When we see large numbers or small fractions we automatically pick float or double floating-point types for our data representation. However, is it always optimal? What other choices we have? Are there any other standards than the IEEE 754 FP?
In this talk I’ll present such topics as:
- Brief intro to the FP IEEE 754 standard, its pros, cons and pitfalls when computing in the FP domain.
- How can we compress floating point data – shorter formats, lossless & lossy compressions, new ZFP and SZ methods and their use in C++.
- A new kid on the block – the POSIT arithmetic and how to use it in C++ programs.
- Intro to the relevant C++ libraries and – as always – a lot of practical examples.
- Few notes on deep learning and FP formats.
Everybody is invited ;)
Please login to comment