American University of Beirut

Talk Abstracts

​​​​​​​​​​​​​April 22, 2024


Track 1: Truths, Lies, and GenAI: Data Fabrication and Detection

  • Lama Nachman on Fortifying Trust in the Age of GenAI: Deep Fake Detection and Safeguarding Personal Content
    Talk Abstract:
    Deepfakes have emerged as a double-edged sword in the modern digital landscape and continue to improve due to the increasing capabilities of generative AI. While they offer unprecedented opportunities for creative expression and entertainment, their potential for misuse threatens to erode trust, manipulate public opinion, and sow confusion on a global scale. As these hyper-realistic manipulations become increasingly accessible and indistinguishable from genuine content, society faces profound ethical, legal, and societal implications, raising urgent questions about privacy, consent, and the very nature of reality in the digital age. In this keynote, I will discuss some of our research in Responsible AI and improving trust in Human/AI systems. I will explore different approaches, key learnings and results for improving deep fake detection focused on measuring human authentic signals.  I will also discuss some preliminary research on protecting user content from unconsented utilization by diffusion models.  

  • Sibel Adali on A Community and Trust Frame for Enhancing Misinformation Analysis​

    Talk Abstract:
    In this talk, Dr. Sibel Adali will provide an overview of the work her group has been conducting on understanding information credibility and its correlation with the veracity of information, particularly within news and social media domains. The dissemination of false and misleading news online can yield significant offline consequences across various domains such as health, public opinion, and safety. The challenges associated with combating misinformation using language analysis methods will be discussed, with emphasis on the potential exacerbation of these challenges due to the rising popularity of large language models. Dr. Adali will then present community-based analysis methods that offer substantial enhancements to language-based tools, offering valuable insights for both users and system developers alike.

  • ​Roaa Al-Feel on Data Science for Social Impact, Policy Development, and Media Analysis​

    ​Talk Abstract:
    In today's digital age, the rampant spread of misinformation presents a challenge to society, especially in conflict zones where reliable information is paramount. This presentation explores the convergence of data science, machine learning, and societal welfare through a case study centered on the Syrian conflict. Journey through the innovative application of data science and machine learning techniques in identifying and combating fake news during the war. Delve into the indispensable role of data-driven analysis in shaping policy decisions and media scrutiny. Join me as we unveil the transformative potential of data science in advancing social good and upholding media integrity in conflict-ridden regions.

  • ​​Maryleen Amaizu on Building Trust and Combating Misinformation with AI in Crises

    Talk Abstract:
    Humanitarian crises are breeding grounds for spread of misinformation.  This talk explores the complex relationship between trust, fake news, and the role of GenAI in navigating this minefield. Drawing on my research on building a "Trust Framework" for information ecosystem, I'll delve into the challenges of quantifying trust and its impact on information consumption. I'll showcase how this framework can be applied to identify reliable sources and mitigate the spread of misinformation. GenAI in this context is a double-edged sword, offering a powerful weapon against misinformation, but also creating its own ("hallucination"). This paves the way for LLM autonomous agents with reasoning and planning abilities, offering a more user-friendly future for fighting misinformation.​​​

  • Zeenat Patrawala on  Generative AI being underestimated in the field of Healthcare - what is the truth of LLMs and Diffusion Models with a specific focus on clinical application tools

    Talk Abstract:
    The rapid advancements in Generative AI, particularly in the form of Large Language Models (LLMs) and Diffusion Models, have sparked significant interest and debate within the healthcare industry. This talk aims to explore whether the true potential of these transformative technologies is being underestimated, with a specific focus on their clinical application tools.

    The discussion will begin by defining the key concepts of AI, LLMs, and Diffusion Models, while also acknowledging the continuously evolving landscape of foundation models. This foundation will set the stage for a deeper examination of what the new era of AI in healthcare might entail.

    Drawing on a case study, the talk will delve into how Generative AI can provide strong clinical application tools for pregnant women in critical need during humanitarian crises. This real-world example will illustrate the tangible benefits and practical applications of these cutting-edge technologies in the healthcare domain.  A second topic will include the power of on demand tools (radiology/AI and ultrasound/AI) that can be utilized on the field and enable critical care professionals to gain access to best in class medical information from the global experts (and training data) to provide immediate care, thereby increasing the ability to save life since time is extremely critical.

