arXiv subjects

Minjun Kim

Publications and source records attributed to Minjun Kim.

At least 19 recordsLinked to original sources

arXiv2026

Distribution-aware Language Neuron Identification in Multilingual Large Language Models

Multilingual large language models (mLLMs) contain a small fraction of feed-forward neurons that are sensitive to particular languages, commonly termed language-specific neurons. Existing work measures language specificity using the entropy of each neuron's language-wise probabilities of being active, where a neuron is considered active when its activation value is positive. However, this approach may not fully capture the multilingual nature of mLLMs, where language representations are distributional and mutually related. We propose Distribution-aware Language Neuron selection, which leverages pairwise relationships between per-language activation distributions over the full activation range, including negative values. Specifically, we quantify each neuron's language specificity by clustering languages using pairwise overlap coefficients between their activation distributions. Across two mLLMs and two held-out corpora, our identifier more effectively isolates language-specific causal effects, yielding up to 4.9$\times$ higher on-target language damage per neuron while preserving off-target language performance.

cs.CL↗

arXiv2026

TELLME: Test-Enhanced Learning for Language Model Enrichment

Continual pre-training (CPT) has been widely adopted as a method for domain adaptation in large language models. However, CPT has consistently been accompanied by challenges, such as the difficulty of acquiring large-scale domain-specific datasets and high computational costs. In this study, we propose a novel method called Test-Enhanced Learning for Language Model Enrichment (TELLME) to alleviate these issues. TELLME leverages the TestEnhanced Learning (TEL) principle, whereby the model's training efficiency is improved using quizzes during training. It integrates this principle with CPT, thereby promoting efficient domain-specific knowledge acquisition and long-term memory retention. Experimental results demonstrate that TELLME outperforms existing methods by up to 23.6% in the financial domain and achieves a 9.8% improvement in long-term memory retention.

cs.CL↗

arXiv2026

Mr.Dec: Daily-Scale Longitudinal Multimodal Modeling for 30-Day Readmission Prediction

Predicting 30-day hospital readmission is essential for assessing patient stability and optimizing healthcare resources. As clinical risk evolves with the accumulation of evidence during hospitalization, capturing these dynamic trajectories is essential. However, many existing approaches compress the complex longitudinal history into fixed representations, often losing the granular, day-level clinical signals that reflect a patient's evolving physiological state. To address this, we propose Mr.Dec (Multimodal Readmission-risk prediction Decoder), which models each admission as a natural chronological sequence of daily multimodal events. By leveraging a Transformer Decoder, Mr.Dec integrates daily Electronic Health Record(EHR) updates and intermittent Chest X-ray(CXR) findings in a time-aligned stream, reflecting the actual clinical workflow. To ensure robustness, we utilize Disease-Specific Supervised Contrastive Learning as an auxiliary regularization to induce a diagnosis-aware structure in the latent space. Evaluations on the MIMIC-IV and MIMIC-CXR datasets show that Mr.Dec achieves state-of-the-art performance by preserving the integrity of the clinical sequence. Furthermore, our model identifies "Critical Days" within an admission, providing actionable and clinically grounded interpretations for real-time risk stratification. Code is available at: https://github.com/yejix-ai/MR.DEC

cs.LG↗

arXiv2026

Clinically-Grounded Hierarchical Classification for Consistent Chest X-ray Interpretation

Accurate chest X-ray interpretation is inherently hierarchical. Clinical decisions depend not only on what abnormality is present but where it is situated, requiring reasoning from broad anatomical systems down to specific pathological findings. Yet existing automated systems largely treat this as a flat classification problem, failing to capture inter-level dependencies or enforce coherence between coarse and fine predictions. We propose CHASE (Classification with Hierarchical Analysis and Structured Enforcement), a unified single-stage framework that mirrors radiologists' coarse-to-fine reasoning through a clinically driven three-level taxonomy of 9 anatomical regions, 17 sub-regions, and 28 pathological findings. CHASE jointly optimizes multi-level supervision, cross-level probability alignment, and a hierarchy-violation penalty within a shared Vision Transformer backbone. This ensures that fine-grained findings are anatomically supported by their coarser-level context rather than predicted in isolation. Experiments demonstrate that CHASE outperforms flat and hierarchical baselines across all levels while achieving superior probabilistic hierarchy consistency, with level-wise attention maps confirming anatomically grounded predictions. Code is available at: https://github.com/yejix-ai/CHASE.

