Open problems in the security of learning

Machine learning has become a valuable tool for detecting and preventing malicious activity. However, as more applications employ machine learning techniques in adversarial decision-making situations, increasingly powerful attacks become possible against machine learning systems. In this paper, we present three broad research directions towards the end of developing truly secure learning. First, we suggest that finding bounds on adversarial influence is important to understand the limits of what an attacker can and cannot do to a learning system. Second, we investigate the value of adversarial capabilities-the success of an attack depends largely on what types of information and influence the attacker has. Finally, we propose directions in technologies for secure learning and suggest lines of investigation into secure techniques for learning in adversarial environments. We intend this paper to foster discussion about the security of machine learning, and we believe that the research directions we propose represent the most important directions to pursue in the quest for secure learning.

Real-time detection of concealed chemical hazards under ambient light conditions using Raman spectroscopy

Current concerns regarding terrorism and international crime highlight the need for new techniques for detecting unknown and hazardous substances. A novel Raman spectroscopy-based technique, spatially offset Raman spectroscopy (SORS), was recently devised for non-invasively probing the contents of diffusely scattering and opaque containers. Here, we demonstrate a modified portable SORS sensor for detecting concealed substances in-field under different background lighting conditions. Samples including explosive precursors, drugs and an organophosphate insecticide (chemical warfare agent surrogate) were concealed inside diffusely scattering packaging including plastic, paper and cloth. Measurements were carried out under incandescent and fluorescent light as well as under daylight to assess the suitability of the probe for different real-life conditions. In each case, it was possible to identify the substances against their reference Raman spectra in less than one minute. The developed sensor has potential for rapid detection of concealed hazardous substances in airports, mail distribution centers and customs checkpoints.

ASICS : authenticated key exchange security incorporating certification systems

Most security models for authenticated key exchange (AKE) do not explicitly model the associated certification system, which includes the certification authority (CA) and its behaviour. However, there are several well-known and realistic attacks on AKE protocols which exploit various forms of malicious key registration and which therefore lie outside the scope of these models. We provide the first systematic analysis of AKE security incorporating certification systems (ASICS). We define a family of security models that, in addition to allowing different sets of standard AKE adversary queries, also permit the adversary to register arbitrary bitstrings as keys. For this model family we prove generic results that enable the design and verification of protocols that achieve security even if some keys have been produced maliciously. Our approach is applicable to a wide range of models and protocols; as a concrete illustration of its power, we apply it to the CMQV protocol in the natural strengthening of the eCK model to the ASICS setting.

On the security of PAS (predicate-based authentication service)

Recently a new human authentication scheme called PAS (predicate-based authentication service) was proposed, which does not require the assistance of any supplementary device. The main security claim of PAS is to resist passive adversaries who can observe the whole authentication session between the human user and the remote server. In this paper we show that PAS is insecure against both brute force attack and a probabilistic attack. In particular, we show that its security against brute force attack was strongly overestimated. Furthermore, we introduce a probabilistic attack, which can break part of the password even with a very small number of observed authentication sessions. Although the proposed attack cannot completely break the password, it can downgrade the PAS system to a much weaker system similar to common OTP (one-time password) systems.

Cryptology and network security

This book constitutes the refereed proceedings of the 11th International Conference on Cryptology and Network Security, CANS 2012, held in Darmstadt, Germany, in December 2012. The 22 revised full papers, presented were carefully reviewed and selected from 99 submissions. The papers are organized in topical sections on cryptanalysis; network security; cryptographic protocols; encryption; and s-box theory.

On the Average Case Communication Complexity for Detection in Sensor Networks

The problem of sensor-network-based distributed intrusion detection in the presence of clutter is considered. It is argued that sensing is best regarded as a local phenomenon in that only sensors in the immediate vicinity of an intruder are triggered. In such a setting, lack of knowledge of intruder location gives rise to correlated sensor readings. A signal-space viewpoint is introduced in which the noise-free sensor readings associated to intruder and clutter appear as surfaces $\mathcal{S_I}$ and $\mathcal{S_C}$ and the problem reduces to one of determining in distributed fashion, whether the current noisy sensor reading is best classified as intruder or clutter. Two approaches to distributed detection are pursued. In the first, a decision surface separating $\mathcal{S_I}$ and $\mathcal{S_C}$ is identified using Neyman-Pearson criteria. Thereafter, the individual sensor nodes interactively exchange bits to determine whether the sensor readings are on one side or the other of the decision surface. Bounds on the number of bits needed to be exchanged are derived, based on communication complexity (CC) theory. A lower bound derived for the two-party average case CC of general functions is compared against the performance of a greedy algorithm. The average case CC of the relevant greater-than (GT) function is characterized within two bits. In the second approach, each sensor node broadcasts a single bit arising from appropriate two-level quantization of its own sensor reading, keeping in mind the fusion rule to be subsequently applied at a local fusion center. The optimality of a threshold test as a quantization rule is proved under simplifying assumptions. Finally, results from a QualNet simulation of the algorithms are presented that include intruder tracking using a naive polynomial-regression algorithm.

