Rethinking floating point for deep learning · HackerPapers discussions

Rethinking floating point for deep learning — mentioned by edwintorok

Sun, 21 Jul 2019 11:06:03 GMT

Have you evaluated alternatives such as posits, Kulisch accumulation, and zfp? https://arxiv.org/pdf/1805.08624.pdf https://arxiv.org/pdf/1811.01721.pdf https://insidehpc.com/2018/05/universal-coding-reals-alterna... In particular the latter describes a generic framework that can be used to generate a lot of different number systems. Could hardware implement this, allowing us to compose and choose the number system by just setting some simple flags?

Paper and discussion history. A mention is not an endorsement.

Rethinking floating point for deep learning — HN submission

Tue, 05 Mar 2019 00:57:29 GMT

Rethinking floating point for deep learning

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Rethinking floating point for deep learning — HN submission

Thu, 08 Nov 2018 16:38:48 GMT

Rethinking floating point for deep learning

3 points; 1 discussion reply on HN. Paper and source history. A submission is not an endorsement.