So I had a skim through the paper, and I'm a bit skeptical about it. First and foremost, the authors used R2 score for measuring performance. IMHO the metric is binary, either it is or isn't the one that "de-obfuscates" the transaction, therefore I would have used F1 or a similar harmonic mean based... (full message at <libera.ems.host/_matrix/media/v3/do…7503dcf6dfdf3a22c9c3454f59d9d52e119>)

* So I had a skim through the paper, and I'm a bit skeptical about it. First and foremost, the authors used R2 score for measuring performance. IMHO the metric is binary, either it is or isn't the one that "de-obfuscates" the transaction, therefore I would have used F1 or a similar harmonic mean... (full message at <libera.ems.host/_matrix/media/v3/do…c540b260800bb022b992423cde34732f96f>)

I see they have published another article: "Correlations of Multi-input Monero Transactions" also in 2020. I'll go through this one as well, since they are discussing denoising mechanisms. I'll get back to you on this one tomorrow.