Poster Session
Join us for an exciting Poster Session where top experts from academia, industry, and government present cutting-edge research in Data Science and AI. This is your chance to connect, engage in insightful discussions, and expand your network in a dynamic, interactive setting. Refreshments and finger food will be served.
Explore the posters and presenters, organized by key themes, below.
Digital Economy
Jorge Zubelli
Professor and Chair of the Mathematics Department at Khalifa University
A personal trajectory leading to foundational aspects of AI-based phygital technology
Abstract
Reminiscing about his Berkeley graduate student days in the 1980s, Professor Jorge Zubelli describes his experience in the field of mathematical and computational modeling, which is relevant to the study of AI-based phygital technology. In the allocated space, he will outline how classical statistical analysis, combined with more recent subjects such as LLM and GPT, can impact practical problems of significance for phygital innovations.
Bio
Professor Zubelli earned his PhD in Applied Mathematics from UC Berkeley (1989), his MSc from the National Institute for Pure and Applied Mathematics (IMPA) in 1984, and his Electrical Engineering degree from IME-RJ in 1983, specializing in Telecommunications. He has served as a Professor of Mathematics at IMPA, headed LAMCA, and coordinated the Mathematical Methods in Finance MSc program at IMPA from 2002 to 2017. His research focuses on Inverse Problems and Mathematical Modeling, with applications to real-world issues. He has published in top journals like Science, Plos One, and SIAM Journal. He has also led numerous academic and industrial projects and serves on the editorial boards of several journals, including IJTAF and Mathematics and Computers in Simulation.
Thomas Hardjono
CTO of Connection Science and Technical Director of the MIT Trust-Data Consortium
Decentralized management of component attestations for cyber-resilient computing
Abstract
Device attestations for cybersecurity require a continuously verifiable supply chain of data regarding the components that constitute a device. The supply-chain data for a given component or product represents an endorsement by its manufacturer regarding the security quality of the product. Several challenges exist today concerning the infrastructure that supports the availability of endorsements from various manufacturers throughout the lifecycle of their components. Thomas Hardjono presents a distributed architecture for the registration, management, and retirement of endorsements for trustworthy and cyber-resilient computing.
Bio
Dr. Hardjono is a pioneer in digital identities and trusted hardware, playing a key role in the adoption of the MIT Kerberos authentication protocol. He has led standards efforts at IETF, IEEE, Trusted Computing Group, and Confidential Computing Alliance. With over 70 publications, several books, and 30+ patents, he is active in startups around MIT. His current focus is on Web3 Digital Assets, particularly the interoperability and cybersecurity of asset networks.
Michael Wolf
Professor of Econometrics and Applied Statistics, University of Zurich
The Hedged Random Forest
Abstract
The random forest is one of the most popular and widely employed tools for supervised machine learning. It can be used for both classification and regression tasks; in this paper, the focus will be on regression only. In its standard form, the crux of the random forest is to use an equal-weighted ensemble of tree-based predictors. Instead, we suggest a more general weighting scheme that borrows certain ideas from the related problem of financial portfolio selection and, in particular, allows for negative weights. Based on a benchmark collection of real-life data sets, we demonstrate the improved predictive performance of our method not only relative to the standard random forest but also relative to two previous proposals for (non-equal) weighting the tree-based predictors. It is noteworthy that our methodology is of a high-level nature and can also be applied to other forecast-combination problems, when forecast methods are of arbitrary nature and not necessarily tree-based.
Bio
Michael Wolf is a Professor of Econometrics and Applied Statistics at the University of Zurich, with a Ph.D. in Statistics from Stanford University. He previously held positions at UCLA, Universidad Carlos III de Madrid, and Universitat Pompeu Fabra in Barcelona. His research focuses on resampling-based inference, multiple testing methods, large-dimensional covariance matrix estimation, and financial econometrics. His work has been published in top journals, including The Annals of Statistics, Biometrika, Econometrica, Journal of the American Statistical Association, and The Review of Financial Studies.
Francesco Grigoli
Senior Economist, IMF and Georgetown University
Monetary Policy Transmission in Emerging Markets: Proverbial Concerns, Novel Evidence
Abstract
Proverbial concerns remain about the effectiveness of monetary policy in emerging markets. The empirical evidence is scarce due to challenges in identifying monetary policy shocks. In this paper, we construct monetary policy shocks using analysts' forecasts of policy rate decisions. Crucial for identification, analysts can update forecasts up to the policy meeting to incorporate any information relevant to the policy rate decision. Using these shocks, we show that monetary transmission wields considerable traction on financial and macroeconomic conditions in emerging markets. Monetary tightening lifts bond yields, curbs real activity, reduces inflation, and impacts leveraged firms more strongly.
