3 ms·
Interesting. Most ML models presumably operate over IEEE double precision floating point values. These have a couple interesting properties like INF and NAN, bu
by flatline 6y ago
Interesting. Most ML models presumably operate over IEEE double precision floating point values. These have a couple interesting properties like INF and NAN, but not infinitesimals.