Domenico Vitali

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

Machine Learning (ML) algorithms are used to train computers to perform a variety of complex tasks and improve with experience. Computers learn how to recognize patterns, make unintended decisions, or react to a dynamic environment. Certain trained machines may be more effective than others because they are based on more suitable ML algorithms or because they were trained through 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. While much research has been performed about the privacy of the elements of training sets, 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. This kind of information leakage can be exploited, for example, by a vendor to build more effective classifiers or to simply acquire trade secrets from a competitors apparatus, potentially violating its intellectual property rights.

Obfuscation of sensitive data in network flows

In the last decade, the release of network flows has gained significant popularity among researchers and networking communities. Indeed, network flows are a fundamental tool for modeling the network behavior, identifying security attacks, and validating research results. Unfortunately, due to the sensitive nature of network flows, security and privacy concerns discourage the publication of such datasets. On the one hand, existing techniques proposed to sanitize network flows do not provide any formal guarantees. On the other hand, microdata anonymization techniques are not directly applicable to network flows. In this paper, we propose a novel obfuscation technique for network flows that provides formal guarantees under realistic assumptions about the adversarys knowledge. Our work is supported by extensive experiments with a large set of real network flows collected at an important Italian Tier II Autonomous System, hosting sensitive government and corporate sites. Experimental results show that our obfuscation technique preserves the utility of network flows for network traffic analysis.

Obfuscation of sensitive data for incremental release of network flows

Large datasets of real network flows acquired from the Internet are an invaluable resource for the research community. Applications include network modeling and simulation, identification of security attacks, and validation of research results. Unfortunately, network flows carry extremely sensitive information, and this discourages the publication of those datasets. Indeed, existing techniques for network flow sanitization are vulnerable to different kinds of attacks, and solutions proposed for microdata anonymity cannot be directly applied to network traces. In our previous research, we proposed an obfuscation technique for network flows, providing formal confidentiality guarantees under realistic assumptions about the adversarys knowledge. In this paper, we identify the threats posed by the incremental release of network flows, we propose a novel defense algorithm, and we formally prove the achieved confidentiality guarantees. An extensive experimental evaluation of the algorithm for incremental obfuscation, carried out with billions of real Internet flows, shows that our obfuscation technique preserves the utility of flows for network traffic analysis.

Bypassing Censorship: A Proven Tool against the Recent Internet Censorship in Turkey

Users of mobile devices are experiencing great difficulties to circumvent Internet censorship technologies that violate human rights. Mobile users do not have full control of their own systems, and in many cases, they cannot even change the configuration imposed by their 3G/4G providers. Such limitations allow the provider acting under the government authority to enforce specific Internet filtering mechanisms, and to prevent access to censored material. In this paper, we survey the events related to the Internet censorship happened in Turkey during the first months of 2014 and we introduce DNSet, an Android app that has been used by Turkish citizens to successfully circumvent the Internet censorship. In particular, DNSet allows mobile users to easily change the DNS server imposed by their 3G/4G providers, without the mobile users have administrative rights on the device (i.e. Without rooting the device). We report on data and information that has been anonymously collected through the DNSet application. Furthermore, we raise up the suspicion that a few censorship activities in Turkey began at least a month before the official ban on Twitter.

Obsidian: A scalable and efficient framework for NetFlow obfuscation

Through this software the authors aim to promote the sharing of network logs within the research community. The (k, j)obfuscation technique opens sundry interesting future directions. In fact, many networking and security tasks can be re-thought based on obfuscated datasets, for instance, quality of service (QoS), traffic classification, anomaly detection and more.

Replication Schemes in Unattended Wireless Sensor Networks

Unattended Wireless Sensor Networks (UWSNs) are networks that operate without on-line data collection entities (Sink Collectors). This configuration enriches their flexibility but, at the same time, introduces new security issues. A challenging issue is data survival: in absence of the sink, in fact, data sensed by network sensors cannot be kept clear from the adversary. Cryptographic schemes, moreover, cannot be easily implemented: cryptographic material as well as sensed data are continuously exposed to the activity of a smart adversary. Several strategies were proposed in literature which collectively attempt to mitigate this challenge. In this paper we investigate the use of replication as a mean to enhance the survival probability of collected data. We propose a scheme that deterministically grants data survival against a focused mobile adversary. We analyze the behavior of our proposal and study its security and efficiency properties to prove its suitability for UWSNs.

