4 ms·
In case anyone wants to check out some pre-history, back in 2002 Manfred Warmuth et al.[0] were using learning (Weighted Majority) to drive systems components l
by frankmcsherry 9y ago
In case anyone wants to check out some pre-history, back in 2002 Manfred Warmuth et al.[0] were using learning (Weighted Majority) to drive systems components like cache replacement policy. I'm not sure where the work went from there, but add it to the pile of techniques.
[0]: https://users.soe.ucsc.edu/~sbrandt/papers/NIPS02.pdf https://users.soe.ucsc.edu/~sbrandt/papers/NIPS02.pdf
- eternalban 9y agoThanks for the link. Very interesting. I found this [1] from 2015. Reading your cite, the practical issue seems to me to be that the optimizer's memory footprint costs may in fact negate any benefit (e.g. ~40% over LRU) obtained in reducing cache misses. My gut feeling is that this approach (for online systems) may work best with a hardware component (a card hosting the 'experts' and their virtual model e.g. the "virtual cache"). The distributed variant also seems worth exploring. [1]: https://arxiv.org/pdf/1403.0388.pdf https://arxiv.org/pdf/1403.0388.pdf