Bio
Francesco Grigoli is a Senior Economist in the Research Department of the International Monetary Fund (IMF) and an adjunct professor at Georgetown University. Previously, he worked in the IMF’s Western Hemisphere Department and Fiscal Affairs Department and was a visiting scholar at Columbia University. While on leave from the IMF, he served as Director of Research at the Central Bank of the UAE. He published extensively in leading academic journals and policy outlets on a wide range of topics in macroeconomics and international economics. His current research focuses on inflation expectations, price formation, monetary policy, and international finance. Francesco received his PhD in Economics from the University of Insubria and holds a Master's in International Economics from the University of Sussex.
Praneeth Vepakomma
Assistant Professor, Department of Machine Learning, Mohamed Bin Zayed University of Artificial Intelligence
DAVED: Data Acquisition via Experimental Design for Data Markets
Abstract
Acquiring training data is essential for machine learning, especially in data-scarce fields like healthcare. Data markets can help by incentivizing data providers. A key challenge for buyers is selecting the most valuable data points. Unlike previous approaches that assume centralized access, we propose a federated data acquisition method inspired by linear experimental design. Our approach reduces prediction error without needing labeled validation data and can be optimized quickly in a federated setting. The core insight is that directly estimating the benefit of new data for test predictions aligns well with a decentralized market.
Bio
Praneeth Vepakomma is a Visiting Assistant Professor at MIT and an Assistant Professor in the Department of Machine Learning at Mohamed Bin Zayed University of Artificial Intelligence in Abu Dhabi. He has extensive industry experience at Meta, Apple, Amazon Web Services, Motorola Solutions, Corning, and various startups, including his role as Principal Staff Scientist at Motorola Solutions. Praneeth has received numerous awards, including the Meta PhD Research Fellowship in Applied Statistics, an OpenDP Academic Fellowship at Harvard, and the ADIA Lab Fellowship. He has also been recognized with best paper awards at several ML conferences. He holds a PhD from MIT and an MS in Mathematical and Applied Statistics from Rutgers University. His research focuses on algorithms for distributed computation in statistics and machine learning, emphasizing privacy, efficiency, and responsible data science.
Nurbek Tastan
PhD Student at Mohamed Bin Zayed University of Artificial Intelligence, Abu Dhabi
Confidential, Private, and Fair Decentralized Learning Based on Encryption-friendly Distillation Loss
Abstract
Training accurate deep neural networks often requires large volumes of data that are not available to any single entity due to privacy concerns and stringent regulations. Cross-silo federated learning (FL) allows for collaborative learning without data sharing, but existing FL algorithms often struggle with an unacceptable utility-privacy trade-off. We propose a framework called Confidential and Private Decentralized (CaPriDe) learning, which utilizes fully homomorphic encryption to enable collaborative learning while preserving data confidentiality. In CaPriDe, participants share their private data in encrypted form, allowing inference in the encrypted domain. The core of CaPriDe is mutual knowledge distillation between local models using a novel distillation loss that approximates the Kullback-Leibler (KL) divergence between local predictions and encrypted inferences, computable in the encrypted domain. Additionally, we introduce a reputation-scoring method based on gradient alignment to address collaborative fairness in decentralized learning.
Bio
Nurbek Tastan is a PhD student specializing in Machine Learning at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE. His research interests include federated learning, trustworthy machine learning, and anomaly detection, and he works under Dr. Karthik Nandakumar at SPriNT-AI Lab and is co-supervised by Dr. Samuel Horvath. Nurbek holds an MSc in Machine Learning from MBZUAI and a BSc in Systems of Information Security from the International IT University, Almaty, Kazakhstan. He has been recognized with several awards from the Ministry of Education of Kazakhstan, including the “Golden Badge” and the “Presidential Scholarship,” alongside medals in national mathematics and computer science competitions.
Abdulla Almansoori
PhD Student at Mohamed Bin Zayed University of Artificial Intelligence, Abu Dhabi
Collaborative and Efficient Personalization with Mixtures of Adaptors
Abstract
Non-iid data is common in real-world federated learning. We propose a parameter-efficient framework, Federated Low-Rank Adaptive Learning (FLoRAL), which allows each client to adapt by mixing between parameter-efficient adaptors. FLoRAL is a model parameterization that casts federated learning as a multi-task learning problem, with weight sharing as an implicit regularizer. It is memory-efficient and does not require stateful clients, as the adaptors are federated across clients. Our results show that FLoRAL can, in some scenarios, outperform mixture of full models and models with locally tuned adaptors, demonstrating “federated personalization” and robustness against overfitting. We derive convergence rates and show theoretically that FLoRAL can lead to improved variance reduction.
