arXiv subjects

Hongseok Kim

Publications and source records attributed to Hongseok Kim.

13 recordsLinked to original sources

arXiv2026

UNION: A Unified AC-OPF Framework for Topology-Varying Real-Time Grid Operation

Secure real-time grid operation requires fast AC optimal power flow (AC-OPF) tools that stay accurate and feasible as operating conditions and topology change. Learning-based methods have advanced, but most are trained per system or per topology, and delivering an operating point that satisfies every operational limit remains challenging. This paper proposes UNION, a unified graph-based AC-OPF framework for heterogeneous systems and topology-varying operation. UNION proposes a shared graph encoder, a scalar-gated aggregation with explicit consensus correction, and a sparse-aware differentiable implicit layer embedding the AC power-flow equations. The remaining inequalities are handled by primal-dual training and the deterministic restoration layer. A single model trained jointly across seven systems, including a real-world 4,492-bus Korean transmission grid, attains a 1.23% mean objective gap and satisfies every operational limit on 99.56% of test instances. It sustains this under zero-shot $N-1$ contingencies, i.e., line and generator outages, and over five days of time-varying Korean topologies; it retains full snapshot coverage at a 2.51% gap under lightweight online fine-tuning. UNION pre-restoration inference takes 55$-$58 ms per instance on the three largest systems, and 108$-$114 ms including restoration. These results indicate that one jointly trained, physics-consistent model can support real-time AC-OPF across heterogeneous systems and evolving topologies.

eess.SY↗

arXiv2026

SABLE: GPU-Based Power Flow Accelerator for Sparsity-Aware Batched Learning

Recent studies have developed GPU-based approaches for solving AC power flow and successfully applied them to standalone power flow problems. However, integrating these approaches into modern differentiable learning frameworks while preserving sparsity remains challenging. To this end, we present SABLE, a GPU-based sparse batched power flow accelerator for differentiable learning via an implicit power flow layer. SABLE leverages a block-diagonal embedding that reformulates batched three-dimensional Jacobians as a fixed-pattern two-dimensional sparse template that is shared across PyTorch, CuPy, and cuDSS. This common template enables zero-copy interoperability and memory-efficient sparse reuse across the software stack. On top of this representation, SABLE accelerates repeated power flow computations through reusable sparse templates, custom GPU kernels, a cuDSS-based sparse-direct LU solver, and mixed-precision techniques. Extensive experiments show that SABLE improves standalone power flow solving throughput by up to 253.4$\times$ over pandapower and 5.7$\times$ over ExaPF. In end-to-end training, evaluated on AC optimal power flow learning models based on DC3 and DeepLDE, SABLE expands the feasible training batch range by up to 64$\times$ and improves training throughput by up to 206.7$\times$ over the corresponding baseline.

eess.SY↗

arXiv2026

Physics-Informed Graph Learning Acceleration for Large-Scale AC-OPF with Topology Changes

In power systems, alternating current optimal power flow (AC-OPF) has been a challenging problem for decades due to its nonconvexity, but fast and efficient solutions are even more needed because of high penetration of large scale renewable generation and load growth. Recently, neural networks (NN) have gained attention in solving AC-OPF, but it is still in an early stage to be applicable for real and large-scale power system operation with topology-changing characteristics. To end this, we propose a novel framework called GraphOPF that considers topology-adaptability, scalability, NN training time, self-supervision, and feasibility altogether. Extensive experiments show that the proposed framework against the baselines is up to 200 times faster in NN training and up to 66 times faster in solving AC-OPF for large-scale power systems including the real Korean power system, while achieving more than 99% feasibility.

eess.SY↗

arXiv2026

Tangram: Unlocking Non-Uniform KV Cache Compression for Efficient Multi-turn LLM Serving

Multi-turn LLM serving accumulates dialogue history whose Key-Value (KV) cache grows with every turn and every user, quickly exceeding the model weights themselves and making memory -- not compute -- the binding constraint on throughput. Non-uniform KV compression, which allocates heterogeneous budgets across attention heads, preserves accuracy far better than uniform schemes, yet remains impractical: modern serving stacks assume identical KV lengths across heads, so heterogeneity traps freed memory as page fragmentation, spends up to 25% of prefill time reclaiming scattered pages, and skews GPU workloads that inflate decode latency by up to $1.7\times$ or burn 15--20% of each decode step on re-planning. We observe that this heterogeneity need not be discovered at runtime: head-wise retention follows a two-level structural regularity -- an input-invariant head ranking with narrowly bounded per-head ratios -- that can be calibrated offline from as few as 50 samples. Building on this insight, we present Tangram, a serving framework that statically resolves what prior systems handle dynamically: Budget Reservation fixes each head's post-compression footprint at scheduling time, eliminating page reclamation; Ragged Paging clusters similar-budget heads into independent page tables, turning fragmentation into reclaimable memory; and Ahead-of-Time Load Balancing precomputes balanced GPU partitions with zero runtime planning. Implemented on vLLM, Tangram serves as a drop-in substrate for existing non-uniform compression methods, matching their accuracy while improving end-to-end throughput by up to $2.6\times$ over the full-KV baseline. Our implementation is publicly available at https://github.com/aiha-lab/TANGRAM.