    Furthermore, the presentation will draw comparisons to past hype cycles, allowing the audience to gain a nuanced understanding of the current state of Generative AI in healthcare and to envision the future of healthcare tools and applications. This comparative analysis will help separate fact from fiction and provide a balanced perspective on the true potential of these transformative technologies.

    By the end of the talk, attendees will have a comprehensive understanding of the current capabilities and limitations of Generative AI in healthcare, as well as a glimpse into the promising future that these technologies hold for improving patient outcomes and revolutionizing clinical practices.​

  • Vida Hamad on Leveraging Data for Risk Mitigation and Crisis Response​​

    Talk Abstract:
    ​In an increasingly interconnected world, data plays a pivotal role in managing risks and responding effectively to crises. The presentation will outline how data-driven approaches enhance risk mitigation, empower crisis response, and safeguard against misinformation risks.​​

  • Sophia Ananiadou on Emotion Detection and Misinformation Harms from Large Language Models

    Talk Abstract:
    The increasing popularity of social media is making it easier than ever to spread misinformation. Through careful wording that stirs strong emotions, rumormongers can ensure that false information is rapidly and widely reshared within and across social networks.

    Misinformation detection is a muti-factorial problem, reliant not only on establishing whether or not a piece of text is factual, but also on determining a variety of features concerning both the textual content and structure of social media posts, which could interact to signal that information is fake.

    Large Language Models (LLMs) with their language understanding capabilities may be used maliciously to intensify the problem, by automatically generating large amounts of highly convincing false information. However, LLMs may be used positively, to fight against the spread of misinformation. I will investigate how to best exploit LLMs for the automated detection and analysis of misinformation, building upon previous approaches based on conventional machine learning and deep learning. In collaboration with social science scholars working on misinformation, disinformation, conspiracy theories, argumentation and trust, we analyse collections of social media posts in topics surrounding global elite conspiracies to identify a range of semantic, lexical and stylistic features that are characteristic of misinformation, including emotions, sentiment and stance, along with structural and discourse level, information such as dialogue acts and temporal dynamics.
  • Wei Xiao on NVIDIA: Powering the AI Revolution

    Talk Abstract:
    AI stands at the forefront of innovation, revolutionizing industries and pushing the boundaries of what's possible in AI-driven creativity and problem-solving. NVIDIA, as a global leader in AI and accelerated computing, has been instrumental in advancing AI technologies, driving breakthroughs across various domains. 

    This concise session offers a comprehensive overview of NVIDIA's key initiatives, breakthrough technologies, and vision for shaping the future of AI and computing. From cutting-edge GPU architectures to AI-driven solutions, attendees will gain valuable insights into NVIDIA's role as a global leader in AI, gaming, data centers, and autonomous systems.

Track 2: Data Science for Humanitarian Logistics and Supply Chains in Conflict Zones​

  • Tanveer Syeda-Mahmood on Responsible Development of Foundational Models

    Keynote Talk Abstract:
    With advances in generative artificial intelligence (AI), it is now possible to produce realistic-looking automated reports for a variety of applications ranging from customer service, to preliminary reads in radiology. Such reports can expedite existing workflows, improve accuracy and reduce overall costs. However, it is also well-known that such models often hallucinate, leading to false findings in the generated reports.

    In this talk,  Dr. Tanveer Syeda-Mahmood​ will describe industry and open source efforts in building responsible, trustworthy, and explainable AI models that begin with responsible data collection and curation, a community-driven consensus knowledge creation and inherent fact-checking. Dr. Tanveer Syeda-Mahmood will highlight some of the research advances in this field for both textual and image-driven fact-checked models. Future generative AI approaches can use these emerging methods to validate their reports leading to a more responsible use of AI in expediting workflows.​​

  • Joana Vasques on Optimizing Aid: The Intersection of Data Science and Humanitarian Supply Chains

    Talk Abstract:
    In the rapidly evolving field of humanitarian aid, the need for efficiency and effectiveness in logistics and supply chain management has never been more critical. "Optimizing Aid: The Intersection of Data Science and Humanitarian Supply Chains" explores the transformative potential of data science in redefining how aid is delivered in the most challenging environments. This presentation delves into the integration of advanced analytical techniques and machine learning algorithms with traditional humanitarian logistics to create resilient, responsive, and adaptable supply chains capable of meeting the dynamic needs of crisis-affected communities.

