Speaker Bios & Session Abstracts
In order to achieve net-zero greenhouse gas (GHG) emissions in the coming decades, the world will need to eliminate the emissions from electricity and heat generation, transportation, the production of all materials such as chemicals, plastics, steel, concrete, and from the entire food supply chain. This talk will outline some of the technical challenges that will have to be overcome to meet our climate goals. Moreover, global GHG emissions continue to rise, and it seems unlikely that we will achieve our climate goals of keeping the rise in the global average temperature to less than 2 degrees Centigrade. We will need to develop methods of direct air capture of CO2 and hydrogen production at low enough costs to allow wide-scale deployment.
Professor Chu is an internationally renowned scientist who was named co-winner of the 1997 Nobel Prize in Physics alongside Claude Cohen-Tannoudji and William D. Phillips “for development of methods to cool and trap atoms with laser light”. Over recent years he has focused on the search for new solutions to energy and climate challenges, both as US Secretary of Energy from 2009 to 2013 and, before that, as Director of the Lawrence Berkeley National Lab, where he explored alternative and renewable energy technologies.
Professor Chu has made important contributions in atomic physics, quantum electronics, polymer and biophysics including tests of fundamental theories in physics, the development of methods to laser cool and trap atoms, atom interferometry, the study of polymers and biological systems at the single molecule level, molecular biology, medical ultrasound imaging, nanoparticle synthesis, batteries and other applications in electrochemistry.
The holder of 20 patents, Professor Chu has published more than 300 scientific and technical papers. He is a member of numerous scientific societies including the National Academy of Sciences, the American Philosophical Society, the Royal Society, the Royal Academy of Engineering, the Academia Sinica, the Korean Academy of Sciences and Technology, and is an honorary member of the Institute of Physics, the Chinese Academy of Sciences, and a Lifetime Member of the Optical Society of America and the Pontifical Academy of Sciences. He received an A.B. degree in mathematics, a B.S. degree in physics from the University of Rochester, and a Ph.D. in physics from the University of California, Berkeley, as well as 34 honorary degrees.
Today we are in an age of many risks. Often these risks impact many assets at the same time, reducing the benefit of diversification. Active risk management requires forecasts of these risks, ideally before others recognize them. This talk examines a passive strategy based on volatility forecasts: when volatility is predicted to be high, the strategy reduces exposure. Volatility target indices developed for insurance annuities have this property — lower risk but higher risk-adjusted returns, and in some cases higher unadjusted returns. Commercial versions exhibit higher Sharpe ratios than the underlying asset in both high and low volatility environments, and significant CAPM alphas; on high common volatility (COVOL) days the Sharpe ratios are dramatically higher. Empirical models of the price of risk reveal that when high volatility is predicted, expected returns are not sufficiently high to compensate risk-averse investors, so simple buy-and-hold strategies are inefficient. This analysis is also applied to sustainable portfolios, allowing investors to reduce risk while achieving hedging returns through dynamic risk management.
Professor Robert F. Engle is Professor Emeritus in the Management of Financial Services at New York University’s Stern School of Business. He received the 2003 Nobel Memorial Prize in Economic Sciences, which he shared with Clive Granger, “for his pioneering work on analyzing economic time series with time-varying volatility.” His creation of the ARCH (Autoregressive Conditional Heteroskedasticity) model transformed financial econometrics and serves as a fundamental component of contemporary risk management.
Professor Engle established the Volatility and Risk Institute at NYU Stern, now co-directed with Richard Berner. The institute operates the V-LAB platform, offering current information on worldwide systemic risk and emerging risks. His current work emphasizes climate risk while continuing to influence financial econometrics, macroeconomic forecasting, and quantitative risk modeling. Together with Eric Ghysels, he established the Society for Financial Econometrics (SoFiE). He earned a Ph.D. in economics from Cornell University and previously taught at UC San Diego and MIT.
Can scientific discovery scale as AI has scaled with data, models, and compute? LineShine is an exascale supercomputer designed to support high-precision HPC and data-intensive AI on a unified architecture. A central case study is a recent ACM Gordon Bell Prize finalist work, with the XLSDFT framework scaling Kohn–Sham DFT to a 100-million-atom silicon system and an 11-million-atom solid-state battery interface — bringing first-principles simulation to experimentally relevant length scales and enabling direct, quantitative comparison with XPS measurements. Two additional examples, numerical–AI climate ensembles and full-machine generative AI training for Earth-observation compression, demonstrate LineShine’s broader ability to support both HPC and AI. Together, these experiences point toward a scaling law for scientific discovery based on the co-scaling of algorithms, data, models, systems, and validation.
Haohuan Fu is a Professor and Vice Dean at Tsinghua Shenzhen International Graduate School. He also serves as Deputy Director of the National Supercomputing Centers in Shenzhen. He received his Ph.D. in Computing from Imperial College London in 2009. His research focuses on high-performance computing, supercomputing architectures, software optimization, and AI for Earth system science. He has received three ACM Gordon Bell Prizes for work in atmospheric modeling, earthquake simulation, and random quantum circuit simulation.
The binding constraint on AI’s economic impact has shifted from model capability to organizational absorption. Evidence from large-scale field experiments at Boston Consulting Group and Procter & Gamble, involving more than 1,500 professionals, shows what happens when generative AI enters real knowledge work. Performance follows a jagged frontier: substantial quality gains on tasks within AI’s capability boundary and systematic degradation beyond it. Teamwork itself is reconfigured, with AI-augmented individuals matching the performance of unaugmented teams. Abundant machine expertise inverts the logic of organizational design: the scarce resource is no longer expertise but the judgment to orchestrate it. Scaling AI is therefore an institutional redesign problem, not a technology adoption problem.
