Predicting minimal error bounds through an algorithm

Add to your list(s) Download to your calendar using vCal

• David Boland ( Imperial College London)
• Thursday 19 February 2009, 12:00-13:00
• Mahanakorn Laboratory, EEE.

If you have a question about this talk, please contact Dr George A Constantinides.

The amount of precision used in an algorithm trades error for silicon area usage and potential parallelism. This talk will explain the cause of floating point error in computations and show how simple polynomials can be used to describe this error. It will then attempt to describe some background theory which is intended to be used on these polynomials to find tight bounds on the final error in any algorithm. Finally, it will then display some simple examples to illustrate the use of this theory & highlight some of the complexities in creating a general algorithm to use this theory to find minimal error bounds.

This talk is part of the CAS FPGA Talks series.

This talk is included in these lists:

• CAS FPGA Talks
• Mahanakorn Laboratory, EEE

Note that ex-directory lists are not shown.