arXiv subjects

Hyomin Kim

Publications and source records attributed to Hyomin Kim.

At least 19 recordsLinked to original sources

arXiv2026

Plan-and-Verify Video Reward Reasoning with Spatio-Temporal Scene Graph Grounding

Reward models for text-to-video (T2V) generation guide post-training but often fail at fine-grained semantic alignment. We trace this to two structural weaknesses in existing reasoning-based reward models: they do not systematically verify every condition described in the prompt, and the visual evidence supporting each judgment remains implicit in their free-form reasoning. We propose SG-PVR, a video reward model that addresses these limitations through plan-and-verify reasoning grounded in spatio-temporal scene graphs. The verification plan decomposes the prompt into atomic claims, ensuring every requirement is checked. The spatio-temporal scene graph, encoding entities, attributes, and temporally-grounded relations, is extracted from the video and maintained as a persistent structured visual reference throughout reasoning. Each claim is verified against both the video and the scene graph, anchoring judgments in explicit visual evidence. SG-PVR achieves strong performance on semantic alignment, including fine-grained temporal semantics. As a test-time reranker, it further enhances compositional alignment in T2V generation.

cs.CV↗

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

Learning Adaptive Perturbation-Conditioned Contexts for Robust Transcriptional Response Prediction

Predicting high-dimensional transcriptional responses to genetic perturbations is challenging because signals are sparse and experimental noise is severe. Existing methods often suffer from mean collapse, achieving high correlation by predicting the global average expression rather than perturbation-specific responses, which yields false positives and poor interpretability. Methods that add biological knowledge graphs typically treat them as dense, static priors shared across perturbations, propagating noise. We propose AdaPert, which counters mean collapse by extracting a sparse, perturbation-specific subgraph via differentiable node selection, then suppressing spurious variation in non-responsive genes while emphasizing differentially expressed ones. Across multiple benchmarks, \textsc{AdaPert} outperforms existing baselines, with the largest gains on DEG-aware metrics.

cs.CE↗

arXiv2026

Progressive Multi-Agent Reasoning for Biological Perturbation Prediction

Predicting gene regulation responses to biological perturbations requires reasoning about underlying biological causalities. While large language models (LLMs) show promise for such tasks, they are often overwhelmed by the entangled nature of high-dimensional perturbation results. Moreover, recent works have primarily focused on genetic perturbations in single-cell experiments, leaving bulk-cell chemical perturbations, which is central to drug discovery, largely unexplored. Motivated by this, we present LINCSQA, a novel benchmark for predicting target gene regulation under complex chemical perturbations in bulk-cell environments. We further propose PBio-Agent, a multi-agent framework that integrates difficulty-aware task sequencing with iterative knowledge refinement. Our key insight is that genes affected by the same perturbation share causal structure, allowing confidently predicted genes to contextualize more challenging cases. The framework employs specialized agents enriched with biological knowledge graphs, while a synthesis agent integrates outputs and specialized judges ensure logical coherence. PBio-Agent outperforms existing baselines on both LINCSQA and PerturbQA, enabling even smaller models to predict and explain complex biological processes without additional training.

cs.AI↗

arXiv2026

DNACHUNKER: Learnable Tokenization for DNA Language Models

DNA language models are increasingly used to represent genomic sequence, yet their effectiveness depends critically on how raw nucleotides are converted into model inputs. Unlike natural language, DNA offers no canonical boundaries, making fixed tokenizations a brittle design choice under shifts, indels, and local repeats. We introduce DNAChunker, a masked DNA language model that incorporates a learnable adaptive segmentation module to produce context-dependent, variable-length units. Building on a dynamic segmentation procedure, DNAChunker learns to allocate finer granularity to functionally enriched regions while compressing repetitive or redundant sequence. We pretrain DNAChunker on the human reference genome and evaluate it across five benchmarks, where it consistently improves over strong fixed-tokenization baselines. Further analyses and ablations indicate that unlike fixed tokenizations, segmentation is learned in a biologically-informed, mutation-resilient manner.

q-bio.GN↗

arXiv2025

Camera Splatting for Continuous View Optimization

We propose Camera Splatting, a novel view optimization framework for novel view synthesis. Each camera is modeled as a 3D Gaussian, referred to as a camera splat, and virtual cameras, termed point cameras, are placed at 3D points sampled near the surface to observe the distribution of camera splats. View optimization is achieved by continuously and differentiably refining the camera splats so that desirable target distributions are observed from the point cameras, in a manner similar to the original 3D Gaussian splatting. Compared to the Farthest View Sampling (FVS) approach, our optimized views demonstrate superior performance in capturing complex view-dependent phenomena, including intense metallic reflections and intricate textures such as text.