High-performance computing is entering a decisive transition driven by forces that are largely external to traditional scientific HPC. The economics of AI and hyperscale cloud now shape leading-edge silicon, system architectures, and software ecosystems, while energy and data movement have become the dominant constraints on performance, facility design, and long-term sustainability. This talk examines how these dynamics shift HPC’s center of gravity from a primarily FP64, node-centric worldview toward accelerator-heavy, rack-scale, and workflow-defined systems.
We argue that the next era of scientific capability will be measured less by peak floating-point rates and more by time–energy–fidelity trade-offs across end-to-end pipelines. The most plausible path to “effective zettascale” is not brute-force FP64, but certified mixed-precision algorithms, communication-avoiding methods, AI-augmented reduced-order models, and hybrid AI+simulation workflows with rigorous error control and uncertainty quantification. We also outline an emerging reference architecture for platforms comprising integrated simulation, AI, and data/workflow partitions, federated with secure cloud resources and instruments.
The talk concludes with concrete technical priorities for 2026–2028: open AI+simulation testbeds with full-stack observability, standardized mixed-precision software layers, energy-aware runtimes and benchmarks, and workforce programs that bridge numerical HPC and AI. The goal is not to resist the AI-driven reshaping of compute, but to reclaim scientific agency within new hardware, economic, and policy realities.
Professor Jack Dongarra holds multiple prestigious positions: he is the recipient of the 2021 ACM A.M. Turing Award, a member of the US National Academy of Engineering, a foreign member of the Royal Society, an Emeritus Professor in the Electrical Engineering and Computer Science Department at the University of Tennessee, a Distinguished Research Participant at the Department of Energy’s Oak Ridge National Laboratory, a Turing Fellow at the University of Manchester’s School of Mathematics, and an Adjunct Professor at Rice University’s Computer Science Department.
Professor Dongarra has been involved in the design and development of high performance mathematical software for the past 40 years, especially regarding linear algebra libraries for parallel machines, vector processors and cloud environments. He has been a major driver in the creation of de facto standards (PVM and MPI) that have been widely used in computer and computational science. His accomplishments have earned him membership in the US National Academy of Engineering, appointment as a foreign fellow of the Royal Society in the UK, and foreign membership in the Russian Academy of Science.
Artificial intelligence is rapidly reshaping climate science. Foundation models are complementing traditional simulations, machine learning is accelerating data analysis and forecasting, and AI systems are increasingly being used to generate scientific hypotheses. Yet a fundamental question remains: where is AI in climate science actually heading? This talk argues that the future of AI for Science lies not only in accurate prediction, but in accelerating scientific discovery itself. While large language models are often criticized for hallucination, science depends on the generation of ideas that are not yet known to be true — the challenge is not to eliminate hallucination, but to transform it into constructive hypothesis generation coupled with rigorous verification. The talk discusses how future AI Scientists may combine language models, scientific literature, observational data, physical constraints, simulations, and automated verification systems to create closed-loop discovery processes, using climate science as a particularly demanding testbed where data are abundant but verification is often difficult given the scale and complexity of Earth systems.
Torsten Hoefler is a Professor of Computer Science at ETH Zürich, a member of Academia Europaea, and a Fellow of the ACM, IEEE, and ELLIS. He received the 2024 ACM Prize in Computing, one of the highest honors in the field. Following a “Performance as a Science” vision, he combines mathematical models of architectures and applications to design optimized computing systems. Before joining ETH Zürich, he led the performance modeling and simulation efforts for the first sustained Petascale supercomputer, Blue Waters, at the University of Illinois at Urbana-Champaign. He is also a key contributor to the Message Passing Interface (MPI) standard, where he chaired the “Collective Operations and Topologies” working group.
Torsten has won best paper awards at his field’s top conference, ACM/IEEE Supercomputing, in 2010, 2013, 2014, 2019, 2022, 2023, and 2024, and at other international conferences. For his work, he received the IEEE CS Sidney Fernbach Memorial Award in 2022, the ACM Gordon Bell Prize in 2019, the ACM Gordon Bell Prize in Climate Modeling in 2025, Germany’s Max Planck-Humboldt Medal, the ISC Jack Dongarra Award, the IEEE TCSC Award of Excellence (MCR), ETH Zürich’s Latsis Prize, the SIAM SIAG/Supercomputing Junior Scientist Prize, the IEEE TCSC Young Achievers in Scalable Computing Award, and the BenchCouncil Rising Star Award. He was elected to the first steering committee of ACM’s SIGHPC in 2013 and has been re-elected for every term since. His research interests revolve around performance-centric system design, including scalable networks, parallel programming techniques, and performance modeling for large-scale simulations and artificial intelligence systems.
We are entering an era in which humanity’s relationship with nature can increasingly be understood, measured, and redesigned through mathematics and artificial intelligence. This talk explores the emergence of an Algorithmic Planet, where algorithms move beyond analyzing information to helping us manage energy, water, natural resources, and ecosystems. Computational optimization is transforming engineering, from low-energy desalination to more efficient vertical-axis wind turbines, while satellites, remote sensing, and AI are revolutionizing our ability to observe the planet — turning satellite imagery into estimates of forest cover, carbon sequestration, biodiversity, and agricultural productivity. The next frontier moves beyond prediction toward causal AI and intervention: identifying which actions actually improve environmental outcomes. The Algorithmic Planet is not about replacing nature with technology, but about using mathematics and AI to understand nature better, manage resources more intelligently, and design systems capable of creating prosperity within planetary limits.
