La seguridad informática siempre ha ido asociada a los avances tecnológicos, desde los años 60 protegiendo los sistemas de forma física impidiendo su acceso hasta nuestros días en los que se utiliza la inteligencia artificial para detectar comportamientos anómalos en redes, detectar malware o dar soporte a sistemas de acceso restringido. Según Cybercrime Magazine, el impacto económico que tendrá el cibercrimen en el mundo en cinco años (2020-2025) será de aproximadamente 10 billones de dólares anuales aumentando un 15% anual. Así mismo, se espera un gasto en defensa mayor con un incremento anual de un 13 %. La IA ha tomado un papel fundamental en la detección, protección y predicción de incidentes a medida que los modelos han ido mejorando y la potencia de sistemas de cómputo han permitido un volumen mayor de datos para el entrenamiento. Actualmente la inteligencia artificial está logrando resultados sorprendentes, podemos ver como ChatGPT es capaz de generar textos, responder a preguntas complejas o generar código de programación. Dall E 2 genera imágenes de alta resolución respondiendo a complejos prompts, desde dibujos animados a todo tipo de representaciones realistas. Existen otros tipos de modelos, quizás menos mediáticos, que también han alcanzado resultados notables. Los modelos de clasificación de imágenes, detección de malware, detección de objetos o identificación biométrica son algunos ejemplos. Si bien es cierto que la IA no está exenta de polémicas relacionadas con la protección de datos o sobre su uso ético. Numerosos países y organizaciones han mostrado sus preocupaciones de cómo estas tecnologías pueden afectar a las sociedades debido a los sesgos que introducen y sobre cómo pueden impactar en los derechos humanos. En este contexto de IA y seguridad se desarrolla esta tesis. Los modelos de inteligencia artificial son vulnerables a distintos tipos ataques como Data Poisoning, Model Stealing, Model Inversion o Adversarial Attacks. En este trabajo nos enfocamos en analizar las vulnerabilidades presentes en los modelos de clasificación de imágenes y concretamente en los ataques adversarios. Dichos ataques fueron descubiertos en 2013 y desde entonces se han desarrollado múltiples técnicas y tipos de algoritmos, así como defensas o métodos para crear modelos más robustos y resistentes. En esta tesis analizamos la capacidad de los ataques adversarios para afectar a un modelo de inteligencia artificial específico y luego transferir ese conocimiento para atacar con éxito a otros modelos similares.

<p>Despite the rapid development of adversarial attacks on machine learning models, many types of new adversarial examples remain unknown. Undiscovered types of adversarial attacks pose a</p><p>serious concern for the safety of the models, which raises the issue about the effectiveness of current adversarial robustness evaluation. Image semantic segmentation is a practical computer</p><p>vision task. However, segmentation networks’ robustness under adversarial attacks receives insufficient attention. Recently, machine learning researchers started to focus on generating</p><p>adversarial examples beyond the norm-bound restriction for segmentation neural networks. In this thesis, a simple and efficient method: AdvDRIT is proposed to synthesize unconstrained controllable adversarial images leveraging conditional-GAN. Simple CGAN yields poor image quality and low attack effectiveness. Instead, the DRIT (Disentangled Representation Image Translation) structure is leveraged with a well-designed loss function, which can generate valid adversarial images in one step. AdvDRIT is evaluated on two large image datasets: ADE20K and Cityscapes. Experiment results show that AdvDRIT can improve the quality of adversarial examples by decreasing the FID score down to 40% compared to state-of-the-art generative models such as Pix2Pix, and also improve the attack success rate 38% compared to other adversarial attack methods including PGD.</p>

Federated Learning (FL) has emerged as a promising approach for decentralized model training across multiple devices, while still preserving data privacy. Previous research has predominantly concentrated on classification tasks in FL settings, leaving  a noticeable gap in FL research specifically for regression models. This thesis addresses this gap by examining the vulnerabilities of Deep Neural Network (DNN) regression models within FL, with a specific emphasis on adversarial attacks. The primary objective is to examine the impact on model performance of two distinct adversarial attacks-output-flipping and random weights attacks. The investigation involves training FL models on three distinct data sets, engaging eight clients in the training process. The study varies the presence of malicious clients to understand how adversarial attacks influence model performance.  Results indicate that the output-flipping attack significantly decreases the model performance with involvement of at least two malicious clients. Meanwhile, the random weights attack demonstrates a substantial decrease even with just one malicious client out of the eight. It is crucial to note that this study's focus is on a theoretical level and does not explicitly account for real-world settings such as non-identically distributed (non-IID) settings,  extensive data sets, and a larger number of clients. In conclusion, this study contributes to the understanding of adversarial attacks in FL, specifically focusing on DNN regression models. The results highlights the importance of defending FL models against adversarial attacks, emphasizing the significance of future research in this domain.

