operation on (*) 0 and the input. (-5[1])The L infinity norm (10[1])of a vector is equal (10[1])to this operation applied to its components. Generalizing the logistic function to multiple dimensions produces a function called “soft [this operation]”. For 10 points, name the larger of the two values that are subtracted to find the range of a dataset. ■END■ (10[1])

Back to tossups