Dr. Behrouz Zolfaghari

Diffusion models are generative models that generate images, videos, and audio through learning samples and have attracted attention in recent years. In this paper, we investigate whether a diffusion model is resistant to membership inference attacks, which evaluate the privacy leakage of a machine learning model. We primarily discuss the diffusion model from the standpoints of comparison with a generative adversarial network (GAN) as a conventional model and hyperparameters unique to the diffusion model, such as timesteps, sampling steps, and sampling variances. We conduct extensive experiments with the denoising diffusion implicit model (DDIM) as a diffusion model and the deep convolutional GAN (DCGAN) as a GAN on the CelebA and CIFAR-10 datasets in both white-box and black-box settings and then show that the diffusion model is comparable to GAN in terms of resistance to membership inference attacks. Next, we demonstrate that the impact of timesteps is significant and that the intermediate steps in a noise schedule the most vulnerable to the attack. We also found two key insights through further analysis. First, we identify that DDIM is more vulnerable to the attack when trained with fewer samples even though it achieves lower Frechet inception distance scores than DCGAN. Second, sampling steps in hyperparameters are important for resistance to the attack, whereas the impact of sampling variances is negligible.

Purpose: To evaluate the impact of statin therapy on warfarin dose requirements in diabetic patients and to assess the performance of various machine learning algorithms in predicting optimal warfarin dosing.
Methods: The datasets available for total participants of 628 (216 diabetics and 412 non-diabetic patients) were analyzed. We categorized the patients according to height, weight, gender, race, and age, plasma international normalized ratio (INR) on reported therapeutic dose of warfarin, target INR, warfarin dose, statin therapy, and indications for warfarin. Various models were tested on data of patients from the International Warfarin Pharmacogenetics Consortium (IWPC). Data preprocessing involves structuring and handling missing values. Six predictive models, including least absolute shrinkage and selection operator (LASSO), k-nearest neighbors (KNN), support vector regression (SVR), linear regression (LR), decision tree, and random forest (RF), were employed in predicting optimal warfarin dosage. The best dose for each patient will be predicted using one of the six regression models.
Results: This comparative study showed that the mean (and the standard deviation) of warfarin dose for diabetic and non-diabetic patients were 38.73 (15.37) and 34.50 (18.27) mg per week, respectively. Furthermore, the impact of various statin they use is considered and patient undergoing atorvastatin and rosuvastatin therapy against the necessity of high dose warfarin if the diabetic patients use lovastatin and fluvastatin.
Conclusion: Diabetic patients under statin therapy, considering the specific statin used, require different warfarin dose. Through the application of advanced machine learning, models as dosing predictors may attenuate the adverse effects of warfarin.

In recent years, we have witnessed the era of IoT. Extensions of the IoT are found almost everywhere in the modern world: under the ground, on the ground, under the sea, in the sky, and in space. Such a rapid proliferation has given rise to a variety of requirements and challenges. As suggested by the recent literature, security is the most critical challenge in this area. A comprehensive survey in this area can pave the way for further research by highlighting current trends and shedding light on less-studied aspects of the area. This paper provides a comprehensive review of the current state of research on IoT extensions, with a focus on security. We start with reviewing existing relevant surveys, noting their shortcomings. We highlight the lack of inclusiveness in existing surveys. Moreover, we show that these surveys do not look closely at security challenges and fail to develop a taxonomy or a solid future roadmap. Then, we provide an overview of the security challenges and mechanisms of IoT extensions. We proceed to develop a taxonomy of these extensions with a focus on security. Lastly, we discuss what the future may hold for IoT extensions, given the role of artificial intelligence in IoT and the advancements of artificial intelligence on the horizon.