Bio
Abdulla Jasem Almansoori is a PhD student in machine learning at Mohamed bin Zayed University of Artificial Intelligence, working under the supervision of Martin Takáč and Samuel Horváth. He earned his master's degree in computer science from the University of Southern California in 2018 and his bachelor's degree in industrial engineering from Purdue University in 2016. His research focuses on optimization in machine learning, particularly in efficient and collaborative learning within the context of personalized federated learning.
Javier Parra Domínguez
Associate Professor, Financial Economics and Accounting, University of Salamanca, Spain
Diego Valdeolmillos Villaverde
PhD Candidate, University of Salamanca, Spain
Building Citizen Trust through Explainable AI and DLT in Public Sector Decisions
Abstract
Artificial Intelligence is increasingly used in public sector decision-making, but a lack of transparency can lead to societal distrust. Explainable AI (XAI) has become essential for enhancing transparency and accountability. Effective XAI implementation requires reliable and verifiable systems. Integrating blockchain—especially with oracles—provides trustworthy external data, boosting explainability and enabling automation. Oracles securely connect external data to the blockchain, ensuring AI models use authentic information. Blockchain's decentralization, cryptography, and traceability ensure integrity, confidentiality, and transparency. Smart contracts automate compliance, improving efficiency and reducing bias. This approach builds citizen trust, promotes participation, and aligns with the Sustainable Development Goals (SDGs), fostering social and environmental sustainability.
Bio
Javier holds a degree in Computer Science from the Wrexham University and a Master's in Macroeconometrics and Finance from the Universidad Internacional Menéndez Pelayo, the Instituto de Estudios Fiscales and the Centro de Estudios Económicos y Comerciales de España. Javier currently combines his research at the BISITE Research Group with coordinating the Quantum Economics & Technology Experience Lab and participating in the Economic Advisory Board at the IoT Digital Innovation Hub. He is also a member of the Executive Committee of the European Network of Excellence in Artificial Intelligence 'dAIEdge'.
Diego, a Computer Engineering graduate from the University of Salamanca, specializes in distributed systems and DLT. He holds three master’s degrees in Intelligent Systems, Information Systems, and IoT. Pursuing a PhD since December 2023, his research focuses on blockchain, decentralization, cybersecurity, and trustworthy AI. Diego has led R&D projects in forest management, lending systems, agriculture, regenerative medicine, and smart platforms. He deployed the first Smart Contract on the TRON blockchain and currently leads DLT projects at the University of Salamanca, also contributing to the Cyberchain International Cybersecurity Chair.
Hamid Arian
Assistant Professor of Finance at York University, Canada
Back-test Overfitting in the Machine Learning Era
Abstract
: This research explores integrating advanced statistical models and machine learning in financial analytics, representing a shift from traditional to advanced, data-driven methods. We address a critical gap in quantitative finance: the need for robust model evaluation and out-of-sample testing methodologies, particularly tailored cross-validation techniques for financial markets. We present a comprehensive framework to assess these methods, considering the unique characteristics of financial data like non-stationarity, autocorrelation, and regime shifts.
Bio
Hamid Arian is an assistant professor of finance at York University, Canada. His academic background is in mathematics. Hamid obtained his PhD in applied mathematics from the University of Toronto and is a CFA charter-holder. In addition to academia, he has industrial experience in both sell-side and buy-side firms. Hamid's current research interests are applications of AI in finance, partial differential equations, and causal inference.
Health Science
Francisco Herrera
Professor of Computer Science and Artificial Intelligence at the University of Granada
Human-AI Decision Making: Synergy Between Explainability and the Role of Human Intuition for AI Reliance—An Applied Perspective in Health Sciences.
Abstract
Appropriate AI reliance means trusting accurate AI recommendations and disregarding incorrect ones. Human intuition is key in AI-assisted decision-making, helping determine when to rely on or override AI advice. Explainability enhances this process by making AI reasoning transparent, fostering trust and effective collaboration. This synergy combines AI’s computational power with human judgment, improving decision-making and promoting fairness. In healthcare, this balance is crucial. AI provides valuable insights from vast medical data, but human intuition is vital for interpreting these recommendations. Explainable AI systems help healthcare professionals understand and trust AI, leading to better diagnoses and treatments.
Bio
Professor Herrera earned his M.Sc. (1988) and Ph.D. (1991) in Mathematics from the University of Granada, Spain. An academician of the Royal Academy of Engineering (Spain), he has published over 600 journal papers with more than 130,000 citations (Google Scholar, H-index 173) and serves on the editorial boards of numerous academic journals. He has been recognized as a Highly Cited Researcher in Computer Science and Engineering by Clarivate Analytics. His research interests include computational intelligence, information fusion, decision-making, explainable artificial intelligence, and data science, encompassing data preprocessing, prediction, and big data.
