A core challenge in scientific machine learning, and scientific computing more generally, is modeling continuous phenomena which (in practice) are represented discretely. Machine-learned operators (MLO) have been introduced as a means to achieve this modeling goal, as this class of architecture can perform inference at arbitrary resolution. In this work, we evaluate whether this architectural innovation is sufficient to perform “zero-shot super-resolution,” namely to enable a model to serve inference on higher-resolution data than that on which it was originally trained. We comprehensively evaluate both zero-shot sub-resolution and super-resolution (i.e., multi-resolution) inference in MLOs. We decouple multi-resolution inference into two key behaviors: 1) extrapolation to varying frequency information; and 2) interpolating across varying resolutions. We empirically demonstrate that MLOs fail to do both of these tasks in a zero-shot manner. Consequently, we find MLOs are not able to perform accurate inference at resolutions different from those on which they were trained, and instead they are brittle and susceptible to aliasing. To address these failure modes, we propose a simple, computationally-efficient, and data-driven multi-resolution training protocol that overcomes aliasing and that provides robust multi-resolution generalization.

One year ago, the AISLE roadmap argued that autonomous laboratories operated as isolated islands and proposed a grassroots network organized around five critical dimensions. The field has since moved faster than anticipated. Multi-agent systems have produced experimentally validated hypotheses, self-driving laboratories have grown more interoperable and orchestrated, reasoning-trained and domain foundation models have raised the capability ceiling, and the Genesis Mission has placed autonomous experimentation at the center of U.S. federal science strategy, with industry emerging as a primary actor. Progress has met a sobering counter-current, including a corrected flagship discovery result, benchmarks showing that agents which rival experts on closed-ended questions still complete only a fraction of open-ended research, and fabricated citations surfacing at leading venues. We read this as the defining tension of the field. Producing a candidate discovery is no longer the hard part, but verifying it is, and this asymmetry now limits autonomous science more than raw model capability. We update the roadmap around seven dimensions, revisiting the original five and elevating two former cross-cutting concerns, trust, verification, and reproducibility, and safety, security, and governance, to first-class status. We assess the original milestones (M1 through M14) as achieved, partially achieved, reframed, or open, add four new milestones (M15 through M18), and scope the path forward to a two-year horizon. The first year concentrates on interfaces, protocol adoption, and the scaffolding of verification, and the second targets federation, zero-trust coordination, and governance. Throughout, we position the grassroots network as the interoperability fabric that lets national programs, international initiatives, and commercial platforms connect rather than re-silo.

The term scientific workflow has evolved over the last two decades to encompass a broad range of compositions of interdependent compute tasks and data movements. It has also become an umbrella term for processing in modern scientific applications. Today, many scientific applications can be considered as workflows made of multiple dependent steps, and hundreds of workflow management systems (WMSs) have been developed to manage and run these workflows. However, no turnkey solution has emerged to address the diversity of scientific processes and the infrastructure on which they are implemented. Instead, new research problems requiring the execution of scientific workflows with some novel feature often lead to the development of an entirely new WMS. A direct consequence is that many existing WMSs share some salient features, offer similar functionalities, and can manage the same categories of workflows but also have some distinct capabilities. This situation makes researchers who develop workflows face the complex question of selecting a WMS. This selection can be driven by technical considerations, to find the system that is the most appropriate for their application and for the resources available to them, or other factors such as reputation, adoption, strong community support, or long-term sustainability. To address this problem, a group of WMS developers and practitioners joined their efforts to produce a community-based terminology of WMSs. This paper summarizes their findings and introduces this new terminology to characterize WMSs. This terminology is composed of fives axes: workflow characteristics, composition, orchestration, data management, and metadata capture. Each axis comprises several concepts that capture the prominent features of WMSs. Based on this terminology, this paper also presents a classification of 23 existing WMSs according to the proposed axes and terms.

Learning and deploying inexpensive replacements for computationally expensive subroutines “on-the-fly” (OTF) during a dynamic simulation offers potential advantages and unique drawbacks. In OTF learning, a machine learned surrogate function is trained to replace a target subroutine in a simulation as the dynamics evolve. The advantages of OTF learning include reducing simulation error and model training costs, but the weaknesses include the possibility of introducing artifacts when adaptively updating the physics of the simulation over time. Here, we enhance an existing control system, Proxima, which ensures that surrogates are used appropriately in time-independent state-sampling based simulations, to time-dependent dynamical simulations by introducing a “blending” procedure that hides discontinuities when transitioning between original subroutine and surrogate. Our new control system, Proxima+Blend, produces a blend of the surrogate and target functions according to the relationship between error and an uncertainty signal observed as the dynamic simulation evolves. We show that while the original control system can shorten application runtime and accurately capture some macro-scale observables of a molecular dynamics simulation, the addition of blending is necessary to avoid unphysical dynamics at shorter time and length scales and to correct observables derived from these dynamics. Proxima+Blend delivers a 1.5× speedup over use of the target subroutine while producing solutions within 5% error in dynamical quantities, while the original Proxima algorithm has up to 80% error. Our implementation of Proxima+Blend can be deployed by simply replacing the existing subroutine with a wrapper that includes a machine learning approach for surrogate training along with specifying control and uncertainty signals for the simulation of interest. In this extended version of our conference paper (Tynes et al., 2025), we provide a more detailed analysis of the behavior of the Proxima+Blend controller and of our uncertainty metric, provide more details on our implementation and open-source release, and further contextualize how our findings contribute to the existing literature.