    Drawing on real-world examples, the talk highlights innovative approaches to planning, executing, and optimizing the delivery of aid. From predictive analytics for demand forecasting to drone technology for last-mile delivery, the presentation showcases how data science empowers decision-makers to anticipate needs, mitigate risks, and maximize the impact of their interventions.

    By bridging the gap between operational challenges and cutting-edge technology, "Optimizing Aid" offers insights into building the next generation of humanitarian supply chains. Attendees will gain an understanding of the practical applications of data science in enhancing operational efficiencies, reducing waste, and ultimately ensuring that aid reaches those who need it most, when they need it. This synthesis of technology and humanitarianism charts a course towards a more effective, data-driven future for aid delivery.

  • Alia Gharaibeh on Leveraging Data Analytics for Humanitarian Supply Chain Resilience

    Talk Abstract:
    Humanitarian actors today face a multitude of operational and environmental risks that can result in supply chain disruptions. Building resilient supply chains is key to ensuring the availability of life-saving commodities and begins with understanding the operating environment and the contextual factors that make them susceptible to disruptions. In this talk, Alia Gharaibeh introduces the Supply Chain Resilience (SCR) tool, an approach to simulate potential disruptions to humanitarian supply chains that can empower organizations to enhance supply chain resilience proactively. Through a System Dynamics model, this simulation fosters systems thinking and provides a comprehensive analysis of supply chain operations, identifying bottlenecks and testing diverse scenarios. These insights inform the creation of data-driven supply strategies that offer a roadmap for organizations to mitigate risks, reduce disruptions, and ensure aid reaches vulnerable populations, even in the most challenging environments.

  • Tina Comes on AI for Crisis Decisions & Humanitarian Logistics

    Talk Abstract:
    The world is confronted with a series of super-wicked problems, ranging from climate change to refugee crises. AI and data science have have been recognized as vital to improve resilience especially in the face of adverse events. At the same time, computational tools also have created new vulnerabilities, and changed the way we interact and make decisions. Major challenges remain in rapidly identifying and analysing different data sources and develop from there meaningful and actionable information. Through several case studies, I will outline key resilience principles, highlight how data and information can be used to improve disaster response and humanitarian logistics. Further, I will highlight the possible dilemmas arising given the different time frames of decision-making and discuss requirements for responsible design, and outline directions for future research​

  • Erica Gralla on Monitoring Complex Agricultural Market Systems: Framework and Insights from Uganda

    Talk Abstract:
    Meeting the United Nations Sustainable Development Goals (SDGs) will require adapting or redirecting a variety of very complex global and local human systems. Tools are needed to understand the dynamics of these systems and the key drivers of their behavior, such as barriers to progress and leverage points for driving sustainable change. System dynamics tools are well suited to address this challenge, but they must first be adapted for the data-poor and fragmented environment. This talk describes a framework for capturing and gaining insight from diverse qualitative and quantitative data, developed through a 4-year engagement with USAID/Uganda. It is illustrated with an application to agricultural financing in Uganda and the management of supply chain shocks during COVID-19.

Track 3: Healthcare in Crises: Augmenting Decision-Making to Optimize Care in Conflict Zones​

  • ​​Dina Katabi on Remote Sensing of Motion and Vital Signs: AI-Enhanced Radio Signal Analysis Beyond Physical Barriers

    Keynote Talk Abstract:
    ​This presentation introduces an innovative sensing technology that harnesses radio signals to track individuals' movements and vital signs, including respiration, heart rate, and sleep patterns, all without direct physical contact. By emitting low-power wireless signals and employing machine learning algorithms to analyze their reflections, this technology enables remote monitoring through walls and obstacles, effectively providing a non-invasive form of "X-ray vision". Its diverse applications range from creating intelligent environments to aiding first responders in disaster scenarios and facilitating remote patient monitoring within the healthcare system.

