arXiv subjects

Jiwoo Lee

Publications and source records attributed to Jiwoo Lee.

At least 19 recordsLinked to original sources

arXiv2026

Scenes as Objects, Not Primitives: Instance-Structured 3D Tokenization from Unposed Views

A 3D scene is understood through its objects, not the primitives that compose them. Yet feed-forward reconstruction methods output dense, unstructured sets of points or Gaussians, leaving object-level structure to be recovered after the fact. We propose a feed-forward framework that decomposes a scene into instance-structured 3D token groups directly from unposed multi-view images -- compact object-centric units from which reconstruction, segmentation, and manipulation all follow. Each token group pairs an instance token capturing entity-level identity with anchor tokens that encode local geometry and appearance, which are decoded into a set of 3D Gaussians. This two-level factorization decouples object identity from local appearance, making object instances a native interface of the representation rather than a derived product. The token groups are learned through differentiable rendering with joint reconstruction and segmentation supervision, requiring no 3D annotations. Our feed-forward model surpasses per-scene optimization baselines in class-agnostic instance segmentation while remaining competitive in novel view synthesis. Beyond these metrics, the same token groups directly unlock instance-level scene editing -- removing, translating, or inserting objects by operating on their groups -- as well as efficient open-vocabulary 3D instance retrieval, where retrieval complexity scales with the number of instances rather than primitives.

cs.CV↗

arXiv2026

Bounded Personas Match Retrieval on Classification but Not Regression for a Frozen Agent

A personalized language agent must convert a user's interaction history into behavior on each new request at inference time. Two strategies dominate. Retrieval pulls a few of the user's most relevant past items into the prompt, which is accurate but pays a per-query selection and context cost that grows with the history. Distillation instead compresses the history once into a compact natural-language persona, which is bounded, query-independent, and interpretable, but is widely assumed to sacrifice accuracy. Whether, and on which tasks, a distilled persona can match retrieval has not been characterized cleanly. We introduce PersonaLink, a training-free method that distills a user's history into a bounded three-field persona and recursively refines it: each pass self-evaluates the frozen agent on a held-out slice of the user's own labeled history, rewrites the persona from its errors, and keeps the result only when it does not regress on that slice. Because every comparison shares one frozen 7B backbone and differs only in what is placed in context, the design isolates the effect of representation from that of the model. The result is a clear task-type asymmetry. On 200 users of LaMP-2 (15-way news categorization), PersonaLink reaches 0.745-0.755 accuracy, statistically indistinguishable from BM25 retrieval (0.760-0.765).

cs.CL↗

arXiv2026

MolDeTox: Evaluating Language Model's Stepwise Fragment Editing for Molecular Detoxification

Large Language Models (LLMs) and Vision Language Models (VLMs) have recently shown promising capabilities in various scientific domain. In particular, these advances have opened new opportunities in drug discovery, where the ability to understand and modify molecular structures is critical for optimizing drug properties such as efficacy and toxicity. However, existing models and benchmarks often overlook toxicity-related challenges, focusing primarily on general property optimization without adequately addressing safety concerns. In addition, even existing toxicity repair benchmarks suffer from limited data diversity, low structural validity of generated molecules, and heavy reliance on proxy models for toxicity assessment. To address these limitations, we propose MolDeTox, a novel benchmark for molecular detoxification, designed to enable fine-grained and reliable evaluation of toxicity-aware molecular optimization across stepwise tasks. We evaluate a wide range of general-purpose LLMs and VLMs under diverse settings, and demonstrate that understanding and generating molecules at the fragment-level improves structural validity and enhances the quality of generated molecules. Moreover, through detailed task-level performance analysis, MolDeTox provides an interpretable benchmark that enables a deeper understanding of the detoxification process. Our dataset is available at : https://huggingface.co/datasets/MolDeTox/MolDeTox

