arXiv subjects

Minho Kim

Publications and source records attributed to Minho Kim.

At least 19 recordsLinked to original sources

arXiv2026

When Should LLMs Search? Counterfactual Supervision for Search Routing

Search-augmented language models can use external evidence to compensate for limitations in parametric knowledge, but search is not uniformly beneficial: models may call search for questions they can already answer, or rely on noisy evidence when correction, clarification, or abstention would be more appropriate. We formulate this as an instance-level search-routing problem: deciding whether search is needed to improve task success relative to a no-search execution. To derive supervision, we compare no-search and forced-search outcomes for the same question and construct an oracle over NO SEARCH, SEARCH, and UNSOLVED based on task-specific success. Using this oracle as both an evaluation criterion and a learning signal, we train search-routing policies with supervised fine-tuning and preference optimization, improving routing macro-F1 on oracle-eligible examples from 0.7082 to 0.8235 for Gemma E2B and from 0.7053 to 0.8365 for Qwen3.5-4B. Further analysis shows that the learned policies reduce model-specific routing failures: Gemma primarily learns no-search restraint, while Qwen further reduces missed search; residual UNSOLVED cases reveal heterogeneous bottlenecks involving model capacity, retrieval budget, evidence use, and policy behavior.

cs.CL↗

arXiv2026

Design Principles for AI-Ready QCD Data with a Barrel Imaging Calorimeter Application

Data from large physics collider experiments in Quantum Chromodynamics (QCD) research differ fundamentally from the modalities used in modern foundation models. The heterogeneity of detector readouts and their technology dependence require principled curation for cross experiment AI applications. We present a design framework for AI-ready QCD data to define a unified data structure that accommodates heterogeneous detector technologies within a single schema. We apply the design principle to the simulated data of the Barrel Imaging Calorimeter (BIC) in the ePIC detector at the Electron--Ion Collider. The BIC simulation data combines AstroPix silicon pixel imaging layers with Pb/ScFi calorimeter layers across different readout types. We describe the schema specialization, data preparation pipeline, and visualization of the curated AI-ready dataset.

physics.data-an↗

arXiv2026

Squid-inspired soft superpropulsion

Squid span four orders of magnitude in size yet rely on pulsed jets. We show that the funnel (siphon) is a compliant nozzle whose dilation and recoil lag mantle contraction, storing and returning energy within each pulse, a mechanism we term superpropulsion. Histology reveals a collagen sheath, and chromatophore tracking in two squid species quantifies a repeatable phase lag. Engineered nozzles, 3D fluid-structure simulations, and a reduced-order mathematical model predict > 300% impulse amplification when nozzle response time matches jet acceleration (tau/T = 0.2-0.4), overlapping in vivo timing. Tuned nozzles extend jet reach, enhance plume dispersion, and improve jet-driven boat transport, with gains persisting after 40x miniaturization. Superpropulsion recasts pulsed jets as impedance matching, with a soft nozzle acting as an elastic capacitor that passively shapes impulse delivery in soft robotic thrusters and fluidic actuators.

physics.flu-dyn↗

arXiv2026

When VLMs 'Fix' Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR

Accurate transcription of handwritten mathematics is crucial for educational AI systems, yet current benchmarks fail to evaluate this capability properly. Most prior studies focus on single-line expressions and rely on lexical metrics such as BLEU, which fail to assess the semantic reasoning across multi-line student solutions. In this paper, we present the first systematic study of multi-line handwritten math Optical Character Recognition (OCR), revealing a critical failure mode of Vision-Language Models (VLMs): over-correction. Instead of faithfully transcribing a student's work, these models often "fix" errors, thereby hiding the very mistakes an educational assessment aims to detect. To address this, we propose PINK (Penalized INK-based score), a semantic evaluation metric that leverages a Large Language Model (LLM) for rubric-based grading and explicitly penalizes over-correction. Our comprehensive evaluation of 15 state-of-the-art VLMs on the FERMAT dataset reveals substantial ranking reversals compared to BLEU: models like GPT-4o are heavily penalized for aggressive over-correction, whereas Gemini 2.5 Flash emerges as the most faithful transcriber. Furthermore, human expert studies show that PINK aligns significantly better with human judgment (55.0% preference over BLEU's 39.5%), providing a more reliable evaluation framework for handwritten math OCR in educational settings.

