-
dartian[m]<Rucknium[m]> "dⱮartian: Here is the current..." <- thank you! I missed this
-
dartian[m]<Rucknium[m]> "It's not obvious that machine..." <- to clarify, I absolutely would prefer to avoid ml entirely and I am very interested in what other statistical methods work. To me, it's just a bit baffling to consider a predictable algorithm that can't be discernable from human behavior.
-
dartian[m]a monero problem indeed. I've got some reading up to do.
-
dartian[m]<ghostway[m]> "I don't feel right using ml..." <- outputs could be tracing of outputs. of course, it would be wrong a lot of the time... but could be refined heavily with off-chain data. basically "sanity checking" leads until plausible, and likely options appear.
-
dartian[m]or it could produce unusable outputs because the decoy algorithms are effective against it
-
dartian[m]a major challenge for researchers trying to use ml for monero should be that they don't have datasets of real outputs to analyze. they don't know. the only data they can get is from those willing to share it, which could only help them analyze those kind of users
-
xmrack[m]dⱮartian: I researched this problem over the last year and collected a dataset suitable for ML/DL applications. You can find the download links here: github.com/ACK-J/Monero-Dataset-Pipeline
-
one-horse-wagon[Seraphis/Jametis protocol meeting in 7 hours 15 minutes--1800 UTC--in Matrix room-No Wallet Left Behind. Everyone is cordially invited.
-
dartian[m]<xmrack[m]> "dⱮartian: I researched this..." <- thank you! heh... you people are always ahead of me, and I love it. I'm going to be making a website that will aim to make things like this more widespread.
-
tempest[m]1@[one-horse-wagon] can you link that room?
-
ofrnxmr[m]#no-wallet-left-behind:monero.social
-
tempest[m]1Thanks!
-
sgp[m]UkoeHB: do you have rough comparison numbers for tx generation and verification relative times for current Monero vs. seraphis?
-
sgp[m]say, seraphis 128 ringsize
-
UkoeHBsgp[m]: can look here monero-project/research-lab #91#issuecomment-1047191259
-
UkoeHBcurrent protocol should match the CLSAG case in those tests pretty close
-
UkoeHByou want to look at the ref set size vs verification cost batch size: 25 chart
-
sgp[m]anything on generation time?
-
UkoeHBno
-
UkoeHBit's probably 2x slower if I had to guess
-
sgp[m]okay thanks