Biblio

Federated learning (FL) is the rapidly developing machine learning technique that is used to perform collaborative model training over decentralized datasets. FL enables privacy-preserving model development whereby the datasets are scattered over a large set of data producers (i.e., devices and/or systems). These data producers train the learning models, encapsulate the model updates with differential privacy techniques, and share them to centralized systems for global aggregation. However, these centralized models are always prone to adversarial attacks (such as data-poisoning and model poisoning attacks) due to a large number of data producers. Hence, FL methods need to ensure fairness and high-quality model availability across all the participants in the underlying AI systems. In this paper, we propose a novel FL framework, called FairFed, to meet fairness and high-quality data requirements. The FairFed provides a fairness mechanism to detect adversaries across the devices and datasets in the FL network and reject their model updates. We use a Python-simulated FL framework to enable large-scale training over MNIST dataset. We simulate a cross-device model training settings to detect adversaries in the training network. We used TensorFlow Federated and Python to implement the fairness protocol, the deep neural network, and the outlier detection algorithm. We thoroughly test the proposed FairFed framework with random and uniform data distributions across the training network and compare our initial results with the baseline fairness scheme. Our proposed work shows promising results in terms of model accuracy and loss.

Pushed by market forces, software development has become fast-paced. As a consequence, modern development projects are assembled from 3rd-party components. Security & privacy assurance techniques once designed for large, controlled updates over months or years, must now cope with small, continuous changes taking place within a week, and happening in sub-components that are controlled by third-party developers one might not even know they existed. In this paper, we aim to provide an overview of the current software security approaches and evaluate their appropriateness in the face of the changed nature in software development. Software security assurance could benefit by switching from a process-based to an artefact-based approach. Further, security evaluation might need to be more incremental, automated and decentralized. We believe this can be achieved by supporting mechanisms for lightweight and scalable screenings that are applicable to the entire population of software components albeit there might be a price to pay.

The goal of this project is to create a realistic VR experience of solitary confinement and study its impact on users. Although there have been active debates and studies on this subject, very few people have personal experience of solitary confinement. Our first aim is to create such an experience in VR to raise the awareness of solitary confinement. We also want to conduct user studies to compare the VR solitary confinement experience with other types of media experiences, such as films or personal narrations. Finally, we want to study people’s sense of time in such a VR environment.

This study focuses on identifying the factors contributing to a sense of personal responsibility that could improve understanding of insecure cybersecurity behavior and guide research toward more effective messaging targeting non-adopting populations. Towards that, we ran a 2(account type) x2(usage scenario) x2(message type) between-group study with 237 United States adult participants on Amazon MTurk, and investigated how the non-adopting population allocates blame, and under what circumstances they blame the end user among the parties who hold responsibility: the software companies holding data, the attackers exposing data, and others. We find users primarily hold service providers accountable for breaches but they feel the same companies should not enforce stronger security policies on users. Results indicate that people do hold end users accountable for their behavior in the event of a breach, especially when the users' behavior affects others. Implications of our findings in risk communication is discussed in the paper.

With the rapid development of smart grids and the continuous deepening of informatization, while realizing remote telemetry and remote control of massive data-based grid operation, electricity information network security problems have become more serious and prominent. A method for electricity information network security situation prediction method based on improved deep belief network is proposed in this paper. Firstly, the affinity propagation clustering algorithm is used to determine the depth of the deep belief network and the number of hidden layer nodes based on sample parameters. Secondly, continuously adjust the scaling factor and crossover probability in the differential evolution algorithm according to the population similarity. Finally, a chaotic search method is used to perform a second search for the best individuals and similarity centers of each generation of the population. Simulation experiments show that the proposed algorithm not only enhances the generalization ability of electricity information network security situation prediction, but also has higher prediction accuracy.