Prof. Luis Seco is Professor of Mathematics, advisor and former director of the Mathematical Finance Program at the University of Toronto, where he also directs RiskLab, a quantitative finance research laboratory established in 1996. He chairs the Centre for Sustainable Development at the Fields Institute and is an Affiliate Faculty Member at the Vector Institute for Artificial Intelligence. His current research focuses on applying mathematics, machine learning, and artificial intelligence to sustainability, climate risk, environmental measurement, carbon emissions, and sustainable finance. He was appointed a visiting ADIA Lab Fellow in 2022 and is co-founder of Feishu, the Fields Institute’s partner in China. He received Canada’s NSERC Synergy Award for Innovation in 2007 and was named Knight of Spain’s Order of Civil Merit in 2011 and of the Order of Isabel la Católica in 2025.
Deploying AI weather models at scale requires more than improving benchmark accuracy: models must adapt efficiently to heterogeneous tasks, retain skill across extended horizons, connect gridded outputs with real-world observations, and support downstream decisions. This talk presents a progressive research pathway from weather foundation models to operational climate intelligence — from WeatherPEFT, which enables task-adaptive, parameter-efficient tuning of weather foundation models, to TianQuan-S2S, which extends prediction to subseasonal-to-seasonal horizons, to observation-aligned generative and correction modules that bridge model outputs to practice. Together, these studies outline a scalable HPC–AI pathway from reusable foundation models, to reliable local forecast products, and ultimately to decision-relevant climate and renewable-energy services.
Juepeng Zheng is an Associate Professor at the School of Artificial Intelligence, Sun Yat-sen University, and a joint researcher at the National Supercomputing Center in Shenzhen. His research lies at the intersection of artificial intelligence, high-performance computing, and Earth system science, developing scalable AI methods for weather and climate prediction, remote sensing interpretation, and scientific foundation models. He has published more than 50 papers as first or corresponding author in top-tier international conferences (ICML, ICLR, NeurIPS) and journals (IEEE TGRS, IEEE GRSM, ISPRS P&RS, GMD).
Global climate discourse increasingly puts equal emphasis on mitigation (reducing emissions) and adaptation (remediation investments), driven both by recognition of rising physical risk and catastrophic loss and by precautionary principle guidance. Deploying adaptation solutions, however, still suffers from a lack of data, a lack of capital, and a lack of coordination. This presentation discusses addressing all three of these gaps through geospatial data science solutions.
Parag Khanna is Founder & CEO of AlphaGeo, the leading AI-powered geospatial analytics platform. He is the internationally bestselling author of seven books, including MOVE: Where People Are Going for a Better Future (2021), The Future is Asian (2019), Connectography (2016), How to Run the World (2011), and The Second World (2008). He was named one of Esquire’s “75 Most Influential People of the 21st Century” and featured in WIRED’s “Smart List.” He holds a Ph.D. from the London School of Economics and degrees from Georgetown University’s School of Foreign Service, and is a Young Global Leader of the World Economic Forum.
The water and resource recovery sectors are rapidly adopting digital technologies and AI-based decision support, from reducing energy and chemical consumption to improving treatment performance, maintenance, and resource recovery. But realizing the value of AI requires more than building a model — it requires reliable data, scalable digital infrastructure, integration with operational systems, and approaches that users can trust. This talk explores how water and resource recovery facilities can move from data collection and visualization to applied AI at scale, through practical examples and lessons from real-world implementation.
Nina Kshetry is Founder and President of Ensaras, Inc., providing data management and artificial intelligence solutions for water, wastewater, and resource recovery facilities through its AI-as-a-service platform, the Ensaras AI Wastewater Stack. She is also Co-founder and Vice President of Circle H2O, a wastewater resource recovery engineering and project management firm. She has two decades of experience spanning water and wastewater treatment, engineering, data analytics, and applied AI. She received her S.B. and S.M. degrees in environmental engineering from MIT, is a licensed professional engineer in Illinois and Florida, and a 2024 Presidential Leadership Scholar.
Climate Co-Pilot aims to turn heterogeneous Earth-system data into grounded, auditable support for climate-resilience decisions. This talk presents progress across three connected components: TerraBench and TerraAgent, an executable benchmark and ReAct-style framework for tool-grounded Earth-science reasoning spanning 403 tasks; EarthAlign, which aligns ERA5 fields with diverse natural-event records and multi-style grounded captions across roughly 554,000 samples from 2000–2024; and an EarthToken-based Evidence Graph that sits between data retrieval and the LLM planner, tracking provenance and uncertainty and returning the strongest supportable claim together with missing evidence.
Salman Khan is a Professor at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), where he works on multimodal learning, vision-language models, remote sensing, and AI for climate and environmental science. He previously served as a Senior Scientist at the Inception Institute of AI and a Research Scientist at CSIRO Data61, and is an Honorary Faculty member at the Australian National University. He has authored more than 150 papers in leading AI and computer-vision venues, is an inventor on 18 US patents, a Fellow with ADIA Lab, and a Senior Fellow with Microsoft AIEI.
