4 ms·

I tend to not believe unbelievable results in machine learning. It's too easy to unintenionally cause some kind of information leakage. I haven't read the paper

I tend to not believe unbelievable results in machine learning. It's too easy to unintenionally cause some kind of information leakage. I haven't read the paper in detail though, so their experimentation setup could be foolproof, this is not a critique of this paper specifically.

sidewndr46 4y ago
This reminds me of the ML research that could predict sex from an iris. It turns out they were using entire photos of eyes to do this. There are so many obvious cues to pick up on in that case, like eyeliner, eyelashes being uniform (or fake), trimmed eyebrows, general makeup on the skin, etc.
dragonwriter 4y ago
> I tend to not believe unbelievable results That seems tautologically true.