Invited Speakers

Dante D. Sanchez-Gallegos
Universidad Carlos III de Madrid, Spain

Modern scientific workflows in the computing continuum face significant security and reliability risks during cross-domain data movement. Tools for enforcing non-functional requirements (NFRs), such as confidentiality, integrity, and fault tolerance, mitigate these risks but introduce performance overhead. In current simulation frameworks, there is a lack of systematic methodologies for analyzing and identifying tradeoffs between enforcing NFRs and preserving workflow efficiency before physical deployment across the computing continuum. To address this gap, this paper introduces a simulation model that estimates the cost of enforcing NFRs in distributed workflows across the computing continuum. The model supports the parallel execution of NFRs as patterns, capturing the impact of data partitioning, synchronization, and load balancing on overall performance. We validated this model across heterogeneous infrastructures, including Raspberry Pi edge nodes, fog PCs, and an HPC cloud cluster, through multiple experiments and a case study based on a real-world medical workflow for CT scan management. The results demonstrate that the model successfully identifies critical architectural bottlenecks before physical deployment; for example, accurately modeling the performance scaling of parallel patterns and evaluating energy consumption tradeoffs. Overall, our simulation model demonstrates its usability as a scalable tool for proactive workflow optimization.

Guido Guarnieri
ENEA, Italy

CRESCO8 is an HPC system based on Intel Xeon Emerald Rapids CPUs and Ponte Vecchio GPUs, designed to support large-scale simulations for magnetic confinement fusion and materials science. The system provides a comprehensive software environment for first-principles plasma modeling, including the major gyrokinetic codes GYSELA,GENE and ORB5, enabling detailed studies of turbulence and transport in fusion-relevant tokamak regimes. In addition, CRESCO8 supports advanced materials simulations through widely used electronic-structure and atomistic codes like Quantum ESPRESSO and CP2K. Selected applications have been ported and optimized for Intel GPU architectures, including Intel Ponte Vecchio, demonstrating effective CPUGPU portability and satisfactory performance. Experimental and simulation data analysis is supported through the IMAS software suite for tokamak workflows. CRESCO8 serves as the reference supercomputing facility for the Divertor Tokamak Test (DTT) facility, providing a key computational infrastructure for integrated nuclear fusion research.

Kyle Chard
University of Chicago and Argonne National Lab, USA

Agents can now externalize experience into memory, consolidating historical traces into semantic knowledge and procedural shortcuts that persist between sessions. Such memory is typically private to a single agent. We argue that a critical class of agentic memory is inherently collective:what an ecosystem of heterogeneous, distributed agents, operating across environments no one of them controls or fully trusts, has learned about the data sources, services, and tools they depend on. We present Cairn, a community reputation layer that illustrates this gap. Cairn enables agents to review the community’s opinion of a resource before use and submit evidence-backed ratings after. Cairn aggregates independent observations and uses a time-decayed Beta model with confidence shrinkage to weigh reviews. Cairn also supports free-text discovery by searching embeddings across reviewers’ scoring rationales. We evaluate Cairn’s reputation engine under adversarial simulation (liars, colluding cliques, camouflage, slander), benchmark discovery retrieval, and report a case study of rating heterogeneous agents in the wild.

Rosa Filgueira
University of St. Andrews, UK

Administrative research-funding records contain valuable evidence for metascience and science-policy analysis, but they are rarely analysis-ready. They are distributed across heterogeneous tables and contain duplicated information, missing references, inconsistent identifiers, and limited metadata for comparing funding instruments. This paper presents a methodology for transforming UK Research and Innovation’s Gateway to Research (GtR) records into an enhanced metascience infrastructure. The approach combines data-quality auditing, schema extension, loading-time repairs, enrichment heuristics, and materialised analytical tables. The resulting database supports two public dashboards:GTR Analyser, for research-question-driven comparison of funding instruments, and GTR Explorer, for inspection and export of enriched records. A case study compares applicant-led and thematic funding across councils, organisations, partners, places, and outcomes.

Manish Parashar
University of Utah, USA

The current data renaissance, accelerated by advances in artificial intelligence, is transforming scientific discovery across disciplines. As data becomes increasingly abundant, the challenge is no longer simply finding data but finding data that is fit for use. Doing so requires looking beyond datasets themselves to the network of relationships that connect data to the people, workflows, publications, and institutions that produce and use it. This talk explores the social networks of data through data usage knowledge graphs, which capture the relationships among datasets, computational workflows, users, publications, and organizations. These graphs expose the contextual signals needed to assess provenance, trustworthiness, relevance, and fitness for purpose, enabling more informed data discovery and responsible reuse. Using the National Data Platform as a systems exemplar, I will present an architectural approach for building open, interoperable, and trustworthy data ecosystems. I will show how graph-based representations of data usage strengthen transparency and reproducibility while enabling AI-driven discovery and reasoning across disciplinary boundaries. Ultimately, I argue that understanding the social networks surrounding data is key to helping both researchers and intelligent systems identify data that is truly fit for use.