cs.CY↗

arXiv2025

MambaEye: A Size-Agnostic Visual Encoder with Causal Sequential Processing

Despite decades of progress, a truly input-size agnostic visual encoder-a fundamental characteristic of human vision-has remained elusive. We address this limitation by proposing \textbf{MambaEye}, a novel, causal sequential encoder that leverages the low complexity and causal-process based pure Mamba2 backbone. Unlike previous Mamba-based vision encoders that often employ bidirectional processing, our strictly unidirectional approach preserves the inherent causality of State Space Models, enabling the model to generate a prediction at any point in its input sequence. A core innovation is our use of relative move embedding, which encodes the spatial shift between consecutive patches, providing a strong inductive bias for translation invariance and making the model inherently adaptable to arbitrary image resolutions and scanning patterns. To achieve this, we introduce a novel diffusion-inspired loss function that provides dense, step-wise supervision, training the model to build confidence as it gathers more visual evidence. We demonstrate that MambaEye exhibits robust performance across a wide range of image resolutions, especially at higher resolutions such as $1536^2$ on the ImageNet-1K classification task. This feat is achieved while maintaining linear time and memory complexity relative to the number of patches.

cs.CV↗

arXiv2025

Impact of AlN buffer thickness on electrical and thermal characteristics of AlGaN/GaN/AlN HEMTs

We investigate the influence of AlN buffer thickness on the structural, electrical, and thermal properties of AlGaN/GaN high-electron mobility transistors (HEMTs) grown on semi-insulating SiC substrates by metal-organic chemical vapor deposition. X-ray diffraction and atomic force microscopy reveal that while thin AlN layers (120 nm) exhibit compressive strain and smooth step-flow surfaces, thicker single-layer buffers (550 nm) develop tensile strain and increased surface roughness. Multi-layer buffer structures up to 2 {\mu}m alleviate strain and maintain surface integrity. Low-temperature Hall measurements confirm that electron mobility decreases with increasing interface roughness, with the highest mobility observed in the structure with a thin AlN buffer. Transient thermoreflectance measurements show that thermal conductivity (ThC) of the AlN buffer increases with the thickness, reaching 188 W/m.K at 300 K for the 2 {\mu}m buffer layer, which is approximately 60% of the bulk AlN ThC value. These results highlight the importance of optimizing AlN buffer design to balance strain relaxation, thermal management, and carrier transport for high-performance GaN-based HEMTs.

cond-mat.mtrl-sci↗

arXiv2025

Thermal transport in GaN/AlN HEMTs on 4H-SiC: Role of layer thickness and hetero-interfaces

Thermal transport in high-electron-mobility-transistor (HEMT) structures grown on 4H-SiC substrates by metalorganic-vapour-phase epitaxy (MOCVD) is systematically investigated. The thermal conductivity of the GaN channel and AlN buffer layers is measured by thermoreflectance (TTR). A pronounced thickness dependence of thermal conductivity as a result of phonon-boundary scattering is observed at low temperatures, while this effect becomes significantly weaker at elevated temperatures. The thermal boundary resistance (TBR) at the AlN/4H-SiC and GaN/AlN interfaces is also examined, showing a substantial reduction and eventual saturation with increasing temperature, indicating elastic phonon transport as the dominant mechanism. Reliable simulations of the temperature profile across the structures based on the measured thermal metrics highlight the critical role of TBR in thin-channel device and the advantage of thicker channel and buffer layers for efficient heat dissipation in the HEMTs.

cond-mat.mtrl-sci↗

arXiv2025

Evaluation of the Response to Electrons and Pions in the Scintillating Fiber and Lead Calorimeter for the Future Electron-Ion Collider

The performance of the Baby Barrel Electromagnetic Calorimeter (Baby BCAL) - a small-scale lead-scintillating-fiber (Pb/ScFi) prototype of the GlueX Barrel Electromagnetic Calorimeter (BCAL) - was tested in a dedicated beam campaign at the Fermilab Test Beam Facility (FTBF). This study provides a benchmark for the Pb/ScFi component of the future Barrel Imaging Calorimeter (BIC) in the ePIC detector at the Electron-Ion Collider (EIC). The detector response to electrons and pions was studied at beam energies between 4 and 10 GeV, extending previous GlueX tests [NIM A 596 (2008) 327-337 and arXiv:1801.03088] to a higher energy regime. The calibrated detector exhibits good linearity within uncertainties, and its electron energy resolution meets EIC requirements. The data further constrain the constant term in the energy resolution to below 1.9%, improving upon previous constraints at lower energies. Simulations reproduce key features of the electron and pion data within the limitations of the collected dataset and the FTBF test environment. Electron-pion separation in the test beam setup was analyzed using multiple methods, incorporating varying degrees of beam-related effects. The inclusion of longitudinal shower profile information enhanced the separation performance, underscoring its relevance for the full-scale BIC in ePIC. These results provide essential benchmarks for the Pb/ScFi section of the future BIC, validating detector simulations and guiding optimization strategies for electron-pion discrimination.