  • ​​​Erica Nelson on Spatial Methods for Health-Oriented Programmatic Decision-Making In Humanitarian Crises

    Talk Abstract:
    In humanitarian crises, rapid and informed decision-making is critical to mitigate health risks and save lives. Spatial methods, including Geographic Information Systems (GIS), remote sensing, and spatial analysis, offer powerful tools to enhance the effectiveness of health-oriented interventions in such contexts -from creating shared situational awareness and understanding population vulnerabilities, coordinating programmatic responses and improving the effectiveness and efficiency of interventions, to creating predictive, spatially-explicit models that guide resilience and preparedness efforts. This presentation will explore the multifaceted ways in which spatial methods can be utilized to inform and improve decision-making processes in humanitarian crises, discuss limitations and challenges, and advocate for the further advancement and adoption of these methodologies to support humanitarian health systems and, ultimately, mitigate mortality and morbidity in humanitarian crises. 
  • Danielle Poole on Data Science Methods for Monitoring Alleged War Crimes

    Talk Abstract:
    Healthcare facilities are civilian objects protected by international humanitarian law. And yet, attacks on healthcare facilities are widely documented in modern conflicts around the world. Whether indiscriminate or intentional, attacks on healthcare may constitute a war crime under the Geneva Conventions and their Additional Protocols. Satellite imagery damage assessments confer new opportunities to investigate the distribution of attacks on healthcare in real-time and at-scale. Geospatial analyses of damage to healthcare facilities may provide statistical evidence of alleged war crimes.​

Track 4: 
Measuring Data-Driven Impact Assessments​

  • Kimberley Abbott on Measuring What Matters – Leveraging Big Data and AI to Automate Impact Assessments.

    Talk Abstract:
    In today's world, understanding the impact of companies' products, services and activities on society and our planet is crucial, especially with increasing regulations that companies and financial institutions must follow. This presentation will explain why measuring impact matters and how by amalgamating and codifying decades of accumulated best practices from the social science and development sectors with cutting-edge technology and harnessing the power of open data, Vested Impact revolutionizes impact assessments by automating them

    Beyond the mechanics of measurement, this presentation navigates the terrain of utilizing impact data to enhance decision-making processes and drive favorable outcomes for both people and the planet. However, amidst the promise of progress lie inherent risks and challenges when dealing with constantly changing and nuanced social and environmental issues. The presentation will discuss the potential pitfalls associated with automated approaches such as Vested Impact's, emphasizing the importance of mitigating risks in pursuit of comprehensive impact assessment.​

  • Dareen Alhyari on YouTube Recommendations System and How We Measure Impact

    Talk Abstract:
    YouTube is considered one of the largest scale and most sophisticated industrial
    recommendation systems in existence. In this talk, Dareen Alhyari will describe the massive recommendation system at a high level, and focus on how we use data to measure impact.

  • Nour Sibai on Legal Disruption and Innovation using LLM

    Talk Abstract:
    This talk explores the impact of Large Language Models (LLMs) on the legal sector, showcasing how they can be utilized to detect errors and biases within legal documents ​and proceedings. By leveraging the advanced analytical capabilities of LLMs, we delve into practical case studies demonstrating their effectiveness in identifying and rectifying gender bias and other forms of injustice. The presentation will highlight the transformative potential of integrating AI technologies like LLMs into legal practices to enhance fairness, efficiency, and transparency, ultimately paving the way for a more just legal system

Workshops and Parallel Sessions

  •  Introduction to ChatGPT: Maneuvering GenAI Tools by Layal Tannous

    Workshop Abstract:
    In the first part of the workshop, attendees will explore various AI tools tailored for content generation. They will learn effective prompting techniques that will boost their creativity. The second part of the workshop will introduce them to an array of research-centered tools. Attendees will be able to conduct literature reviews on almost any topic within seconds! They will also “interact” with research papers, extracting key insights and results with an unparalleled efficiency. Finally, the workshop will briefly provide insights into installing and running language models locally. Overall, the workshop aims to familiarize the audience with a myriad of free, reliable, and user-friendly AI tools while emphasizing the ethical considerations and implications that come with their utilization.


Contact Us

For various questions, please try contacting us via social media first!
read more

Privacy Statement

We take data privacy seriously and adhere to all applicable data privacy laws and regulations.
read more

Copyright and Disclaimer

Written permission is needed to copy or disseminate all or part of the materials on the AUB website.
read more

Title IX, Non-Discrimination, and Anti-Discriminatory Harassment

AUB is committed to providing a safe, respectful, and inclusive environment to all members of its community.
read more