It makes it very clear why MERA of a few layers would be useful for language models. Because in a sentence the important correlations are not necessarily nearest neighbor, but maybe 10th neighbor. It is still short range. A few layers of MERA gets you the potential to capture such finite length cells structure.
@Max-eo6vx4 жыл бұрын
Thank you, please do repeat the questions as they cannot be heard.
@faiyazhasan47973 жыл бұрын
Fantastic presentation!
@weituliao37124 жыл бұрын
very nice
@hubstrangers34504 жыл бұрын
computing = code, don't see that, slides, and theory have accumulated more than thousands years is it another book or accumulation infor??