Re: sizes of floats, was: +-Inf and NaN

James Cloos <[email protected]> Wed, 28 Feb 2024 21:27:22 -0500

Newsgroups	gmane.comp.mathematics.maxima.general,gmane.lisp.gcl.devel
Message-ID	<[email protected]>

>>>>> "SM" == Stavros Macrakis <[email protected]> writes:

SM> The application these days that really benefits from smaller floats
SM> (float16 or smaller) is ML, both because it saves space and because
SM> it speeds up computation.

There was a post today (I think it was on HN) about a paper which
concluded that using only the sign of floats was enough.

Ie replacing each float with a single balanced trit.

They also reported that replacing dot products with a simple sum of all
of the components proved sufficient as well.

-JimC
-- 
James Cloos <[email protected]>
            OpenPGP: https://jhcloos.com/0x997A9F17ED7DAEA6.asc