Awards: UMN-SNU Research Partnership Seed Grants

The University of Minnesota – Seoul National University Partnership, administered through the Global Programs and Strategy Alliance, supports collaborative research between the University of Minnesota (UMN) and Seoul National University (SNU).

This partnership is designed to strengthen international partnerships in research, education, and workforce development, grounded in the belief that addressing global challenges and maintaining competitiveness in science and technology require robust, cross-border collaboration.

The following are brief descriptions of the projects (taken directly from the original proposals) selected for the UMN-SNU Research Partnership seed grants program in 2026. 
 

Cross-Country iEEG Foundation Modeling and Clinical QA/QC: DIVER-2 and ABCD-v2

UMN PI: David Darrow, Neurosurgery 
SNU PI: Jiook Cha, Psychology/Brain and Cognitive Sciences

Abstract: Intracranial electroencephalography (iEEG) offers unparalleled access to human neural dynamics, but current iEEG foundation models are constrained by small, institutionally homogeneous pretraining corpora, the lack of scalable quality assurance/quality control (QA/QC) frameworks, and the absence of rigorous cross-site benchmarking. This project establishes a UMN–SNU research partnership to address these gaps through two integrated aims. First, we will develop DIVER-2, an iEEG foundation model pretrained on a cross-country corpus combining the University of Minnesota's 26,000-hour American clinical dataset (92 patients) with SNU's Korean iEEG recordings from the HBF Lab—extending the SNU Connectome Lab's DIVER-1 architecture (13M–1.82B parameters, pretrained on 59,300 hours across 17,700+ subjects). Second, we will develop ABCD-v2, a next-generation automated channel QA/QC framework that replaces hand-crafted spectral features with DIVER-2 embeddings, validated against 9,689 expert-annotated UMN channel labels and evaluated for cross-country generalization to Korean iEEG data. UMN contributes large-scale clinical data, validated quality labels, and neurosurgical expertise; the SNU Connectome Lab contributes foundation model infrastructure and approximately 1,191 PF-days of pretraining compute; and the SNU HBF Lab contributes Korean clinical iEEG recordings. Deliverables include open-source model weights, a cross-site benchmarking suite, and a joint manuscript, providing preliminary data for future NIH R01 and NSF collaborative grant applications. This seed investment directly advances the development of internationally validated, clinically deployable neural foundation models.

Your Personal AI Assistant, Built to Last a Lifetime — A UMN-SNU Frontier AI Center

UMN PI: Dongyeop Kang, Computer Science and Engineering 
SNU PIs: Gunhee Kim, Computer Science and Engineering; Jonghyun Choi, Electrical and Computer Engineering; Yohan Jo, Data Science

Abstract: Imagine a personal AI companion that knows you as well as a trusted friend—remembering your goals, learning from your habits, adapting to your needs across years, not just hours. Within five to ten years, always-on personal AI agents will be as ubiquitous as smartphones. Yet today's leading systems (ChatGPT, Claude, Gemini and etc.) are fundamentally ill-suited for this role: they forget every conversation when it ends, cannot be customized to individual users, route personal data through corporate servers, and charge fees that exclude many. The only viable path to truly personal AI is open-source, on-premise agents that run locally and belong to their users. This project launches LCEA (Lifelong Co-Evolving Agent), an open-source platform developed jointly by University of Minnesota (UMN) and Seoul National University (SNU) to solve three interconnected scientific problems. Thrust 1 (PI Kim, SNU) builds a scalable hierarchical memory system that stores, organizes, and retrieves years of personal multimodal experience without becoming overwhelmed. Thrust 2 (PI Choi, SNU) develops cognitively-inspired continual learning so the agent integrates new experiences without catastrophic forgetting—translating how human memory consolidates over time into practical AI algorithms. Thrust 3 (PIs Jo/SNU, Kang/UMN) constructs a persistent, evolving user model tracking preferences, goals, cognitive-affective states, and long-term behavioral patterns to enable genuinely personalized long-horizon assistance. Thrust 4 (PI Kang, UMN) integrates all three thrusts into a unified platform with standardized APIs, benchmarks, and deployment pipelines, validated through educational tutoring. SNU contributes world-class strengths in memory architectures, cognitive AI, and embodied learning; UMN brings expertise in user modeling and human-AI interaction. This seed grant generates the preliminary data, benchmarks, and infrastructure to anchor joint NSF, NIH, and Korea NRF proposals and a lasting research partnership.

