IEEE International Symposium onWomen in Services Computing (WISC 2023) 2023 WISC PhD Forum Sunday July 2Soldier Field The PhD Forum was started this year by Hadjer Benmeziane as a way for PhD students to get feedback on their work from peers and industry professionals. Date/Time Session PhD Forum Presentations Sunday 7/210:50 - 12:00Location: Soldier Field Presented during WISC 2 Efficient Hardware Implementation of AI AlgorithmsMeriem BeteyabArtificial Intelligence (AI) solutions are driving pioneering research and the development of next-generation technologies, offering improved robustness, reliability, security, logical thinking, and enhanced energy, data, and performance efficiency. However, the adoption of AI applications is hindered by computing and memory limitations, particularly for edge devices. Traditional technologies and architectures, used for the past four decades, struggle to meet the demands of the AI and big data eras. Researchers are now exploring novel technologies and hardware architectures to accelerate AI algorithms. One promising candidate is memristor-based in-memory computing, which overcomes the traditional von-Neumann bottleneck. Memristors integrate storage and computation into a single physical element, similar to biological systems. Their appealing features include non-volatile analog behavior, nano-scale size, fast read and write times, a high impedance ratio, and potential compatibility with CMOS technology. These characteristics make memristors suitable for common digital processing operations. The crossbar architecture of memristor arrays enables efficient parallel vector-matrix operations, commonly used in AI and signal processing algorithms, leading to potential energy, area, and execution time savings. This project aims to explore popular AI algorithms like Convolutional Neural Network (CNN), Transformer, and Retinex and map them onto efficient hardware implementations. The investigation will focus on data reuse, data preprocessing, and leveraging the memristor crossbar architecture. The goal is to achieve lower latency and higher efficiency while maintaining comparable accuracy to traditional methods. By unlocking the full potential of AI applications, this project contributes to the advancement of AI technologies in various domains. Securing Wireless Sensor Networks Using Blockchain and Machine LearningShereen Subhi IsmailAs an Internet of Things (IoT) technological key enabler, Wireless Sensor Networks (WSNs) are prone to different kinds of cyberattacks. WSNs have unique characteristics and several limitations which harden the design of effective attacks prevention and detection techniques. My work aims to provide a comprehensive understanding of the fundamental principles underlying cybersecurity in WSNs. In addition to current and envisioned solutions that are studied in detail, focusing primarily on state-of-the-art Machine Learning (ML) and Blockchain (BC) security techniques. Then, we propose integrating BC and ML towards developing a lightweight security framework that consists of two lines of defense, i.e., cyberattacks detection and prevention in WSNs, emphasizing their design insights and challenges. The work concludes by implementing a proposed integrated BC and ML solution highlighting potential underpinning BC and ML algorithms for less computationally demanding solution. Sunday 7/214:00 - 15:10Location: Soldier Field Presented during WISC 3 A Path Towards Fair and Equitable AI: Advancing Bias Mitigation in Federated LearningLynda FerragguigMachine learning is nowadays used in practically every domain to analyze data and guide the decision-making process. With the advent of big data, machine learning has evolved towards decentralized solutions to be more efficient.As technology advances, several regulations have been introduced with respect to data privacy, such as the GDPR.To address the security and privacy issues surrounding data, Google introduced federated learning in 2016 [McMahan et al., 2016]. FL is a promising approach for privacy preserving ML but also brings various challenges, notably bias and fairness in AI models. A Multi-tier Concurrent Privacy Preserving Federated Learning Architecture for Resource Constraint DevicesFatema SiddikaAs federated learning (FL) becomes more and more popular, it is common for a large number of FL processes to run concurrently. This paper studies concurrent FL in a hierarchical system where edge servers lie between edge devices and FL servers. One challenge in such a setting is the communication constraint at the edge, i.e., the limited uplink bandwidth from edge devices to edge servers. Thus, efficiently and fairly allocating the bandwidth to support simultaneous FL processes is an important problem. We propose a game-theoretic approach to model the bandwidth allocation problem as a Stackelberg Game and develop a heuristic scheme to find an approximate Nash Equilibrium of the game. Through experimentation, we demonstrate that our scheme efficiently and fairly assigns the uplink bandwidth to the FL processes, and it outperforms a baseline scheme where each edge server assigns bandwidth proportionally to the FL servers' requests that it receives. Sunday 7/215:25 - 16:35 Presented during WISC 4 Quality in Use Evaluation of Smart Environment Applications by Agent ApproachesMaria Paula