Dima Al-Absi
PhD candidate in Management Science and Engineering at Khalifa University, UAE
Building Trust in AI for Healthcare: A Study on Human-AI Collaboration for AKI Risk Prediction
Abstract
Establishing trust in AI tools is crucial for their acceptance in healthcare, especially in high-stakes areas like perioperative care. This study examines how transparency, explainability, and human-in-the-loop design foster trust in an AI tool for predicting Acute Kidney Injury (AKI) risk. Co-designed with physicians, the tool aligns input parameters with clinical practices and offers actionable insights. Usability testing highlighted the importance of intuitive design, real-time updates, and visual explainability, such as SHAP plots, to make AI predictions understandable for healthcare providers. Human-in-the-loop design and expert involvement ensure the AI respects clinician expertise, promoting collaboration. Continuous validation and seamless integration into Electronic Health Record (EHR) systems are vital for building trust. The tool’s user-friendly interface and clear visual outputs enhance real-time decision-making, improving patient safety and operational efficiency. Ultimately, AI tools must provide accurate predictions while meeting physicians' practical needs to ensure transparency, usability, and trust in healthcare settings.
Bio
Dima Tareq Al-Absi is a PhD candidate in Management Science and Engineering at Khalifa University, UAE. Her research focuses on integrating AI into healthcare to enhance risk-based decision-making, with a particular emphasis on explainable AI for patient safety. Dima has collaborated extensively with Sheikh Shakhbout Medical City (SSMC), where she evaluated the impact of technological interventions on healthcare outcomes. She was also a visiting PhD scholar at UCL School of Management, conducting research on the use of explainable AI in healthcare settings. Her work centers on transforming healthcare.
Firda Rahmadani
PhD candidate in Management Science and Engineering at Khalifa University, UAE
Decision-making Enhancement Through Human-AI Decision Support Framework: The Case of Streamlining Sepsis Pathways within the Critical Golden Hour
Abstract
Sepsis is a severe condition resulting from an abnormal response to infection and often leads to poor patient outcomes in hospitals. The existing early warning systems to identify sepsis risk frequently generate false positive alarms, contributing to alert fatigue and being ignored by healthcare providers. This delay in sepsis recognition is worsened by heavy workloads and low staffing, making timely intervention difficult. To address this issue, this study proposed a Human-AI Decision Support Framework (HADS) that integrates human expertise and AI to identify high-risk patients and improve sepsis management. Using data from 74,567 patient records in the MIMIC-IV database, the framework developed a prediction model enhanced with physician input. It applied the Bayesian Belief Network and outperformed traditional alert systems by reducing false alarms by 48%. Critical interventions included timely antibiotic administration, blood culture collection, and frequent patient monitoring. The framework streamlined sepsis pathways and reduced response times during the golden hour, essential for improving patient outcomes. The study also provided insights into future research and practical applications to enhance patient care using HADS.
Bio
Firda Rahmadani is a PhD candidate in Management Science and Engineering from Khalifa University of Science & Technology. Her research focused on developing a Human-AI Decision Support (HADS) Framework to enhance sepsis management.
Khaled Toffaha
PhD candidate in Management Science and Engineering at Khalifa University, UAE
Leveraging Data-Driven Approaches in Predicting Inpatient Falls
Abstract
Patient falls pose a major challenge in healthcare, impacting safety and costs, with approximately 2,740 falls occurring daily in U.S. hospitals. By 2030, the financial burden could exceed $101 billion, emphasizing the need for effective prevention strategies. This research integrates predictive modeling, causal analysis, and intervention evaluation to develop innovative fall prevention methods. Machine learning algorithms identify key risk factors for early interventions, achieving high accuracy by combining clinical and operational data. A Bayesian Belief Network (BBN) models complex causal relationships among these factors, while Chain Event Diagrams (CEGs) visualize events leading to falls, pinpointing intervention opportunities. Using Average Treatment Effect (ATE) methodology, the study evaluates patient education initiatives, demonstrating a significant reduction in falls and related costs. By leveraging systems engineering and machine learning, healthcare providers can enhance patient safety, improve operational efficiency, and lower healthcare costs.
Bio
Khaled Toffaha is a Ph.D. researcher at Khalifa University specializing in Engineering Systems and Management. His doctoral thesis focuses on the critical areas of patient safety and hospital avoidable risk prediction and mitigation. By blending systems engineering techniques with machine learning and causal networks, Khaled aims to develop advanced tools that integrate organizational, clinical, and demographic data to eliminate risk factors within healthcare settings. This approach has the potential to improve patient outcomes and enhance overall healthcare management across various applications.