Ilkay Altintas
University of California, San Diego, USA

Scientific workflows have been a defining contribution of e-Science, enabling complex computational investigations to become more automated, scalable, reproducible, and reusable. Together with advances in distributed computing, cyberinfrastructure, data management, and open science, workflows have transformed research across disciplines. However, many contemporary challenges, e.g., wildfire resilience, climate adaptation, public health, sustainable agriculture, and disaster preparedness, cannot be addressed through scientific workflows alone. Addressing them requires infrastructure that persistently connects observations, models, artificial intelligence, operational systems, institutions, and human expertise. This paper describes how such infrastructure can be built. We introduce societal computing as an evolution of e-Science that extends reproducible workflows into programmable innovation ecosystems, and identify the infrastructure characteristics it requires:composability, interoperability, extensibility, continuous learning, human-centered AI, trustworthy operation, and integration with real-world decision processes. We ground this account in the WIFIRE program, which over roughly fifteen years grew from data-driven scientific workflows for wildfire modeling into an operational ecosystem linking fire science, data, simulations, cyberinfrastructure, emergency agencies, utilities, and communities. Building on lessons from WIFIRE, the National Data Platform (NDP) generalizes these lessons into reusable, federated infrastructure—spanning data discovery, computational workspaces, collaborative environments, education, and AI-ready services that other domains can adopt to build their own societal computing ecosystems. Together, WIFIRE and NDP offer a concrete, tested blueprint for the infrastructure the next generation of e-Science needs:not only reproducible science, but sustained, reusable, and trustworthy pathways from scientific knowledge to actionable use of knowledge.

José Luis Gonzales Compeán
Cinvestav Tamaulipas, México

This paper presents STORI-Jub, a data science service hub whose architecture is inspired by el Caracol, the ancient observatory in Chichen Itzá, Yucatán, alongside STORI, a novel retrieval methodology designed to enhance LLM agent performance. STORI vectorizes Spatial–Temporal, Observational, Reference, and Interest profiles, allowing agents within the Jub ecosystem to automatically ingest profiled vectors for tools and skills via the Model Context Protocol (MCP). This setup facilitates the recursive generation of massive, searchable information products and multi-view analytical perspectives. To evaluate STORI, we compared local Jub agents using conventional RAG (on-demand processing of raw and cured data) against agents leveraging STORI profile-based vectorization. Both configurations were tested on real-world case studies featuring disease mortality and toxic substance databases provided by participating healthcare institutions under Mexican government agency support (Secihti, MADTeC-2025-478). Agents were evaluated on efficacy and efficiency in handling heterogeneous queries and questions. Experimental results revealed that STORI improves retrieval accuracy and efficiency when conducting exploratory studies, prospective research, and situational decision-making.

Ravi Madduri
Argonne National Laboratory

The American Science Cloud (AmSC), established under the Genesis Mission of the U.S. Department of Energy (DOE), aims to integrate DOE high-performance computing systems, experimental facilities, and data resources into a single, coordinated, AI-driven discovery platform. AmSC’s early services focus on curated artifacts, such as gated inference access to hosted models, experiment tracking, and function execution across computing facilities. However, what these services lack is a means to train a model across organizational boundaries where data cannot be centralized due to policy, privacy, or scale. This is, by definition, a use case for federated learning (FL) and a growing class of scientific AI. In this paper, we show that this gap can be bridged by deploying the orchestration logic of the Advanced Privacy-Preserving Federated Learning (APPFL) framework as a scalable cloud service on top of the primitives AmSC already provides:project-scoped authentication that supports secure and reliable federation membership, function execution that drives distributed training at each site, experiment tracking that records round-level performance, and finally, the model-hosting and inference infrastructure that can be leveraged to distribute the federated trained models to authorized participants. We argue that offering federated computing as an important AmSC service would unlock privacy-constrained scientific collaborations, enabling public-private partnerships in model building while exercising and enhancing the platform’s own federated infrastructure.

Kris Bubendorfer
Victoria University of Wellington, New Zealand

Highly Pathogenic Avian Influenza (HPAI) H5N1 reached Aotearoa New Zealand in 2026, creating an urgent need to understand its likely impact on vulnerable bird populations before comprehensive monitoring data becomes available. Building on our earlier work, in which we correctly predicted the Brown Skua as the most likely vector of H5N1 into New Zealand, now that H5N1 has been confirmed, we extend the modelling to purely terrestrial native raptors. We use community science observations from eBird to infer a first estimate of mortality for the raptors K\u0101hu (Swamp Harrier), and K\u0101rearea (New Zealand Falcon) by comparison with ecologically similar overseas species affected by H5N1 and matched control species. In this paper we demonstrate one use case of how large-scale community science datasets can be repurposed to support rapid modelling and decision making when conventional data collection cannot keep pace with emerging events.