cs.AI↗

arXiv2026

Teaching Language Models to Think in Code

Tool-integrated reasoning (TIR) has emerged as a dominant paradigm for mathematical problem solving in language models, combining natural language (NL) reasoning with code execution. However, this interleaved setup has three key limitations: code often acts as a post-hoc verifier, intermediate NL computations are error-prone, and NL and code play overlapping rather than clearly distinct roles. We propose ThinC (Thinking in Code), a framework in which code itself serves as the reasoner rather than as a tool invoked by NL. A ThinC trajectory begins with a brief NL planning step, after which all reasoning unfolds through code blocks connected only by their execution outputs. We distill 12.2k code-centric trajectories from a teacher model and train ThinC-1.7B and ThinC-4B with supervised fine-tuning followed by reinforcement learning. ThinC-4B consistently outperforms every TIR baseline on five competition-level math benchmarks and even surpasses the much larger Qwen3-235B-A22B-Thinking. Further analysis shows that ThinC reasons through code: 99.2% of its final answers are grounded in interpreter output, and the model recovers reliably from code execution failures without intermediate NL reasoning. Our code and models will be released soon.

cs.CL↗

arXiv2026

A PMP-inspired Evaluation Framework for Assessing Deep-Learning Earth System Models

In recent years, Deep-Learning Earth System Models (DL-ESMs) have emerged as promising, computationally efficient complements to traditional Earth system models. Here, we present an evaluation framework for testing DL-ESMs from an Earth system model-development perspective using standardized diagnostics from the PCMDI Metrics Package (PMP). This framework allows DL-ESMs, including Ai2's ACE2 and Google's NeuralGCM, to be assessed with metrics that quantify their ability to reproduce climatology, major modes of variability, monsoons, and precipitation variability relative to observational reference datasets and CMIP-class benchmarks. By evaluating DL-ESMs with tools commonly used for traditional models, we extend their assessment beyond short-range forecast skill and toward longer Earth System-relevant applications. The results identify encouraging strengths in several large-scale fields and modes of variability, while also highlighting persistent challenges in precipitation, tropical variability, and long-run stability for some model versions. This evaluation is a critical step toward building trust in DL-ESMs, guiding future model development, and clarifying their fit-for-purpose for Earth system science applications.

physics.ao-ph↗

arXiv2026

Benchmarking Direct Preference Optimization for Medical Large Vision-Language Models

Large Vision-Language Models (LVLMs) hold significant promise for medical applications, yet their deployment is often constrained by insufficient alignment and reliability. While Direct Preference Optimization (DPO) has emerged as a potent framework for refining model responses, its efficacy in high-stakes medical contexts remains underexplored, lacking the rigorous empirical groundwork necessary to guide future methodological advances. To bridge this gap, we present the first comprehensive examination of diverse DPO variants within the medical domain, evaluating nine distinct formulations across two medical LVLMs: LLaVA-Med and HuatuoGPT-Vision. Our results reveal several critical limitations: current DPO approaches often yield inconsistent gains over supervised fine-tuning, with their efficacy varying significantly across different tasks and backbones. Furthermore, they frequently fail to resolve fundamental visual misinterpretation errors. Building on these insights, we present a targeted preference construction strategy as a proof-of-concept that explicitly addresses visual misinterpretation errors frequently observed in existing DPO models. This design yields a 3.6% improvement over the strongest existing DPO baseline on visual question-answering tasks. To support future research, we release our complete framework, including all training data, model checkpoints, and our codebase at https://github.com/dmis-lab/med-vlm-dpo.

cs.CV↗

arXiv2025

Spinal Line Detection for Posture Evaluation through Train-ing-free 3D Human Body Reconstruction with 2D Depth Images

The spinal angle is an important indicator of body balance. It is important to restore the 3D shape of the human body and estimate the spine center line. Existing mul-ti-image-based body restoration methods require expensive equipment and complex pro-cedures, and single image-based body restoration methods have limitations in that it is difficult to accurately estimate the internal structure such as the spine center line due to occlusion and viewpoint limitation. This study proposes a method to compensate for the shortcomings of the multi-image-based method and to solve the limitations of the sin-gle-image method. We propose a 3D body posture analysis system that integrates depth images from four directions to restore a 3D human model and automatically estimate the spine center line. Through hierarchical matching of global and fine registration, restora-tion to noise and occlusion is performed. Also, the Adaptive Vertex Reduction is applied to maintain the resolution and shape reliability of the mesh, and the accuracy and stabil-ity of spinal angle estimation are simultaneously secured by using the Level of Detail en-semble. The proposed method achieves high-precision 3D spine registration estimation without relying on training data or complex neural network models, and the verification confirms the improvement of matching quality.