cs.CV↗

arXiv2026

Engineering Disordered Many-particle Plasmonic Nanoclusters for Wafer-scale Uniform and Giant Electromagnetic Field Enhancement

Scalable plasmonic technologies face a critical trade-off: few-body architectures offer high enhancement but are sensitive to fabrication flaws, while scalable methods like solid-state dewetting yield large, low-enhancement gaps. We introduce a paradigm shift using a many-body plasmonic architecture inspired by statistical mechanics. By moving toward the continuum limit (N>>1), local geometric variations are statistically averaged out, effectively decoupling optical performance from microscopic disorder. We implement this concept via a lithography- and etching-free, multi-step dewetting strategy, creating wafer-scale nanoclusters. This process strategically forms a robust many-body system by introducing numerous small satellite nanoparticles between larger particles. Crucially, this design achieves a high collective enhancement that surpasses even optimized few-body systems, despite having larger individual gaps. Under optimized conditions, these substrates exhibit a surface-enhanced Raman scattering enhancement factor approaching 4 x 108 with unprecedented reproducibility (RSD of ~10%). This scalable, low-cost concept establishes a practical route toward reproducible wafer-scale nanophotonic platforms for sensing, spectroscopy, and quantum technologies.

physics.optics↗

arXiv2026

Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMs

Following the recent achievement of gold-medal performance on the IMO by frontier LLMs, the community is searching for the next meaningful and challenging target for measuring LLM reasoning. Whereas olympiad-style problems measure step-by-step reasoning alone, research-level problems use such reasoning to advance the frontier of mathematical knowledge itself, emerging as a compelling alternative. Yet research-level math benchmarks remain scarce because such problems are difficult to source (e.g., Riemann Bench and FrontierMath-Tier 4 contain 25 and 50 problems, respectively). To support reliable evaluation of next-generation frontier models, we introduce Soohak, a 439-problem benchmark newly authored from scratch by 64 mathematicians. Soohak comprises two subsets. On the Challenge subset, frontier models including Gemini-3-Pro, GPT-5, and Claude-Opus-4.5 reach 30.4%, 26.4%, and 10.4% respectively, leaving substantial headroom, while leading open-weight models such as Qwen3-235B, GPT-OSS-120B, and Kimi-2.5 remain below 15%. Notably, beyond standard problem solving, Soohak introduces a refusal subset that probes a capability intrinsic to research mathematics: recognizing ill-posed problems and pausing rather than producing confident but unjustified answers. On this subset, no model exceeds 50%, identifying refusal as a new optimization target that current models do not directly address. To prevent contamination, the dataset will be publicly released in late 2026, with model evaluations available upon request in the interim.

cs.CL↗

arXiv2026

A Target-Free Harmonization Method for MRI

In MRI, variations in scan parameters, sequence, or hardware can lead to discrepancies in image appearance, even for the same subject. These inconsistencies, known as domain shifts, can hinder image analysis and degrade the performance of deep learning models trained on data from specific target domains. MRI image harmonization aims to address these issues by aligning source domain images to the target domain images while preserving biological information such as anatomical structures. However, most existing harmonization approaches require access to both source and target domain data in training or test time. This dependence induces data sharing between institutions, raising concerns about patient privacy and substantially limiting the harmonization approaches that can be practically deployed in clinical settings. To overcome these limitations, we introduce TgtFreeHarmony, the harmonization framework tailored for target-free scenarios, eliminating the need for target domain data and any data sharing, enabling privacy-preserving harmonization directly within the source institution. Our approach estimates the target domain style by searching the manifold of MRI domain style constructed via a disentanglement-based generator using Bayesian optimization guided by the performance of a downstream task model, which is trained on target domain data. We evaluated our method on the brain tissue segmentation task across multiple institutes and demonstrated that it effectively harmonizes source images into target images, leading to improved downstream task performance. By enabling harmonization without any access to target-domain data, TgtFreeHarmony establishes a new direction of harmonization preserving data privacy that can be realistically deployed within clinical environments.

