arXiv subjects

Minjae Kim

Publications and source records attributed to Minjae Kim.

At least 55 records · Page 3Linked to original sources

arXiv2025

Microscopic origin of the spin-splitting in altermagnets

Altermagnets, characterized by spin-split bands without net magnetization, have recently emerged as a promising platform for spintronics. However, their microscopic mechanisms remain elusive, often relying on abstract group theory. In this work, we present an intuitive and pedagogical framework to understand the origin of spin splitting in altermagnets. We identify two essential ingredients: (1) alternating spin-polarized wavefunction localization on sublattices, and (2) broken translational symmetry caused by distortions in non-magnetic ion cages. We discuss a minimal model Hamiltonian based on an atomic exchange-driven spin splitting and anisotropic hopping that captures these effects and reproduces the hallmark features of altermagnetic band structures, including nodal spin degeneracies and large spin splittings. Our model is further validated by ab initio calculations on MnF2. By demystifying the microscopic origins of altermagnetism, our work bridges symmetry analysis and material realizations, shedding light on practical designs of altermagnetic spintronic devices.

cond-mat.mtrl-sci↗

arXiv2025

Universal Fabrication of Graphene/Perovskite Oxide Hybrid Heterostructures

Hybrid heterostructures composed of graphene and perovskite oxides provide a promising platform for exploiting synergetic interfacial functionalities. Conventional fabrication methods of the hybrid heterostructures rely on transferring graphene grown on metallic substrates-- a process that is time-consuming, labor-intensive, and prone to introducing numerous defects. In this study, we present a universal, catalyst-free method for the direct growth of graphene on insulating substrates by employing three different perovskite oxide substrates (SrTiO$_3$, LaAlO$_3$, and (La$_{0.18}$Sr$_{0.82}$)(Al$_{0.59}$Ta$_{0.41}$)O$_3$) using atmospheric chemical vapor deposition. Comprehensive characterization via Raman spectroscopy, X-ray spectroscopy, scanning probe microscopy, and electron microscopy confirmed the formation of a uniform, continuous monolayer graphene on all substrates. We identified that growth temperature critically governs graphene quality, as excessive active species may lead to secondary nucleation and the formation of multilayer graphene. Notably, all substrates shared the same optimal growth conditions. Low-temperature Raman spectroscopy and scanning tunneling microscopy of the graphene/SrTiO$_3$ hybrid heterostructure revealed cooperative phenomena, including substrate-induced lattice-phonon and electron-phonon coupling. Our work establishes a reproducible, transfer-free fabrication route for graphene/perovskite oxide hybrid heterostructures and provides empirical support for the universal growth of graphene on insulating substrates.

cond-mat.mtrl-sci↗

arXiv2025

VLM in a flash: I/O-Efficient Sparsification of Vision-Language Model via Neuron Chunking

Edge deployment of large Vision-Language Models (VLMs) increasingly relies on flash-based weight offloading, where activation sparsification is used to reduce I/O overhead. However, conventional sparsification remains model-centric, selecting neurons solely by activation magnitude and neglecting how access patterns influence flash performance. We present Neuron Chunking, an I/O-efficient sparsification strategy that operates on chunks (i.e., groups of contiguous neurons in memory) and couples neuron importance with storage access cost. The method models I/O latency through a lightweight abstraction of access contiguity and selects chunks with high utility, defined as neuron importance normalized by estimated latency. By aligning sparsification decisions with the underlying storage behavior, Neuron Chunking improves I/O efficiency by up to 4.65x and 5.76x on Jetson Orin Nano and Jetson AGX Orin, respectively.

cs.LG↗

arXiv2025

Motif 2 12.7B technical report

We introduce Motif-2-12.7B, a new open-weight foundation model that pushes the efficiency frontier of large language models by combining architectural innovation with system-level optimization. Designed for scalable language understanding and robust instruction generalization under constrained compute budgets, Motif-2-12.7B builds upon Motif-2.6B with the integration of Grouped Differential Attention (GDA), which improves representational efficiency by disentangling signal and noise-control attention pathways. The model is pre-trained on 5.5 trillion tokens spanning diverse linguistic, mathematical, scientific, and programming domains using a curriculum-driven data scheduler that gradually changes the data composition ratio. The training system leverages the MuonClip optimizer alongside custom high-performance kernels, including fused PolyNorm activations and the Parallel Muon algorithm, yielding significant throughput and memory efficiency gains in large-scale distributed environments. Post-training employs a three-stage supervised fine-tuning pipeline that successively enhances general instruction adherence, compositional understanding, and linguistic precision. Motif-2-12.7B demonstrates competitive performance across diverse benchmarks, showing that thoughtful architectural scaling and optimized training design can rival the capabilities of much larger models.