physics.ins-det↗

arXiv2024

Topological chiral superconductivity beyond pairing in a Fermi liquid

We investigate a mechanism to produce superconductivity by strong purely repulsive interactions for flat dispersion $\varepsilon \sim k^4$, without using pairing instability in Fermi-liquid. The resulting superconductors break both time-reversal and reflection symmetries in the orbital motion of electrons, and exhibit non-trivial topological order. Our findings suggest that this topological chiral superconductivity is more likely to emerge near or between fully spin-valley polarized metallic phase and Wigner crystal phase. These topological chiral superconductors can be fully or partially spin-valley polarized. For partial spin-valley polarization, the ratios of electron densities associated with different spin-valley quantum numbers are quantized as simple rational numbers. Furthermore, many of these topological chiral superconductors exhibit charge-4 or higher condensation, neutral quasiparticles with fractional statistics, and/or gapless chiral edge states. Two of the topological chiral superconductors are in the same phases as the ``spin''-triplet or spinless $p+ \textrm{i} p$ BCS superconductor, while others are in different phases than any BCS superconductors. The same mechanism is also used to produce anyon superconductivity between fractional anomalous quantum Hall states in the presence of a periodic potential.

cond-mat.str-el↗

arXiv2024

An interdisciplinary data-science approach to managing natural hazards risk

Natural hazard risk management is a demanding interdisciplinary task. It requires domain knowledge, integration of robust computational methods, and effective use of complex datasets. However, existing solutions tend to focus on specific aspects, data, or methods, limiting their impact and applicability. Here, we present a general data-driven framework to support risk assessment and policy making illustrating its usage in the context of fire hazard by integrating three unique datasets of fire behavior, street network, and census data for the whole state of California. We show that integrating spatial complexity by including a fire behavior layer and a socio-demographic layer changes the universal function observed in previous optimization frameworks that only work with the accessibility of facilities. These results open avenues for the future development of flexible interdisciplinary frameworks in natural hazards management using complex large-scale data.

math.OC↗

arXiv2024

LatentSwap: An Efficient Latent Code Mapping Framework for Face Swapping

We propose LatentSwap, a simple face swapping framework generating a face swap latent code of a given generator. Utilizing randomly sampled latent codes, our framework is light and does not require datasets besides employing the pre-trained models, with the training procedure also being fast and straightforward. The loss objective consists of only three terms, and can effectively control the face swap results between source and target images. By attaching a pre-trained GAN inversion model independent to the model and using the StyleGAN2 generator, our model produces photorealistic and high-resolution images comparable to other competitive face swap models. We show that our framework is applicable to other generators such as StyleNeRF, paving a way to 3D-aware face swapping and is also compatible with other downstream StyleGAN2 generator tasks. The source code and models can be found at \url{https://github.com/usingcolor/LatentSwap}.

cs.CV↗

arXiv2023

Using Exact Tests from Algebraic Statistics in Sparse Multi-way Analyses: An Application to Analyzing Differential Item Functioning

Asymptotic goodness-of-fit methods in contingency table analysis can struggle with sparse data, especially in multi-way tables where it can be infeasible to meet sample size requirements for a robust application of distributional assumptions. However, algebraic statistics provides exact alternatives to these classical asymptotic methods that remain viable even with sparse data. We apply these methods to a context in psychometrics and education research that leads naturally to multi-way contingency tables: the analysis of differential item functioning (DIF). We explain concretely how to apply the exact methods of algebraic statistics to DIF analysis using the R package algstat, and we compare their performance to that of classical asymptotic methods.

stat.ME↗

arXiv2023

Algebraic Characterization of the Voronoi Cell Structure of the $A_n$ Lattice

We characterized the combinatorial structure of the Voronoi cell of the $A_n$ lattice in arbitrary dimensions. Based on the well-known fact that the Voronoi cell is the disjoint union of $(n+1)!$ congruent simplices, we show that it is the disjoint union of $(n+1)$ congruent hyper-rhombi, which are the generalized rhombi or trigonal trapezohedra. The explicit structure of the faces is investigated, including the fact that all the $k$-dimensional faces, $2\le k\le n-1$, are hyper-rhombi. We show it to be the vertex-first orthogonal projection of the $(n+1)$-dimensional unit cube. Hence the Voronoi cell is a zonotope. We prove that in low dimensions ($n\le 3$) the Voronoi cell can be understood as the section of that of the $D_{n+1}$ lattice with the hyperplane orthogonal to the diagonal direction. We provide all the explicit coordinates and transformation matrices associated with our analysis. Most of our analysis is algebraic and easily accessible to those less familiar with the Coxeter-Dynkin diagrams.