Large-Scale Traffic Anomaly Detection: Analysis of Real Netflow Datasets

The analysis of large amount of traffic data is the daily routine of Autonomous Systems and ISP operators. The detection of anomalies like denial-of-service (DoS) or distributed denial-of-service (DDoS) is also one of the main issues for critical services and infrastructures. The suitability of metrics coming from the information theory for detecting DoS and DDoS episodes has been widely analyzed in the past. Unfortunately, their effectiveness are often evaluated on synthetic data set, or, in other cases, on old and unrepresentative data set, e.g. the DARPA network dump. This paper presents the evaluation by means of main metrics proposed in the literature of a real and large network flow dataset, collected from an Italian transit tier II Autonomous System (AS) located in Rome. We show how we effectively detected and analyzed several attacks against Italian critical IT services, some of them also publicly announced. We further report the study of others legitimate and malicious activities we found by ex-post analysis.

Relieve Internet Routing Security of Public Key Infrastructure

Lack of security mechanisms expose the Border Gateway Protocol (BGP) to a wide range of threats that are constantly undermining security of the Internet. Most prominent attacks include prefix hijacking and announcement of false routes to maliciously attract or divert traffic. A number of cryptographic solutions to prevent both attacks have been proposed but have not been adopted due to involved operations and considerable overhead. Most of them rely on digital signatures to authorize Autonomous Systems to propagate route announcements. Surprisingly, the scientific community has devoted only little interest to the problem of revocation in BGP. In particular, BGP systems based on Public Key Infrastructure allow to revoke an Autonomous System by revoking its public key certificate. However, there seem to be no solution for selective revocation of AS-path announcements. This paper introduces reBGP, an enhanced version of BGP that leverages Identity Based Cryptography to secure BGP with minimal overhead. reBGP prevents prefix hijacking and false route announcement through Aggregate Identity Based Signatures and provides an effective revocation means to invalidate AS-path announcements. reBGP enjoys a constant overhead to verify authenticity of routes and does not require a Public Key Infrastructure. Extensive testing of our implementation, show that our proposal represents a practical solution to secure BGP.

MhRep: Multi-hop Replication Scheme for Data Survival in Unattended Wireless Sensor Networks

Unattended Wireless Sensors Networks (UWSNs) are emerging Wireless Sensor Networks (WSNs) characterized by periodic absence of trusted entities, like the Sink Collectors. The time periods with no sink attendance, force the network sensors to locally store collected data until the next sink visit. This means that data collection is not performed in real time, but it falls at regular intervals. Such conditions define a different paradigm regarding traditional WSNs and introduce several new security issues, data survival above all. At the best of our knowledge, only two strategies were proposed to mitigate mobile adversary data deletion attack : encryption protocols (implementing backward and/or forward secrecy) and replication. While encryption protocols can impose an extensive computational effort and experience several security issues (like key exposure), replication schemes seem to capture a renewed interest from researchers, since their power requirement seem compatible with the constraints of wireless sensors. In this paper we focus on a new replication scheme and compare its performances with some other proposals, previously presented in literature. The schemes share some model assumptions, in terms of network and adversary attack strategy and objectives.

Firmware Enhancements for BYOD-Aware Network Security

In today’s connected world, users migrate within a complex set of networks, including, but not limited to, 3G and 4G (LTE) services provided by mobile operators, Wi-Fi hotspots in private and public places, as well as wireless and/or wired LAN access in business and home environments. Following the widely expanding Bring Your Own Device (BYOD) approach, many public and educational institutions have begun to encourage customers and students to use their own devices at all times. While this may be cost-effective in terms of decreased investments in hardware and consequently lower maintenance fees on a long-term basis, it may also involve some security risks. In particular, many users are often connected to more than one network and/or communication service provider at the same time, for example to a 3G/4G mobile network and to a Wi-Fi. In a BYOD setting, an infected device or a rogue one can turn into an unwanted gateway, causing a security breach by leaking information across networks. Aiming at investigating in greater detail the implications of BYOD on network security in private and business settings we are building a framework for experiments with mobile routers both in home and business networks. This is a continuation of our earlier work on communications and services with enhanced security for network appliances.