Machine learning now touches every stage of the climate science pipeline, from emulating Earth system models and downscaling projections to reading the literature and running the analysis itself — but progress has outpaced our ability to say whether it is real. This talk frames the problem through three benchmarks: ClimateBench, which asks whether emulators generalise across emission pathways; ClimaQA, which tests whether language models hold graduate-level climate knowledge; and Zephyrus, which asks whether agents can carry out weather and climate analysis end to end. The talk argues for evaluation built around physical consistency, calibrated uncertainty, and genuine extrapolation, and for treating benchmarks as living infrastructure maintained alongside the models they judge.
Duncan Watson-Parris is an Assistant Professor at Scripps Institution of Oceanography and the Halıcıoğlu Data Science Institute, UC San Diego, where he leads the Climate Analytics Lab. His research focuses on aerosol–cloud interactions and their representation in global climate models, combining satellite observations, machine learning, and differentiable modeling to reduce uncertainty in anthropogenic forcing. He is a lead developer of JEM, the first fully differentiable Earth System Model, and leads the GAIA Initiative at UCSD. His research is supported by an NSF CAREER award, a Google Academic Research Award, and DARPA.
Shafi Goldwasser received the ACM Turing Award in 2012, and the Gödel Prize in 1993 and 2001 for her work on interactive proofs and connections to hardness of approximation. Her other honors include the ACM Grace Murray Hopper Award (1996), the RSA Award in Mathematics (1998), the ACM Athena Award (2008), the Benjamin Franklin Medal in Computer and Cognitive Science (2010), the IEEE Emanuel R. Piore Award (2011), and the L’Oréal-UNESCO International Women in Science Award (2021), among others. She is a member of the AAAS, ACM, NAE, NAS, and the Israeli Academy of Sciences, a foreign member of the Royal Society, and holds honorary degrees from Oxford, Carnegie Mellon, and several other universities.
Many attempts to deploy AI tools result in mixed or poor results. Successful deployments require understanding the strengths and weaknesses of both the technology and organization, so that productivity improvement doesn’t come at the cost of accuracy, risk, or unsustainable computational cost. This talk will present a strategy for successful AI adoption, and evidence of how well it works in real organizations.
Alex Pentland is a global expert on AI, data analytics, and secure distributed information systems. He began his career at Stanford before moving to MIT to help create the Media Lab and later the Institute for Data, Systems and Society. He has advised the OECD and, formerly, the UN Secretary-General, the EU Presidency, the World Economic Forum, and companies including Google, AT&T, and Nissan. He is a member of the US National Academy of Engineering and has received numerous awards, including DARPA’s 40th Anniversary of the Internet award, the Brandeis Privacy Award, and an AI Influencer Lifetime Achievement Award.
Can current AI tools answer research questions from healthcare databases on their own? The answer depends on the learning task the question implies: description, prediction (pattern recognition), or counterfactual prediction (causal inference). The first two are largely in hand. Given structured and labeled data: descriptive quantities are directly observable, and held-out data let us quantify how well a prediction algorithm has learned. Causal tasks have no such criterion. Their bottleneck is not the algorithm but the causal model that licenses the estimate. Humans acquire causal models through experimentation or literature searches; the machine equivalents are reinforcement learning and ontology-constrained language models. This talk assesses how far these substitutes get us, and where they still fail.
Miguel Hernán is a member of the Advisory Board of ADIA Lab. He is the Director of CAUSALab, the Kolokotrones Professor of Biostatistics and Epidemiology at the Harvard T.H. Chan School of Public Health, and faculty at the Harvard-MIT Division of Health Sciences and Technology. His team repurposes real world data into actionable evidence for the prevention and treatment of infectious diseases, cancer, cardiovascular disease, and mental illness. Their work has contributed to shape health research methodology worldwide.
Professor Hernán has received several awards, including the Rousseeuw Prize for Statistics, the Rothman Epidemiology Prize, and a MERIT award from the U.S. National Institutes of Health. He is elected Fellow of the American Association for the Advancement of Science and the American Statistical Association, and an Associate Editor of Annals of Internal Medicine. With about 200,000 citations, he has received the Highly Cited Researcher distinction for the last 15 years. Professor Hernán is a co-founder of Adigens Health.
Edward Jung is a global expert in innovation ecosystems, with 40 years of experience in software, R&D, entrepreneurship, and startups across multiple countries, and scientific expertise in mathematical physics and biophysics. An inventor and entrepreneur, he holds more than 1,200 issued patents and has founded more than 40 organizations spanning biomedicine, computing, networking, energy, and materials science. He founded Intellectual Ventures in 1999 after leaving Microsoft, where he was chief architect and co-founder of Microsoft Research. He has advised organizations including the National Academy of Sciences, Harvard Medical School, the Bill and Melinda Gates Foundation, and the World Health Organization.
Konstantin Novoselov is internationally recognized for his pioneering work on graphene — the thinnest and strongest material ever identified — for which he was awarded the 2010 Nobel Prize in Physics alongside Andre Geim. He is currently Director of the Institute for Functional Intelligent Materials at the National University of Singapore, and has authored more than 600 peer-reviewed publications. He has played a pivotal role in founding major research centers, including the National Graphene Institute, and is a Fellow of the Royal Society, the Royal Society of Chemistry, and the Institute of Physics. He joined ADIA Lab’s Advisory Board in 2025.