Intruder detection over sensor placement in a hexagonal lattice

The problem of intrusion detection and location identification in the presence of clutter is considered for a hexagonal sensor-node geometry. It is noted that in any practical application,for a given fixed intruder or clutter location, only a small number of neighboring sensor nodes will register a significant reading. Thus sensing may be regarded as a local phenomenon and performance is strongly dependent on the local geometry of the sensor nodes. We focus on the case when the sensor nodes form a hexagonal lattice. The optimality of the hexagonal lattice with respect to density of packing and covering and largeness of the kissing number suggest that this is the best possible arrangement from a sensor network viewpoint. The results presented here are clearly relevant when the particular sensing application permits a deterministic placement of sensors. The results also serve as a performance benchmark for the case of a random deployment of sensors. A novel feature of our analysis of the hexagonal sensor grid is a signal-space viewpoint which sheds light on achievable performance.Under this viewpoint, the problem of intruder detection is reduced to one of determining in a distributed manner, the optimal decision boundary that separates the signal spaces SI and SC associated to intruder and clutter respectively. Given the difficulty of implementing the optimal detector, we present a low-complexity distributive algorithm under which the surfaces SI and SC are separated by a wellchosen hyperplane. The algorithm is designed to be efficient in terms of communication cost by minimizing the expected number of bits transmitted by a sensor.

Two timescale convergent Q-learning for sleep-scheduling in wireless sensor networks

In this paper, we consider an intrusion detection application for Wireless Sensor Networks. We study the problem of scheduling the sleep times of the individual sensors, where the objective is to maximize the network lifetime while keeping the tracking error to a minimum. We formulate this problem as a partially-observable Markov decision process (POMDP) with continuous stateaction spaces, in a manner similar to Fuemmeler and Veeravalli (IEEE Trans Signal Process 56(5), 2091-2101, 2008). However, unlike their formulation, we consider infinite horizon discounted and average cost objectives as performance criteria. For each criterion, we propose a convergent on-policy Q-learning algorithm that operates on two timescales, while employing function approximation. Feature-based representations and function approximation is necessary to handle the curse of dimensionality associated with the underlying POMDP. Our proposed algorithm incorporates a policy gradient update using a one-simulation simultaneous perturbation stochastic approximation estimate on the faster timescale, while the Q-value parameter (arising from a linear function approximation architecture for the Q-values) is updated in an on-policy temporal difference algorithm-like fashion on the slower timescale. The feature selection scheme employed in each of our algorithms manages the energy and tracking components in a manner that assists the search for the optimal sleep-scheduling policy. For the sake of comparison, in both discounted and average settings, we also develop a function approximation analogue of the Q-learning algorithm. This algorithm, unlike the two-timescale variant, does not possess theoretical convergence guarantees. Finally, we also adapt our algorithms to include a stochastic iterative estimation scheme for the intruder's mobility model and this is useful in settings where the latter is not known. Our simulation results on a synthetic 2-dimensional network setting suggest that our algorithms result in better tracking accuracy at the cost of only a few additional sensors, in comparison to a recent prior work.

Adaptive Sleep-Wake Control using Reinforcement Learning in Sensor Networks

The aim in this paper is to allocate the `sleep time' of the individual sensors in an intrusion detection application so that the energy consumption from the sensors is reduced, while keeping the tracking error to a minimum. We propose two novel reinforcement learning (RL) based algorithms that attempt to minimize a certain long-run average cost objective. Both our algorithms incorporate feature-based representations to handle the curse of dimensionality associated with the underlying partially-observable Markov decision process (POMDP). Further, the feature selection scheme used in our algorithms intelligently manages the energy cost and tracking cost factors, which in turn assists the search for the optimal sleeping policy. We also extend these algorithms to a setting where the intruder's mobility model is not known by incorporating a stochastic iterative scheme for estimating the mobility model. The simulation results on a synthetic 2-d network setting are encouraging.

Lógica nebulosa aplicada a um sistema de detecção de intrusos em computação em nuvem.