Coupled Reactive Transport and Geomechanics for Subsurface Water and Energy Systems

UMN PI: Peter Kang, Earth and Environmental Science 
SNU PI: Jinhyun Choo, Civil, Urban, and Environmental Engineering

Abstract: Sustainable management of subsurface systems is central to global challenges in clean energy, carbon storage, and water security. Applications such as geothermal energy production, geologic carbon storage, and in situ critical mineral recovery depend on tightly coupled processes linking fluid flow, chemical reactions, and rock deformation. Despite significant advances in reactive transport and geomechanics individually, a fundamental gap remains: the lack of an integrated predictive framework that connects reaction-driven changes in flow to their mechanical consequences, particularly under non-isothermal conditions. This project addresses this gap by developing a unified framework linking temperature-driven flow reorganization, spatially heterogeneous mineral precipitation, and fracture mechanical evolution.  The University of Minnesota team, led by Peter K. Kang, will conduct microfluidic experiments and three-dimensional reactive transport simulations to map precipitation regimes across key dimensionless parameters (temperature contrast, flow rate, and Damköhler number). The Seoul National University team, led by Jinhyun Choo, will develop coupled thermo-poro-mechanical simulations to quantify stress evolution arising from non-uniform precipitation and thermal loading. By integrating these efforts, the project will produce unified regime diagrams linking subsurface conditions to system-level outcomes—fracture sealing versus sustained permeability and mechanical damage. Outcomes include predictive, dimensionless regime maps and precipitation–stress correlations with direct relevance to geothermal scaling, CO₂ mineralization, and critical mineral recovery. This collaboration leverages complementary expertise in reactive transport and geomechanics and will establish the foundation for future DOE, NSF, and NRF-funded research.

AI-Enhanced 3D Multi-scale Analysis of Gas Venting Dynamics for Sustainable Water and Energy Systems

UMN PI: Sungyun Lee, Mechanical Engineering 
SNU PI: Hyungmin Park, Mechanical Engineering

Abstract: Buoyancy-driven gas migration and venting in liquid-saturated granular media is central to many natural processes. The generation and emission of biogenic methane from sediments plays an important role in local ecosystems and the global carbon cycle. Hence, predicting the net rate of gas venting has huge environmental implications, directly relevant to Water and Environmental Sustainability. In addition, understanding gas dynamics in deformable media is of great importance to Future Energy Landscape, as the success of Carbon Capture and Storage and the sequestration of green hydrogen in deep aquifers depends on our ability to predict and control gas-liquid-grain flows. Despite their environmental and technological significance, a fundamental understanding of gas venting processes is still lacking, due to the difficulty in visualizing three-dimensional interfacial flows in the subsurface environment. To overcome this limitation, we propose to combine multi-scale experiments with AI-enhanced data processing to gain a complete physical picture of gas migration and venting from deformable media. The team led by PI Lee (University of Minnesota) will conduct experiments ranging from microfluidics for grain-scale visualization, table-top for meso-scale flows, to meter-scale flumes for global flow patterns. The data produced by PI Lee will be processed by PI Park’s team (Seoul National University) using their AI tools to reconstruct highly localized information, such as interfacial curvatures, essential for predictive modeling. Enabled by this seed funding, we aim to establish an interdisciplinary collaboration that will transform our understanding of gas venting and develop a novel paradigm: AI-Driven Digital Multiphase and Multiphysics Mechanics.