eess.IV↗

arXiv2026

FiRe: Fine-grained Multimodal Reasoning for Enhanced Image Generation

With the rapid progress of Multimodal Large Language Models (MLLMs), unified MLLMs that jointly perform image understanding and generation have advanced significantly. However, despite the inherent reasoning capabilities of unified MLLMs for self-reflection and self-refinement, their use in text-to-image generation remains largely underexplored. Meanwhile, existing multimodal reasoning-based image generation methods mostly rely on prompt augmentation or holistic image-text alignment judgments, without fine-grained reflection and refinement of detailed prompt attributes, leading to limited fine-grained control. To address this limitation, we propose FiRe, a Fine-grained Multimodal Reasoning method for enhanced image generation by MLLM. In specific, FiRe performs a fine-grained multi-step reasoning by first decomposing the prompt into key visual requirements and then self-judging their satisfaction in the generated image, followed by localized refinement according to self-generated precise feedback. In addition, to further strengthen the MLLM's multimodal reasoning ability, we introduce FiRe-GRPO, a reinforcement learning method tailored to FiRe. Since standard Group Relative Policy Optimization (GRPO) suffers from sparse, outcome-based rewards in multi-step reasoning, we formulate our reasoning process as a step-level decision-making problem, design step-specific rewards, and compute step-level advantages for granular credit assignment within GRPO. Extensive experiments demonstrate that FiRe consistently outperforms competitive text-to-image baselines, including existing reasoning-based methods, with particularly substantial gains on compositional text-to-image benchmarks.

cs.CV↗

arXiv2026

CRIT: Graph-Based Automatic Data Synthesis to Enhance Cross-Modal Multi-Hop Reasoning

Real-world reasoning often requires combining information across modalities, connecting textual context with visual cues in a multi-hop process. Yet, most multimodal benchmarks fail to capture this ability: they typically rely on single images or set of images, where answers can be inferred from a single modality alone. This limitation is mirrored in the training data, where interleaved image-text content rarely enforces complementary, multi-hop reasoning. As a result, Vision-Language Models (VLMs) frequently hallucinate and produce reasoning traces poorly grounded in visual evidence. To address this gap, we introduce CRIT, a new dataset and benchmark built with a graph-based automatic pipeline for generating complex cross-modal reasoning tasks. CRIT consists of diverse domains ranging from natural images, videos, and text-rich sources, and includes a manually verified test set for reliable evaluation. Experiments on this benchmark reveal that even state-of-the-art models struggle on such reasoning tasks. Models trained on CRIT show significant gains in cross-modal multi-hop reasoning, including strong improvements on SPIQA and other standard multimodal benchmarks.

cs.LG↗

arXiv2026

HEART-PFL: Stable Personalized Federated Learning under Heterogeneity with Hierarchical Directional Alignment and Adversarial Knowledge Transfer

Personalized Federated Learning (PFL) aims to deliver effective client-specific models under heterogeneous distributions, yet existing methods suffer from shallow prototype alignment and brittle server-side distillation. We propose HEART-PFL, a dual-sided framework that (i) performs depth-aware Hierarchical Directional Alignment (HDA) using cosine similarity in the early stage and MSE matching in the deep stage to preserve client specificity, and (ii) stabilizes global updates through Adversarial Knowledge Transfer (AKT) with symmetric KL distillation on clean and adversarial proxy data. Using lightweight adapters with only 1.46M trainable parameters, HEART-PFL achieves state-of-the-art personalized accuracy on CIFAR-100, Flowers-102, and Caltech-101 (63.42%, 84.23%, and 95.67%, respectively) under Dirichlet non-IID partitions, and remains robust to out-of-domain proxy data. Ablation studies further confirm that HDA and AKT provide complementary gains in alignment, robustness, and optimization stability, offering insights into how the two components mutually reinforce effective personalization. Overall, these results demonstrate that HEART-PFL simultaneously enhances personalization and global stability, highlighting its potential as a strong and scalable solution for PFL(code available at https://github.com/danny0628/HEART-PFL).

cs.CV↗

arXiv2026

SynQ: Accurate Zero-shot Quantization by Synthesis-aware Fine-tuning

How can we accurately quantize a pre-trained model without any data? Quantization algorithms are widely used for deploying neural networks on resource-constrained edge devices. Zero-shot Quantization (ZSQ) addresses the crucial and practical scenario where training data are inaccessible for privacy or security reasons. However, three significant challenges hinder the performance of existing ZSQ methods: 1) noise in the synthetic dataset, 2) predictions based on off-target patterns, and the 3) misguidance by erroneous hard labels. In this paper, we propose SynQ (Synthesis-aware Fine-tuning for Zero-shot Quantization), a carefully designed ZSQ framework to overcome the limitations of existing methods. SynQ minimizes the noise from the generated samples by exploiting a low-pass filter. Then, SynQ trains the quantized model to improve accuracy by aligning its class activation map with the pre-trained model. Furthermore, SynQ mitigates misguidance from the pre-trained model's error by leveraging only soft labels for difficult samples. Extensive experiments show that SynQ provides the state-of-the-art accuracy, over existing ZSQ methods.