Eman Ouda
PhD candidate in Management Science and Engineering at Khalifa University, UAE
Optimizing Emergency Department Operations for Greater Efficiency and Resilience
Abstract
Effective emergency department management is crucial for timely patient care, especially during peak demand. This research presents a framework that integrates design thinking and discrete event simulation to optimize operations and assess resilience. By identifying bottlenecks and evaluating resource strategies through design thinking phases, the study develops simulation models to measure key metrics like patient wait times. Findings show that adding a physician during peak hours improved resilience, as indicated by enhanced resistance and recovery metrics. Utilizing data from Sheikh Shakhbout Medical City Hospital, the study offers recommendations for improving operational efficiency and patient outcomes. This work emphasizes the role of data-driven approaches in emergency department performance, aligning with the Health 5.0 framework for personalized care and sustainability.
Bio
Eman Ouda is a Ph.D. candidate in the Department of Management Science and Engineering at Khalifa University. Her research interests lie in the design and analysis of stochastic simulations under input uncertainty, as well as the applications of data analytics and simulation-based optimization in operations management, with a focus on healthcare systems. Eman’s work specifically focuses on developing mathematical models and algorithms for emergency departments with limited resources and uncertainty. Eman’s work has the potential to significantly improve the efficiency and effectiveness of emergency department operations.
Leopoldo Julián Lechuga López
PhD Candidate at NYU Abu Dhabi
MedCertAIn: Uncertainty Quantification Framework using Multimodal Data for Enhancing Reliability of AI-driven In-Hospital Mortality Prediction
Abstract
Integrating Artificial Intelligence (AI) into clinical decision support systems aims to enhance patient care, but effective communication of prediction uncertainty is essential for reliable AI use. However, principled uncertainty quantification is challenging, especially in advanced multimodal learning, limiting real-world deployment. MedCertAIn is a framework designed to improve uncertainty quantification in healthcare applications. By utilizing tailored data-driven prior distributions over neural network parameters, it enhances the predictive performance and reliability of clinical decision-making models. Evaluated on MIMIC-IV clinical time-series data and MIMIC-CXR chest X-ray images for ICU mortality prediction, MedCertAIn outperformed state-of-the-art models and existing Bayesian methods in both predictive performance and uncertainty metrics. The success of MedCertAIn demonstrates its potential to elevate clinical decision-making and sets a new standard for AI-driven healthcare research, advocating for the integration of advanced uncertainty quantification methods for more reliable patient care.
Bio
L. Julian Lechuga Lopez has a double Master’s degree in Mathematics and Computer Science with specialization in Data Science from Université Paris Cité (2021). Born and raised in Mexico, he obtained a BSc degree in Mechatronics Engineering from Instituto Tecnológico y de Estudios Superiores de Monterrey ITESM (2016). Currently, he is a 2nd year PhD candidate in CSE at NYU Tandon School of Engineering and NYU Abu Dhabi. His research focuses on improving the reliability of multimodal learning using model robustness and uncertainty quantification of AI-assisted clinical decision-making applications. He is very passionate about the use of machine learning and deep learning in different areas of healthcare to develop applications that can positively impact the lives of people across the globe.
Shaza Elsharief
PhD Candidate at NYU Abu Dhabi
MedMod: Multimodal Benchmark for Medical Prediction Tasks with Electronic Health Records and Chest X-Ray Image
Abstract
Multimodal machine learning presents a myriad of opportunities for developing models that integrate multiple modalities and mimic decision-making in the real-world, such as in medical settings. However, benchmarks involving multimodal medical data are scarce, especially routinely collected modalities such as Electronic Health Records (EHR) and Chest X-ray images (CXR). To contribute towards advancing multimodal learning in tackling real-world prediction tasks, we present MedMod, a multimodal medical benchmark with EHR and CXR using publicly available datasets MIMIC-IV and MIMIC-CXR, respectively. MedMod comprises five clinical prediction tasks: clinical conditions, in-hospital mortality, decompensation, length of stay, and radiological findings. We extensively evaluate several multimodal supervised learning models and self-supervised learning frameworks.
Bio
Shaza Elsharief is a research assistant at the Clinical Artificial Intelligence Lab, where she is working on developing self-supervised learning methods for clinical applications. Prior to joining the lab, Shaza obtained a BSc in Computer Engineering from New York University Abu Dhabi in 2023. Her research interests lie at the intersection of machine learning and healthcare, with a particular focus on real-world impact. Shaza will be presenting her work on a multimodal benchmark for medical prediction tasks, which includes an open-source codebase and public leaderboard, at the ADIA Lab Symposium poster session in November 2024. She aims for this work to enhance accessibility and reproducibility within the research community.