Reliable intelligence is not a property that agentic systems simply inherit from their underlying models. Once a model begins to plan, use tools, maintain memory, delegate tasks, and interact with other agents, reliability becomes a system-level challenge. Errors can accumulate across multi-step trajectories, safeguards can weaken outside conversational settings, and individually capable components can produce unsafe or unpredictable collective behaviour. This talk introduces Trustworthy Agentic Intelligence as an engineering approach to closing this reliability gap. It argues that agentic reliability must be designed into architectures, demonstrated through verifiable evidence, continuously assessed at runtime, and dynamically revised as models, tools, permissions, and operating conditions change. Moving beyond static benchmarks and one-time certification, the talk outlines a continuous-assurance paradigm based on trajectory-level evaluation, sandbox testing, evidence-based authorization, monitoring, and revocable trust. The central challenge is therefore clear: when reliable models become unreliable agents, trust must be engineered back into the system.
Professor Herrera received his M.Sc. in Mathematics in 1988 and Ph.D. in Mathematics in 1991, both from the University of Granada, Spain. He is an academician of the Royal Academy of Engineering (Spain). He has published over 600 journal papers, received more than 177,000 citations (Scholar Google, H-index 194), and is an editorial member of a dozen academic journals. Professor Herrera has been nominated as a Highly Cited Researcher in Computer Science and Engineering areas by Clarivate Analytics. His current research interests include computational intelligence, trustworthy AI (explainability, federated learning and safety), and general purpose AI.
Connecting genetic variation to disease-relevant genes remains a central challenge in human genetics and therapeutic target discovery. This talk introduces FM-GWAS, a framework that uses GeneUNet, a genomic foundation model, to transform personalized gene sequences into multimodal functional embeddings for gene-level association testing. Applied to 8,856 Human Phenotype Project participants across eight cardiovascular, metabolic, and musculoskeletal traits, FM-GWAS identifies more significant gene–trait associations than MAGMA, with reduced variant-density bias and support from pathway analyses and established disease-gene resources. The embeddings also outperform raw genotype features in predicting five disease traits. A case study of CACNA1H, a candidate osteoporosis-associated gene, illustrates how the framework links sequence variation to predicted functional effects and disease associations. These findings highlight the potential of genomic foundation models to advance disease gene discovery and generate mechanistic hypotheses.
Prior to joining MBZUAI, Professor Segal published more than 200 peer-reviewed publications, which have been cited more than 60,000 times. He received several awards and honors for his work, including the Overton Prize, awarded annually by the International Society for Bioinformatics (ICSB) to one scientist for outstanding accomplishments in computational biology, and the Michael Bruno Award. He heads the Human Phenotype Project, a large-scale (more than 10,000 participants) deep-phenotype prospective longitudinal cohort and biobank that his lab established, aimed at identifying novel molecular markers with diagnostic, prognostic and therapeutic value, and at developing prediction models for disease onset and progression. The deep profiling includes medical history, lifestyle and nutritional habits, vital signs, anthropometrics, blood tests, continuous glucose and sleep monitoring, and molecular profiling of the transcriptome, genetics, gut and oral microbiome, metabolome and immune system. Segal’s research focuses on developing and fine-tuning robust foundation models via self-supervised learning techniques based on the data of the Human Phenotype Project, extending large language model approaches to diverse clinical and multi-omics data modalities, including imaging, time-series, tabular, and sequencing-based data. He was elected as an EMBO member and as a member of the Young Israeli Academy of Science. During the COVID-19 pandemic, Professor Segal developed models for analyzing the dynamics of the pandemic and served as a senior advisor to the government of Israel. Segal was awarded a B.Sc. in Computer Science summa cum laude from Tel-Aviv University, and a Ph.D. in Computer Science and Genetics from Stanford University. Before joining the Weizmann Institute, Professor Segal held an independent research position at Rockefeller University, New York.
Farah Shamout is an Assistant Professor of Computer Engineering at NYU Abu Dhabi, where she leads the Clinical Artificial Intelligence Lab. She is also affiliated with NYU Tandon School of Engineering (Computer Science and Biomedical Engineering) and NYU Langone Health (Radiology). At the Clinical AI Lab, her research focuses on developing machine learning methods and systems using heterogeneous real-world data for applications in precision health, including electronic health records and medical imaging. Her work emphasizes multi-modal learning, foundation models, and trustworthy AI, with the goal of improving performance and clinical utility in real-world settings.
Dr. Shamout completed her DPhil (PhD) in Engineering Science at the University of Oxford as a Rhodes Scholar, where she was a member of Balliol College. She holds a BSc in Computer Engineering (cum laude) from NYU Abu Dhabi.
Health systems generate vast amounts of data that could be used to estimate the effects of sustained treatment strategies, but commonly used causal inference tools may be sensitive to model misspecification. The Non-Iterative Conditional Expectation (NICE) g-formula typically relies on parametric models that may not adequately capture complex dependencies among time-varying confounders, treatments, and outcomes. We developed an LSTM-based NICE g-formula estimator that uses recurrent neural networks to learn these temporal dependencies and incorporates sample splitting and cross-fitting to support uncertainty quantification. We evaluated the estimator in simulated datasets modeled on longitudinal healthcare data from people with HIV and compared it with classical parametric estimators under simpler and more complex data-generating scenarios. In complex settings, the LSTM-based estimator reduced bias and produced confidence intervals with coverage close to the nominal level, although it had greater variance and computational cost. These findings suggest that incorporating deep learning into the g-formula can improve causal effect estimation in complex longitudinal data while supporting approximately valid statistical inference.
