nocturne protocol

arxiv.org/abs/2201.04018 @gg159136

ML models memorize training data, a major #privacy problem for data collaboration, even w/ #federatedlearning @gg159136 and @vectorinversion gave a fantastic preso at @opendp_org this morning on applying #differentialprivacy to #splitlearning to solve this More to come!

@lukefshaw datatable: > tic() ;x <- fread("stuff.csv") ;toc() 0.569 sec elapsed > gc() used (Mb) gc trigger (Mb) max used (Mb) Ncells 352321 18.9 654548 35.0 654548 35.0 Vcells 143416782 1094.2 174806664 1333.7 143591916 1095.6

I'm about to start analysing a 1.1GB, 20 million row, data set in #rstats. When dealing with data that size or bigger before I've never worked in R memory but am about to for this project - any tips? #tidyverse ("don't do it" is an acceptable response...)

@lukefshaw dataframe: > tic() ;x <- read.csv("stuff.csv") ;toc() 116.381 sec elapsed

@lukefshaw > system("ls -alh stuff.csv") -rw-r--r-- 1 phil phil 1.5G Dec 6 13:55 stuff.csv > system("wc -l stuff.csv") 21658045 stuff.csv

noaa-19 storm francis

nice noaa 19 pass

image-super-resolution with Artefact Cancelling GANS model: github.com/idealo/image-s…