Baraa Al Jorf
PhD Candidate at NYU Abu Dhabi
Advancing Multimodal Foundation Models in Healthcare: A Framework of Guidelines, Reviews, and Novel Architectures
Abstract
This work aims to advance multimodal foundation models in healthcare through three main objectives. First, it conducts a literature review to assess the current landscape, focusing on architectural advancements, data-processing techniques, and identifying gaps for future research. Second, it establishes best practices for developing clinically applicable multimodal machine learning systems by surveying clinicians on how they manage various clinical data types. This survey will inform the creation of models that accurately reflect real-world clinical workflows. Finally, the study will design novel multimodal foundation model architectures that integrate electronic health records, chest X-rays, radiology reports, and discharge notes using hybrid fusion techniques. Extensive experiments will demonstrate that this approach outperforms traditional state-of-the-art models in in-hospital mortality prediction, clinical label classification, and radiology label classification.
Bio
Baraa Al Jorf is an NYU Abu Dhabi Global PhD fellow working on multimodal foundation models for precision medicine. He completed his BSc in Computer Engineering at NYU Abu Dhabi in 2023. Baraa's previous experience includes working in the Clinical AI lab as part of the Postgraduate Training Program, where he focused on preprocessing mammograms and ultrasound datasets. Passionate about the potential of AI to revolutionize healthcare, Baraa envisions a future where technology empowers both medical professionals and patients, enhancing diagnostic capabilities and transforming treatment strategies.
Saleh Ibrahim
Associate Dean for Research
Professor, Department of Medical Sciences
College of Medicine and Health Sciences
Khalifa University
The Genetics of Complex Diseases
Abstract
Chronic diseases are influenced by multiple genes and environmental factors. Systems genetics links DNA variations to clinical traits through intermediate phenotypes like RNA, protein, and metabolites, offering insights into conditions such as inflammatory and metabolic disorders. While GWAS have identified many genetic loci, identifying causal genes is difficult, especially as most SNPs are in non-coding regions. Environmental factors like climate, diet, and smoking also shape disease risks, and underrepresentation of diverse populations limits insights in regions like the MENA. New sequencing technologies offer opportunities to explore these gaps. Saleh Ibrahim's research focuses on gene interactions and the relationship between inflammation, metabolism, and environmental risks. Using systems genetics in mouse, cellular, and human models, he aims to identify genetic variants that contribute to disease susceptibility, with a focus on the UAE.
Bio
Saleh Ibrahim earned his Bachelor of Medicine in Egypt, then obtained an MD in Immunology from Helsinki University. He completed postdoctoral training in molecular genetics at Princeton University, studying the genetic basis of autoimmunity, before leading a group at the University of Rostock in 1997. In 2008, he became a professor of Genetics at Luebeck University and joined Khalifa University in 2022. Dr. Ibrahim's research focuses on investigating population differences in genetic susceptibility to complex diseases, as well as exploring the role of gene-microbiota interactions in the pathogenesis of age-related diseases.
Climate Science
Youngjin Nam
Post-Doc at the Korea Advanced Institute of Science & Technology (KAIST), Daejeon, Korea
CliFin: Climate-related Financial Risk Assessment Framework
Abstract
This study introduces CliFin, a Python framework designed for assessing financial risks in a variety of investment assets in response to climate change. Addressing the technical limitations of conventional climate risk assessment methods—which often depend on hypothetical scenarios and spreadsheet models—CliFin enhances flexibility and extensibility.
Bio
Dr. Youngjin Nam is a post-doctoral scholar at the Korea Advanced Institute of Science & Technology (KAIST). He is interested in evaluating the economic impact of climate risks. His research is both theoretical and practical, as he plans to develop various climate risk assessment tools, including a financial model that considers cash flow and a large-scale macro-econometric model. These tools will be instrumental in mitigating climate risks at both the asset and macroeconomic levels. He is confident that interdisciplinary research combining engineering and economic approaches will help mitigate climate risks. He believes sustainable economic growth will be achieved by transitioning to a low-carbon economy and dreams that a small effort will bring a brighter tomorrow.
Franziska Stickling
Master Student, Technical University of Munich, Germany
Optimizing Global Carbon Sequestration through Tree Growth
Abstract
This study explores the potential of global forests as carbon sinks under future climate conditions. The research uses the Canadian IPCC Tier 3 Carbon Budget Model (CBM-CFS3) to identify climate-resilient tree species that maximize carbon sequestration. This climate-responsive model helps optimize future forest management strategies to mitigate climate change risks by selecting the most suitable tree species for specific regions based on tree, soil, and climate data.
Bio
Currently pursuing a Master's in Finance and Information at the Technical University of Munich, Franziska research focuses on developing a model to assess global forest carbon sequestration potential under climate change. The aim is to account for climate risks more accurately, contributing to global sustainable strategies for maximizing forest carbon capture in the face of shifting climate conditions.