UAVs have found their applications in numerous applications from recreational activities to business in addition to military and strategic fields. However, research on UAVs is not going on as quickly as the technology. Especially, when it comes to the security of these devices, the academia is lagging behind the industry. This gap motivates our work in this article as a stepping stone for future research in this area. A comprehensive survey on the security of UAVs and UAV-based systems can help the research community keep pace with, or even lead the industry. Although there are several reviews on UAVs or related areas, there is no recent survey broadly covering various aspects of security. Moreover, none of the existing surveys highlights current and future trends with a focus on the role of an omnipresent technology such as AI. This article endeavors to overcome these shortcomings. We conduct a comprehensive review on security challenges of UAVs as well as the related security controls. Then we develop a future roadmap for research in this area with a focus on the role of AI. The future roadmap is established based on the identified current trends, under-researched topics, and a future look-ahead.

This book studies the intersection between cryptography and AI, highlighting the significant cross-impact and potential between the two technologies. The authors first study the individual ecosystems of cryptography and AI to show the omnipresence of each technology in the ecosystem of the other one. Next, they show how these technologies have come together in collaborative or adversarial ways. In the next section, the authors highlight the coevolution being formed between cryptography and AI. Throughout the book, the authors use evidence from state-of-the-art research to look ahead at the future of the crypto-AI dichotomy. The authors explain how they anticipate that quantum computing will join the dichotomy in near future, augmenting it to a trichotomy. They verify this through two case studies highlighting another scenario wherein crypto, AI and quantum converge. The authors study current trends in chaotic image encryption as well as information-theoretic cryptography and show how these trends lean towards quantum-inspired artificial intelligence (QiAI). After concluding the discussions, the authors suggest future research for interested researchers. -the publishers

Ensemble methods try to improve performance via integrating different kinds of input data, features, or learning algorithms. In addition to other areas, they are finding their applications in cancer prognosis and diagnosis. However, in this area, the research community is lagging behind the technology. A systematic review along with a taxonomy on ensemble methods used in cancer prognosis and diagnosis can pave the way for the research community to keep pace with the technology and even lead trend. In this article, we first present an overview on existing relevant surveys and highlight their shortcomings, which raise the need for a new survey focusing on Ensemble Classifiers (ECs) used for the diagnosis and prognosis of different cancer types. Then, we exhaustively review the existing methods, including the traditional ones as well as those based on deep learning. The review leads to a taxonomy as well as the identification of the best-studied cancer types, the best ensemble methods used for the related purposes, the prevailing input data types, the most common decision-making strategies, and the common evaluating methodologies. Moreover, we establish future directions for researchers interested in following existing research trends or working on less-studied aspects of the area.

A large amount of lignocellulosic waste is generated every day in the world, and their accumulation in the agroecosystems, integration in soil compositions, or incineration for energy production has severe environmental pollution effects. Using enzymes as biocatalysts for the biodegradation of lignocellulosic materials, especially in harsh processing conditions, is a practical step towards green energy and environmental biosafety. Hence, the current study focuses on enzyme computationally screened from camel rumen metagenomics data as specialized microbiota that have the capacity to degrade lignocellulosic-rich and recalcitrant materials. The novel hyperthermostable xylanase named PersiXyn10 with the performance at extreme conditions was proper activity within a broad temperature (30-100 degrees C) and pH range (4.0-11.0) but showed the maximum xylanolytic activity in severe alkaline and temperature conditions, pH 8.0 and temperature 90 degrees C. Also, the enzyme had highly resistant to metals, surfactants, and organic solvents in optimal conditions. The introduced xylanase had unique properties in terms of thermal stability by maintaining over 82% of its activity after 15 days of incubation at 90 degrees C. Considering the crucial role of hyperthermostable xylanases in the paper industry, the PersiXyn10 was subjected to biodegradation of paper pulp. The proper performance of hyperthermostable PersiXyn10 on the paper pulp was confirmed by structural analysis (SEM and FTIR) and produced 31.64 g/L of reducing sugar after 144 h hydrolysis. These results proved the applicability of the hyperthermostable xylanase in biobleaching and saccharification of lignocellulosic biomass for declining the environmental hazards.

