Dr. Achraf Cohen

Modern foundational models are often compressed via a combination of structured pruning and re-training to meet the strict compute, memory, and connectivity constraints of edge deployments. While state-of-the-art (SoTA) pruning schemes target the entire Transformer, we adopt a simple, layer-wise L 2 -norm pruning on only the multi-layer perceptron (MLP) blocks as a fixed baseline. Our focus is not on achieving maximal compression, but on isolating the impact of the re-training loss function: (i) L2-norm Pruning with Cross-Entropy Fine-Tuning (L2PFT), which relies on labeled data, versus (ii) L2-norm Pruning with KL-Divergence Self-Distillation (L2PSD), which utilizes only teacher logits without requiring labeled data. We evaluate both pipelines on the OLMo2-7B-SFT model for CommonsenseQA, suitable for intermittent or denied connectivity scenarios typical of edge networks. Under identical pruning schedules, L2PSD achieves comparable or superior test accuracy to L2PFT, indicating that the choice of loss function has a significant impact on compressed model recovery in resource-constrained environments.

Transfer learning is commonly utilized in various fields such as computer vision, natural language processing, and medical imaging due to its impressive capability to address sub-tasks and work with different datasets. However, its application in cybersecurity has not been thoroughly explored. In this paper, we present an innovative neurosymbolic AI framework designed for network intrusion detection systems, which play a crucial role in combating malicious activities in cybersecurity. Our framework leverages transfer learning and uncertainty quantification. The findings indicate that transfer learning models, trained on large and well-structured datasets, outperform neural-based models that rely on smaller datasets, paving the way for a new era in cybersecurity solutions.

Academic advising plays a crucial role in fostering student success. However, challenges such as limited advisor availability can hinder effective support. Generative AI, particularly AI-powered chatbots, offers the potential to enhance student advising in higher education by providing personalized guidance. These technologies help college students find the information and resources needed to create degree plans aligned with their academic goals. This research introduces ARGObot, an intelligent advising system that facilitates student navigation of university policies through automated interpretation of the student handbook as its primary knowledge base. ARGObot enhances accessibility to critical academic policies and procedures, supporting incoming students' success through personalized guidance. Our system integrates a multifunctional agent enhanced by a Large Language Model (LLM). The architecture employs multiple external tools to enhance its capabilities: a Retrieval-Augmented Generation (RAG) system accesses verified university sources; email integration facilitates Human-in-the-Loop (HITL) interaction; and a web search function expands the system's knowledge base beyond predefined constraints. This approach enables the system to provide contextually relevant and verified responses to various student queries. This architecture evolved from our initial implementation based on Gemini 1 Pro, which revealed significant limitations due to its lack of agent-based functionality, resulting in hallucination issues and irrelevant responses. Subsequent evaluation demonstrated that our enhanced version, integrating GPT-4 with the text-embedding-ada-002 model, achieved superior performance across all metrics. This paper also presents a comparative analysis of both implementations, highlighting the architectural improvements and their impact on system performance.

IoT systems such as smart homes or cities capture large amounts of data in real time using a wide range of interconnected sensor devices linked to edge nodes and their associated cloud services. In such systems, sensors must operate continuously and reliably in their deployed environment to deliver meaningful data to the users. As time passes, sensors may fail or become less effective in collecting raw data, resulting in erroneous data being delivered to the end user, who relies on their accuracy for decision-making. In this paper, we study detection methods of anomalies in sensor data based on Hotelling's T^{2} statistic. In simulation experiments, we calculate the lowest possible shift in data streams that are detectable by our proposed methods using sensor data from a smart home. Our results show that the proposed method successfully identifies the low shift in the means, given a confidence level with an accuracy of 90%. These results are based on data and do not assume a specific distribution, which opens up new potential applications for IoT data and anomaly detection. Additionally, some low shifts can be difficult to detect visually, but our proposed method will be able to detect those low shifts.

This comprehensive study delves into the examination and application of novel statistical distributions, namely the Ristić-Balakrishnan-Topp-Leone-Exponentiated half Logistic-G (RB-TL-EHL-G) family of distributions, emphasizing their paramount importance in reliability and medical data modeling. We meticulously explore a multitude of this family of novel distributions, accentuating their respective features, properties, and real-world applicability. The probability density, the cumulative distribution, the hazard rate, and the quantile functions are provided. The density functions of the RB-TL-EHL-G family are expanded, enabling a deeper understanding of their statistical properties, including various moments, generating functions, order statistics, stochastic orderings, probability weighted moments, and the Rényi entropy. A significant portion of the investigation is dedicated to the intensive analysis of various data sets, to which these distributions are fitted, unveiling noteworthy insights into their behavior and performance. Furthermore, the discussions extend to a comparative study, delineating the advantages and limitations of each distribution, fostering a deeper understanding and selection criteria for practitioners.

This paper presents a new methodology for generating continuous statistical distributions, integrating the exponentiated odds ratio within the framework of survival analysis. This new method enhances the flexibility and adaptability of distribution models to effectively address the complexities inherent in contemporary datasets. The core of this advancement is illustrated by introducing a particular subfamily, the “Type 2 Gumbel Weibull-G family of distributions”. We provide a comprehensive analysis of the mathematical properties of these distributions, including statistical properties such as density functions, moments, hazard rate and quantile functions, Rényi entropy, order statistics, and the concept of stochastic ordering. To test the robustness of our new model, we apply five distinct methods for parameter estimation. The practical applicability of the Type 2 Gumbel Weibull-G distributions is further supported through the analysis of three real-world datasets. These real-life applications illustrate the exceptional statistical precision of our distributions compared to existing models, thereby reinforcing their significant value in both theoretical and practical statistical applications.