Nicolas Mauricio Cuadrado Avila
PhD Student at Mohamed Bin Zayed University of Artificial Intelligence, Abu Dhabi
Generalized Policy Learning for Smart Grids: FL TRPO Approach
Abstract
This work introduces a framework combining FL with Trust Region Policy Optimization (FL TRPO) to reduce energy-associated emissions and costs. Our approach reveals latent interconnections and employs personalized encoding methods to capture unique insights, understanding the relationships between features and optimal strategies, allowing our model to generalize to previously unseen data.
Bio
Nicolas is a proud Colombian with a passion for technology, entrepreneurship, and sustainability. Since childhood, he has enjoyed learning about science, which inspired him to study electronics engineering. After completing his undergraduate degree, he founded a CleanTech startup in Colombia with the support of the UAE government. Facing various challenges, he decided to attend MBZUAI to further his education and explore the local entrepreneurial environment, seeking the tools to continue his journey. He has recently completed his Master's in Machine Learning and is currently pursuing a PhD in the same field.
Enrique Herrera Viedma
Professor of Computer Science and Artificial Intelligence at the University of Granada
Safe and secure Large Language Models: a risk analysis and technological solutions. An applied perspective in climate science.
Abstract
The rise of large language models (LLMs) brings security and safety concerns, such as data breaches, misuse, and algorithmic bias. This poster explores these risks and outlines solutions like robust behavior, continuous monitoring, updates, and adherence to ethical guidelines. LLMs also hold potential for addressing climate change by analyzing social media data to gauge public sentiment on the economic impact of climate policies. Their ability to extract causal links from diverse sources supports informed, adaptive policies that enhance economic resilience to extreme weather events.
Bio
Enrique Herrera Viedma is a Professor in the Department of Computer Science and Artificial Intelligence at the University of Granada (UGR), where he is also Vice-Rector for Research and Knowledge Transfer. A Fellow of IEEE and IFSA with an honorary doctorate from Oradea University, he has served as Vice-President for Publications in the IEEE Systems, Man, and Cybernetics Society and co-founded the IEEE Transactions on Artificial Intelligence. Recognized as a Highly Cited Researcher (2014-2023) with over 350 papers and an h-index of 121, he published in Science on digital libraries' role in the information society. He has served on expert panels for project evaluations in several countries and as an Associate Editor for multiple AI journals since 2017.
Cross-Vertical
Alexei Kondratyev
Visiting Professor at the Department of Mathematics, Imperial College London.
Density Matrix Classifier
Abstract
We introduce a new type of quantum machine learning classifier based on the estimation of Frobenius distance between density matrices. The central idea is to encode all samples in the dataset belonging to a particular class into a corresponding density matrix using the quantum feature map methodology. The same quantum feature map is used to encode every new test sample into its own corresponding density matrix. Frobenius distances between the sample density matrix and the density matrices encoding all classes present in the training dataset are calculated. The class label is assigned by selecting the class with minimum distance. The density matrix classifier benefits from the expressive power of parameterised quantum circuits and performs well in comparison with the standard classical classifiers. The model is resistant to certain types of noise and is suitable for execution on NISQ computers.
Bio
Oleksiy holds MSc in Theoretical Physics from Taras Shevchenko National University of Kyiv and PhD in Mathematical Physics from the Institute for Mathematics, National Academy of Sciences of Ukraine. His primary research interests are in quantitative finance, machine learning and quantum computing. Oleksiy has over 20 years of quantitative finance experience in both risk management and front office roles and has been recognized as Quant of the Year (2019) by Risk magazine for his research on the application of machine learning techniques to risk factor analysis and portfolio optimisation.
Emilio Porcu
Department of Mathematics, Khalifa University and Visiting Professor, Trinity College Dublin
Generalized Networks
Abstract
This poster examines generalized networks, defined by nonlinear edges and stochastic processes continuously indexed over vertices and edges. These structures are represented as graphs with Euclidean edges, allowing for dynamic topology changes over time, where vertices and edges can disappear and edges may alter in shape and length. We analyze both linear and circular time cases, with the latter exhibiting a periodic structure. Our findings highlight the advantages and disadvantages of each setting. Generalized networks can be viewed as semi-metric spaces when equipped with a proper semi-metric, enabling the development of appropriate semi-metrics for their evolving structures. We conclude by guiding the reader in selecting function classes for constructing proper reproducing kernels in conjunction with these temporally evolving semi-metric topologies.
Bio
Emilio Porcu earned his Ph.D. in statistics in 2005 and became a full professor in 2012. He holds the position of Professor of Statistics and Data Science at Khalifa University since August 2020, where he is also part of the Biotechnology Research Center. He is an adjunct professor at Trinity College, Dublin, and previously served as Chair of Statistics at Newcastle University and Trinity College, and Senior Scientist at MIDAS Research Center in Chile. His research focuses on Statistical and Machine Learning, Data Science, and Spatial Statistics, with over 160 peer-reviewed publications. His work spans applications in climate change, weather forecasting, spatial criminology, and more recently, Health Data Science, particularly in computational genetics, genomics, and aging.