In this work, novel metaheuristic algorithms are proposed to address the network coding (NC)-based routing and spectrum allocation (RSA) problem in elastic optical networks, aiming to increase the level of security against eavesdropping attacks for the network's confidential connections. A modified simulated annealing, a genetic algorithm, as well as a combination of the two techniques are examined in terms of confidentiality and spectrum utilization. Performance results demonstrate that using metaheuristic techniques can improve the performance of NC-based RSA algorithms and thus can be utilized in real-world network scenarios.

In this work, we propose a new concept of multiple support thresholds to sanitize the database for specific sensitive itemsets. The proposed method assigns a stricter threshold to the sensitive itemset for data sanitization. Furthermore, a genetic-algorithm (GA)-based model is involved in the designed algorithm to minimize side effects. In our experimental results, the GA-based PPDM approach is compared with traditional compact GA-based model and results clearly showed that our proposed method can obtain better performance with less computational cost.

Cryptography is a powerful means of delivering information in a secure manner. Over the years, many image encryption algorithms have been proposed based on the chaotic system to protect the digital image against cryptography attacks. In chaotic encryption, it jumbles the image to vary the framework of the image. This makes it difficult for the attacker to retrieve the original image. This paper introduces an efficient image encryption algorithm incorporating the genetic algorithm, bit plane slicing and bit plane rotation of the digital image. The digital image is sliced into eight planes and each plane is well rotated to give a fully encrypted image after the application of the Genetic Algorithm on each pixel of the image. This makes it less prone to attacks. For decryption, we perform the operations in the reverse order. The performance of this algorithm is measured using various similarity measures like Structural Similarity Index Measure (SSIM). The results exhibit that the proposed scheme provides a stronger level of encryption and an enhanced security level.

Security services and share information is provided by the computer network. The computer network is by default there is not security. The Attackers can use this provision to hack and steal private information. Confidentiality, creation, changes, and truthful of data is will be big problems in the network. Many types of research have given many methods regarding this, from these methods Generating Initial Chromosome Key called Generating Dynamic Chromosome Key (GDCK), which is a novel approach. With the help of the RSA (Rivest Shamir Adleman) algorithm, GICK and GDCK have created an initial key. The proposed method has produced new techniques using genetic fitness function for the sender and receiver. The outcome of GICK and GDCK has been verified by NIST (National Institute of Standards Technology) tools and analyzes randomness of auto-generated keys with various methods. The proposed system has involved three examines; it has been yield better P-Values 6.44, 7.05, and 8.05 while comparing existing methods.

These days the digitization process is everywhere, spreading also across central governments and local authorities. It is hoped that, using open government data for scientific research purposes, the public good and social justice might be enhanced. Taking into account the European General Data Protection Regulation recently adopted, the big challenge in Portugal and other European countries, is how to provide the right balance between personal data privacy and data value for research. This work presents a sensitivity study of data anonymization procedure applied to a real open government data available from the Brazilian higher education evaluation system. The ARX k-anonymization algorithm, with and without generalization of some research value variables, was performed. The analysis of the amount of data / information lost and the risk of re-identification suggest that the anonymization process may lead to the under-representation of minorities and sociodemographic disadvantaged groups. It will enable scientists to improve the balance among risk, data usability, and contributions for the public good policies and practices.

For solving multi-objective optimization problems, this paper firstly combines a multi-objective evolutionary algorithm based on decomposition (MOEA/D) with good convergence and non-dominated sorting genetic algorithm II (NSGA-II) with good distribution to construct. Thus we propose a hybrid multi-objective optimization solving algorithm. Then, we consider that the population diversity needs to be improved while applying multi-objective particle swarm optimization (MOPSO) to solve the multi-objective optimization problems and an improved MOPSO algorithm is proposed. We give the distance function between the individual and the population, and the individual with the largest distance is selected as the global optimal individual to maintain population diversity. Finally, the simulation experiments are performed on the ZDT\textbackslashtextbackslashDTLZ test functions and track planning problems. The results indicate the better performance of the improved algorithms.