cs.CL↗

arXiv2025

Where and How to Perturb: On the Design of Perturbation Guidance in Diffusion and Flow Models

Recent guidance methods in diffusion models steer reverse sampling by perturbing the model to construct an implicit weak model and guide generation away from it. Among these approaches, attention perturbation has demonstrated strong empirical performance in unconditional scenarios where classifier-free guidance is not applicable. However, existing attention perturbation methods lack principled approaches for determining where perturbations should be applied, particularly in Diffusion Transformer (DiT) architectures where quality-relevant computations are distributed across layers. In this paper, we investigate the granularity of attention perturbations, ranging from the layer level down to individual attention heads, and discover that specific heads govern distinct visual concepts such as structure, style, and texture quality. Building on this insight, we propose "HeadHunter", a systematic framework for iteratively selecting attention heads that align with user-centric objectives, enabling fine-grained control over generation quality and visual attributes. In addition, we introduce SoftPAG, which linearly interpolates each selected head's attention map toward an identity matrix, providing a continuous knob to tune perturbation strength and suppress artifacts. Our approach not only mitigates the oversmoothing issues of existing layer-level perturbation but also enables targeted manipulation of specific visual styles through compositional head selection. We validate our method on modern large-scale DiT-based text-to-image models including Stable Diffusion 3 and FLUX.1, demonstrating superior performance in both general quality enhancement and style-specific guidance. Our work provides the first head-level analysis of attention perturbation in diffusion models, uncovering interpretable specialization within attention layers and enabling practical design of effective perturbation strategies.

cs.CV↗

arXiv2025

Strain Engineering of van Hove Singularity and Coupled Itinerant Ferromagnetism in Quasi-2D Oxide Superlattices

Engineering van Hove singularities (vHss) near the Fermi level, if feasible, offers a powerful route to control exotic quantum phases in electronic and magnetic behaviors. However, conventional approaches, which rely primarily on chemical and electrical doping, focus mainly on local electrical or optical measurements, limiting their applicability to coupled functionalities. In this study, a vHs-induced insulator-metal transition coupled with a ferromagnetic phase transition was empirically achieved in atomically designed quasi-2D SrRuO3 (SRO) superlattices via epitaxial strain engineering, which has not been observed in conventional 3D SRO systems. Theoretical calculations revealed that epitaxial strain effectively modulates the strength and energy positions of vHs of specific Ru orbitals, driving correlated phase transitions in the electronic and magnetic ground states. X-ray absorption spectroscopy confirmed the anisotropic electronic structure of quasi-2D SRO modulated by epitaxial strain. Magneto-optic Kerr effect and electrical transport measurements demonstrated modulated magnetic and electronic phases. Furthermore, magneto-electrical measurements detected significant anomalous Hall effect signals and ferromagnetic magnetoresistance, indicating the presence of magnetically coupled charge carriers in the 2D metallic regime. This study establishes strain engineering as a promising platform for tuning vHss and resultant itinerant ferromagnetism of low-dimensional correlated quantum systems.

cond-mat.mtrl-sci↗

arXiv2025

FOSSIL: Regret-Minimizing Curriculum Learning for Metadata-Free and Low-Data Mpox Diagnosis

Deep learning in small and imbalanced biomedical datasets remains fundamentally constrained by unstable optimization and poor generalization. We present the first biomedical implementation of FOSSIL (Flexible Optimization via Sample-Sensitive Importance Learning), a regret-minimizing weighting framework that adaptively balances training emphasis according to sample difficulty. Using softmax-based uncertainty as a continuous measure of difficulty, we construct a four-stage curriculum (Easy-Very Hard) and integrate FOSSIL into both convolutional and transformer-based architectures for Mpox skin lesion diagnosis. Across all settings, FOSSIL substantially improves discrimination (AUC = 0.9573), calibration (ECE = 0.053), and robustness under real-world perturbations, outperforming conventional baselines without metadata, manual curation, or synthetic augmentation. The results position FOSSIL as a generalizable, data-efficient, and interpretable framework for difficulty-aware learning in medical imaging under data scarcity.

cs.LG↗

arXiv2025

FinAgentBench: A Benchmark Dataset for Agentic Retrieval in Financial Question Answering