cs.CV↗

arXiv2025

Clothing agnostic Pre-inpainting Virtual Try-ON

With the development of deep learning technology, virtual try-on technology has devel-oped important application value in the fields of e-commerce, fashion, and entertainment. The recently proposed Leffa technology has addressed the texture distortion problem of diffusion-based models, but there are limitations in that the bottom detection inaccuracy and the existing clothing silhouette persist in the synthesis results. To solve this problem, this study proposes CaP-VTON (Clothing Agnostic Pre-Inpainting Virtual Try-On). CaP-VTON integrates DressCode-based multi-category masking and Stable Diffu-sion-based skin inflation preprocessing; in particular, a generated skin module was in-troduced to solve skin restoration problems that occur when long-sleeved images are con-verted to short-sleeved or sleeveless ones, introducing a preprocessing structure that im-proves the naturalness and consistency of full-body clothing synthesis, and allowing the implementation of high-quality restoration considering human posture and color. As a result, CaP-VTON achieved 92.5%, which is 15.4% better than Leffa, in short-sleeved syn-thesis accuracy, and consistently reproduced the style and shape of the reference clothing in visual evaluation. These structures maintain model-agnostic properties and are appli-cable to various diffusion-based virtual inspection systems; they can also contribute to applications that require high-precision virtual wearing, such as e-commerce, custom styling, and avatar creation.

cs.CV↗

arXiv2025

Conditional Temporal Neural Processes with Covariance Loss

We introduce a novel loss function, Covariance Loss, which is conceptually equivalent to conditional neural processes and has a form of regularization so that is applicable to many kinds of neural networks. With the proposed loss, mappings from input variables to target variables are highly affected by dependencies of target variables as well as mean activation and mean dependencies of input and target variables. This nature enables the resulting neural networks to become more robust to noisy observations and recapture missing dependencies from prior information. In order to show the validity of the proposed loss, we conduct extensive sets of experiments on real-world datasets with state-of-the-art models and discuss the benefits and drawbacks of the proposed Covariance Loss.

cs.LG↗

arXiv2025

Exploring Multimodal Perception in Large Language Models Through Perceptual Strength Ratings

This study investigated whether multimodal large language models can achieve human-like sensory grounding by examining their ability to capture perceptual strength ratings across sensory modalities. We explored how model characteristics (size, multimodal capabilities, architectural generation) influence grounding performance, distributional factor dependencies (word frequency, embeddings, feature distances), and human-model processing differences. We evaluated 21 models from four families (GPT, Gemini, LLaMA, Qwen) using 3,611 words from the Lancaster Sensorimotor Norms through correlation, distance metrics, and qualitative analysis. Results showed that larger (6 out of 8 comparisons), multimodal (5 of 7), and newer models (5 of 8) generally outperformed their smaller, text-based, and older counterparts. Top models achieved 85-90% accuracy and 0.58-0.65 correlations with human ratings, demonstrating substantial similarity. Moreover, distributional factors showed minimal impact, not exceeding human dependency levels. However, despite strong alignment, models were not identical to humans, as even top performers showed differences in distance and correlation measures, with qualitative analysis revealing processing patterns related to absent sensory grounding. Additionally, it remains questionable whether introducing multimodality resolves this grounding deficit. Although multimodality improved performance, it seems to provide similar information as massive text rather than qualitatively different data, as benefits occurred across unrelated sensory dimensions and massive text-only models achieved comparable results. Our findings demonstrate that while advanced LLMs can approximate human sensory-linguistic associations through statistical learning, they still differ from human embodied cognition in processing mechanisms, even with multimodal integration.