Particle swarm optimization (PSO), as a kind of swarm intelligence algorithm, has the advantages of simple algorithm principle, less programmable parameters and easy programming. Many scholars have applied particle swarm optimization (PSO) to various fields through learning it, and successfully solved linear problems, nonlinear problems, multiobjective optimization and other problems. However, the algorithm also has obvious problems in solving problems, such as slow convergence speed, too early maturity, falling into local optimization in advance, etc., which makes the convergence speed slow, search the optimal value accuracy is not high, and the optimization effect is not ideal. Therefore, many scholars have improved the particle swarm optimization algorithm. Taking into account the improvement ideas proposed by scholars in the early stage and the shortcomings still existing in the improvement, this paper puts forward the idea of improving particle swarm optimization algorithm in the future.

In this thesis, an modified algorithm for grey wolf optimization in swarm intelligence optimization algorithm is proposed, which is called an adaptive grey wolf algorithm (AdGWO) based on population system and bacterial foraging optimization algorithm (BFO). In view of the disadvantages of premature convergence and local optimization in solving complex optimization problems, the AdGWO algorithm uses a three-stage nonlinear change function to simulate the decreasing change of the convergence factor, and at the same time integrates the half elimination mechanism of the BFO. These improvements are more in line with the actual situation of natural wolves. The algorithm is based on 23 famous test functions and compared with GWO. Experimental results demonstrate that this algorithm is able to avoid sinking into the local optimum, has good accuracy and stability, is a more competitive algorithm.

In this work, a new particle swarm optimization (PSO)-based optimization algorithm, and the idea of a running match is introduced and employed in a non-linear system PID controller design. This algorithm aims to modify the formula of velocity calculating of the general PSO method to increase the diversity of the searching process. In this process of designing an optimal PID controller for a non-linear system, the three gains of the PID controller form a particle, which is a parameter vector and will be updated iteratively. Many of those particles then form a population. To reach the PID gains which are optimum, using modified velocity updating formula and position updating formula, the position of all particles of the population will be moved into the optimization direction. In the meanwhile, an objective function may be minimized as the performance of the controller get improved. To corroborate the controller functioning of this method, a non-linear system known as inverted pendulum will be controlled by the designed PID controller. The results confirm that the new method can show excellent performance in the non-linear PID controller design task.

Groups often struggle to reach decisions, especially when populations are strongly divided by conflicting views. Traditional methods for collective decision-making involve polling individuals and aggregating results. In recent years, a new method called Artificial Swarm Intelligence (ASI) has been developed that enables networked human groups to deliberate in real-time systems, moderated by artificial intelligence algorithms. While traditional voting methods aggregate input provided by isolated participants, Swarm-based methods enable participants to influence each other and converge on solutions together. In this study we compare the output of traditional methods such as Majority vote and Borda count to the Swarm method on a set of divisive policy issues. We find that the rankings generated using ASI and the Borda Count methods are often rated as significantly more satisfactory than those generated by the Majority vote system (p\textbackslashtextless; 0.05). This result held for both the population that generated the rankings (the “in-group”) and the population that did not (the “out-group”): the in-group ranked the Swarm prioritizations as 9.6% more satisfactory than the Majority prioritizations, while the out-group ranked the Swarm prioritizations as 6.5% more satisfactory than the Majority prioritizations. This effect also held even when the out-group was subject to a demographic sampling bias of 10% (i.e. the out-group was composed of 10% more Labour voters than the in-group). The Swarm method was the only method to be perceived as more satisfactory to the “out-group” than the voting group.

