Matt Krause

# 1,540 Actions

 Jul 11 revised Are there mathematical reasons for convolution in neural networks beyond expediency? added 31 characters in body Jun 26 awarded Nice Answer Jun 15 answered How to handle timeseries extremes (sigma > 20) in deep learning? Jun 15 awarded Benefactor Jun 15 accepted Using indistinguishable subjects as predictors/random effects Jun 14 reviewed Leave Open How to apply VAR model for a I(1) and other I(0) variable? Objective is to forecast Jun 13 answered Using indistinguishable subjects as predictors/random effects Jun 10 revised Using indistinguishable subjects as predictors/random effects edited title Jun 7 awarded Promoter Jun 4 comment Why do we take the square root of variance to create standard deviation? We have that already, in the form of the mean (or median) absolute deviation, L1 norms, and the like. However, the major advantage of the traditional approach is that, unlike absolute values, it's differentiable, which allows you to analytically minimize and maximize things. Jun 3 awarded Curious Jun 3 revised Using indistinguishable subjects as predictors/random effects Added more about indistinguishability. Jun 3 comment Using indistinguishable subjects as predictors/random effects Right, but what about a pair of examples where one's id_i equals the other's id_j? In other words, is correlated with , even though neither the first ids nor the second match. Jun 3 comment Using indistinguishable subjects as predictors/random effects It's certainly possible that I'm overthinking this and your solution is perfect, but ideally I'd like a slightly more rigorous argument than "well, it runs". (Not that I don't appreciate the effort!) Jun 3 comment Using indistinguishable subjects as predictors/random effects This will definitely run, and I'm reasonably convinced it's the right solution for distinguishable subjects (e.g., $i$ indexes over buyers, $j$ over sellers). I'm worried that it's not quite right for indistinguishable subjects though: the observations where $i=1,j=2$, for example, are not independent of $i=2,j=3$, since both involve subject 2. In other words, your model has two effects of subjects (id_i, id_j), but I'm shooting for something where there's just one (I think). Jun 1 asked Using indistinguishable subjects as predictors/random effects Apr 30 answered Are there mathematical reasons for convolution in neural networks beyond expediency? Apr 30 answered Why is there a change in the number of degrees of freedom when the following modification is made? Apr 20 awarded Enlightened Apr 20 awarded Nice Answer