Accurate information retrieval (IR) is critical in the financial domain, where investors must identify relevant information from large collections of documents. Traditional IR methods -- whether sparse or dense -- often fall short in retrieval accuracy, as it requires not only capturing semantic similarity but also performing fine-grained reasoning over document structure and domain-specific knowledge. Recent advances in large language models (LLMs) have opened up new opportunities for retrieval with multi-step reasoning, where the model ranks passages through iterative reasoning about which information is most relevant to a given query. However, there exists no benchmark to evaluate such capabilities in the financial domain. To address this gap, we introduce FinAgentBench, the first large-scale benchmark for evaluating retrieval with multi-step reasoning in finance -- a setting we term agentic retrieval. The benchmark consists of 26K expert-annotated examples on S&P-500 listed firms and assesses whether LLM agents can (1) identify the most relevant document type among candidates, and (2) pinpoint the key passage within the selected document. Our evaluation framework explicitly separates these two reasoning steps to address context limitations. This design enables to provide a quantitative basis for understanding retrieval-centric LLM behavior in finance. We evaluate a suite of state-of-the-art models and further demonstrated how targeted fine-tuning can significantly improve agentic retrieval performance. Our benchmark provides a foundation for studying retrieval-centric LLM behavior in complex, domain-specific tasks for finance.

cs.IR↗

arXiv2025

Optical properties of popular dielectric substrate materials in a wide spectral range from far-infrared to ultraviolet

We investigated the optical properties of 13 different dielectric materials (slide glass, quartz, Al2O3 (c-cut), DyScO3 (110), KTaO3 (001), LaAlO3 (001), (LaAlO3)0.3-(Sr2AlTaO6)0.7 (001) (LSAT), MgF2 (100), MgO (100), SiC, SrTiO3 (001), TbScO3 (110), and TiO2). The single-bounce reflectance spectra of the bulk samples were measured using Fourier transform infrared (FTIR) and monochromatic spectrometers across a wide spectral range, from far infrared to ultraviolet (80-50,000 cm-1). Using the Kramers-Kronig analysis, we obtained the optical conductivity and dielectric function of the dielectric materials from their measured reflectance spectra. Moreover, we measured the transmittance spectra of the materials to obtain their bandgaps. We fitted the measured reflectance spectra using the Lorentz model to obtain phononic structures. Each dielectric material exhibits unique phononic structures and optical bandgaps, associated with the composition and crystal structure of the material. The observed optical properties of these dielectric materials provide valuable information for the optical analysis of thin films grown on them.

cond-mat.mtrl-sci↗

arXiv2025

Layer controlled orbital selective Mott transition in monolayer nickelate

Dimensionality and electronic correlations are crucial elements of many quantum material properties. An example is the change of the electronic structure accompanied by the loss of quasiparticles when a metal is reduced from three dimensions to a lower dimension, where the Coulomb interaction between carriers becomes poorly screened. Here, using angle-resolved photoemission spectroscopy (ARPES), we report an orbital-selective decoherence of spectral density in the perovskite nickelate LaNiO3 towards the monolayer limit. The spectral weight of the dz2 band vanishes much faster than that of the dx2-y2 band as the thickness of the LaNiO3 layer is decreased to a single unit cell, indicating a stronger correlation effect for the former upon dimensional confinement. Dynamical mean-field theory (DMFT) calculations show an orbital-selective Mott transition largely due to the localization of dz2 electrons along the c axis in the monolayer limit. This orbital-selective correlation effect underpins many macroscopic properties of nickelates, such as metal-to-insulator transition and superconductivity, where most theories are built upon a dx2-y2-dz2 two-band model.

cond-mat.str-el↗

arXiv2025

Uncertainty Quantification of Drag Reduction over Superhydrophobic Surfaces by Unified Parameterizing Structure Spacing

Superhydrophobic surfaces (SHS) have demonstrated significant potential in reducing turbulent drag by introducing slip conditions through micro-structured geometries. While previous studies have examined individual SHS configurations such as post-type, ridge-type, and transverse ridge-type surfaces, a unified analysis that connects these patterns through geometric parameterization remains limited. In this study, we propose a systematic framework to explore the drag reduction characteristics by varying the streamwise and spanwise spacing ($d_1, d_2$) of post-type patterns, effectively encompassing a range of SHS geometries. High-fidelity direct numerical simulations (DNS) were performed using NekRS, a GPU-accelerated spectral element solver, to resolve incompressible turbulent channel flows over these SHSs. To account for variability in the geometric parameters and quantify their influence, we construct a surrogate model based on polynomial chaos expansion (PCE) using Latin hypercube sampling (LHS) method. The resulting model enables efficient uncertainty quantification (UQ) and sensitivity analysis, revealing the relative importance of $d_1$ and $d_2$ in drag reduction performance. This unified UQ framework provides both predictive capability and design guidance for optimizing SHS configurations under uncertain geometric conditions.

