Searcharxiv⌕ Search

arXiv subjects

Eunsu Kim

Publications and source records attributed to Eunsu Kim.

At least 19 recordsLinked to original sources

K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts

Frontier model evaluations are shifting from foundational capabilities (e.g., instruction following and reasoning) toward compositional, agentic ones, but Korean agentic benchmarks remain scarce. We introduce K-BrowseComp, a web-browsing agent benchmark grounded in Korean contexts, consisting of 400 problems. The 300-problem K-BrowseComp-Verified subset is manually constructed and validated by native Korean speakers. On this subset, frontier LLMs, including GPT-5.5, DeepSeek-V4-Pro, and GLM-5.1, reach only 30.00--45.67\%, a substantial drop from BrowseComp, while Korean LLMs released through Korea's Proprietary AI Foundation Model program obtain only 0.00--10.33\%. We further construct a 100-problem synthetic split using hard few-shot exemplars and failure-mode-targeted generation to exploit the asymmetry between solving and creating web browsing problems. On the adversarially filtered synthetic diagnostic split, the strongest model reaches only 26.00\%, and we report this split separately as a targeted stress test. We publicly release our data and code.

cs.CL↗

C-to-Rust Translation Validation at the LLVM IR Level via AI-Assisted Refinement Proof

Migrating C codebases to Rust is a promising approach to securing legacy software, but existing translators do not guarantee that a translated Rust program preserves the behavior of the C program. Neither rule-based nor large language model (LLM)-based translators have been formally verified, and existing approaches to checking translation correctness are limited to testing and bounded checks, which cannot reason about all possible executions. In this work, we present R2C2, which validates C-to-Rust translation by proving in Rocq that the Rust program refines the C program, i.e., every behavior on the Rust side is also a behavior on the C side. Since no formal semantics defines the behaviors of C and Rust programs in the same terms, we prove refinement between the two programs compiled to the LLVM intermediate representation (LLVM IR). In addition, to resolve the discrepancies between the data representations of C and Rust, we introduce a common abstraction that relates two corresponding representations through an abstract value, and specify the behavior of each library function over abstract values. We construct the refinement proof as a chain of refinement relations, where only one relation is program-specific and proved at validation time, while the others are proved for all programs in advance. Proof construction is automated by AI agents, and since Rocq checks the proofs, the agents need not be trusted. We evaluate R2C2 on LLM translations of 519 C programs drawn from a benchmark suite used in prior work on C-to-Rust translation validation. R2C2 proves the translations of 486 programs correct without bounding their executions and discovers three incorrect translations, providing valid counterexamples that existing translation validation tools cannot find. We also conduct a case study on a small real-world image compression library and demonstrate how R2C2 can handle file operations.

cs.PL↗

Expandable Modality Experts for Medical Vision-Language Learning

Learning across medical imaging modalities requires making use of images without paired text and accommodating new domains while retaining existing representations. We propose a medical vision-language technique built around an expandable bank of modality experts. Its Multiway Transformer combines shared visual computation with deterministically routed feed-forward experts and a separate text pathway. A modality-aware visual tokenizer supplies targets for masked pretraining on 19.4 million unpaired image rows across seven modalities; contrastive alignment then uses 3.23 million image-text pairs. Additional fundus and optical coherence tomography experts learn from paired data while the original image and text embedding paths remain frozen. With a ViT-Large backbone, the proposed technique achieves the best or tied-best result in 24 of 25 benchmark settings across nine modalities: 10 of 11 zero-shot classification benchmarks, all 10 linear probes, and all four retrieval benchmarks. Mean balanced accuracy is 0.674 for zero-shot classification and 0.856 for linear probing, and mean bidirectional retrieval recall is 20.70%. A separate three-seed sequential experiment with matched adaptation data yields fundus and OCT retrieval gains of 14.95 and 9.31 percentage points, with no measured loss on preceding fundus or pathology tasks. The preservation argument requires unchanged existing computation paths and inference conditions, and feature-identity checks verify retention on evaluated inputs. The proposed technique provides a modular approach to learning and expanding medical vision-language representations.