Correa AngeloniSmart environments are characterized by the physical and virtual interaction between occupants and the built environment. In this paradigm, the user does not need to explicitly interact with applications (via touchscreen, mouse, etc.) but instead performs everyday activities (moving around, using objects, etc.) that the smart environment interprets as implicit inputs in order to provide proactive and relevant results. To ensure the adoption of these new applications, it is important to assess the quality of the interaction with users, or in other words, the Quality in Use (QinU). When these users are senior citizens or someone with disabilities, it is better to ensure the QinU before allowing them in such habitat. The objective of this thesis is to build an approach based on Artificial Intelligence (AI) techniques to measure the Quality in Use in smart environments by using agent approaches and data mining. Designing Complex Logistics Distribution Systems using Optimization ToolsAfaf AloullalDistribution systems play a vital role in delivering essential goods and fostering economic growth. To meet increasing demand, these systems must be efficient, robust, resilient, and sustainable. My work research focuses on designing distribution systems using optimization tools, specifically addressing hub location-routing problems. These problems involve the design of effective transportation and distribution networks to ship commodities from multiple nodes to multiple destinations. The selection of hub locations enables consolidation and redistribution of flow, resulting in cost savings and improved distribution times. While most literature focuses on hub location problems, the variant hub location-routing problem (HLRP) is gaining attention for its relevance in various applications. This research investigates time-dependent decisions in hub location-routing by partitioning the planning horizon into multiple periods. A four-phase matheuristic is developed, combining relax-and-fix, variable neighborhood descent, and local branching schemes. Three objective functions are tested, providing insights into cost choices, and the value of the multi-period solution is assessed. Additionally, uncertainty in flow is addressed through a chance constraints model. A comparison between deterministic counterparts and algorithms based on Monte Carlo Simulation and Sample Average Approximation (SAA) is conducted. A Variable Neighborhood Search (VNS) is developed to efficiently obtain high-quality solutions. Future work research involves extending the problem to incorporate multi-period settings into the stochastic hub location problem. See also: WISC 2023 Program PhD Forum Scholarships
2023 WISC PhD Forum Sunday July 2Soldier Field The PhD Forum was started this year by Hadjer Benmeziane as a way for PhD students to get feedback on their work from peers and industry professionals. Date/Time Session PhD Forum Presentations Sunday 7/210:50 - 12:00Location: Soldier Field Presented during WISC 2 Efficient Hardware Implementation of AI AlgorithmsMeriem BeteyabArtificial Intelligence (AI) solutions are driving pioneering research and the development of next-generation technologies, offering improved robustness, reliability, security, logical thinking, and enhanced energy, data, and performance efficiency. However, the adoption of AI applications is hindered by computing and memory limitations, particularly for edge devices. Traditional technologies and architectures, used for the past four decades, struggle to meet the demands of the AI and big data eras. Researchers are now exploring novel technologies and hardware architectures to accelerate AI algorithms. One promising candidate is memristor-based in-memory computing, which overcomes the traditional von-Neumann bottleneck. Memristors integrate storage and computation into a single physical element, similar to biological systems. Their appealing features include non-volatile analog behavior, nano-scale size, fast read and write times, a high impedance ratio, and potential compatibility with CMOS technology. These characteristics make memristors suitable for common digital processing operations. The crossbar architecture of memristor arrays enables efficient parallel vector-matrix operations, commonly used in AI and signal processing algorithms, leading to potential energy, area, and execution time savings. This project aims to explore popular AI algorithms like Convolutional Neural Network (CNN), Transformer, and Retinex and map them onto efficient hardware implementations. The investigation will focus on data reuse, data preprocessing, and leveraging the memristor crossbar architecture. The goal is to achieve lower latency and higher efficiency while maintaining comparable accuracy to traditional methods. By unlocking the full potential of AI applications, this project contributes to the advancement of AI technologies in various domains. Securing Wireless Sensor Networks Using Blockchain and Machine LearningShereen Subhi IsmailAs an Internet of Things (IoT) technological key enabler, Wireless Sensor Networks (WSNs) are prone to different kinds of cyberattacks. WSNs have unique characteristics and several limitations which harden the design of effective attacks prevention and detection techniques. My work aims to provide a comprehensive