[2/2] Under FRA-debatten 2008 skrev FRA:s generaldirektör att FRA inte skulle läsa svenskars e-post och sms. Det vore enligt honom ”en vidrig tanke”. Han undrade vi kunde tro ”att en demokratiskt vald riksdag skulle vilja sitt folk så illa”. www.svd.se/a/f23bba52-9...#ChatControl#svpol#eupol
DEBATT. MISSFÖRSTÅDD LAGSTIFTNING. FRA ska signalspana för att kartlägga yttre hot mot landet. Ändå odlas uppfattningen att FRA skall lyssna på alla svenskars telefonsamtal, läsa deras e-post och sms....
Yes-pairwise is a significantly smaller calculation- one line versus a matrix! Interesting how it ‘holds’ up. A few small matrices randomly chosen to do column wise attention breaks the problem down. Moving pairwise means the problem is ‘embarrassingly parallelisable’ and 16 GPU can be used.
I guess the pairformer has significantly fewer parameters than the evoformer and even with AF2.3 inference is often done in a minute or so (per recycle), so I am not that surprised by the speed for the GPU/TPU part - might well be run on a single GPU (that is what the size-limitation indicates)
According to Martin Steinegger they are just using (n) hmmer and likely only one iteration. This could easily be parallelized to make it fast.
The good thing is that even a full professor does not make more than twice the salary of a PhD student. I think that should be the rule in the US as well. Either raise the salary of PhD students or lower the professors.
15% higher than the median salary for an assistant professor in Sweden.
It is a paper with many relevant measures..
Cool idea
Who uses excel to analyze their data. Remember all the autocorrected gene names..
Nej, det gör man inte.