cs.CV↗

Benchmarks Are Not Atomic: Composition-Aware LLM Evaluation using BenchHub

LLM benchmarks are often treated as coherent measurement units, yet they are heterogeneous collections of instances spanning diverse domains, skills, formats, and contexts. As a result, aggregate benchmark scores can conflate model capability with benchmark composition, obscuring what existing benchmarks actually cover. We introduce BenchHub, a composition-aware evaluation framework that represents benchmarks as distributions over instance-level attributes. BenchHub integrates 54 benchmarks comprising 839K samples across 10 languages. It enables researchers to inspect benchmark contents, uncover reusable coverage hidden behind benchmark names, transparently compare benchmarks, and construct controllable evaluation sets for application-aligned model selection. Using BenchHub, we show that similarly motivated benchmarks can differ substantially in internal composition, existing benchmarks often contain reusable coverage beyond their stated purposes, models exhibit fine-grained category-level performance variation hidden by aggregate scores, and model rankings can shift under different reweighting and resampling configurations. Our results motivate evaluation practices that make benchmark composition explicit, inspectable, and controllable.

cs.LG↗

Patch-to-Global: Random Patch Diffusion for Globally Consistent Megapixel Artifact Inpainting in Whole Slide Images

Although deep learning has advanced Whole Slide Image (WSI) Analysis, tissue artifacts like bubbles and folds often cause silent failures by concealing essential morphology. Current pathology image restoration methods are mostly restricted to small patches, struggling to maintain global structural coherence at a megapixel scale. We introduce RestorePath, a framework for globally consistent megapixel scale inpainting that reconstructs diagnostic structures in histological image to prevent incorrect high-confidence predictions and lower error rates. Our model utilizes a Latent Diffusion Model (LDM) conditioned on Pathology Foundation Model (PFM) embeddings, integrating Large Kernel Attention (LKA) to manage long-range dependencies during random patch diffusion. Enhanced by Distance-Weighted Interpolation (DWI) and an Adaptive Guidance Scale (AGS), RestorePath ensures structural consistency and fidelity by modulating information from surrounding patches. Evaluations across TCGA-BRCA, BACH, and Camelyon16 datasets for images ranging from 512 to 4608 pixels demonstrate state-of-the-art performance in maintaining histological consistency. RestorePath significantly improves downstream Computational Pathology (CP) tasks, outperforming both raw artifact images and the conventional Detect-and-Discard (D&D) approach. The code is available at https://github.com/PathfinderLab/RestorePath

cs.CV↗

Benchmarking Vision-Language Models for Automated Pathology Diagnosis and Report Generation

The rapid advancement of vision-language models (VLMs) has accelerated progress in computational pathology; however, whole-slide image (WSI)-based pathology report generation remains limited by the scarcity of large-scale WSI--report datasets and the complexity of mapping spatially distributed visual patterns to structured clinical text. To address this, we introduce a clinically curated Pan-Asia WSI--report dataset of approximately 10,500 pairs from five institutions and establish the REG 2025 benchmark through a MICCAI challenge for systematic evaluation of multimodal models. We analyze submitted methods spanning pretrained VLMs, multiple-instance learning frameworks, hierarchical expert models, retrieval-augmented generation, and cross-modal Transformers. Rather than indicating that VLM use alone was sufficient for superior performance, the results suggest that top-performing methods benefited from structured report representations, hierarchical diagnostic decomposition, and effective multimodal grounding. We identify key limitations, including instability in quantitative attribute estimation (e.g., numeric hallucination) and a tendency toward diagnostic overspecification, with some errors resembling known diagnostic pitfalls in routine pathology. These findings establish REG 2025 as a benchmark for evaluating WSI-based structured report generation and vision-language understanding in computational pathology, providing insights for the design of clinically grounded multimodal pathology models.

cs.CV↗

"I Didn't Make the Micro Decisions": Measuring, Inducing, and Exposing Goal-Level AI Contributions in Collaboration

As large language models (LLMs) increasingly shape how users form, refine, and extend their goals, attributing contributions in human-AI collaboration becomes critical for users calibrating their own reliance and for evaluators assessing AI-assisted work. Yet existing methods focus on final artifacts, missing the process through which goals themselves are jointly shaped. We introduce a goal-level attribution framework, CoTrace, that decomposes explicit goals into verifiable requirements and traces both direct contributions and indirect influences across dialogue turns. Applying CoTrace to 638 real-world collaboration logs, we find that while models account for only 11-26% of goal-shaping contribution, they contribute substantially more on introducing lower-level concrete requirements, and make various kinds of indirect contributions. Through controlled simulations, we show that interaction design choices significantly affect model goal-shaping behavior. In a user study, exposing participants to goal-level analyses shifts their perceived contributions by nearly 2 points on a 5-point scale, revealing systematic miscalibration in how users understand their own AI-assisted work.

