Least squares and exploding variance in means of exponentials
Machine Learning Research Blog explores estimating or optimizing “log-sum-exp” functions, which it describes as useful for normalizing probabilistic models and smoothly approximating a maximum.
TLDR
Machine Learning Research Blog presents least squares as a way to tackle exploding variance in means of exponentials. It discusses log-sum-exp functions and their applications in data science, including normalizing probabilistic models. The author says the article also offers a spectral closed-form formula for approximate multinomial logistic regression.
Least squares and exploding variance in means of exponentials
Machine Learning Research Blog explores estimating or optimizing “log-sum-exp” functions, which it describes as useful for normalizing probabilistic models and smoothly approximating a maximum.