Khaled Hamad Saeed AlNuaimi
Research Specialist, ADIA, Abu Dhabi
Enriching Datasets with Demographics through Large Language Models
Abstract
Enriching datasets with demographic details, like gender, race, and age from names, is essential in healthcare, public policy, and social sciences for better population engagement. Prior methods using hidden Markov models and recurrent neural networks have faced challenges, such as limited large-scale, unbiased datasets and lack of robustness across various data. This has restricted traditional supervised learning development. Our work shows that the zero-shot capabilities of Large Language Models (LLMs) can match or surpass specialized models. We test these LLMs on diverse datasets, including a real-life set of licensed financial professionals in Hong Kong, and examine demographic biases. This research advances demographic enrichment techniques and paves the way for addressing biases in LLMs.
Bio
As a Research Specialist at the Abu Dhabi Investment Authority (ADIA), I focus on developing data-driven investment signals to guide investment decision-making. With over a decade of financial expertise, a Chartered Financial Analyst (CFA) designation, and a Master’s in Computational Data Science, I bring a robust analytical foundation to my work. Currently pursuing a PhD in Engineering at Khalifa University, my research bridges Data Science, Natural Language Processing, and Finance to advance investment insights and strategy development.
Raghav Singhal
Research Assistant at Mohamed Bin Zayed University of Artificial Intelligence, Abu Dhabi
FedEx-LoRA: Exact Aggregation for Federated and Efficient Fine-Tuning of Foundation Models
Abstract
Low-Rank Adaptation (LoRA) is a popular technique for efficient fine-tuning of foundation models. However, applying LoRA in federated learning environments, where data is distributed across multiple clients, presents unique challenges. Existing methods rely on traditional federated averaging of LoRA adapters, resulting in inexact updates. To address this, we propose Federated Exact LoRA, or FedEx-LoRA, which adds a residual error term to the pretrained frozen weight matrix. Our approach achieves exact updates with minimal computational and communication overhead, preserving LoRA’s efficiency. We evaluate the method on various Natural Language Understanding (NLU) and Natural Language Generation (NLG) tasks, showing consistent performance gains over state-of-the-art methods across multiple settings. Through extensive analysis, we quantify that the deviations in updates from the ideal solution are significant, highlighting the need for exact aggregation. Our method’s simplicity, efficiency, and broad applicability position it as a promising solution for accurate and effective federated fine-tuning of foundation models.
Bio
Raghav Singhal is currently a Research Assistant at MBZUAI, Abu Dhabi. He completed an interdisciplinary Dual Degree Program at IIT Bombay, with a bachelor's in electrical engineering and a Masters in the Centre for Machine Intelligence and Data Science. He was selected for a semester exchange program at KTH Royal Institute of Technology and did internships at Google and Aditya Birla Group. He was the recipient of Undergraduate Research Award (URA 02) for one of the best bachelor’s theses in the EE department, a recipient of Undergraduate Research Award (URA 01) for exceptional undergraduate research work and was awarded the DAAD WISE scholarship by the German Government for a summer research internship in Germany.
Kaustubh Ponkshe
Research Assistant at Mohamed Bin Zayed University of Artificial Intelligence, Abu Dhabi
Power Learning: Differentially private and model agnostic embeddings
Abstract
Traditional collaborative learning typically involves sharing model weights between clients and a server, but sharing activations can enhance resource efficiency. While various differentially private methods exist for weight sharing, mechanisms for activation sharing are lacking. We propose Power-Learning, which employs a privacy encoding network alongside a small utility generation network to ensure the final activations have formal differential privacy guarantees. These privatized activations are shared with a powerful server that processes them, improving accuracy in machine learning tasks. Our approach requires only one round of privatized communication and less client computation than traditional methods. Additionally, the shared privatized activations are agnostic to the model type (e.g., deep learning, random forests, XGBoost) used by the server.
Bio
Kaustubh Ponkshe is currently a Research Assistant at MBZUAI, Abu Dhabi. He completed both his bachelor's in electrical engineering and Master's in AI and Data Science at IIT Bombay. He was selected for a semester exchange program at Denmark Technical University. He has completed internships at Core AI Engine, AWL Inc. Japan, TCS Research and collaborated with Adobe Research India. He was the Recipient of Undergraduate Research Award (URA 02) for one of the best bachelor’s thesis among 80+ projects. He has participated in the IIT Bombay Racing team that built a driverless race car to compete in FS Germany. This became the first Indian team to win the Engineering Design event in the history of FSUK.