cs.CL↗

LoCar: Localization-Aware Evaluation of In-Vehicle Assistants through Fine-Grained Sociolinguistic Control

While Large Language Models (LLMs) are increasingly integrated into in-vehicle conversational systems, identifying the optimal model remains challenging due to the lack of domain-specific evaluation standards tailored to real-world deployment requirements. In this paper, we propose a novel evaluation framework for in-vehicle assistants, with a particular focus on Korean-language localization. Our empirical analysis reveals notable patterns in model behavior. First, fine-grained Korean honorific control remains unstable in current LLMs, indicating that precise speech-level realization must be explicitly evaluated in localization settings. Second, models exhibit weaker performance in strategic conversational metrics like clarification and proactivity. Our analysis suggests this stems from the inherent subjective complexity of these tasks, where our framework adopts a conservative evaluation stance to prioritize reliability. Together, our findings underscore that automotive AI must move beyond general competence toward precise linguistic tailoring and reliable, safety-oriented interaction management.

cs.CL↗

The DAMSA Experiment

DAMSA (DArk Messenger Searches at an Accelerator) is a novel short-baseline accelerator/beam dump experiment aimed at probing short-lived physics processes, including searches for evidence of a dark sector of particle physics and well-motivated rare Standard Model signals. Motivated by open questions in neutrino physics and the absence of conclusive evidence for conventional weakly interacting massive particles, DAMSA targets MeV-to-sub-GeV dark-sector messengers with feeble couplings that can be produced in abundance at a beam dump/target. By employing an ultra-short baseline, DAMSA is uniquely positioned to overcome the beam-dump "ceiling" that limits sensitivity to fast decaying particles in longer-baseline experiments. The conceptual design emphasizes a beam-dump production scheme combined with a compact detector optimized for rare decays while mitigating intense neutron-induced backgrounds, inherent to high-power proton beams. To validate the experimental strategy and detector technologies, the DAMSA Path-Finder (DPF) proof-of-concept experiment is also proposed, focusing on axion-like particles decaying to two photons, as the benchmark physics case and operating with 8 GeV electron beams at SLAC Linac-to-ESA (LESA) facility. Successful realization of DPF will establish the feasibility of the DAMSA approach, enabling a broad and powerful program to explore short-lived new physics and precision Standard Model processes in a previously inaccessible regime. This paper outlines the technical details of DAMSA's physics goals, key experimental challenges, and how to overcome them.

hep-ex↗

Are they lovers or friends? Evaluating LLMs' Social Reasoning in English and Korean Dialogues

As LLMs are increasingly deployed in real-world interactions, their social reasoning in interpersonal communication becomes critical. To explore their capabilities, we introduce SCRIPTS, a 1.1k-dialogue dataset in English and Korean, sourced from movie scripts and propose a social reasoning task based on SCRIPTS that evaluates the capacity of LLMs to infer the social relationships (e.g., friends, lovers) between speakers in each dialogue. Evaluating nine models on our task, current LLMs achieve around 75--80% on the English dataset and 58--69% in Korean, and models predict an Unlikely relationship in 10--25% of responses in both languages. Furthermore, we find that thinking models and chain-of-thought prompting provide minimal benefits for social reasoning and occasionally amplify social biases. In sum, there are significant limitations in current LLMs' social reasoning capabilities, especially for Korean, highlighting the need for efforts to develop socially-aware LLMs across languages.

cs.CL↗

RoleConflictBench: A Benchmark of Role Conflict Scenarios for Evaluating LLMs' Contextual Sensitivity

