Doina Bucur

An evolutionary framework for routing protocol analysis in wireless sensor networks

Wireless Sensor Networks (WSNs) are widely adopted for applications ranging from surveillance to environmental monitoring. While powerful and relatively inexpensive, they are subject to behavioural faults which make them unreliable. Due to the complex interactions between network nodes, it is difficult to uncover faults in a WSN by resorting to formal techniques for verification and analysis, or to testing. This paper proposes an evolutionary framework to detect anomalous behaviour related to energy consumption in WSN routing protocols. Given a collection protocol, the framework creates candidate topologies and evaluates them through simulation on the basis of metrics measuring the radio activity on nodes. Experimental results using the standard Collection Tree Protocol show that the proposed approach is able to unveil topologies plagued by excessive energy depletion over one or more nodes, and thus could be used as an offline debugging tool to understand and correct the issues before network deployment and during the development of new protocols.

Applying time series analysis and neighbourhood voting in a decentralised approach for fault detection and classification in WSNs

In pervasive computing environments, wireless sensor networks play an important infrastructure role, collecting reliable and accurate context information so that applications are able to provide services to users on demand. In such environments, sensors should be self-adaptive by taking correct decisions based on sensed data in real-time in a decentralised manner; however, sensed data is often faulty. We thus design a decentralised scheme for fault detection and classification in sensor data in which each sensor node does localised fault detection. A combination of neighbourhood voting and time series data analysis techniques are used to detect faults. We also study the comparative accuracy of both the union and the intersection of the two techniques. Then, detected faults are classified into known fault categories. An initial evaluation with SensorScope, an outdoor temperature dataset, confirms that our solution is able to detect and classify faulty readings into four fault types, namely, 1) random, 2) mal-function, 3) bias, and 4) drift with accuracy up to 95%. The results also show that, with the experimental dataset, the time series data analysis technique performs comparable well in most of the cases, whilst in some other cases the support from neighbourhood voting technique and histogram analysis helps our hybrid solution to successfully detects the faults of all types.

Influence Maximization in Social Networks with Genetic Algorithms

We live in a world of social networks. Our everyday choices are often influenced by social interactions. Word of mouth, meme diffusion on the Internet, and viral marketing are all examples of how social networks can affect our behaviour. In many practical applications, it is of great interest to determine which nodes have the highest influence over the network, i.e., which set of nodes will, indirectly, reach the largest audience when propagating information. These nodes might be, for instance, the target for early adopters of a product, the most influential endorsers in political elections, or the most important investors in financial operations, just to name a few examples. Here, we tackle the NP-hard problem of influence maximization on social networks by means of a Genetic Algorithm. We show that, by using simple genetic operators, it is possible to find in feasible runtime solutions of high-influence that are comparable, and occasionally better, than the solutions found by a number of known heuristics (one of which was previously proven to have the best possible approximation guarantee, in polynomial time, of the optimal solution). The advantages of Genetic Algorithms show, however, in them not requiring any assumptions about the graph underlying the network, and in them obtaining more diverse sets of feasible solutions than current heuristics.

The tradeoffs between data delivery ratio and energy costs in wireless sensor networks: a multi-objectiveevolutionary framework for protocol analysis

Wireless sensor network (WSN) routing protocols, e.g., the Collection Tree Protocol (CTP), are designed to adapt in an ad-hoc fashion to the quality of the environment. WSNs thus have high internal dynamics and complex global behavior. Classical techniques for performance evaluation (such as testing or verification) fail to uncover the cases of extreme behavior which are most interesting to designers. We contribute a practical framework for performance evaluation of WSN protocols. The framework is based on multi-objective optimization, coupled with protocol simulation and evaluation of performance factors. For evaluation, we consider the two crucial functional and non-functional performance factors of a WSN, respectively: the ratio of data delivery from the network (DDR), and the total energy expenditure of the network (COST). We are able to discover network topological configurations over which CTP has unexpectedly low DDR and/or high COST performance, and expose full Pareto fronts which show what the possible performance tradeoffs for CTP are in terms of these two performance factors. Eventually, Pareto fronts allow us to bound the state space of the WSN, a fact which provides essential knowledge to WSN protocol designers.

Black Holes and Revelations: Using Evolutionary Algorithms to Uncover Vulnerabilities in Disruption-Tolerant Networks

A challenging aspect in open ad hoc networks is their resilience against malicious agents. This is especially true in complex, urban-scale scenarios where numerous moving agents carry mobile devices that create a peer-to-peer network without authentication. A requirement for the proper functioning of such networks is that all the peers act legitimately, forwarding the needed messages, and concurring to the maintenance of the network connectivity. However, few malicious agents may easily exploit the movement patterns in the network to dramatically reduce its performance. We propose a methodology where an evolutionary algorithm evolves the parameters of different malicious agents, determining their types and mobility patterns in order to minimize the data delivery rate and maximize the latency of communication in the network. As a case study, we consider a fine-grained simulation of a large-scale disruption-tolerant network in the city of Venice. By evolving malicious agents, we uncover situations where even a single attacker can hamper the network performance, and we correlate the performance decay to the number of malicious agents.