physics.flu-dyn↗

arXiv2025

THEME: Enhancing Thematic Investing with Semantic Stock Representations and Temporal Dynamics

Thematic investing, which aims to construct portfolios aligned with structural trends, remains a challenging endeavor due to overlapping sector boundaries and evolving market dynamics. A promising direction is to build semantic representations of investment themes from textual data. However, despite their power, general-purpose LLM embedding models are not well-suited to capture the nuanced characteristics of financial assets, since the semantic representation of investment assets may differ fundamentally from that of general financial text. To address this, we introduce THEME, a framework that fine-tunes embeddings using hierarchical contrastive learning. THEME aligns themes and their constituent stocks using their hierarchical relationship, and subsequently refines these embeddings by incorporating stock returns. This process yields representations effective for retrieving thematically aligned assets with strong return potential. Empirical results demonstrate that THEME excels in two key areas. For thematic asset retrieval, it significantly outperforms leading large language models. Furthermore, its constructed portfolios demonstrate compelling performance. By jointly modeling thematic relationships from text and market dynamics from returns, THEME generates stock embeddings specifically tailored for a wide range of practical investment applications.

q-fin.PM↗

arXiv2025

Motif 2.6B Technical Report

Recent advancements in Large Language Models (LLMs) have revolutionized artificial intelligence, yet developing an effective foundational LLM that balances high performance with computational efficiency remains challenging, especially for emerging research groups. To address this gap, we introduce Motif-2.6B, a 2.6-billion-parameter foundation model designed to democratize advanced LLM capabilities. Motif-2.6B incorporates several innovative architectural enhancements, including Differential Attention and PolyNorm activation functions, which improve long-context comprehension, reduce hallucination, and enhance in-context learning capabilities. We rigorously tested multiple novel architectural components through extensive experimentation to determine the optimal architecture for Motif-2.6B. Comprehensive evaluations demonstrate that Motif-2.6B consistently meets or exceeds the performance of similarly sized state-of-the-art models across diverse benchmarks, showcasing its effectiveness, scalability, and real-world applicability. Through detailed experiments and tailored techniques, Motif-2.6B significantly advances the landscape of efficient, scalable, and powerful foundational LLMs, offering valuable insights and a robust foundation for future research and deployment.

cs.LG↗

arXiv2025

Structuring the Unstructured: A Multi-Agent System for Extracting and Querying Financial KPIs and Guidance

Extracting structured and quantitative insights from unstructured financial filings is essential in investment research, yet remains time-consuming and resource-intensive. Conventional approaches in practice rely heavily on labor-intensive manual processes, limiting scalability and delaying the research workflow. In this paper, we propose an efficient and scalable method for accurately extracting quantitative insights from unstructured financial documents, leveraging a multi-agent system composed of large language models. Our proposed multi-agent system consists of two specialized agents: the \emph{Extraction Agent} and the \emph{Text-to-SQL Agent}. The \textit{Extraction Agent} automatically identifies key performance indicators from unstructured financial text, standardizes their formats, and verifies their accuracy. On the other hand, the \textit{Text-to-SQL Agent} generates executable SQL statements from natural language queries, allowing users to access structured data accurately without requiring familiarity with the database schema. Through experiments, we demonstrate that our proposed system effectively transforms unstructured text into structured data accurately and enables precise retrieval of key information. First, we demonstrate that our system achieves approximately 95\% accuracy in transforming financial filings into structured data, matching the performance level typically attained by human annotators. Second, in a human evaluation of the retrieval task -- where natural language queries are used to search information from structured data -- 91\% of the responses were rated as correct by human evaluators. In both evaluations, our system generalizes well across financial document types, consistently delivering reliable performance.

cs.AI↗

arXiv2025

HARDMath2: A Benchmark for Applied Mathematics Built by Students as Part of a Graduate Class