Federated Learning (FL) has emerged as a promising approach for decentralized model training across multiple devices, while still preserving data privacy. Previous research has predominantly concentrated on classification tasks in FL settings, leaving  a noticeable gap in FL research specifically for regression models. This thesis addresses this gap by examining the vulnerabilities of Deep Neural Network (DNN) regression models within FL, with a specific emphasis on adversarial attacks. The primary objective is to examine the impact on model performance of two distinct adversarial attacks-output-flipping and random weights attacks. The investigation involves training FL models on three distinct data sets, engaging eight clients in the training process. The study varies the presence of malicious clients to understand how adversarial attacks influence model performance.  Results indicate that the output-flipping attack significantly decreases the model performance with involvement of at least two malicious clients. Meanwhile, the random weights attack demonstrates a substantial decrease even with just one malicious client out of the eight. It is crucial to note that this study's focus is on a theoretical level and does not explicitly account for real-world settings such as non-identically distributed (non-IID) settings,  extensive data sets, and a larger number of clients. In conclusion, this study contributes to the understanding of adversarial attacks in FL, specifically focusing on DNN regression models. The results highlights the importance of defending FL models against adversarial attacks, emphasizing the significance of future research in this domain.

Existing adversarial algorithms for Deep Reinforcement Learning (DRL) have largely focused on identifying an optimal time to attack a DRL agent. However, little work has been explored in injecting efficient adversarial perturbations in DRL environments. We propose a suite of novel DRL adversarial attacks, called ACADIA, representing AttaCks Against Deep reInforcement leArning. ACADIA provides a set of efficient and robust perturbation-based adversarial attacks to disturb the DRL agent's decision-making based on novel combinations of techniques utilizing momentum, ADAM optimizer (i.e., Root Mean Square Propagation or RMSProp), and initial randomization. These kinds of DRL attacks with novel integration of such techniques have not been studied in the existing Deep Neural Networks (DNNs) and DRL research. We consider two well-known DRL algorithms, Deep-Q Learning Network (DQN) and Proximal Policy Optimization (PPO), under Atari games and MuJoCo where both targeted and non-targeted attacks are considered with or without the state-of-the-art defenses in DRL (i.e., RADIAL and ATLA). Our results demonstrate that the proposed ACADIA outperforms existing gradient-based counterparts under a wide range of experimental settings. ACADIA is nine times faster than the state-of-the-art Carlini and Wagner (CW) method with better performance under defenses of DRL. / Master of Science / Artificial Intelligence (AI) techniques such as Deep Neural Networks (DNN) and Deep Reinforcement Learning (DRL) are prone to adversarial attacks. For example, a perturbed stop sign can force a self-driving car's AI algorithm to increase the speed rather than stop the vehicle. There has been little work developing attacks and defenses against DRL. In DRL, a DNN-based policy decides to take an action based on the observation of the environment and gets the reward in feedback for its improvements. We perturb that observation to attack the DRL agent. There are two main aspects to developing an attack on DRL. One aspect is to identify an optimal time to attack (when-to-attack?). The second aspect is to identify an efficient method to attack (how-to-attack?). To answer the second aspect, we propose a suite of novel DRL adversarial attacks, called ACADIA, representing AttaCks Against Deep reInforcement leArning. We consider two well-known DRL algorithms, Deep-Q Learning Network (DQN) and Proximal Policy Optimization (PPO), under DRL environments of Atari games and MuJoCo where both targeted and non-targeted attacks are considered with or without state-of-the-art defenses. Our results demonstrate that the proposed ACADIA outperforms state-of-the-art perturbation methods under a wide range of experimental settings. ACADIA is nine times faster than the state-of-the-art Carlini and Wagner (CW) method with better performance under the defenses of DRL.