O objetivo deste trabalho é avaliar os riscos de ocorrências de intrusos em um sistema de computação em nuvem para sistemas distribuídos utilizando lógica nebulosa. A computação em nuvem é um tema que vem sendo bastante abordado e vem alavancando discussões calorosas, tanto na comunidade acadêmica quanto em palestras profissionais. Embora essa tecnologia esteja ganhando mercado, alguns estudiosos encontram-se céticos afirmando que ainda é cedo para se tirar conclusões. Isto se deve principalmente por causa de um fator crítico, que é a segurança dos dados armazenados na nuvem. Para esta dissertação, foi elaborado um sistema distribuído escrito em Java com a finalidade de controlar um processo de desenvolvimento colaborativo de software na nuvem, o qual serviu de estudo de caso para avaliar a abordagem de detecção de intrusos proposta. Este ambiente foi construído com cinco máquinas (sendo quatro máquinas virtuais e uma máquina real). Foram criados dois sistemas de inferência nebulosos, para análise de problemas na rede de segurança implementados em Java, no ambiente distribuído. Foram realizados diversos testes com o intuito de verificar o funcionamento da aplicação, apresentando um resultado satisfatório dentro dessa metodologia.

Measuring inconsistency in network intrusion rules

In this preliminary case study, we investigate how inconsistency in a network intrusion detection rule set can be measured. To achieve this, we first examine the structure of these rules which incorporate regular expression (Regex) pattern matching. We then identify primitive elements in these rules in order to translate the rules into their (equivalent) logical forms and to establish connections between them. Additional rules from background knowledge are also introduced to make the correlations among rules more explicit. Finally, we measure the degree of inconsistency in formulae of such a rule set (using the Scoring function, Shapley inconsistency values and Blame measure for prioritized knowledge) and compare the informativeness of these measures. We conclude that such measures are useful for the network intrusion domain assuming that incorporating domain knowledge for correlation of rules is feasible.

Android Malware Detection Using Parallel Machine Learning Classifiers

Mobile malware has continued to grow at an alarming rate despite on-going mitigation efforts. This has been much more prevalent on Android due to being an open platform that is rapidly overtaking other competing platforms in the mobile smart devices market. Recently, a new generation of Android malware families has emerged with advanced evasion capabilities which make them much more difficult to detect using conventional methods. This paper proposes and investigates a parallel machine learning based classification approach for early detection of Android malware. Using real malware samples and benign applications, a composite classification model is developed from parallel combination of heterogeneous classifiers. The empirical evaluation of the model under different combination schemes demonstrates its efficacy and potential to improve detection accuracy. More importantly, by utilizing several classifiers with diverse characteristics, their strengths can be harnessed not only for enhanced Android malware detection but also quicker white box analysis by means of the more interpretable constituent classifiers.

High accuracy android malware detection using ensemble learning

With over 50 billion downloads and more than 1.3 million apps in Google’s official market, Android has continued to gain popularity amongst smartphone users worldwide. At the same time there has been a rise in malware targeting the platform, with more recent strains employing highly sophisticated detection avoidance techniques. As traditional signature based methods become less potent in detecting unknown malware, alternatives are needed for timely zero-day discovery. Thus this paper proposes an approach that utilizes ensemble learning for Android malware detection. It combines advantages of static analysis with the efficiency and performance of ensemble machine learning to improve Android malware detection accuracy. The machine learning models are built using a large repository of malware samples and benign apps from a leading antivirus vendor. Experimental results and analysis presented shows that the proposed method which uses a large feature space to leverage the power of ensemble learning is capable of 97.3 % to 99% detection accuracy with very low false positive rates.

Eliciting Security Requirements and Tracing them to Design: An Integration of Common Criteria, Heuristics, and UMLsec

Building secure systems is difficult for many reasons. This paper deals with two of the main challenges: (i) the lack of security expertise in development teams, and (ii) the inadequacy of existing methodologies to support developers who are not security experts. The security standard ISO 14508 (Common Criteria) together with secure design techniques such as UMLsec can provide the security expertise, knowledge, and guidelines that are needed. However, security expertise and guidelines are not stated explicitly in the Common Criteria. They are rather phrased in security domain terminology and difficult to understand for developers. This means that some general security and secure design expertise are required to fully take advantage of the Common Criteria and UMLsec. In addition, there is the problem of tracing security requirements and objectives into solution design,which is needed for proof of requirements fulfilment. This paper describes a security requirements engineering methodology called SecReq. SecReq combines three techniques: the Common Criteria, the heuristic requirements editorHeRA, andUMLsec. SecReqmakes systematic use of the security engineering knowledge contained in the Common Criteria and UMLsec, as well as security-related heuristics in the HeRA tool. The integrated SecReq method supports early detection of security-related issues (HeRA), their systematic refinement guided by the Common Criteria, and the ability to trace security requirements into UML design models. A feedback loop helps reusing experiencewithin SecReq and turns the approach into an iterative process for the secure system life-cycle, also in the presence of system evolution.