Artificial Intelligence for scientific applications increasingly requires training large models on data that cannot be centralized due to privacy constraints, data sovereignty, or the sheer volume of data generated. Federated learning (FL) addresses this by enabling collaborative training without centralizing raw data, but scientific applications demand model scales that requires extensive computing resources, typically offered at High Performance Computing (HPC) facilities. Deploying FL experiments across HPC facilities introduces challenges beyond cloud or enterprise settings. We present a comprehensive cross-facility FL framework for heterogeneous HPC environments, built on Advanced Privacy-Preserving Federated Learning (APPFL) framework with Globus Compute and Transfer orchestration, and evaluate it across four U.S. Department of Energy (DOE) leadership-class supercomputers. We demonstrate that FL experiments across HPC facilities are practically achievable, characterize key sources of heterogeneity impacting the training performance, and show that algorithmic choices matter significantly under realistic HPC scheduling conditions. We validate the scientific applicability by fine-tuning a large language model on a chemistry instruction dataset, and identify scheduler-aware algorithm design as a critical open challenge for future deployments.

High-performance computing (HPC) has evolved over decades through multiple architectural transitions, from vector supercomputers to massively parallel CPU clusters and GPU-accelerated systems, continuously expanding the frontier of scientific discovery. With the emergence of quantum processing units (QPUs) as practical computational accelerators, a new opportunity arises to further extend this trajectory by integrating quantum and classical computing paradigms. This paper presents Quantum Integrated High-Performance Computing (QHPC), a visionary architectural framework that unifies CPUs, GPUs, FPGAs, and QPUs as first-class heterogeneous resources. We propose a layered system design comprising unified resource management, quantum-aware scheduling, hybrid workflow orchestration, middleware and programming abstraction, interconnect technologies, and a tiered execution model enabling seamless workload partitioning across classical and quantum backends. A central aspect of our vision is a strong user requests abstraction layer that exposes heterogeneous resources through a unified job submission interface, similar in spirit to existing schedulers such as Slurm, allowing users to describe workloads in a consistent template independent of underlying compute type or location. Drawing insights from prior accelerator integration eras, we outline how QHPC can support emerging workloads in quantum chemistry, materials discovery, combinatorial optimization, and climate modeling. We conclude by highlighting open challenges in building scalable, reliable, and programmable quantum-classical infrastructures that seamlessly connect global users to heterogeneous compute resources for future quantum-classical HPC ecosystems.

Adding reliable checkpoint/restart support to an MPI scientific application is a time-consuming expert effort that requires deep knowledge of both the application and resilience. We ask whether a frontier large language model can perform this work end-to-end without human intervention. We assemble a benchmark suite of MPI applications spanning diverse domains and computation patterns, and drive an iterative code-generation loop for each application using Anthropic's Claude Opus 4.7 invoked through the OpenCode CLI. Across six scientific applications, the LLM generates working checkpoint/restart code in 50 minutes on average while consuming 3.4 M tokens per application. The generated code adds negligible overhead during normal failure-free execution on five of six applications and recovers from injected process failures with efficiency comparable to human-engineered checkpoint/restart implementations. These results suggest that automated end-to-end LLM-driven resilience engineering is technically viable today for a meaningful fraction of HPC applications.

Digital twins of networked infrastructures, known as Virtual Infrastructure Twins (VITs), are increasingly used for software development, pre-deployment testing, and design space exploration. While VITs avoid the costs and potential disruptions associated with experiments on operational networks, their throughput measurements are typically not sufficiently accurate for performance profiling of wide-area networks that they emulate. Here, machine learning (ML) methods are developed to transform these inaccurate VIT network throughput measurements to closely match in peak and overall profile of those from a physical testbed or production network. First, a micro kernel network reflecting a physical network is utilized to collect one-time measurements on a host to support this ML transformation. Then, a generic multi-modal ML method is developed to learn a map that transforms measurements from subsequent VITs on the same host to match past, current and follow-on testbed and cloud networks. ML generalization equations are derived to establish its correctness and probabilistically guarantee its generalization accuracy. Experimental results are presented for a variety of VIT hosts with target testbed and cloud networks; they include a case study of a four-site science ecosystem wherein inaccurate convex VIT measurement profiles are transformed into accurate concave profiles of target networks.