Iacopo Colonnelli
University of Turin, Italy

The Centre of Excellence on Workflow Orchestration, Regularisation, Knowledge Enablement, and Retrospection (CoWORKER) is a European cross-disciplinary competence centre for large-scale scientific workflows that aims to improve the productivity, portability, and sustainability of complex HPC- and AI-enabled applications. By integrating technical services, co-design activities, training, and community-driven standards, CoWORKER will support the complete workflow lifecycle across heterogeneous computing environments, including HPC, cloud, AI, and quantum platforms. The initiative promotes interoperability, reproducibility, and open software ecosystems while strengthening European digital sovereignty. Through collaboration with the EuroHPC ecosystem and the broader scientific community, CoWORKER seeks to accelerate workflow adoption, software modernisation, and the long-term impact of workflow-based scientific applications.

Anirban Mandal
Renaissance Computing Institute, UNC Chapel Hill, USA

Scientific workflows increasingly span sensors, instruments, edge devices, programmable networks, clouds, and HPC facilities. In this computing continuum, resources appear and disappear, network conditions change, failures propagate across layers, and decisions often must be made close to where data is produced. Treating the continuum as a collection of static resources managed by a centralized scheduler is becoming increasingly inadequate.This talk traces a journey from conventional scientific workflows to adaptive edge-to-core execution and, ultimately, distributed, decentralized, multi-agent workflow management. I will describe experiments in dynamic computation placement, programmable network services, controlled anomaly injection, high-throughput data transport, and decentralized workload selection. A recurring theme is the role of research testbeds - particularly FABRIC - as “digital wind tunnels” where new infrastructure ideas can be subjected to realistic network conditions, controlled failures, and repeatable experiments before real-world deployment. I will outline a vision for an adaptive computing continuum in which distributed agents observe, negotiate, and act while remaining explainable, reproducible, and aligned with scientific intent.

Michele Di Capua
US srl, Italy

Cytoplasmic dynamics in the human zygote are a promising non-invasive biomarker of embryo competence, but the cytoplasmic wave (CW) is subtle, brief, and easily confused with the global movement of the embryo and with illumination changes, so that reliable visual assessment is hard even for expert embryologists. We present a two-stage pipeline for the automatic detection of the CW in time-lapse sequences. The first stage consists of a self-supervised spatial foundation model, a hybrid CNN–Transformer encoder pretrained on millions of embryo frames, which segments the embryo in every frame. The resulting masks drive a rigid registration and photometric correction of the sequence, providing a spatially aligned and appearance- normalized input for the second stage. Optical flow is computed on the registered segmented frames and concatenated with the corresponding frame intensities to construct an appearance– motion tensor. The second stage extends the same backbone to the spatio-temporal domain and is trained from scratch on this tensor to classify the presence of the wave while reconstructing two physically grounded temporal descriptors:wave motion magnitude and directional entropy. These provide both auxiliary supervision during training and interpretable cues supporting the final prediction. On 53 held-out sequences the model attains a ROC-AUC of 0.90 and a macro-F1 of 0.82, and localizes the event in time.

Daniel S. Katz
University of Illinois Urbana Champaign, USA

The Journal of Open Source Software (JOSS) is a diamond open-access journal with an open, collaborative, peer-review model for reusable open-source research software packages and associated short papers. Over its ten-year history, the journal has had a consistent aim of improving research software packages and providing scholarly credit to the people behind them. However, research culture and technology have changed over this time, most dramatically in the last few years with the rise of the use of Generative AI for both text and then software. This has led to both positive and negatives changes for JOSS, which are described in this talk.

Ioan Raicu
Illinois Institute of Technology, USA

Blockchain technology offers compelling properties for scientific computing, including decentralization, immutability, transparency, and verifiable provenance. Yet today’s blockchain systems face a fundamental mismatch with science:many are too slow, energy-intensive, and expensive to support data- and compute-intensive scientific workflows at scale. What if blockchain infrastructure could approach the performance and energy efficiency of centralized systems while preserving the trust and resilience of decentralization? This talk explores how advances in sustainable, high-throughput blockchain architectures could enable a new generation of scientific computing infrastructure. We present lessons from MEMO, a high-throughput blockchain architecture designed around the Proof-of-Space consensus mechanism that uses memoization and efficient indexing to reduce the computational, memory, storage I/O, and energy costs traditionally associated with blockchain consensus. MEMO targets thousands of transactions per second, with the potential to scale toward millions of transactions per second through sharding, while operating across platforms ranging from Raspberry Pis to high-performance servers. Low-cost, high-throughput blockchains could provide decentralized infrastructure for scientific data provenance, reproducible workflows, federated resource sharing, incentive mechanisms, and auditable AI and computational pipelines. Scientific datasets, computations, and results could become independently verifiable without requiring a single trusted institution. By making blockchain faster, cheaper, and dramatically more energy efficient, sustainable blockchain research can transform the technology from an expensive mechanism for digital currencies into practical infrastructure for open, reproducible, and collaborative science.