Article: Hacking smart machines with smarter ones: How to extract meaningful data from machine learning classifiers Journal: International Journal of Security and Networks (IJSN) 2015 Vol.10 No.3 pp.137 - 150 Abstract: Machine-learning (ML) enables computers to learn how to recognise patterns, make unintended decisions, or react to a dynamic environment. The effectiveness of trained machines varies because of more suitable ML algorithms or because superior training sets. Although ML algorithms are known and publicly released, training sets may not be reasonably ascertainable and, indeed, may be guarded as trade secrets. In this paper we focus our attention on ML classifiers and on the statistical information that can be unconsciously or maliciously revealed from them. We show that it is possible to infer unexpected but useful information from ML classifiers. In particular, we build a novel meta-classifier and train it to hack other classifiers, obtaining meaningful information about their training sets. Such information leakage can be exploited, for example, by a vendor to build more effective classifiers or to simply acquire trade secrets from a competitor's apparatus, potentially violating its intellectual property rights. Inderscience Publishers - linking academia, business and industry through research

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

Authors: Giuseppe Ateniese; Luigi V. Mancini; Angelo Spognardi; Antonio Villani; Domenico Vitali; Giovanni Felici

Addresses: Computer Science Department, Sapienza Università di Roma, Via Salaria 113, 00198 Rome, Italy ' Computer Science Department, Sapienza Università di Roma, Via Salaria 113, 00198 Rome, Italy ' Institute of Informatics and Telematics, CNR, Italy, Via G. Moruzzi 1m, 56124 Pisa, Italy ' Computer Science Department, Sapienza Università di Roma, via Salaria 113, 00198 Rome, Italy ' Computer Science Department, Sapienza Università di Roma, via Salaria 113, 00198 Rome, Italy ' Institute for Systems Analysis and Computer Science "Antonio Ruberti", CNR, Italy, Via dei Taurini 19, 00185 Rome, Italy

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

Keywords: machine learning classifiers; information leakages; attack methodology; unauthorised access; trade secrets; intellectual property rights; IPR; security; meta-classifiers; classifier hacking; training sets; hacking attacks.

DOI: 10.1504/IJSN.2015.071829

International Journal of Security and Networks, 2015 Vol.10 No.3, pp.137 - 150

Received: 08 Jul 2014
Accepted: 18 Feb 2015
Published online: 19 Sep 2015 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

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

Keep up-to-date