Sophia Rein is an Instructor at CAUSALab and the Department of Epidemiology at the Harvard T.H. Chan School of Public Health. Her research focuses on the development and application of causal inference and machine learning methods for the analysis of real-world healthcare data. Her primary interests include integrating deep learning into causal inference methods, and comparative effectiveness research to evaluate the safety and effectiveness of treatments using healthcare data. She holds a Ph.D. in Biostatistics and Epidemiology from University College London (UCL).
Artificial intelligence is changing how financial information is analyzed, forecast, and acted upon. This talk presents recent work along three directions. The first concerns large language models for corporate disclosure analysis. Fin-RATE is a benchmark built from real SEC filings that evaluates models on within-filing reasoning, cross-company comparison, and longitudinal tracking of firms over time, together with a fine-grained error taxonomy that reveals where models hallucinate figures, confuse entities, or mis-assign reporting periods. NumCache then addresses efficiency, compressing filings into number-aware KV caches and retrieving directly in cache space, retaining most of full-context accuracy at a fraction of the inference cost.
The second turns to decentralized finance, where multichain transaction histories from major lending protocols are used to assess borrower credit risk and predict liquidation events. Behavioral features extracted from on-chain activity feed gradient-boosting and neural classifiers, pointing to on-chain behavior as an alternative to traditional, centralized credit scoring.
The third examines financial forecasting with large and multimodal models, from adapting pre-trained LLMs to cryptocurrency price series — where selectively freezing transformer layers matches or outperforms established time-series baselines in short-term and few-shot settings — to multimodal approaches that combine numerical series with textual and visual market signals. The talk closes with open challenges in numerical faithfulness, reasoning across documents and time, and evaluation that reflects real-world financial analysis workflows.
Leandros Tassiulas’ research interests are in the field of computer and communication networks, with emphasis on fundamental mathematical models and algorithms of complex networks, architectures and protocols of wireless systems, sensor networks, emerging economies in networked communities, novel internet architectures, and experimental platforms for network research. His most notable contributions include the max-weight scheduling algorithm and the back-pressure network control policy, the maximum lifetime approach for wireless network energy management, and the consideration of joint access control and antenna transmission management in multiple-antenna wireless systems.
Dr. Tassiulas has been a Fellow of IEEE since 2007, and his research has been recognized by several awards, including the inaugural INFOCOM 2007 Achievement Award “for fundamental contributions to resource allocation in communication networks,” the INFOCOM 1994 Best Paper Award, a National Science Foundation (NSF) Research Initiation Award (1992), an NSF CAREER Award (1995), an Office of Naval Research Young Investigator Award (1997), and a Bodossaki Foundation Award (1999).
Algorithms increasingly supply liquidity in modern securities markets, raising a central question for market design and regulation: do learning algorithms intensify competition and improve liquidity, or can they sustain persistent non-competitive outcomes? A growing literature finds that reinforcement-learning agents can generate supra-competitive spreads and elevated quotes even without explicit communication or collusive intent. A key limitation of existing work, however, is that algorithm design is imposed exogenously: learning rates, exploration schedules, and evaluation horizons are fixed by the researcher rather than chosen by profit-maximizing dealers facing competitive incentives. In practice, dealers and technology providers actively select these parameters, and this strategic dimension has been largely absent from the literature.
Agostino Capponi is a Professor of Industrial Engineering and Operations Research at Columbia University, with a courtesy appointment at the Columbia Business School. He is also the Director of the Center for Digital Finance and Technologies and a member of the Data Science Institute. Agostino’s research interests broadly span financial technology, market microstructure, financial networks, machine learning in finance, supply chains, energy systems, and sustainability. His research has been published in major journals of his field.
Agostino is a fellow of the Crypto and Blockchain Economic Research Forum, an academic fellow of the Luohan Academy established by the Alibaba Group, and an external research fellow at the Fintech@Cornell Center. He serves as an Area Editor at Operations Research and a Co-Editor of Mathematics and Financial Economics, and has been on the editorial boards of major journals in his field. Agostino is a member of the Council of the Bachelier Finance Society and also served as Chair of the SIAM Activity Group in Financial Mathematics and Engineering. He received a Presidential Early Career Award for Scientists and Engineers (PECASE) in 2025.
Integration of distributed ledger systems with the existing digital finance ecosystem presents several challenges, primarily due to the complex nature of many traditional asset transaction workflows. While ledger-based digital currencies may offer faster payment settlement times, the broad adoption of digital currencies requires a standardized model for their transfers based on strong roots of trust. In this presentation we discuss a newly proposed model for coordinating transfers among gateways that enables seamless integration with traditional systems and provides protection against dishonest actors in the workflow.
Dr. Thomas Hardjono is an early pioneer in the field of digital identities and trusted hardware, and has been instrumental in the development and broad adoption of the MIT Kerberos authentication protocol. His activities include leading standards development efforts, notably at the IETF (Internet Engineering Task Force), IEEE, the Trusted Computing Group, the Confidential Computing Alliance, and others.
He has published more than 70 technical conference and journal papers, several books, and more than 30 patents. He is currently involved in several startups around the MIT community. His current area of interest is Web3 digital assets, with a focus on the interoperability of asset networks and the survivability of these networks against cybersecurity attacks.