Large language models (LLMs) have shown remarkable progress in mathematical problem-solving, but evaluation has largely focused on problems that have exact analytical solutions or involve formal proofs, often overlooking approximation-based problems ubiquitous in applied science and engineering. To fill this gap, we build on prior work and present HARDMath2, a dataset of 211 original problems covering the core topics in an introductory graduate applied math class, including boundary-layer analysis, WKB methods, asymptotic solutions of nonlinear partial differential equations, and the asymptotics of oscillatory integrals. This dataset was designed and verified by the students and instructors of a core graduate applied mathematics course at Harvard. We build the dataset through a novel collaborative environment that challenges students to write and refine difficult problems consistent with the class syllabus, peer-validate solutions, test different models, and automatically check LLM-generated solutions against their own answers and numerical ground truths. Evaluation results show that leading frontier models still struggle with many of the problems in the dataset, highlighting a gap in the mathematical reasoning skills of current LLMs. Importantly, students identified strategies to create increasingly difficult problems by interacting with the models and exploiting common failure modes. This back-and-forth with the models not only resulted in a richer and more challenging benchmark but also led to qualitative improvements in the students' understanding of the course material, which is increasingly important as we enter an age where state-of-the-art language models can solve many challenging problems across a wide domain of fields.

cs.LG↗

arXiv2025

Euler-Lagrange study of Microbubble-Laden Turbulent Flow over Superhydrophobic surfaces

For slow-speed ships, underwater vehicles, and pipe transportation systems, viscous resistance accounts for a large proportion of the total energy losses. As such, various technologies have been developed to reduce viscous resistance and enhance energy efficiency in these applications. Air injection and surface treatment are two representative drag reduction techniques. Additionally, efforts to combine multiple drag-reduction techniques have been the subject of extensive research. In this study, the synergistic effects of integrating microbubble injection and superhydrophobic Surface(SHS) drag reduction approaches were analyzed. A 2-way coupling Euler-Lagrange approach was used alongside direct numerical simulation, based on the spectral element method, to investigate the synergistic effects of applying two separate drag reduction methods. Three types of SHS were investigated in our simulations; post type, transverse ridge type, and ridge type. The drag reduction performances and flow characteristics of the various configurations, with and without microbubble injection, were compared in a turbulent horizontal channel flow with $Re_τ=180$. The results of these tests showed that, combining post-type SHS with microbubbles was the most effective, producing a synergistic drag reduction effect. However, combining microbubble injection with ridge-type SHS increased drag relative to ridge-type SHS alone, showing the importance of carefully selecting wall type for the best possible performance.

physics.flu-dyn↗

arXiv2024

A Noise is Worth Diffusion Guidance

Diffusion models excel in generating high-quality images. However, current diffusion models struggle to produce reliable images without guidance methods, such as classifier-free guidance (CFG). Are guidance methods truly necessary? Observing that noise obtained via diffusion inversion can reconstruct high-quality images without guidance, we focus on the initial noise of the denoising pipeline. By mapping Gaussian noise to `guidance-free noise', we uncover that small low-magnitude low-frequency components significantly enhance the denoising process, removing the need for guidance and thus improving both inference throughput and memory. Expanding on this, we propose \ours, a novel method that replaces guidance methods with a single refinement of the initial noise. This refined noise enables high-quality image generation without guidance, within the same diffusion pipeline. Our noise-refining model leverages efficient noise-space learning, achieving rapid convergence and strong performance with just 50K text-image pairs. We validate its effectiveness across diverse metrics and analyze how refined noise can eliminate the need for guidance. See our project page: https://cvlab-kaist.github.io/NoiseRefine/.

cs.CV↗

arXiv2024

The characterisation of water ice in debris discs: implications for JWST scattered light observations

Water ice plays a crucial role throughout the different stages of planetary evolution and is abundant in the Universe. However, its presence and nature in debris discs of exoplanetary systems are not yet strongly established observationally. In this study, we quantify and discuss the impact of ice parameters such as volume fraction ${\mathcal{F}}_{\rm ice}$, blow-out grain size, size distribution, and its phase on the observational appearance of debris discs, considering the diverse nature of these systems around stellar spectral types ranging from A to M. Our findings reveal that the prominent ice features at approximately 2.7 and 3.3\,$μ$m depend on both the water ice fraction ${\mathcal{F}}_{\rm ice}$ and the scattering angle, with backscattering geometries yielding the most prominent signatures. When the phase function is considered and data are not background limited, strong forward and backward scattering (near edge-on discs) are expected to yield the strongest detections in images/spectra for A or F-type stars, while scattering angle matters less for later type stars. The Fresnel peak at 3.1\,$μ$m serves as a viable discriminant for the transitional phase (crystalline/amorphous), while simultaneously constraining the water ice temperature. For JWST imaging, we find that the F356W and F444W filter combination is most effective for constraining the grain size distribution, while the F356W and F277W filter combination provides better constraints on the ice fraction ${\mathcal{F}}_{\rm ice}$ in debris discs. However, degeneracy between the grain size distribution and ice fraction when using photometric flux ratios means that obtaining robust constraints will likely require more than two filters, or spectroscopic data.

astro-ph.EP↗