math.CO↗

arXiv2023

A Practical Box Spline Compendium

Box splines provide smooth spline spaces as shifts of a single generating function on a lattice and so generalize tensor-product splines. Their elegant theory is laid out in classical papers and a summarizing book. This compendium aims to succinctly but exhaustively survey symmetric low-degree box splines with special focus on two and three variables. Tables contrast the lattices, supports, analytic and reconstruction properties, and list available implementations and code.

math.NA↗

arXiv2022

Anderson localization of emergent quasiparticles: Spinon and vison interplay at finite temperature in a $\mathbb{Z}_2$ gauge theory in three dimensions

Fractional statistics of quasiparticle excitations often plays an important role in the detection and characterization of topological systems. In this paper, we investigate the case of a three-dimensional (3D) Z2 gauge theory, where the excitations take the form of bosonic spinon quasiparticle and vison flux tubes, with mutual semionic statistics. We focus on an experimentally relevant intermediate temperature regime, where sparse spinons hop coherently on a dense quasistatic and stochastic vison background. The effective Hamiltonian reduces to a random-sign bimodal tight-binding model, where both the particles and the disorder are borne out of the same underlying quantum spin liquid (QSL) degrees of freedom, and the coupling between the two is purely driven by the mutual fractional statistics. We study the localization properties and observe a mobility edge located close to the band edge, whose transition belongs to the 3D Anderson model universality class. Spinons allowed to propagate through the quasistatic vison background appear to display quantum diffusive behavior. When the visons are allowed to relax, in response to the presence of spinons in equilibrium, we observe the formation of vison depletion regions slave to the support of the spinon wavefunction. We discuss how this behavior can give rise to measurable effects in the relaxation, response and transport properties of the system and how these may be used as signatures of the mutual semionic statistics and as precursors of the QSL phase arising in the system at lower temperatures.

cond-mat.dis-nn↗

arXiv2022

Mesoscopic transport in KSTAR plasmas: avalanches and the $E \times B$ staircase

The self-organization is one of the most interesting phenomena in the non-equilibrium complex system, generating ordered structures of different sizes and durations. In tokamak plasmas, various self-organized phenomena have been reported, and two of them, coexisting in the near-marginal (interaction dominant) regime, are avalanches and the $E \times B$ staircase. Avalanches mean the ballistic flux propagation event through successive interactions as it propagates, and the $E \times B$ staircase means a globally ordered pattern of self-organized zonal flow layers. Various models have been suggested to understand their characteristics and relation, but experimental researches have been mostly limited to the demonstration of their existence. Here we report detailed analyses of their dynamics and statistics and explain their relation. Avalanches influence the formation and the width distribution of the $E \times B$ staircase, while the $E \times B$ staircase confines avalanches within its mesoscopic width until dissipated or penetrated. Our perspective to consider them the self-organization phenomena enhances our fundamental understanding of them as well as links our findings with the self-organization of mesoscopic structures in various complex systems.

physics.plasm-ph↗

arXiv2021

Leveraging Quantum Annealer to identify an Event-topology at High Energy Colliders

With increasing energy and luminosity available at the Large Hadron collider (LHC), we get a chance to take a pure bottom-up approach solely based on data. This will extend the scope of our understanding about Nature without relying on theoretical prejudices. The required computing resource, however, will increase exponentially with data size and complexities of events if one uses algorithms based on a classical computer. In this letter we propose a simple and well motivated method with a quantum annealer to identify an event-topology, a diagram to describe the history of particles produced at the LHC. We show that a computing complexity can be reduced significantly to the order of polynomials which enables us to decode the "Big" data in a very clear and efficient way. Our method achieves significant improvements in finding a true event-topology, more than by a factor of two compared to a conventional method.

hep-ph↗

arXiv2021

Stochastic fluctuation and transport of tokamak edge plasmas with the resonant magnetic perturbation field

We present that a statistical method known as the Complexity-Entropy analysis is useful to characterize a state of plasma turbulence and flux in the resonant magnetic perturbation (RMP) edge localized mode (ELM) control experiment. The RMP ELM suppression phase with the stochastic pedestal top temperature fluctuation can be distinguished from the natural ELM free phase with the chaotic fluctuation. It is discussed that the stochastic temperature fluctuation localized near the pedestal top can be originated from the narrow layer of the field penetration near the pedestal top. The forced magnetic island can emit the resonant drift wave of comparable sizes (relatively low-k) in the RMP ELM suppression phase, and it can results in the generation of stochastic higher wavenumber fluctuations coupled to tangled fields around the island. The analysis of the ion saturation current measurement around the main outer striking point on the divertor shows that it also becomes more stochastic as the stronger plasma response to the RMP field is expected.

physics.plasm-ph↗