Neural networks for computer virus recognition

摘要: We have developed a neural network for generic detection of particular class computer viruses-the so called boot sector viruses that infect the floppy disk or hard drive. This is an important and relatively tractable subproblem virus detection. Only about 5% all known are viruses, yet they account nearly 90% incidents. successfully deployed our as commercial product, distributing it to millions PC users worldwide part IBM AntiVirus software package. faced several challenges in taking from research idea product. These included designing appropriate input representation scheme; dealing with scarcity available training data; finding trade off point between false positives negatives conform user expectations; making strict constraints on memory speed computation needed run PCs. The article discusses methods handling these challenges.