Discrete logarithmic pseudorandom number generators are a prevailing class of cryptographically-secure pseudorandom number generators (CSPRNGs). In generators of this type, the security parameter affects both security and performance. This adds to the design complexity via creating a critical tradeoff between security and performance. This research is an attempt at shifting the security-performance tradeoff paradigm in this realm. To this end, we propose a modification to Gennaro’s pseudorandom number generator via replacing word-wise arithmetic operations with bit-wise logical operations in trapdoor and hard-core functions. The security of our generator (like that of Gennaro’s) is based on the hardness of a special variant of the discrete logarithm problem. We establish an equivalence between the specific variant of the discrete logarithm problem with the standard problem. Moreover, we demonstrate that in the modified generator, performance will be almost independent of the security parameter as logical operations can be performed in register level without the interference of the Arithmetic-Logic Unit (ALU). This relaxes the security-performance tradeoff and allows designers to maneuver more flexibly in the tradeoff space. We implement and evaluate our proposed generator and prove its security. Our CSPRNG is deemed random by all randomness tests in NIST SP 800-22 suite.

Some enzymes can catalyze more than one chemical conversion for which they are physiologically specialized. This secondary function, which is called underground, promiscuous, metabolism, or cross activity, is recognized as a valuable feature and has received much attention for developing new catalytic functions in industrial applications. In this study, a novel bifunctional xylanase/β-glucosidase metagenomic-derived enzyme, PersiBGLXyn1, with underground β-glucosidase activity was mined by in-silico screening. Then, the corresponding gene was cloned, expressed and purified. The PersiBGLXyn1 improved the degradation efficiency of organic solvent pretreated coffee residue waste (CRW), and subsequently the production of bioethanol during a separate enzymatic hydrolysis and fermentation (SHF) process. After characterization, the enzyme was immobilized on a nanocellulose (NC) carrier generated from sugar beet pulp (SBP), which remarkably improved the underground activity of the enzyme up to four-fold at 80°C and up to two-fold at pH 4.0 compared to the free one. The immobilized PersiBGLXyn1 demonstrated 12 to 13-fold rise in half-life at 70 and 80°C for its underground activity. The amount of reducing sugar produced from enzymatic saccharification of the CRW was also enhanced from 12.97 g/l to 19.69 g/l by immobilization of the enzyme. Bioethanol production was 29.31 g/l for free enzyme after 72 h fermentation, while the immobilized PersiBGLXyn1 showed 51.47 g/l production titre. Overall, this study presented a cost-effective in-silico metagenomic approach to identify novel bifunctional xylanase/β-glucosidase enzyme with underground β-glucosidase activity. It also demonstrated the improved efficacy of the underground activities of the bifunctional enzyme as a promising alternative for fermentable sugars production and subsequent value-added products.

Protecting widely used deep classifiers against black-box adversarial attacks is a recent research challenge in many security-related areas, including malware classification. This class of attacks relies on optimizing a sequence of highly similar queries to bypass given classifiers. In this article, we leverage this property and propose a history-based method named, stateful query analysis (SQA) , which analyzes sequences of queries received by a malware classifier to detect black-box adversarial attacks on an industrial Internet of Things (IIoT). In the SQA pipeline, there are two components, namely the similarity encoder and the classifier, both based on convolutional neural networks. Unlike the state-of-the-art methods, which aim to identify individual adversarial examples, tracking the history of queries allows our method to identify adversarial scenarios and abort attacks before their completion. We optimize SQA using different combinations of hyperparameters on an advanced risc machine (ARM)-based IIoT malware dataset, widely adopted for malware threat hunting in industry 4.0. The use of a novel distance metric in calculating the loss function of the similarity encoder results in more disentangled representations and improves the performance of our method. Our evaluations demonstrate the validity of SQA via a detection rate of 93.1% over a wide range of adversarial examples.