cs.CL↗

arXiv2025

Evaluating Large language models on Understanding Korean indirect Speech acts

To accurately understand the intention of an utterance is crucial in conversational communication. As conversational artificial intelligence models are rapidly being developed and applied in various fields, it is important to evaluate the LLMs' capabilities of understanding the intentions of user's utterance. This study evaluates whether current LLMs can understand the intention of an utterance by considering the given conversational context, particularly in cases where the actual intention differs from the surface-leveled, literal intention of the sentence, i.e. indirect speech acts. Our findings reveal that Claude3-Opus outperformed the other competing models, with 71.94% in MCQ and 65% in OEQ, showing a clear advantage. In general, proprietary models exhibited relatively higher performance compared to open-source models. Nevertheless, no LLMs reached the level of human performance. Most LLMs, except for Claude3-Opus, demonstrated significantly lower performance in understanding indirect speech acts compared to direct speech acts, where the intention is explicitly revealed through the utterance. This study not only performs an overall pragmatic evaluation of each LLM's language use through the analysis of OEQ response patterns, but also emphasizes the necessity for further research to improve LLMs' understanding of indirect speech acts for more natural communication with humans.

cs.CL↗

arXiv2024

Recommendations for Comprehensive and Independent Evaluation of Machine Learning-Based Earth System Models

Machine learning (ML) is a revolutionary technology with demonstrable applications across multiple disciplines. Within the Earth science community, ML has been most visible for weather forecasting, producing forecasts that rival modern physics-based models. Given the importance of deepening our understanding and improving predictions of the Earth system on all time scales, efforts are now underway to develop forecasting models into Earth-system models (ESMs), capable of representing all components of the coupled Earth system (or their aggregated behavior) and their response to external changes. Modeling the Earth system is a much more difficult problem than weather forecasting, not least because the model must represent the alternate (e.g., future) coupled states of the system for which there are no historical observations. Given that the physical principles that enable predictions about the response of the Earth system are often not explicitly coded in these ML-based models, demonstrating the credibility of ML-based ESMs thus requires us to build evidence of their consistency with the physical system. To this end, this paper puts forward five recommendations to enhance comprehensive, standardized, and independent evaluation of ML-based ESMs to strengthen their credibility and promote their wider use.

cs.LG↗

arXiv2024

Learning from Negative Samples in Biomedical Generative Entity Linking

Generative models have become widely used in biomedical entity linking (BioEL) due to their excellent performance and efficient memory usage. However, these models are usually trained only with positive samples, i.e., entities that match the input mention's identifier, and do not explicitly learn from hard negative samples, which are entities that look similar but have different meanings. To address this limitation, we introduce ANGEL (Learning from Negative Samples in Biomedical Generative Entity Linking), the first framework that trains generative BioEL models using negative samples. Specifically, a generative model is initially trained to generate positive entity names from the knowledge base for given input entities. Subsequently, both correct and incorrect outputs are gathered from the model's top-k predictions. Finally, the model is updated to prioritize the correct predictions through preference optimization. Our models outperform the previous best baseline models by up to an average top-1 accuracy of 1.4% on five benchmarks. When incorporating our framework into pre-training, the performance improvement increases further to 1.7%, demonstrating its effectiveness in both the pre-training and fine-tuning stages. The code and model weights are available at https://github.com/dmis-lab/ANGEL.

cs.CL↗

arXiv2024

LAPIS: Language Model-Augmented Police Investigation System

Crime situations are race against time. An AI-assisted criminal investigation system, providing prompt but precise legal counsel is in need for police officers. We introduce LAPIS (Language Model Augmented Police Investigation System), an automated system that assists police officers to perform rational and legal investigative actions. We constructed a finetuning dataset and retrieval knowledgebase specialized in crime investigation legal reasoning task. We extended the dataset's quality by incorporating manual curation efforts done by a group of domain experts. We then finetuned the pretrained weights of a smaller Korean language model to the newly constructed dataset and integrated it with the crime investigation knowledgebase retrieval approach. Experimental results show LAPIS' potential in providing reliable legal guidance for police officers, even better than the proprietary GPT-4 model. Qualitative analysis on the rationales generated by LAPIS demonstrate the model's reasoning ability to leverage the premises and derive legally correct conclusions.