Intrinsically disordered proteins (IDPs) represent crucial therapeutic targets due to their significant role in disease – approximately 80% of cancer-related proteins contain long disordered regions – but their lack of stable secondary/tertiary structures makes them "undruggable". While recent computational advances, such as diffusion models, can design high-affinity IDP binders, translating these to practical drug discovery requires autonomous systems capable of reasoning across complex conformational ensembles and orchestrating diverse computational tools at scale.To address this challenge, we designed and implemented StructBioReasoner, a scalable multi-agent system for designing biologics that can be used to target IDPs. StructBioReasoner employs a novel tournament-based reasoning framework where specialized agents compete to generate and refine therapeutic hypotheses, naturally distributing computational load for efficient exploration of the vast design space. Agents integrate domain knowledge with access to literature synthesis, AI-structure prediction, molecular simulations, and stability analysis, coordinating their execution on HPC infrastructure via an extensible federated agentic middleware, Academy. We benchmark StructBioReasoner across Der f 21 and NMNAT-2 and demonstrate that over 50% of 787 designed and validated candidates for Der f 21 outperformed the human-designed reference binders from literature, in terms of improved binding free energy. For the more challenging NMNAT-2 protein, we identified three binding modes from 97,066 binders, including the well-studied NMNAT2:p53 interface. Thus, StructBioReasoner lays the groundwork for agentic reasoning systems for IDP therapeutic discovery on Exascale platforms.

Lens methods interpret large language models (LLMs) by mapping intermediate activations to the output vocabulary, revealing how next-token predictions develop through the network. Trained lenses remain expensive: affine-translator parameters grow quadratically with model width, while exact, full-vocabulary Kullback–Leibler (KL) training dominates memory. Consequently, prior trained lenses have been applied to models of at most 20B parameters and remain tied to particular component types. We present OmniLens, which applies a single lens family to any model-width activation, whether residual stream, attention, or MLP, and combines two independent scaling techniques. First, low-rank translators make per-lens parameter growth linear in model width and reduce trainable parameters by up to 98.4

Vector databases have been designed and optimized for cloud environments; however, emerging scientific AI workloads (e.g., molecular search, meteorological trajectory detection, and literature-driven hypothesis generation) demand efficient, scalable execution on HPC systems. We present a large-scale evaluation of three state-of-the-art vector databases – Qdrant, Milvus, and Weaviate – on two production supercomputers, scaling to 256 distributed workers across 64 compute nodes. We evaluate representative workload patterns – mixed read/write and write-then-read – using popular benchmarks, multimodal embeddings, and a novel real-world scientific dataset. Our results reveal that workload characteristics can limit latency reduction, additional cores can reduce query throughput by up to 30.67

Federated Learning (FL) is a decentralized machine learning paradigm where models are trained on distributed devices and are aggregated at a central server. Existing FL frameworks assume simple two-tier network topologies where end devices are directly connected to the aggregation server. While this is a practical mental model, it does not exploit the inherent topology of real-world distributed systems like the Internet-of-Things. We present Flight, a novel FL framework that supports complex hierarchical multi-tier topologies, asynchronous aggregation, and decouples the control plane from the data plane. We compare the performance of Flight against Flower, a state-of-the-art FL framework. Our results show that Flight scales beyond Flower, supporting up to 2048 simultaneous devices, and reduces FL makespan across several models. Finally, we show that Flight's hierarchical FL model can reduce communication overheads by more than 60

Modern HPC file systems can contain billions of files and hundreds of petabytes of data, making even simple questions increasingly intractable to answer. Traditional file system utilities such as find and du fail to scale to these sizes. While external indexing tools like GUFI and Brindexer improve query performance, they remain batch-oriented and unsuitable for heterogeneous, rapidly evolving environments. We present Icicle, a scalable framework for continuous file system metadata indexing and monitoring. Icicle maintains a unified, up-to-date, and queryable view of file system state while supporting both periodic snapshot-based ingestion for bulk metadata updates and event-based ingestion for real-time synchronization from production systems such as Lustre and IBM Storage Scale. Built on Apache Kafka and Apache Flink, Icicle provides high-throughput, fault-tolerant, and horizontally scalable ingestion of metadata events into two complementary search indexes, enabling both individual file discovery and aggregate summary statistics by user, group, and directory. This architecture enables efficient support for both coarse-grained administrative queries and interactive analytics over billions of objects. Our experimental evaluation on production-scale HPC datasets demonstrates order-of-magnitude throughput improvements over existing monitoring and indexing approaches, with tunable options for balancing consistency, latency, and metadata freshness.