This session shares the practical experience of building an AI-enabled platform offering a unified “ask me anything” interface into the organization’s complete data assets: a custom-built orchestrator that manages reasoning, memory, and content retrieval together. It looks in more depth at two of its novel components: a memory methodology that maintains what the platform has learned in a deterministic, auditable way, and a set of retrieval techniques designed to surface reliable evidence from complex investment documents. It closes with practical reflections on performance, maintenance, uptake, and the ongoing challenges of running the platform in production.
Edward Barker is Head of the AI Lab at Mubadala. The AI Lab researches and prototypes innovative AI and machine learning tools tailored to Mubadala’s unique business problems. Edward has more than fifteen years of experience across AI, data science and quantitative analysis. Before joining Mubadala, he held senior roles at Integrated Data Intelligence, ANZ, TAL, Liberty Financial and Medibank Private, developing systems for private-market intelligence, commercial loan pricing, insurance underwriting and credit-risk management. He holds a PhD and a Master of Mathematics from the University of Melbourne, as well as bachelor’s degrees in law and science from the University of Queensland. His research focuses principally on reinforcement learning and has been published in the Journal of Machine Learning Research.
Debates on trustworthy AI focus on model bias, alignment, and explainability. Yet trust in AI is only as strong as the security and resilience of the infrastructure it runs on. As the world concentrates AI compute into a handful of hyperscale, gigawatt-scale data centers, it creates critical national infrastructure without the protection doctrine such infrastructure demands. In this talk, Prof. Di Pietro draws on his work in critical-infrastructure protection and distributed-systems security to set out the threat surface unique to physical AI infrastructure: supply-chain and hardware risks, model-weight theft, and others specific to the Gulf region. He will also address the financial dimension, total cost of ownership and return on investment, that shapes how securely this infrastructure is actually built. He then turns to the resilience challenges that determine whether AI capability stays online under stress.
Roberto Di Pietro is a Full Professor of Computer Science at King Abdullah University of Science and Technology (KAUST), Saudi Arabia, where his research spans AI-driven cybersecurity, security for distributed systems, critical-infrastructure protection, and the security of blockchain and emerging technologies. With three decades of experience across academia, government, and industry, he has founded and led research and innovation labs across academia and the private sector, and previously led global security research at Bell Labs. He also founded and successfully exited a startup built on his research results.
He is a Fellow of the IEEE, a Fellow of the International Core Academy of Sciences and Humanities (CORE), a Distinguished Scientist of the ACM, and a member of Academia Europaea. His honors include the Jean-Claude Laprie Award.
Generative AI can significantly increase the productivity of knowledge workers. However, its conversational fluency may also encourage overreliance and the unseen delegation of cognitive work. Building on Agency Theory, this talk examines how employees may sub-delegate work to AI outside formal organisational oversight, creating a “shadow agent” that remains invisible to the organisation while firms continue to bear the consequences of potential errors. It also explores the resulting risks to human judgement and considers how structured friction in AI-enabled workflows can preserve verification, accountability, and human agency where decisions carry significant operational, financial, or strategic consequences.
Dr. Carlos Galera-Zarco is an Associate Professor at University College London (UCL), affiliated with UCL’s Centre for Organisational Network Analysis (CONA), and a member of the UCL Turing University Team, contributing to research groups at The Alan Turing Institute. He is also a member of the Organisation Theory Research Group (OTREG), a cross-university forum of organisation theorists from leading European universities and business schools.
His research focuses on how organisations adopt and integrate AI to create economic and strategic value while managing the challenges of technological transformation. He examines how AI reshapes organisational capabilities, business models, and ways of working, with a particular focus on human-AI trust, AI governance, cognitive delegation, and overreliance. His current work explores how organisations can realise the benefits of AI while preserving human agency, accountability, critical judgement, and resilience.
With an interdisciplinary background combining engineering, economics, and management, Carlos has led and contributed to research, advisory, and knowledge-transfer initiatives with major industry and public-sector organisations, including GSK and the UK National Physical Laboratory, as well as projects related to major international sporting events. He also delivers executive education on AI and digital transformation for senior professional audiences.
We introduce AIDO Cell, a cell biology simulator from GenBio AI built on a world model that maintains a coherent cell state across the full course of an in silico experiment. Users can apply unlimited sequences of perturbations, including genetic knockouts, knockdowns, overexpressions, and pharmacological treatments, to a single persistent cell instance and decode any resulting state into multiscale, multimodal readouts. These readouts span gene regulation, epigenomics, RNA isoform expression, protein structure, interaction, abundance, and localization, and cell morphology, combinations that would be difficult or impossible to measure simultaneously in a physical experiment. The same stateful representation supports branching trajectories and in-context molecular design, where AIDO Cell proposes molecules that induce a specified change in cell state. Our version 1.0 release provides prototype virtual cells for the K-562 and Hep-G2 lines, intended to demonstrate continuous, compositional experimentation rather than to serve as definitive high-fidelity models of those lines. AIDO Cell is adaptive by design, allowing users to refine existing virtual cells or build new ones from their own data. We close by positioning the virtual cell as a composable unit toward a complete digital organism.
Le Song, Co-founder and CTO of GenBio AI, is an accomplished machine learning researcher with a passion for developing innovative AI methods to tackle complex challenges in healthcare and drug design. His expertise spans structured prediction, neuro-symbolic integration, and scalable algorithms for dynamic, multi-modal data. At GenBio AI, Le leads the development of advanced foundation models that bridge the gap between artificial intelligence and biology, driving cutting-edge solutions for transformative scientific discovery. Before co-founding GenBio AI, Le held prominent academic and research roles, including Associate Professor at the Georgia Institute of Technology and Associate Director of its Center for Machine Learning. He has also contributed to leading institutions like Google Research and Carnegie Mellon University. Recognized for his contributions to the field, Le has received multiple best paper awards at premier conferences such as NeurIPS, ICML, and AISTATS.