cs.CL↗

arXiv2024

MultiPragEval: Multilingual Pragmatic Evaluation of Large Language Models

As the capabilities of Large Language Models (LLMs) expand, it becomes increasingly important to evaluate them beyond basic knowledge assessment, focusing on higher-level language understanding. This study introduces MultiPragEval, the first multilingual pragmatic evaluation of LLMs, designed for English, German, Korean, and Chinese. Comprising 1200 question units categorized according to Grice's Cooperative Principle and its four conversational maxims, MultiPragEval enables an in-depth assessment of LLMs' contextual awareness and their ability to infer implied meanings. Our findings demonstrate that Claude3-Opus significantly outperforms other models in all tested languages, establishing a state-of-the-art in the field. Among open-source models, Solar-10.7B and Qwen1.5-14B emerge as strong competitors. By analyzing pragmatic inference, we provide valuable insights into the capabilities essential for advanced language comprehension in AI systems.

cs.CL↗

arXiv2024

KU-DMIS at EHRSQL 2024:Generating SQL query via question templatization in EHR

Transforming natural language questions into SQL queries is crucial for precise data retrieval from electronic health record (EHR) databases. A significant challenge in this process is detecting and rejecting unanswerable questions that request information beyond the database's scope or exceed the system's capabilities. In this paper, we introduce a novel text-to-SQL framework that robustly handles out-of-domain questions and verifies the generated queries with query execution.Our framework begins by standardizing the structure of questions into a templated format. We use a powerful large language model (LLM), fine-tuned GPT-3.5 with detailed prompts involving the table schemas of the EHR database system. Our experimental results demonstrate the effectiveness of our framework on the EHRSQL-2024 benchmark benchmark, a shared task in the ClinicalNLP workshop. Although a straightforward fine-tuning of GPT shows promising results on the development set, it struggled with the out-of-domain questions in the test set. With our framework, we improve our system's adaptability and achieve competitive performances in the official leaderboard of the EHRSQL-2024 challenge.

cs.DB↗

arXiv2024

Small Language Models Learn Enhanced Reasoning Skills from Medical Textbooks

While recent advancements in commercial large language models (LM) have shown promising results in medical tasks, their closed-source nature poses significant privacy and security concerns, hindering their widespread use in the medical field. Despite efforts to create open-source models, their limited parameters often result in insufficient multi-step reasoning capabilities required for solving complex medical problems. To address this, we introduce Meerkat, a new family of medical AI systems ranging from 7 to 70 billion parameters. The models were trained using our new synthetic dataset consisting of high-quality chain-of-thought reasoning paths sourced from 18 medical textbooks, along with diverse instruction-following datasets. Our systems achieved remarkable accuracy across six medical benchmarks, surpassing the previous best models such as MediTron and BioMistral, and GPT-3.5 by a large margin. Notably, Meerkat-7B surpassed the passing threshold of the United States Medical Licensing Examination (USMLE) for the first time for a 7B-parameter model, while Meerkat-70B outperformed GPT-4 by an average of 1.3%. Additionally, Meerkat-70B correctly diagnosed 21 out of 38 complex clinical cases, outperforming humans' 13.8 and closely matching GPT-4's 21.8. Our systems offered more detailed free-form responses to clinical queries compared to existing small models, approaching the performance level of large commercial models. This significantly narrows the performance gap with large LMs, showcasing its effectiveness in addressing complex medical challenges.

cs.CL↗

arXiv2024

Pragmatic Competence Evaluation of Large Language Models for the Korean Language

Benchmarks play a significant role in the current evaluation of Large Language Models (LLMs), yet they often overlook the models' abilities to capture the nuances of human language, primarily focusing on evaluating embedded knowledge and technical skills. To address this gap, our study evaluates how well LLMs understand context-dependent expressions from a pragmatic standpoint, specifically in Korean. We use both Multiple-Choice Questions (MCQs) for automatic evaluation and Open-Ended Questions (OEQs) assessed by human experts. Our results show that GPT-4 leads with scores of 81.11 in MCQs and 85.69 in OEQs, closely followed by HyperCLOVA X. Additionally, while few-shot learning generally improves performance, Chain-of-Thought (CoT) prompting tends to encourage literal interpretations, which may limit effective pragmatic inference. Our findings highlight the need for LLMs to better understand and generate language that reflects human communicative norms.

cs.CL↗