Hacking smart machines with smarter ones: How to extract meaningful data from machine learning classifiers

摘要: Machine-learning ML enables computers to learn how recognise patterns, make unintended decisions, or react a dynamic environment. The effectiveness of trained machines varies because more suitable algorithms superior training sets. Although are known and publicly released, sets may not be reasonably ascertainable and, indeed, guarded as trade secrets. In this paper we focus our attention on classifiers the statistical information that can unconsciously maliciously revealed from them. We show it is possible infer unexpected but useful classifiers. particular, build novel meta-classifier train hack other classifiers, obtaining meaningful about their Such leakage exploited, for example, by vendor effective simply acquire secrets competitor's apparatus, potentially violating its intellectual property rights.