cs.CV↗

arXiv2026

Prune-then-Quantize or Quantize-then-Prune? Understanding the Impact of Compression Order in Joint Model Compression

What happens when multiple compression methods are combined-does the order in which they are applied matter? Joint model compression has emerged as a powerful strategy to achieve higher efficiency by combining multiple methods such as pruning and quantization. A central but underexplored factor in joint model compression is the compression order, or the sequence of different methods within the compression pipeline. Most prior studies have either sidestepped the issue by assuming orthogonality between techniques, while a few have examined them only in highly constrained cases. Consequently, the broader role of compression order in shaping model performance remains poorly understood. In this paper, we address the overlooked problem of compression order and provide both theoretical and empirical analysis. We formulate the problem of optimizing the compression order and introduce the Progressive Intensity Hypothesis, which states that weaker perturbations should precede stronger ones. We provide theoretical guarantees showing that the relative benefit of one order increases with the underlying performance gap. Extensive experiments on both language and vision models validate the hypothesis, and further show its generality to broader setups such as multi-stage compression and mixed-precision quantization.

cs.AI↗

arXiv2026

Density-Dependent Transition in Bacterial Self-Organization Driven by Confinement and Aerotaxis

We experimentally investigate how aerotactic bacteria, confined within a thin liquid film between two solid substrates, respond to a controlled oxygen gradient. We find that the total bacterial number density dictates which mechanism dominates the steady-state spatial distribution: wall accumulation or aerotaxis. At low densities, despite receiving oxygen only from one substrate, motile bacteria accumulate at both walls, forming a symmetric distribution. In contrast, pronounced aerotactic migration toward the oxygen-supplying wall emerges as the density increases. Analyzing the temporal evolution of this bacterial distribution reveals that the aerotactic response is driven by a self-generated oxygen gradient induced by collective respiration. Our diffusion-advection model of bacteria and oxygen, accounting for aerotactic migration, hydrodynamic attraction to the walls, and respiration, quantitatively reproduces our experimental observations and provides valuable insights into bacterial self-organization within complex environments.

cond-mat.soft↗

arXiv2026

ELO: Efficient Layer-Specific Optimization for Continual Pretraining of Multilingual LLMs

We propose an efficient layer-specific optimization (ELO) method designed to enhance continual pretraining (CP) for specific languages in multilingual large language models (MLLMs). This approach addresses the common challenges of high computational cost and degradation of source language performance associated with traditional CP. The ELO method consists of two main stages: (1) ELO Pretraining, where a small subset of specific layers, identified in our experiments as the critically important first and last layers, are detached from the original MLLM and trained with the target language. This significantly reduces not only the number of trainable parameters but also the total parameters computed during the forward pass, minimizing GPU memory consumption and accelerating the training process. (2) Layer Alignment, where the newly trained layers are reintegrated into the original model, followed by a brief full fine-tuning step on a small dataset to align the parameters. Experimental results demonstrate that the ELO method achieves a training speedup of up to 6.46 times compared to existing methods, while improving target language performance by up to 6.2\% on qualitative benchmarks and effectively preserving source language (English) capabilities.

cs.CL↗

arXiv2025

Activation Boundary Matching: Task-Informed Initialization for Low-Rank Adaptation