This thesis tackles the problem of understanding deep neural network with piece- wise linear activation functions. We leverage tropical geometry, a relatively new field in algebraic geometry to characterize the decision boundaries of a single hidden layer neural network. This characterization is leveraged to understand, and reformulate three interesting applications related to deep neural network. First, we give a geo- metrical demonstration of the behaviour of the lottery ticket hypothesis. Moreover, we deploy the geometrical characterization of the decision boundaries to reformulate the network pruning problem. This new formulation aims to prune network pa- rameters that are not contributing to the geometrical representation of the decision boundaries. In addition, we propose a dual view of adversarial attack that tackles both designing perturbations to the input image, and the equivalent perturbation to the decision boundaries.

Cyber-physical systems, built from the integration of cyber and physical components, are being used in multiple domains ranging from manufacturing and healthcare to traffic con- trol and safety. Ensuring the security of cyber-physical systems is crucial because they provide the foundation of the critical infrastructure, and security incidents can result in catastrophic failures. Recent publications report that machine learning models are vul- nerable to adversarial examples, crafted by adding small perturbations to input data. For the past decade, machine learning security has become a growing interest area, with a significant number of systematic reviews and surveys that have been published. Secu- rity of artificial intelligence in cyber-physical systems is more challenging in comparison to machine learning security, because adversaries have a wider possible attack surface, in both cyber and physical domains. However, comprehensive systematic literature re- views in this research field are not available. Therefore, this work presents a systematic literature review of the adversarial attacks on artificial intelligence in cyber-physical sys- tems, examining 45 scientific papers, selected from 134 publications found in the Scopus database. It provides the classification of attack algorithms and defense methods, the sur- vey of evaluation metrics, an overview of the state of the art in methodologies and tools, and, as the main contribution, identifies open problems and research gaps and highlights future research challenges in this area of interest.

Indiana University-Purdue University Indianapolis (IUPUI) / Temporal point processes (TPP) are mathematical approaches for modeling asynchronous event sequences by considering the temporal dependency of each event on past events and its instantaneous rate. Temporal point processes can model various problems, from earthquake aftershocks, trade orders, gang violence, and reported crime patterns, to network analysis, infectious disease transmissions, and virus spread forecasting. In each of these cases, the entity’s behavior with the corresponding information is noted over time as an asynchronous event sequence, and the analysis is done using temporal point processes, which provides a means to define the generative mechanism of the sequence of events and ultimately predict events and investigate causality. Among point processes, Hawkes process as a stochastic point process is able to model a wide range of contagious and self-exciting patterns. One of Hawkes process’s well-known applications is predicting the evolution of viral processes on networks, which is an important problem in biology, the social sciences, and the study of the Internet. In existing works, mean-field analysis based upon degree distribution is used to predict viral spreading across networks of different types. However, it has been shown that degree distribution alone fails to predict the behavior of viruses on some real-world networks. Recent attempts have been made to use assortativity to address this shortcoming. This thesis illustrates how the evolution of such a viral process is sensitive to the underlying network’s structure. In Chapter 3 , we show that adding assortativity does not fully explain the variance in the spread of viruses for a number of real-world networks. We propose using the graphlet frequency distribution combined with assortativity to explain variations in the evolution of viral processes across networks with identical degree distribution. Using a data-driven approach, by coupling predictive modeling with viral process simulation on real-world networks, we show that simple regression models based on graphlet frequency distribution can explain over 95% of the variance in virality on networks with the same degree distribution but different network topologies. Our results highlight the importance of graphlets and identify a small collection of graphlets that may have the most significant influence over the viral processes on a network. Due to the flexibility and expressiveness of deep learning techniques, several neural network-based approaches have recently shown promise for modeling point process intensities. However, there is a lack of research on the possible adversarial attacks and the robustness of such models regarding adversarial attacks and natural shocks to systems. Furthermore, while neural point processes may outperform simpler parametric models on in-sample tests, how these models perform when encountering adversarial examples or sharp non-stationary trends remains unknown. In Chapter 4 , we propose several white-box and black-box adversarial attacks against deep temporal point processes. Additionally, we investigate the transferability of whitebox adversarial attacks against point processes modeled by deep neural networks, which are considered a more elevated risk. Extensive experiments confirm that neural point processes are vulnerable to adversarial attacks. Such a vulnerability is illustrated both in terms of predictive metrics and the effect of attacks on the underlying point process’s parameters. Expressly, adversarial attacks successfully transform the temporal Hawkes process regime from sub-critical to into a super-critical and manipulate the modeled parameters that is considered a risk against parametric modeling approaches. Additionally, we evaluate the vulnerability and performance of these models in the presence of non-stationary abrupt changes, using the crimes and Covid-19 pandemic dataset as an example. Considering the security vulnerability of deep-learning models, including deep temporal point processes, to adversarial attacks, it is essential to ensure the robustness of the deployed algorithms that is despite the success of deep learning techniques in modeling temporal point processes. In Chapter 5 , we study the robustness of deep temporal point processes against several proposed adversarial attacks from the adversarial defense viewpoint. Specifically, we investigate the effectiveness of adversarial training using universal adversarial samples in improving the robustness of the deep point processes. Additionally, we propose a general point process domain-adopted (GPDA) regularization, which is strictly applicable to temporal point processes, to reduce the effect of adversarial attacks and acquire an empirically robust model. In this approach, unlike other computationally expensive approaches, there is no need for additional back-propagation in the training step, and no further network isrequired. Ultimately, we propose an adversarial detection framework that has been trained in the Generative Adversarial Network (GAN) manner and solely on clean training data. Finally, in Chapter 6 , we discuss implications of the research and future research directions.