Towards Context Consistency in a Rule-Based Activity Recognition Architecture

Accurate human activity recognition (AR) is crucial for intelligent pervasive environments, e.g., energy-saving buildings. In order to gain precise and fine-grained AR results, a system must overcome partial observability of the environment and noisy, imprecise, and corrupted sensor data. In this work, we propose a rule-based AR architecture that effectively handles multiple-user, multiple-area situations, recognizing real-time office activities. The proposed solution is based on an ontological approach, using low-cost, binary, wireless sensors. We employ context consistency diagrams (CCD) as a key component for fault correction. A CCD is a data structure that provides a mechanism for probabilistic reasoning about the current situation and determines the most probable current situation in the presence of inconsistencies, conflicts, and ambiguities in sensor readings. The implementation of the system and its evaluation in a living lab environment show that the CCD corrects up to 46.8% of sensor data faults, improving overall recognition accuracy by up to 11.1%, thus achieving reliable recognition results from unreliable sensor data.

Characterizing topological bottlenecks for data delivery in CTP using simulation-based stress testing with natural selection

Routing protocols for ad-hoc networks, e.g., the Collection Tree Protocol (CTP), are designed with simple node-local behaviour, but are deployed on testbeds with uncontrollable physical topology; exhaustively verifying the protocol on all possible topologies at design time is not tractable. We obtain topological insights on CTP performance, to answer the question: Which topological patterns cause CTP data routing to fail? We stress-test CTP with a quantitative testing method which searches for topologies using evolutionary algorithms combined with protocol simulation. The method iteratively generates new test topologies, such that the execution of the protocol over these topologies shows increasingly worse data-delivery ratios (DDR). We obtain a large set of example topologies of different network sizes up to 50 nodes, network densities, data rates, table sizes, and radio-frequency noise models, which, although connected, trigger a data delivery of nearly zero. We summarize these topologies into three types of topological problems, the root cause of which is the presence of certain asymmetric links and cycles, combined with a certain size of the routing table. We verify causality, i.e., show that randomly generated topologies having these particular features do cause low DDR in CTP. This testing methodology, while computationally intensive, is sound, fully automated and has better coverage over the corner cases of protocol behaviour than testing a protocol over manually crafted or random topologies.

Improving Multi-objective Evolutionary Influence Maximization in Social Networks

In the context of social networks, maximizing influence means contacting the largest possible number of nodes starting from a set of seed nodes, and assuming a model for influence propagation. The real-world applications of influence maximization are of uttermost importance, and range from social studies to marketing campaigns. Building on a previous work on multi-objective evolutionary influence maximization, we propose improvements that not only speed up the optimization process considerably, but also deliver higher-quality results. State-of-the-art heuristics are run for different sizes of the seed sets, and the results are then used to initialize the population of a multi-objective evolutionary algorithm. The proposed approach is tested on three publicly available real-world networks, where we show that the evolutionary algorithm is able to improve upon the solutions found by the heuristics, while also converging faster than an evolutionary algorithm started from scratch.

Towards accurate de novo assembly for genomes with repeats

De novo genome assemblers designed for short k-mer length or using short raw reads are unlikely to recover complex features of the underlying genome, such as repeats hundreds of bases long. We implement a stochastic machine-learning method which obtains accurate assemblies with repeats and self-validates assemblies via consensus. For this, a prior assembler is extended with the ability to (a) assemble variable-length raw reads, which may span and unambiguously recover interspersed repeats in the genome, and (b) recognize long, direct terminal repeats during the assembly, then report an unambiguous circular assembly. Consensus is obtained via stochastically independent runs of the assembler on the same read library. We experiment on viral and mitochondrial genomes of up to 41 kbp, with synthetic raw-read libraries, to be able to evaluate the assembly against a reference. We show the prerequisites for obtaining accurate assemblies. For genomes with interspersed repeats, using raw reads of average length comparable to the repeat length likely gives an accurate genome. Genomes with long direct terminal repeats can be assembled accurately also with reads shorter than the repeat length. In both cases, a simple majority forms consensus, since over 70 % of independent runs on this set of genomes yield a correct assembly.

Improved search methods for assessing Delay-Tolerant Networks vulnerability to colluding strong heterogeneous attacks

Blackhole fast-moving attackers are effective against the First-Contact protocol.Flooding fast-moving attackers are effective against the Epidemic protocol.Mixed blackhole and flooding are effective against the Spray-and-Wait protocol.MaxProp is resilient; flooding attackers will mildly affect the protocol. Increasingly more digital communication is routed among wireless, mobile computers over ad-hoc, unsecured communication channels. In this paper, we design two stochastic search algorithms (a greedy heuristic, and an evolutionary algorithm) which automatically search for strong insider attack methods against a given ad-hoc, delay-tolerant communication protocol, and thus expose its weaknesses. To assess their performance, we apply the two algorithms to two simulated, large-scale mobile scenarios (of different route morphology) with 200 nodes having free range of movement. We investigate a choice of two standard attack strategies (dropping messages and flooding the network), and four delay-tolerant routing protocols: First Contact, Epidemic, Spray and Wait, and MaxProp. We find dramatic drops in performance: replicative protocols (Epidemic, Spray and Wait, MaxProp), formerly deemed resilient, are compromised to different degrees (delivery rates between 24% and 87%), while a forwarding protocol (First Contact) is shown to drop delivery rates to under 5% in all cases by well-crafted attack strategies and with an attacker group of size less than 10% the total network size. Overall, we show that the two proposed methods combined constitute an effective means to discover (at design-time) and raise awareness about the weaknesses and strengths of existing ad-hoc, delay-tolerant communication protocols against potential malicious cyber-attacks.