Low-Rank Adaptation (LoRA) is highly sensitive to initialization, yet existing schemes construct the initial subspace from statistics at the pretrained point, capturing pre-adaptation geometry rather than how the adapter must move during learning. We examine the early adaptation trajectory and uncover a temporal asymmetry: task-induced activation boundaries---the signs of layer-wise pre-activations---recover markedly faster than activation values or effective low-rank updates and become reusable well before the probe adapter converges. Motivated by this, we propose Activation Boundary Matching for LoRA (ABM-LoRA), which briefly trains a standard adapter as an early task probe, then uses its pre-activation signs as layer-wise targets for a fresh adapter under a margin-based hinge objective. Both are discarded before otherwise unchanged downstream fine-tuning. We show that this boundary-supervision lifting can expose LoRA directions attenuated or locally unobservable through the downstream Jacobian, and that what transfers is the boundary side---not the exact activation value or the output-level signal. Because a partial boundary trace suffices, ABM-LoRA recovers most of the benefit of a fully trained reference at a fraction of the overhead, requiring only probe forward passes. It improves over standard LoRA on T5-base/GLUE, ConvNeXt-T and Swin-T fine-grained classification, and instruction tuning with Qwen2.5-1.5B and LLaMA2-7B, and matches or surpasses SVD- and gradient-based initializers without their preprocessing.

cs.CV↗

arXiv2025

LampQ: Towards Accurate Layer-wise Mixed Precision Quantization for Vision Transformers

How can we accurately quantize a pre-trained Vision Transformer model? Quantization algorithms compress Vision Transformers (ViTs) into low-bit formats, reducing memory and computation demands with minimal accuracy degradation. However, existing methods rely on uniform precision, ignoring the diverse sensitivity of ViT components to quantization. Metric-based Mixed Precision Quantization (MPQ) is a promising alternative, but previous MPQ methods for ViTs suffer from three major limitations: 1) coarse granularity, 2) mismatch in metric scale across component types, and 3) quantization-unaware bit allocation. In this paper, we propose LampQ (Layer-wise Mixed Precision Quantization for Vision Transformers), an accurate metric-based MPQ method for ViTs to overcome these limitations. LampQ performs layer-wise quantization to achieve both fine-grained control and efficient acceleration, incorporating a type-aware Fisher-based metric to measure sensitivity. Then, LampQ assigns bit-widths optimally through integer linear programming and further updates them iteratively. Extensive experiments show that LampQ provides the state-of-the-art performance in quantizing ViTs pre-trained on various tasks such as image classification, object detection, and zero-shot quantization.

cs.CV↗

arXiv2025

Classification for dynamics of Markov chains on non-negative integers with arbitrary transition rates and its application

Continuous-time Markov chains on non-negative integers can be used for modeling biological systems, population dynamics, and queueing models. Qualitative behaviors of birth-and-death models, typical examples of such one-dimensional continuous-time Markov chains, have been substantially studied. For one-dimensional Markov chains with polynomial transition rates, recent studies provided criteria for their long-term behavior. In this paper, we provide a classification of Markov chains on non-negative integers when the transition rates are arbitrary functions. The criteria are written with asymptotics of the transition rates. This classification implies their dynamical properties, including explosivity, recurrence, positive recurrence, and exponential ergodicity. As an application, we derive a complete classification (if and only if conditions) for those dynamical features when the transition rates have certain expansion forms, which include all rational functions, so that our classifications cover mass-action kinetics, Michaelis-Menten kinetics, and Haldane equations. Our classification solely relies on easily computable quantities: the maximal degree of the expansion of the transition rates, the mean and the variance of the transition rates. We demonstrate the utility of this classification framework using the approximation of high-dimensional mass-action systems by a one-dimensional reaction system with general rational kinetics.

math.PR↗

arXiv2025

KORMo: Korean Open Reasoning Model for Everyone

This work presents the first large-scale investigation into constructing a fully open bilingual large language model (LLM) for a non-English language, specifically Korean, trained predominantly on synthetic data. We introduce KORMo-10B, a 10.8B-parameter model trained from scratch on a Korean-English corpus in which 68.74% of the Korean portion is synthetic. Through systematic experimentation, we demonstrate that synthetic data, when carefully curated with balanced linguistic coverage and diverse instruction styles, does not cause instability or degradation during large-scale pretraining. Furthermore, the model achieves performance comparable to that of contemporary open-weight multilingual baselines across a wide range of reasoning, knowledge, and instruction-following benchmarks. Our experiments reveal two key findings: (1) synthetic data can reliably sustain long-horizon pretraining without model collapse, and (2) bilingual instruction tuning enables near-native reasoning and discourse coherence in Korean. By fully releasing all components including data, code, training recipes, and logs, this work establishes a transparent framework for developing synthetic data-driven fully open models (FOMs) in low-resource settings and sets a reproducible precedent for future multilingual LLM research.

cs.CL↗