Aubin Ramon
speaker
346 appearances
1 recordings
1 series
first heard Jul 2026
last heard 6 Jul
Aubin Ramon’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Jul 2026 with 1.
Appearances
And this is what people need need to do when they test uh any machine model.
Um and we then use that Heldas test set which is clean and meant meant to be representative.
And then we measure sperman uh correlation.
So this is important for ranking.
So it's gonna correlate with ranking.
Uh we're gonna see if the NOMA uh keeps the same ranking order.
And
Um, it's important for its application because at the end of the day you wanna select, you wanna optimize, so you wanna see you you're okay if the absolute value is a bit different from the ground truth, so pears and correlation, but if the ranking is the same, this is what you wanna do.
um at the end uh of the project.
So we had the sperm correction of open seventy, seventy four on this adult test set, which is not too bad.
And then we also look at the mean average error of uh
This held up test set around five degrees.
Um and something that we also look which is quite important and people don't really look at it or don't even quantify it, it's this um progression to the mean problem.
So we quantify it through the standard ratio um
standard deviation ratio.
So what it's quantifies is um
Because the issue when we have a regression model is that it's trained on a like that one which is trained on mean square error, um, is that uh as a loss is that for sequences he's unsure or he has never seen.
It will learn to like put them in the middle of a distribution.
Because then it's that's gonna uh optimize and minimize the mean square error.
Even though like it's
Showing 61–80 of 346 · page 4 of 18
← Previous
Next →