In this paper, we investigate the waiving of GRE/GMAT for admission to master’s degree programs in a state university in Florida, USA. Standardized tests, such as GRE/GMAT, were required for admission to the master’s degree programs in 2019/2020, waived in 2020/2021, and removed or modified in 2021/2022. We analyzed the application, enrollment, and performance data to assess the impact of these changes. The data show that the number of applicants and enrolled students exhibit an upward trend from 2019 to 2021. The undergraduate GPA of new applicants who did not submit the GRE in 2021 tends to be statistically significantly higher than for those who did submit the GRE in 2019 (p < 0.001). The new students’ first-semester graduate GPA in 2021 (no GRE requirement) tends also to be statistically significantly higher than the new students’ first-semester graduate GPA in 2019 (GRE requirement) (p< 0.01). The study employed random forest feature importance using the Gini index to analyze the predictive power of GRE and undergraduate GPA for forecasting first-semester graduate GPA. The results show that undergraduate GPA is a more significant factor than GRE. Overall, the study’s statistical evidence indicates that waiving GRE/GMAT requirements for master’s degree programs did not affect applicants’ performance, as measured by their undergraduate GPA, nor did it lead to a decline in student performance, as measured by first-semester graduate GPA.

The objective of this study was to evaluate the impact of three different methods of handling missing data on the performance of Phase I Hotelling’s T² multivariate control chart. Using a Monte Carlo simulation, we studied the average, median, and standard deviation of the run length performance of multivariate data imputed using mean substitution, regression imputation, and predictive mean matching at three different levels of missingness (1%, 10%, and 25%) and three levels of variable correlation coefficients (0.2, 0.4, and 0.8). We found that predictive mean matching has average run length performance results comparable to that of the complete in-control data set at all levels of missingness and variable correlation, while the performance of mean substitution was adversely affected by high levels of missingness and by strong variable correlation. Based on the simulation (multivariate normal data), we concluded that predictive mean matching is superior to both regression imputation and mean substitution as a method for imputing missing values for the analysis of Phase I Hotelling’s T² control chart. Two applications were presented using the Altenrhein wastewater treatment plant and Olive oil datasets.

This paper introduces the Lomax-exponentiated odds ratio–G (L-EOR–G) distribution, a novel framework designed to adeptly navigate the complexities of modern datasets. It blends theoretical rigor with practical application to surpass the limitations of traditional models in capturing complex data attributes such as heavy tails, shaped curves, and multimodality. Through a comprehensive examination of its theoretical foundations and empirical data analysis, this study lays down a systematic theoretical framework by detailing its statistical properties and validates the distribution’s efficacy and robustness in parameter estimation via Monte Carlo simulations. Empirical evidence from real-world datasets further demonstrates the distribution’s superior modeling capabilities, supported by compelling various goodness-of-fit tests. The convergence of theoretical precision and practical utility heralds the L-EOR–G distribution as a groundbreaking advancement in statistical modeling, significantly enhancing precision and adaptability. The new model not only addresses a critical need within statistical modeling but also opens avenues for future research, including the development of more sophisticated estimation methods and the adaptation of the model for various data types, thereby promising to refine statistical analysis and interpretation across a wide array of disciplines.

Background: It is theorized that the lack of a synovial lining after anterior cruciate ligament (ACL) injury and ACL reconstruction (ACLR) contributes to slow ligamentization and possible graft failure. Whether graft maturation and incorporation can be improved with the use of a scaffold requires investigation. Purpose: To evaluate the safety and efficacy of wrapping an ACL autograft with an amnion collagen matrix and injecting bone marrow aspirate concentrate (BMAC), quantify the cellular content of the BMAC samples, and assess 2-year postoperative patient-reported outcomes. Study Design: Randomized controlled trial; Level of evidence, 2. Methods: A total of 40 patients aged 18 to 35 years who were scheduled to undergo ACLR were enrolled in a prospective single-blinded randomized controlled trial with 2 arms based on graft type: bone–patellar tendon–bone (BTB; n = 20) or hamstring (HS; n = 20). Participants in each arm were randomized into a control group who underwent standard ACLR or an intervention group who had their grafts wrapped with an amnion collagen matrix during graft preparation, after which BMAC was injected under the wrap layers after implantation. Postoperative magnetic resonance imaging (MRI) mapping/processing yielded mean T2* relaxation time and graft volume values at 3, 6, 9, and 12 months. Participants completed the Single Assessment Numeric Evaluation Score, Knee injury and Osteoarthritis Outcome Score, and pain visual analog scale. Statistical linear mixed-effects models were used to quantify the effects over time and the differences between the control and intervention groups. Adverse events were also recorded. Results: No significant differences were found at any time point between the intervention and control groups for BTB T2* (95% CI, –1.89 to 0.63; P = .31), BTB graft volume (95% CI, –606 to 876.1; P = .71), HS T2* (95% CI, –2.17 to 0.39; P = .162), or HS graft volume (95% CI, –11,141.1 to 351.5; P = .28). No significant differences were observed between the intervention and control groups of either graft type on any patient-reported outcome measure. No adverse events were reported after a 2-year follow-up. Conclusion: In this pilot study, wrapping a graft with an amnion collagen matrix and injecting BMAC appeared safe. MRI T2* values and graft volume of the augmented ACL graft were not significantly different from that of controls, suggesting that the intervention did not result in improved graft maturation. Registration: NCT03294759 (ClinicalTrials.gov identifier).