understanding of the fundamental principles underlying cybersecurity in WSNs. In addition to current and envisioned solutions that are studied in detail, focusing primarily on state-of-the-art Machine Learning (ML) and Blockchain (BC) security techniques. Then, we propose integrating BC and ML towards developing a lightweight security framework that consists of two lines of defense, i.e., cyberattacks detection and prevention in WSNs, emphasizing their design insights and challenges. The work concludes by implementing a proposed integrated BC and ML solution highlighting potential underpinning BC and ML algorithms for less computationally demanding solution. Sunday 7/214:00 - 15:10Location: Soldier Field Presented during WISC 3 A Path Towards Fair and Equitable AI: Advancing Bias Mitigation in Federated LearningLynda FerragguigMachine learning is nowadays used in practically every domain to analyze data and guide the decision-making process. With the advent of big data, machine learning has evolved towards decentralized solutions to be more efficient.As technology advances, several regulations have been introduced with respect to data privacy, such as the GDPR.To address the security and privacy issues surrounding data, Google introduced federated learning in 2016 [McMahan et al., 2016]. FL is a promising approach for privacy preserving ML but also brings various challenges, notably bias and fairness in AI models. A Multi-tier Concurrent Privacy Preserving Federated Learning Architecture for Resource Constraint DevicesFatema SiddikaAs federated learning (FL) becomes more and more popular, it is common for a large number of FL processes to run concurrently. This paper studies concurrent FL in a hierarchical system where edge servers lie between edge devices and FL servers. One challenge in such a setting is the communication constraint at the edge, i.e., the limited uplink bandwidth from edge devices to edge servers. Thus, efficiently and fairly allocating the bandwidth to support simultaneous FL processes is an important problem. We propose a game-theoretic approach to model the bandwidth allocation problem as a Stackelberg Game and develop a heuristic scheme to find an approximate Nash Equilibrium of the game. Through experimentation, we demonstrate that our scheme efficiently and fairly assigns the uplink bandwidth to the FL processes, and it outperforms a baseline scheme where each edge server assigns bandwidth proportionally to the FL servers' requests that it receives. Sunday 7/215:25 - 16:35 Presented during WISC 4 Quality in Use Evaluation of Smart Environment Applications by Agent ApproachesMaria Paula Correa AngeloniSmart environments are characterized by the physical and virtual interaction between occupants and the built environment. In this paradigm, the user does not need to explicitly interact with applications (via touchscreen, mouse, etc.) but instead performs everyday activities (moving around, using objects, etc.) that the smart environment interprets as implicit inputs in order to provide proactive and relevant results. To ensure the adoption of these new applications, it is important to assess the quality of the interaction with users, or in other words, the Quality in Use (QinU). When these users are senior citizens or someone with disabilities, it is better to ensure the QinU before allowing them in such habitat. The objective of this thesis is to build an approach based on Artificial Intelligence (AI) techniques to measure the Quality in Use in smart environments by using agent approaches and data mining. Designing Complex Logistics Distribution Systems using Optimization ToolsAfaf AloullalDistribution systems play a vital role in delivering essential goods and fostering economic growth. To meet increasing demand, these systems must be efficient, robust, resilient, and sustainable. My work research focuses on designing distribution systems using optimization tools, specifically addressing hub location-routing problems. These problems involve the design of effective transportation and distribution networks to ship commodities from multiple nodes to multiple destinations. The selection of hub locations enables consolidation and redistribution of flow, resulting in cost savings and improved distribution times. While most literature focuses on hub location problems, the variant hub location-routing problem (HLRP) is gaining attention for its relevance in various applications. This research investigates time-dependent decisions in hub location-routing by partitioning the planning horizon into multiple periods. A four-phase matheuristic is developed, combining relax-and-fix, variable neighborhood descent, and local branching schemes. Three objective functions are tested, providing insights into cost choices, and the value of the multi-period solution is assessed. Additionally, uncertainty in flow is addressed through a chance constraints model. A comparison between deterministic counterparts and algorithms based on Monte Carlo Simulation and Sample Average Approximation (SAA) is conducted. A Variable Neighborhood Search (VNS) is developed to efficiently obtain high-quality solutions. Future work research involves extending the problem to incorporate multi-period settings into the stochastic hub location problem. See also: WISC 2023 Program PhD Forum Scholarships