With the increasing performance of neural networks on many security-critical tasks, the security concerns of machine learning have become increasingly prominent. Recent studies have shown that neural networks are vulnerable to adversarial examples: carefully crafted inputs with negligible perturbations on legitimate samples could mislead a neural network to produce adversary-selected outputs while humans can still correctly classify them. Therefore, we need an additional measurement on the trustworthiness of the results of a machine learning model, especially in adversarial settings. In this paper, we analyse the root cause of adversarial examples, and propose a new property, namely fidelity, of machine learning models to describe the gap between what a model learns and the ground truth learned by humans. One of its benefits is detecting adversarial attacks. We formally define fidelity, and propose a novel approach to quantify it. We evaluate the quantification of fidelity in adversarial settings on two neural networks. The study shows that involving the fidelity enables a neural network system to detect adversarial examples with true positive rate 97.7%, and false positive rate 1.67% on a studied neural network.

Genetic Programming Hyper-heuristic (GPHH) has been successfully applied to automatically evolve effective routing policies to solve the complex Uncertain Capacitated Arc Routing Problem (UCARP). However, GPHH typically ignores the interpretability of the evolved routing policies. As a result, GP-evolved routing policies are often very complex and hard to be understood and trusted by human users. In this paper, we aim to improve the interpretability of the GP-evolved routing policies. To this end, we propose a new Multi-Objective GP (MOGP) to optimise the performance and size simultaneously. A major issue here is that the size is much easier to be optimised than the performance, and the search tends to be biased to the small but poor routing policies. To address this issue, we propose a simple yet effective Two-Stage GPHH (TS-GPHH). In the first stage, only the performance is to be optimised. Then, in the second stage, both objectives are considered (using our new MOGP). The experimental results showed that TS-GPHH could obtain much smaller and more interpretable routing policies than the state-of-the-art single-objective GPHH, without deteriorating the performance. Compared with traditional MOGP, TS-GPHH can obtain a much better and more widespread Pareto front.

In open network environments, since there is no centralized authority to monitor misbehaving entities, malicious entities can easily cause the degradation of the service quality. Trust has become an important factor to ensure network security, which can help entities to distinguish good partners from bad ones. In this paper, trust in open network environment is regarded as a self-organizing system, using self-organization principle of human social trust propagation, a genetic trust evaluation method with self-optimization and family attributes is proposed. In this method, factors of trust evaluation include time, IP, behavior feedback and intuitive trust. Data structure of access record table and trust record table are designed to store the relationship between ancestor nodes and descendant nodes. A genetic trust search algorithm is designed by simulating the biological evolution process. Based on trust information of the current node's ancestors, heuristics generate randomly chromosome populations, whose structure includes time, IP address, behavior feedback and intuitive trust. Then crossover and mutation strategy is used to make the population evolutionary searching. According to the genetic searching termination condition, the optimal trust chromosome in the population is selected, and trust value of the chromosome is computed, which is the node's genetic trust evaluation result. The simulation result shows that the genetic trust evaluation method is effective, and trust evaluation process of the current node can be regarded as the process of searching for optimal trust results from the ancestor nodes' information. With increasing of ancestor nodes' genetic trust information, the trust evaluation result from genetic algorithm searching is more accurate, which can effectively solve the joint fraud problem.

The prevalence of social botnets has increased public distrust of social media networks. Current methods exist for detecting bot activity on Twitter, Reddit, Facebook, and other social media platforms. Most of these detection methods rely upon observing user behavior for a period of time. Unfortunately, the behavior observation period allows time for a botnet to successfully propagate one or many posts before removal. In this paper, we model the post propagation patterns of normal users and social botnets. We prove that a botnet may exploit deterministic propagation actions to elevate a post even with a small botnet population. We propose a probabilistic model which can limit the impact of social media botnets until they can be detected and removed. While our approach maintains expected results for non-coordinated activity, coordinated botnets will be detected before propagation with high probability.