The global vaccine manufacturing ecosystem is bifurcating between hyper-connected ‘Industry 4.0’ facilities and brownfield environments in low- and middle-income countries (LMICs) constrained by legacy infrastructure, unreliable connectivity and high bandwidth costs. This Commentary argues that achieving vaccine independence in the Global South requires an offline-first, sovereign edge architecture that enables ‘collective intelligence without data pooling’. By combining TinyML, hierarchical federated learning and stateful agentic middleware (software agents that retain persistent memory of local system states across network disruptions), LMIC manufacturers can deploy AI-driven process optimization and surveillance while retaining full data custody. Drawing on established governance blueprints such as MELLODDY and OpenSAFELY, and on emerging initiatives including CEPI’s Pandemic Preparedness Engine for Disease X, we outline how data product management units, versioned federation and byzantine-robust aggregation can align regulatory requirements with continuously learning systems. We conclude with practical next steps and highlight technical, financial and political challenges that must be addressed for sovereign, AI-enabled vaccine systems to move from pilots to field-ready infrastructure.

Real-time scientific workflows operate on continuous data streams and must produce timely, high-quality results despite executing on complex, failure-prone infrastructure. Hardware faults, network disruptions, and performance anomalies caused by resource contention or system heterogeneity can severely degrade performance and violate real-time constraints. We focus on strengthening the resilience of the producer-consumer streaming pattern, a fundamental building block of scientific streaming workflows. We present two complementary techniques: (i) a dynamic, asynchronous, non-blocking checkpointing mechanism that preserves progress without interrupting computation, and (ii) a progress-aware load redistribution strategy that detects slow workers and proactively rebalances tasks. Together, these mechanisms maintain forward progress and balanced execution even in highly error-prone environments. Experimental results show that our approach reduces the impact of failures and performance anomalies by up to 6x, while introducing less than 1

Agentic systems, in which diverse agents cooperate to tackle challenging problems, are exploding in popularity in the AI community. However, existing agentic frameworks take a relatively narrow view of agents, apply a centralized model, and target conversational, cloud-native applications (e.g., LLM-based AI chatbots). In contrast, scientific applications require myriad agents be deployed and managed across diverse cyberinfrastructure. Here we introduce Academy, a modular and extensible middleware designed to deploy autonomous agents across the federated research ecosystem, including HPC systems, experimental facilities, and data repositories. To meet the demands of scientific computing, Academy supports asynchronous execution, heterogeneous resources, high-throughput data flows, and dynamic resource availability. It provides abstractions for expressing stateful agents, managing inter-agent coordination, and integrating computation with experimental control. We present microbenchmark results that demonstrate high performance and scalability in HPC environments. To explore the breadth of applications that can be supported by agentic workflow designs, we also present case studies in materials discovery, astronomy, decentralized learning, and information extraction in which agents are deployed across diverse HPC systems.

We report on our experiences replicating 7.3 petabytes (PB) of Earth System Grid Federation (ESGF) computational simulation data from Lawrence Livermore National Laboratory (LLNL) in California to Argonne National Laboratory (ANL) in Illinois and Oak Ridge National Laboratory (ORNL) in Tennessee-a task motivated by a need for increased reliability, capacity, and performance. This task presented significant challenges: the need to move 29 million files twice under time pressure from aging storage hardware; a source file system bottleneck limiting throughput to 1.5 GB/s; frequent site maintenance windows; and the need for complete reliability at scale. We addressed these challenges using a simple replication tool that invoked Globus to transfer large bundles of files while tracking progress in a database, dynamically rerouting transfers to work around maintenance periods and file system limitations. Under the covers, Globus organized transfers to make efficient use of the high-speed Energy Sciences network (ESnet) and the data transfer nodes deployed at participating sites, and also addressed security, integrity checking, and recovery from a variety of transient failures. This success demonstrates the considerable benefits that can accrue from the adoption of performant data replication infrastructure. The replication tool is available at https://github.com/esgf2-us/data-replication-tools.

Modern research computing cyberinfrastructure, such as ACCESS-CI and NAIRR Pilot, offers GPU resources across geographically distributed clusters to accommodate the increasing needs of scientific deep learning (DL) workloads. Even for high-performance computing (HPC) experts, configuring environments and managing DL workloads across supercomputers remain significant barriers. To address these obstacles, we present Diamond, an open-source platform to simplify and streamline the DL lifecycle on HPC. Diamond provides an intuitive graphical interface that abstracts system-level complexity, enabling users to develop, debug, and deploy DL models with minimal overhead. We identify several challenges in building such a platform, including portability, security, and usability, and propose effective architectural solutions to each. Notably, Diamond enables users to share and reuse DL workload environments across systems and collaborators, reducing redundant setup efforts. Experimental results demonstrate that Diamond reduces the time to first successful deployment by an average of 68%, compared to manual configuration with command lines. The Diamond service is available at https://diamondhpc.ai.