People often encounter role conflicts -- social dilemmas where the expectations of multiple roles clash and cannot be simultaneously fulfilled. As large language models (LLMs) increasingly navigate these social dynamics, a critical research question emerges. When faced with such dilemmas, do LLMs prioritize dynamic contextual cues or the learned preferences? To address this, we introduce RoleConflictBench, a novel benchmark designed to measure the contextual sensitivity of LLMs in role conflict scenarios. To enable objective evaluation within this subjective domain, we employ situational urgency as a constraint for decision-making. We construct the dataset through a three-stage pipeline that generates over 13,000 realistic scenarios across 65 roles in five social domains by systematically varying the urgency of competing situations. This controlled setup enables us to quantitatively measure contextual sensitivity, determining whether model decisions align with the situational contexts or are overridden by the learned role preferences. Our analysis of 10 LLMs reveals that models substantially deviate from this objective baseline. Instead of responding to dynamic contextual cues, their decisions are predominantly governed by the preferences toward specific social roles.

cs.CL↗

DAMSA Experiment Conceptual Design White Paper

DAMSA (DArk Messenger Searches at an Accelerator) is a novel short-baseline accelerator experiment aimed at probing short-lived physics processes, including searches for evidence of a dark sector of particle physics and well-motivated Standard Model signals. Motivated by open questions in neutrino physics and the absence of conclusive evidence for conventional weakly interacting massive particles, DAMSA targets MeV-to-sub-GeV dark-sector messengers with feeble couplings that can be produced in abundance at the PIP-II LINAC. By employing an ultra-short baseline of order one meter, DAMSA is uniquely positioned to overcome the beam-dump "ceiling" that limits sensitivity to promptly decaying particles in longer-baseline experiments. The conceptual design emphasizes a beam-dump production scheme combined with a compact detector optimized for rare decays while mitigating intense neutron-induced backgrounds inherent to high-power proton beams. To validate the experimental strategy and detector technologies, the Little DAMSA Path-Finder (LDPF) proof-of-concept experiment is proposed, focusing on axion-like particles decaying to two photons and operating with 300 MeV electron beams at FAST. Successful realization of LDPF will establish the feasibility of the DAMSA approach, enabling a broad and powerful program to explore short-lived new physics and precision Standard Model processes in a previously inaccessible regime. This conceptual design document outlines the technical details of DAMSA's physics goals, the beam facility proposals, key experimental challenges and how to overcome them, and the proposed experimental staging campaigns.

hep-ex↗

BLUCK: A Benchmark Dataset for Bengali Linguistic Understanding and Cultural Knowledge

In this work, we introduce BLUCK, a new dataset designed to measure the performance of Large Language Models (LLMs) in Bengali linguistic understanding and cultural knowledge. Our dataset comprises 2366 multiple-choice questions (MCQs) carefully curated from compiled collections of several college and job level examinations and spans 23 categories covering knowledge on Bangladesh's culture and history and Bengali linguistics. We benchmarked BLUCK using 6 proprietary and 3 open-source LLMs - including GPT-4o, Claude-3.5-Sonnet, Gemini-1.5-Pro, Llama-3.3-70B-Instruct, and DeepSeekV3. Our results show that while these models perform reasonably well overall, they, however, struggles in some areas of Bengali phonetics. Although current LLMs' performance on Bengali cultural and linguistic contexts is still not comparable to that of mainstream languages like English, our results indicate Bengali's status as a mid-resource language. Importantly, BLUCK is also the first MCQ-based evaluation benchmark that is centered around native Bengali culture, history, and linguistics.

cs.CL↗

World in a Frame: Understanding Culture Mixing as a New Challenge for Vision-Language Models

In a globalized world, cultural elements from diverse origins frequently appear together within a single visual scene. We refer to these as culture mixing scenarios, yet how Large Vision-Language Models (LVLMs) perceive them remains underexplored. We investigate culture mixing as a critical challenge for LVLMs and examine how current models behave when cultural items from multiple regions appear together. To systematically analyze these behaviors, we construct CultureMix, a food Visual Question Answering (VQA) benchmark with 23k diffusion-generated, human-verified culture mixing images across four subtasks: (1) food-only, (2) food+food, (3) food+background, and (4) food+food+background. Evaluating 10 LVLMs, we find consistent failures to preserve individual cultural identities in mixed settings. Models show strong background reliance, with accuracy dropping 14% when cultural backgrounds are added to food-only baselines, and they produce inconsistent predictions for identical foods across different contexts. To address these limitations, we explore three robustness strategies. We find supervised fine-tuning using a diverse culture mixing dataset substantially improve model consistency and reduce background sensitivity. We call for increased attention to culture mixing scenarios as a critical step toward developing LVLMs capable of operating reliably in culturally diverse real-world environments.