The great advance of communication technology comes with a rapid increase of disinformation in many kinds and shapes; manipulated images are one of the primary examples of disinformation that can affect many users. Such activity can severely impact public behavior, attitude, and belief or sway the viewers' perception in any malicious or benign direction. Additionally, adversarial attacks targeting deep learning models pose a severe risk to computer vision applications. This dissertation explores ways of detecting and resisting manipulated or adversarial attack images. The first contribution evaluates perceptual hashing (pHash) algorithms for detecting image manipulation on social media platforms like Facebook and Twitter. The study demonstrates the differences in image processing between the two platforms and proposes a new approach to find the optimal detection threshold for each algorithm. The next contribution develops a new pHash authentication to detect fake imagery on social media networks, using a self-supervised learning framework and contrastive loss. In addition, a fake image sample generator is developed to cover three major image manipulating operations (copy-move, splicing, removal). The proposed authentication technique outperforms the state-of-the-art pHash methods. The third contribution addresses the challenges of adversarial attacks to deep learning models. A new adversarial-aware deep learning system is proposed using a classical machine learning model as the secondary verification system to complement the primary deep learning model in image classification. The proposed approach outperforms current state-of-the-art adversarial defense systems. Finally, the fourth contribution fuses big data from Extra-Military resources to support military decision-making. The study proposes a workflow, reviews data availability, security, privacy, and integrity challenges, and suggests solutions. A demonstration of the proposed image authentication is introduced to prevent wrong decisions and increase integrity. Overall, the dissertation provides practical solutions for detecting manipulated and adversarial attack images and integrates our proposed solutions in supporting military decision-making workflow.

