Dmitriy

I got an Email that said it will be Dec 17th

Has a date been set for the 2022 event? Wondering if I'll be in the bay day-of or traveling for Christmas

I tried to sketch a toy problem with tune-able “factorizability”

Draw n samples each from two equivariant normal distributions with different means *a* and *b*. The task is to estimate the median of the 2n combined samples.

If *a* << *b* it factorizes completely - the network just needs to estimate max(A) and min(B) and in the last step output their midpoint. As we move *a* and *b *closer together more and more comparisons across A and B are important for accuracy, so the problem is partially factorizable. When *a* = *b* the problem doesn’t (naively) factorize at all.

(I assume small enough neural networks can’t find the fancy recursive solution*, so the circuit complexity will behave like the number of comparisons in the naive algorithm.)

*https://en.wikipedia.org/wiki/Median_of_medians

Some forms of biased recall are Bayesian. This is because "recall" is actually a process of reconstruction from noisy data, so naturally priors play a role.

Here's a fun experiment showing how people's priors on fruit size (pineapples > apples > raspberries ...) influenced their recollection of synthetic images where the sizes were manipulated: A Bayesian Account of Reconstructive Memory

I think this framework captures about half of the examples of biased recall mentioned in the Wikipedia article.

I read his thesis as