cs.LG↗

arXiv2026

LUMINA: A Grid Foundation Model for Benchmarking AC Optimal Power Flow Surrogate Learning

AC optimal power flow (ACOPF) is foundational yet computationally expensive in power grid operations, driving learning-based surrogates for large-scale grid analysis. These surrogates, however, often fail to generalize across network topologies, a critical gap for deployment on grids not seen during training and for routine operational what-if studies. We introduce LUMINA-Bench, a comprehensive benchmark suite for ACOPF surrogate learning covering multi-topology pretraining, transfer, and adaptation. The benchmark evaluates homogeneous and heterogeneous architectures under single- and multi-topology learning settings using unified metrics that capture both predictive accuracy and physics-informed constraint violations. We additionally compare constraint-aware training objectives, including MSE, augmented Lagrangian, and violation-based Lagrangian losses, to characterize accuracy-robustness trade-offs across settings. Data processing, training, and evaluation frameworks are open-sourced as the LUMINA suite to support reproducibility and accelerate future research on feasibility-aware OPF surrogates.

cs.LG↗

arXiv2026

LUMINA: Foundation Models for Topology Transferable ACOPF

Foundation models in general promise to accelerate scientific computation by learning reusable representations across problem instances, yet constrained scientific systems, where predictions must satisfy physical laws and safety limits, pose unique challenges that stress conventional training paradigms. We derive design principles for constrained scientific foundation models through systematic investigation of AC optimal power flow (ACOPF), a representative optimization problem in power grid operations where power balance equations and operational constraints are non-negotiable. Through controlled experiments spanning architectures, training objectives, and system diversity, we extract three empirically grounded principles governing scientific foundation model design. These principles characterize three design trade-offs: learning physics-invariant representations while respecting system-specific constraints, optimizing accuracy while ensuring constraint satisfaction, and ensuring reliability in high-impact operating regimes. We present the LUMINA framework, including data processing and training pipelines to support reproducible research on physics-informed, feasibility-aware foundation models across scientific applications.

cs.LG↗

arXiv2025

Macdonald index from 3d TQFT

We propose a new fermionic sum formula for the Macdonald index of a class of Argyres-Douglas theories. The formula arises naturally from a three-dimensional topological field theory obtained via a twisted dimensional reduction of the 4d theory. Such a reduction often gives rise to a 3d ${\mathcal N}=2$ abelian Chern-Simons matter theory, which is expected to flow to an ${\mathcal N}=4$ superconformal fixed point. After performing a topological twist, we obtain a 3d TFT admitting boundary conditions that support the vertex operator algebra associated with the original 4d theory. In this framework, the Macdonald index appears as a half-index of the 3d gauge theory, with the Macdonald grading determined by a distinguished $U(1)_A$ symmetry in the infrared ${\mathcal N}=4$ superconformal algebra. We present a general procedure to identify this $U(1)_A$ symmetry and, whenever possible, show that it reproduces the refined character of the associated vertex operator algebra, thereby recovering the Macdonald index. Our construction also gives a hint towards the IR formula for the Macdonald index in terms of 4d BPS particles on the Coulomb branch.

hep-th↗

arXiv2025

Fixed Point Neural Acceleration and Inverse Surrogate Model for Battery Parameter Identification

The rapid expansion of electric vehicles has intensified the need for accurate and efficient diagnosis of lithium-ion batteries. Parameter identification of electrochemical battery models is widely recognized as a powerful method for battery health assessment. However, conventional metaheuristic approaches suffer from high computational cost and slow convergence, and recent machine learning methods are limited by their reliance on constant current data, which may not be available in practice. To overcome these challenges, we propose deep learning-based framework for parameter identification of electrochemical battery models. The proposed framework combines a neural surrogate model of the single particle model with electrolyte (NeuralSPMe) and a deep learning-based fixed-point iteration method. NeuralSPMe is trained on realistic EV load profiles to accurately predict lithium concentration dynamics under dynamic operating conditions while a parameter update network (PUNet) performs fixed-point iterative updates to significantly reduce both the evaluation time per sample and the overall number of iterations required for convergence. Experimental evaluations demonstrate that the proposed framework accelerates the parameter identification by more than 2000 times, achieves superior sample efficiency and more than 10 times higher accuracy compared to conventional metaheuristic algorithms, particularly under dynamic load scenarios encountered in practical applications.

cs.LG↗

arXiv2024

Locational Scenario-based Pricing in a Bilateral Distribution Energy Market under Uncertainty