Recurrent neural networks (RNNs) have been widely used for supervised pattern recognition and exploring the underlying spatio-temporal correlation. However, due to the vanishing/exploding gradient problem, training a fully connected RNN in many cases is very difficult or even impossible. The difficulties of training traditional RNNs, led us to reservoir computing (RC) which recently attracted a lot of attention due to its simple training methods and fixed weights at its recurrent layer. There are three different categories of RC systems, namely, echo state networks (ESNs), liquid state machines (LSMs), and delayed feedback reservoirs (DFRs). In this dissertation a novel structure of RNNs which is inspired by dynamic delayed feedback loops is introduced. In the reservoir (recurrent) layer of DFR, only one neuron is required which makes DFRs extremely suitable for hardware implementations. The main motivation of this dissertation is to introduce an energy efficient, and easy to train RNN while this model achieves high performances in different tasks compared to the state-of-the-art. To improve the energy efficiency of our model, we propose to adopt spiking neurons as the information processing unit of DFR. Spiking neural networks (SNNs) are the most biologically plausible and energy efficient class of artificial neural networks (ANNs). The traditional analog ANNs have marginal similarity with the brain-like information processing. It is clear that the biological neurons communicate together through spikes. Therefore, artificial SNNs have been introduced to mimic the biological neurons. On the other hand, the hardware implementation of SNNs have shown to be extremely energy efficient. Towards achieving this overarching goal, this dissertation presents a spiking DFR (SDFR) with novel encoding schemes, and defense mechanisms against adversarial attacks. To verify the effectiveness and performance of the SDFR, it is adopted in three different applications where there exists a significant Spatio-temporal correlations. These three applications are attack detection in smart grids, spectrum sensing of multi-input-multi-output(MIMO)-orthogonal frequency division multiplexing (OFDM) Dynamic Spectrum Sharing (DSS) systems, and video-based face recognition. In this dissertation, the performance of SDFR is first verified in cyber attack detection in Smart grids. Smart grids are a new generation of power grids which guarantee a more reliable and efficient transmission and delivery of power to the costumers. A more reliable and efficient power generation and distribution can be realized through the integration of internet, telecommunication, and energy technologies. The convergence of different technologies, brings up opportunities, but the challenges are also inevitable. One of the major challenges that pose threat to the smart grids is cyber-attacks. A novel method is developed to detect false data injection (FDI) attacks in smart grids. The second novel application of SDFR is the spectrum sensing of MIMO-OFDM DSS systems. DSS is being implemented in the fifth generation of wireless communication systems (5G) to improve the spectrum efficiency. In a MIMO-OFDM system, not all the subcarriers are utilized simultaneously by the primary user (PU). Therefore, it is essential to sense the idle frequency bands and assign them to the secondary user (SU). The effectiveness of SDFR in capturing the spatio-temporal correlation of MIMO-OFDM time-series and predicting the availability of frequency bands in the future time slots is studied as well. In the third application, the SDFR is modified to be adopted in video-based face recognition. In this task, the SDFR is leveraged to recognize the identities of different subjects while they rotate their heads in different angles. Another contribution of this dissertation is to propose a novel encoding scheme of spiking neurons which is inspired by the cognitive studies of rats. For the first time, the multiplexing of multiple neural codes is introduced and it is shown that the robustness and resilience of the spiking neurons is increased against noisy data, and adversarial attacks, respectively. Adversarial attacks are small and imperceptible perturbations of the input data, which have shown to be able to fool deep learning (DL) models. So far, many adversarial attack and defense mechanisms have been introduced for DL models. Compromising the security and reliability of artificial intelligence (AI) systems is a major concern of government, industry and cyber-security researchers, in that insufficient protections can compromise the security and privacy of everyone in society. Finally, a defense mechanism to protect spiking neurons against adversarial attacks is introduced for the first time. In a nutshell, this dissertation presents a novel energy efficient deep spiking recurrent neural network which is inspired by delayed dynamic loops. The effectiveness of the introduced model is verified in several different applications. At the end, novel encoding and defense mechanisms are introduced which improve the robustness of the model against noise and adversarial attacks. / Doctor of Philosophy / The ultimate goal of artificial intelligence (AI) is to mimic the human brain. Artificial neural networks (ANN) are an attempt to realize that goal. However, traditional ANNs are very far from mimicking biological neurons. It is well-known that biological neurons communicate with one another through signals in the format of spikes. Therefore, artificial spiking neural networks (SNNs) have been introduced which behave more similarly to biological neurons. Moreover, SNNs are very energy efficient which makes them a suitable choice for hardware implementation of ANNs (neuromporphic computing). Despite the many benefits that are brought about by SNNs, they are still behind traditional ANNs in terms of performance. Therefore, in this dissertation, a new structure of SNNs is introduced which outperforms the traditional ANNs in three different applications. This new structure is inspired by delayed dynamic loops which exist in biological brains. The main objective of this novel structure is to capture the spatio-temporal correlation which exists in time-series while the training overhead and power consumption is reduced. Another contribution of this dissertation is to introduce novel encoding schemes for spiking neurons. It is clear that biological neurons leverage spikes, but the language that they use to communicate is not clear. Hence, the spikes require to be encoded in a certain language which is called neural spike encoding scheme. Inspired by the cognitive studies of rats, a novel encoding scheme is presented. Lastly, it is shown that the introduced encoding scheme increases the robustness of SNNs against noisy data and adversarial attacks. AI models including SNNs have shown to be vulnerable to adversarial attacks. Adversarial attacks are minor perturbations of the input data that can cause the AI model to misscalassify the data. For the first time, a defense mechanism is introduced which can protect SNNs against such attacks.