Compensating for Bias in the SDM Fast Activation Mechanism

The Kanerva Sparse Distributed Memory (SDM) is a mechanism for implementing a memory with a huge address space. The physical memory consists of substantially fewer locations than the virtual address space. The plain Kanerva SDM memory assumes address and data to be random bit vectors with equal probability for 0 and 1. To compensate for any bias in address or data, several more or less elaborate methods can be used. In this paper we describe one simple method for achieving this compensation within the framework of the ``Fast activation Mechanism'' for SDM