Can we predict what the human immune system will do and learn to program it? In this talk, I will present recent work from my group and our collaborators that brings together machine learning and experimental biology to understand immune function across biological scales. I will first show how models trained jointly across autoimmune and inflammatory diseases use population-scale genetic data to improve risk prediction and point to shared immune mechanisms. I will then describe how we combine protein language models with large-scale experiments to predict T cell receptor–antigen recognition and identify disease-relevant targets. Building on these advances, I will discuss our recent work on generative design of immune receptors against defined antigens. These examples motivate a closed loop in which models propose informative experiments, their results update and refine the models, and each cycle drives both prediction and design of immune function. I will close with a few open problems in AI and immunology and their implications for new therapies and diagnostics.
Aly A. Khan, Ph.D., is an Assistant Professor at the University of Chicago and Senior Director of AI Science at Biohub, Chicago. His laboratory develops foundational AI/ML methods to decode inflammation, immune recognition, and cell-state transitions across biological scales. He pioneered machine learning approaches for mapping immune repertoire specificity and high-dimensional cellular dynamics. Dr. Khan previously led research projects at Merck, Genentech, Tempus AI, and 23andMe, and has contributed to three FDA-approved clinical oncology diagnostics. His research is supported by an NIH DP2 NIAID New Innovator Award, a Chan Zuckerberg Biohub Investigator Award, and the Common Mechanisms of Autoimmunity Insight Award.
Foundation models have a key role to play in transforming healthcare by supporting clinical practice with carefully-constructed systems — but where they must be sufficiently robust for use in healthcare systems. This talk presents recent work in this field, with a particular emphasis on the creation of novel tools for use in hospitals, primary care systems, and in public health settings.
Professor David Clifton is the Royal Academy of Engineering Chair of Clinical Machine Learning at the University of Oxford, where he leads the Computational Health Informatics (CHI) Lab, and NIHR Research Professor, the first non-medical scientist appointed to the NIHR flagship chair. He is a Fellow of the IET, a Fellow of the Alan Turing Institute, a Research Fellow of the Royal Academy of Engineering, Visiting Chair in AI for Health at the University of Manchester, and a Fellow of Fudan University. Trained in machine learning at Oxford under Lord Tarassenko, his early work produced patented jet-engine health monitoring systems used on the Airbus A380, Boeing 787, and Eurofighter Typhoon, before he turned to AI for healthcare.
His research has been commercialised through the spin-outs OBS Medical, Oxehealth, Biobeats, Sensyne Health, and Marley Health, and the CHI Lab now spans a second site in Suzhou, the Wellcome Trust Flagship Centre with the Oxford University Clinical Research Unit in Vietnam, the Oxford-CityU Centre for Cardiovascular Engineering in Hong Kong, the AI theme of the Oxford Pandemic Sciences Institute, and the Oxford-GSK Centre for Biostatistics and AI. His work has won over 40 awards, including an EPSRC Grand Challenge award, the inaugural Oxford Vice-Chancellor’s Innovation Prize, and the 2022 IEEE Early Career Award.
Dialysis generates dense longitudinal data on millions of treatments, which makes it a natural setting for machine learning in routine care. This talk describes how Fresenius Medical Care is building AI on its global clinical dataset, from predictive models for complications and anemia management to the operational and governance work required to take such models from pilot to daily use across dialysis networks.
Dr. Milind Nikam is Chief Clinical Officer and Head of Digital Health, Global Medical Office, at Fresenius Medical Care (FMC), based in Dubai, UAE, where he leads medical governance and digital health in Asia Pacific. He joined FMC in 2017 as Medical Director for Singapore. He graduated in medicine from Mumbai with a gold medal, completed postgraduate training in internal medicine and nephrology in the United Kingdom, earned a doctorate in dialysis vascular access from the University of Manchester, and trained in interventional nephrology in the UK. His interests include vascular access and device development, hemodialysis medical program management, and the application of digital health, including AI, to improve dialysis care.
Most discussions of clinical AI end at retrospective validation; this talk starts there. Drawing on first-hand experience deploying several AI systems in real clinical practice in the UAE, it explores the long and often invisible journey from a model that works to one that is trusted, approved, and adopted. The talk looks at the realities that rarely make it into papers: navigating regulatory pathways that were not designed with AI in mind, earning the confidence of clinicians and health authorities, deciding who is accountable when AI is part of a clinical decision, and putting responsible AI principles into day-to-day practice rather than leaving them on paper. It also reflects on the new questions generative AI raises for healthcare governance and oversight, and closes with practical lessons for anyone trying to turn promising AI into safe, lasting clinical impact.
Dr. Ronnie Rajan is Vice President of AI & Applied Sciences at M42, a global health organization that harnesses artificial intelligence, technology, and genomics to transform healthcare. A physician with over eight years of clinical experience and a postgraduate degree specializing in AI and clinical data science from IIT Kharagpur, India, he has over nine years of experience bridging medicine and data science to enable the responsible adoption of AI in healthcare. At M42, he leads AI R&D, advancing the translation of cutting-edge AI research into clinically meaningful applications. He also leads enterprise AI Governance at M42 and serves as the Responsible AI Ambassador on the G42 RAI Council.