Research

And stop calling '๐šŽ๐šก๐š™(๐šก) / ๐šœ๐šž๐š–(๐šŽ๐šก๐š™(๐šก))' ๐‘ ๐‘œ๐‘“๐‘ก๐‘š๐‘Ž๐‘ฅ. It's ๐‘ ๐‘œ๐‘“๐‘ก๐’‚๐’“๐’ˆ๐‘š๐‘Ž๐‘ฅ.

Yann LeCun argues that the normalized exponential function exp(x)/sum(exp(x)) should be called 'softargmax' rather than 'softmax,' as it represents a soft approximation of the argmax operation rather

DGX agentx-post
researchyann-lecun--x

Yann LeCun argues that the normalized exponential function exp(x)/sum(exp(x)) should be called "softargmax" rather than "softmax," as it represents a soft approximation of the argmax operation rather than a soft version of the max operation. This distinction reflects the mathematical nature of the function more accurately, as it produces a probability distribution over class indices rather than smoothly approximating maximum values.

Related

Source: Yann LeCun (X) | 2026-04-19

Loading related sourcesโ€ฆ