As automatic speech recognition (ASR) systems have been integrated into a diverse set of devices around us in recent years, security vulnerabilities of them have become an increasing concern for the public. Existing studies have demonstrated that deep neural networks (DNNs), acting as the computation core of ASR systems, is vulnerable to deliberately designed adversarial attacks. Based on the gradient descent algorithm, existing studies have successfully generated adversarial samples which can disturb ASR systems and produce adversary-expected transcript texts designed by adversaries. Most of these research simulated white-box attacks which require knowledge of all the components in the targeted ASR systems. In this work, we propose the first semi-black-box attack against the ASR system - Kaldi. Requiring only partial information from Kaldi and none from DNN, we can embed malicious commands into a single audio chip based on the gradient-independent genetic algorithm. The crafted audio clip could be recognized as the embedded malicious commands by Kaldi and unnoticeable to humans in the meanwhile. Experiments show that our attack can achieve high attack success rate with unnoticeable perturbations to three types of audio clips (pop music, pure music, and human command) without the need of the underlying DNN model parameters and architecture.

The application of deep recurrent networks to audio transcription has led to impressive gains in automatic speech recognition (ASR) systems. Many have demonstrated that small adversarial perturbations can fool deep neural networks into incorrectly predicting a specified target with high confidence. Current work on fooling ASR systems have focused on white-box attacks, in which the model architecture and parameters are known. In this paper, we adopt a black-box approach to adversarial generation, combining the approaches of both genetic algorithms and gradient estimation to solve the task. We achieve a 89.25% targeted attack similarity, with 35% targeted attack success rate, after 3000 generations while maintaining 94.6% audio file similarity.

In order to protect individuals' privacy, data have to be "well-sanitized" before sharing them, i.e. one has to remove any personal information before sharing data. However, it is not always clear when data shall be deemed well-sanitized. In this paper, we argue that the evaluation of sanitized data should be based on whether the data allows the inference of sensitive information that is specific to an individual, instead of being centered around the concept of re-identification. We propose a framework to evaluate the effectiveness of different sanitization techniques on a given dataset by measuring how much an individual's record from the sanitized dataset influences the inference of his/her own sensitive attribute. Our intent is not to accurately predict any sensitive attribute but rather to measure the impact of a single record on the inference of sensitive information. We demonstrate our approach by sanitizing two real datasets in different privacy models and evaluate/compare each sanitized dataset in our framework.

The problem of fast items retrieval from a fixed collection is often encountered in most computer science areas, from operating system components to databases and user interfaces. We present an approach based on hash tables that focuses on both minimizing the number of comparisons performed during the search and minimizing the total collection size. The standard open-addressing double-hashing approach is improved with a non-linear transformation that can be parametrized in order to ensure a uniform distribution of the data in the hash table. The optimal parameter is determined using a genetic algorithm. The paper results show that near-perfect hashing is faster than binary search, yet uses less memory than perfect hashing, being a good choice for memory-constrained applications where search time is also critical.

This paper addresses analysis and design of multiple fault diagnosis for a class of Lipschitz nonlinear system. In order to automatically estimate multi-fault parameters efficiently, a new method of multi-fault diagnosis based on the differential evolution algorithm (DE) is proposed. Finally, a series of experiments validate the feasibility and effectiveness of the proposed method. The simulation show the high accuracy of the proposed strategies in multiple abrupt faults diagnosis.

To gain strategic insight into defending against the network reconnaissance stage of advanced persistent threats, we recreate the escalating competition between scans and deceptive views on a Software Defined Network (SDN). Our threat model presumes the defense is a deceptive network view unique for each node on the network. It can be configured in terms of the number of honeypots and subnets, as well as how real nodes are distributed across the subnets. It assumes attacks are NMAP ping scans that can be configured in terms of how many IP addresses are scanned and how they are visited. Higher performing defenses detect the scanner quicker while leaking as little information as possible while higher performing attacks are better at evading detection and discovering real nodes. By using Artificial Intelligence in the form of a competitive coevolutionary genetic algorithm, we can analyze the configurations of high performing static defenses and attacks versus their evolving adversary as well as the optimized configuration of the adversary itself. When attacks and defenses both evolve, we can observe that the extent of evolution influences the best configurations.