Abstract

Some of the most important algorithms today have computational requirements that far outstrip what is practically achievable using general purpose processors like Intel’s Pentium or Xeon chips. Important algorithms from areas like security, financial transactions and machine intelligence are increasingly implemented using highly specialized parallel processors like GPUs or by developing application specific circuits implemented on FPG As and custom silicon. There is an arms race raging at the moment to accelerate how we design such hardware systems quicker than has been possible using conventional hardware design techniques. Furthermore, we need to more quickly map or compile algorithms onto such specialized processor to give us the agility we need to adapt to constantly changing and improving models. This presentation will argue for the use of high level abstractions for implementing such low level systems, taking much inspiration from the world of functional programming.