In recent years, there has been a significant focus on advancing the next generation of power systems. Despite these efforts, persistent challenges revolve around addressing the operational impact of uncertainty on predicted data, especially concerning economic dispatch and optimal power flow. To tackle these challenges, we introduce a stochastic day-ahead scheduling approach for a community. This method involves iterative improvements in economic dispatch and optimal power flow, aiming to minimize operational costs by incorporating quantile forecasting. Then, we present a real-time market and payment problem to handle optimization in real-time decision-making and payment calculation. We assess the effectiveness of our proposed method against benchmark results and conduct a test using data from 50 real households to demonstrate its practicality. Furthermore, we compare our method with existing studies in the field across two different seasons of the year. In the summer season, our method decreases optimality gap by 60% compared to the baseline, and in the winter season, it reduces optimality gap by 67%. Moreover, our proposed method mitigates the congestion of distribution network by 16.7\% within a day caused by uncertain energy, which is a crucial aspect for implementing energy markets in the real world.

math.OC↗

arXiv2024

Printing Mosaics of Magnetically Programmed Liquid Crystal Directors for Reversibly Morphing Soft Matter

Liquid crystal elastomer (LCE) has been intensively utilized in 4D printing techniques to fabricate smart structures with reversible actuation on the basis of appropriate alignment of liquid crystal (LC) molecules. As a non-contact alignment strategy with a controllability of orientation, magnetic-field alignment has been rarely adapted in 4D printing of LCE because of its poor printing efficiency and demand on large field strength. Here, we report a digital light projection (DLP) system integrated with reorientable magnetic field to facilely print smart LCE structures. We propose a new LCE precursor solution that maintains a liquid crystalline nematic phase and an adequate flowability at room temperature. The resin prior to photopolymerization can be sufficiently aligned by a magnetic field with a strength of 500 mT in seconds without temperature elevating or cycling. Consequential printed structures are capable of presenting an impressive reversible thermal actuation of more than 30 %. The local and arbitrary magnetic-field alignment in layers during DLP printing is characterized, which renders us the ability to construct smart structures with more delicate LC alignments. Furthermore, we introduce the selective deformation of LCE structures with programmed molecular orientation using the photo-thermal effect. Our reported approach reveals the significant potential for fabricating morphing structures in various fields, including soft robotics, biomedical structures, and microelectronics. ({\dag}These authors contributed equally.)

cond-mat.soft↗

arXiv2023

Fully Decentralized Peer-to-Peer Community Grid with Dynamic and Congestion Pricing

Peer-to-peer (P2P) electricity markets enable prosumers to minimize their costs, which has been extensively studied in recent research. However, there are several challenges with P2P trading when physical network constraints are also included. Moreover, most studies use fixed prices for grid power prices without considering dynamic grid pricing, and equity for all participants. This policy may negatively affect the long-term development of the market if prosumers with low demand are not treated fairly. An initial step towards addressing these problems is the design of a new decentralized P2P electricity market with two dynamic grid pricing schemes that are determined by consumer demand. Futhermore, we consider a decentralized system with physical constraints for optimizing power flow in networks without compromising privacy. We propose a dynamic congestion price to effectively address congestion and then prove the convergence and global optimality of the proposed method. Our experiments show that P2P energy trade decreases generation cost of main grid by 56.9% compared with previous works. Consumers reduce grid trading by 57.3% while the social welfare of consumers is barely affected by the increase of grid price.

eess.SY↗

arXiv2023

Self-supervised Equality Embedded Deep Lagrange Dual for Approximate Constrained Optimization

Conventional solvers are often computationally expensive for constrained optimization, particularly in large-scale and time-critical problems. While this leads to a growing interest in using neural networks (NNs) as fast optimal solution approximators, incorporating the constraints with NNs is challenging. In this regard, we propose deep Lagrange dual with equality embedding (DeepLDE), a framework that learns to find an optimal solution without using labels. To ensure feasible solutions, we embed equality constraints into the NNs and train the NNs using the primal-dual method to impose inequality constraints. Furthermore, we prove the convergence of DeepLDE and show that the primal-dual learning method alone cannot ensure equality constraints without the help of equality embedding. Simulation results on convex, non-convex, and AC optimal power flow (AC-OPF) problems show that the proposed DeepLDE achieves the smallest optimality gap among all the NN-based approaches while always ensuring feasible solutions. Furthermore, the computation time of the proposed method is about 5 to 250 times faster than DC3 and the conventional solvers in solving constrained convex, non-convex optimization, and/or AC-OPF.

math.OC↗

arXiv2022

Markov Chain Score Ascent: A Unifying Framework of Variational Inference with Markovian Gradients

Minimizing the inclusive Kullback-Leibler (KL) divergence with stochastic gradient descent (SGD) is challenging since its gradient is defined as an integral over the posterior. Recently, multiple methods have been proposed to run SGD with biased gradient estimates obtained from a Markov chain. This paper provides the first non-asymptotic convergence analysis of these methods by establishing their mixing rate and gradient variance. To do this, we demonstrate that these methods-which we collectively refer to as Markov chain score ascent (MCSA) methods-can be cast as special cases of the Markov chain gradient descent framework. Furthermore, by leveraging this new understanding, we develop a novel MCSA scheme, parallel MCSA (pMCSA), that achieves a tighter bound on the gradient variance. We demonstrate that this improved theoretical result translates to superior empirical performance.

cs.LG↗