cs.CV↗

Uncovering Factor Level Preferences to Improve Human-Model Alignment

Large language models (LLMs) often exhibit tendencies that diverge from human preferences, such as favoring certain writing styles or producing overly verbose outputs. While crucial for improvement, identifying the factors driving these misalignments remains challenging due to existing evaluation methods' reliance on coarse-grained comparisons and lack of explainability. To address this, we introduce PROFILE, an automated framework to uncover and measure factor-level preference alignment of humans and LLMs. Using PROFILE, we analyze preference alignment across three key tasks: summarization, instruction-following, and document-based QA. We find a significant discrepancy: while LLMs show poor factor-level alignment with human preferences when generating texts, they demonstrate strong alignment in discrimination tasks. We demonstrate how leveraging the identified generation-discrimination gap can be used to improve LLM alignment through multiple approaches, including fine-tuning with self-guidance. Our work highlights the value of factor-level analysis for identifying hidden misalignments and provides a practical framework for improving LLM-human preference alignment.

cs.CL↗

MUG-Eval: A Proxy Evaluation Framework for Multilingual Generation Capabilities in Any Language

Evaluating text generation capabilities of large language models (LLMs) is challenging, particularly for low-resource languages where methods for direct assessment are scarce. We propose MUG-Eval, a novel framework that evaluates LLMs' multilingual generation capabilities by transforming existing benchmarks into conversational tasks and measuring the LLMs' accuracies on those tasks. We specifically designed these conversational tasks to require effective communication in the target language. Then, we simply use task success rate as a proxy for successful conversation generation. Our approach offers two key advantages: it is independent of language-specific NLP tools or annotated datasets, which are limited for most languages, and it does not rely on LLMs-as-judges, whose evaluation quality degrades outside a few high-resource languages. We evaluate 8 LLMs across 30 languages spanning high, mid, and low-resource categories, and we find that MUG-Eval correlates strongly with established benchmarks ($r$ > 0.75) while enabling standardized comparisons across languages and models. Our framework provides a robust and resource-efficient solution for evaluating multilingual generation that can be extended to thousands of languages.

cs.CL↗

Diffusion Models Through a Global Lens: Are They Culturally Inclusive?

Text-to-image diffusion models have recently enabled the creation of visually compelling, detailed images from textual prompts. However, their ability to accurately represent various cultural nuances remains an open question. In our work, we introduce CultDiff benchmark, evaluating state-of-the-art diffusion models whether they can generate culturally specific images spanning ten countries. We show that these models often fail to generate cultural artifacts in architecture, clothing, and food, especially for underrepresented country regions, by conducting a fine-grained analysis of different similarity aspects, revealing significant disparities in cultural relevance, description fidelity, and realism compared to real-world reference images. With the collected human evaluations, we develop a neural-based image-image similarity metric, namely, CultDiff-S, to predict human judgment on real and generated images with cultural artifacts. Our work highlights the need for more inclusive generative AI systems and equitable dataset representation over a wide range of cultures.

cs.CV↗

Spotting Out-of-Character Behavior: Atomic-Level Evaluation of Persona Fidelity in Open-Ended Generation

Ensuring persona fidelity in large language models (LLMs) is essential for maintaining coherent and engaging human-AI interactions. However, LLMs often exhibit Out-of-Character (OOC) behavior, where generated responses deviate from an assigned persona, leading to inconsistencies that affect model reliability. Existing evaluation methods typically assign single scores to entire responses, struggling to capture subtle persona misalignment, particularly in long-form text generation. To address this limitation, we propose an atomic-level evaluation framework that quantifies persona fidelity at a finer granularity. Our three key metrics measure the degree of persona alignment and consistency within and across generations. Our approach enables a more precise and realistic assessment of persona fidelity by identifying subtle deviations that real users would encounter. Through our experiments, we demonstrate that our framework effectively detects persona inconsistencies that prior methods overlook. By analyzing persona fidelity across diverse tasks and personality types, we reveal how task structure and persona desirability influence model adaptability, highlighting challenges in maintaining consistent persona expression.

cs.CL↗