Computer Immunology

Jan 24, 1997

Abstract

Natural immune systems are sophisticated information processors. They learn to recognize relevant patterns, they remember patterns that have been seen previously, and they use combinatorics to construct pattern detectors efficiently. Further, the individual cells and molecules that comprise the immune system are distributed throughout our bodies, encoding and controlling the system in parallel with no central control mechanism. Understanding the immune system is important, both because of its role in complex diseases such as AIDS and because of potential applications to computational problems. If we can understand the immune system from a computational viewpoint, we may gain insights about how to engineer massively parallel adaptive computations.

The talk will describe a project which is aimed at building computer security systems that incorporate the principles and mechanisms used by natural immune systems. It will emphasize our current work on an anomaly detection system in which "normal" is defined by short-range correlations in a Unix process's system calls. Initial experiments suggest that the definition is stable during normal behavior and that it can detect several common intrusions involving sendmail and other processes.