cs.CV↗

arXiv2025

OSPO: Object-Centric Self-Improving Preference Optimization for Text-to-Image Generation

Recent advances in Multimodal Large Language Models (MLLMs) have enabled unified multimodal understanding and generation. However, they still struggle with fine-grained text-image alignment, often failing to faithfully depict objects with correct attributes such as color, shape, and spatial relations. To mitigate this issue, previous studies have explored preference optimization methods such as DPO and GRPO, but these approaches incur substantial computational cost, both in constructing preference data and in performing optimization. This has motivated self-improving preference optimization approaches, in which the MLLM autonomously generates its own training data, self-estimates preference feedback, and self-optimizes using the resulting self-constructed preference pairs. However, existing self-improving methods still overlook fine-grained, object-level semantics, allowing object hallucination to persist. To tackle this problem, we propose Object-centric Self-improving Preference Optimization (OSPO), a self-improving framework designed to enhance object-level text-image alignment. OSPO explicitly constructs object-centric preference data without relying on any external data and external models. We also introduce a new approach that leverages attention-based object masks together with an object-weighted SimPO loss to enhance object-specific fidelity. Extensive experiments on three compositional image generation benchmarks demonstrate that OSPO significantly improves fine-grained alignment and reduces object hallucination, outperforming prior self-improving methods and even specialized diffusion-based text-to-image models.

cs.CV↗

arXiv2025

MT-Mol:Multi Agent System with Tool-based Reasoning for Molecular Optimization

Large language models (LLMs) have large potential for molecular optimization, as they can gather external chemistry tools and enable collaborative interactions to iteratively refine molecular candidates. However, this potential remains underexplored, particularly in the context of structured reasoning, interpretability, and comprehensive tool-grounded molecular optimization. To address this gap, we introduce MT-Mol, a multi-agent framework for molecular optimization that leverages tool-guided reasoning and role-specialized LLM agents. Our system incorporates comprehensive RDKit tools, categorized into five distinct domains: structural descriptors, electronic and topological features, fragment-based functional groups, molecular representations, and miscellaneous chemical properties. Each category is managed by an expert analyst agent, responsible for extracting task-relevant tools and enabling interpretable, chemically grounded feedback. MT-Mol produces molecules with tool-aligned and stepwise reasoning through the interaction between the analyst agents, a molecule-generating scientist, a reasoning-output verifier, and a reviewer agent. As a result, we show that our framework shows the state-of-the-art performance of the PMO-1K benchmark on 17 out of 23 tasks.

cs.AI↗

arXiv2025

Slot-MLLM: Object-Centric Visual Tokenization for Multimodal LLM

Recently, multimodal large language models (MLLMs) have emerged as a key approach in achieving artificial general intelligence. In particular, vision-language MLLMs have been developed to generate not only text but also visual outputs from multimodal inputs. This advancement requires efficient image tokens that LLMs can process effectively both in input and output. However, existing image tokenization methods for MLLMs typically capture only global abstract concepts or uniformly segmented image patches, restricting MLLMs' capability to effectively understand or generate detailed visual content, particularly at the object level. To address this limitation, we propose an object-centric visual tokenizer based on Slot Attention specifically for MLLMs. In particular, based on the Q-Former encoder, diffusion decoder, and residual vector quantization, our proposed discretized slot tokens can encode local visual details while maintaining high-level semantics, and also align with textual data to be integrated seamlessly within a unified next-token prediction framework of LLMs. The resulting Slot-MLLM demonstrates significant performance improvements over baselines with previous visual tokenizers across various vision-language tasks that entail local detailed comprehension and generation. Notably, this work is the first demonstration of the feasibility of object-centric slot attention performed with MLLMs and in-the-wild natural images.

cs.CV↗

arXiv2024

HyperCLOVA X Technical Report

We introduce HyperCLOVA X, a family of large language models (LLMs) tailored to the Korean language and culture, along with competitive capabilities in English, math, and coding. HyperCLOVA X was trained on a balanced mix of Korean, English, and code data, followed by instruction-tuning with high-quality human-annotated datasets while abiding by strict safety guidelines reflecting our commitment to responsible AI. The model is evaluated across various benchmarks, including comprehensive reasoning, knowledge, commonsense, factuality, coding, math, chatting, instruction-following, and harmlessness, in both Korean and English. HyperCLOVA X exhibits strong reasoning capabilities in Korean backed by a deep understanding of the language and cultural nuances. Further analysis of the inherent bilingual nature and its extension to multilingualism highlights the model's cross-lingual proficiency and strong generalization ability to untargeted languages, including machine translation between several language pairs and cross-lingual inference tasks. We believe that HyperCLOVA X can provide helpful guidance for regions or countries in developing their sovereign LLMs.

cs.CL↗

arXiv2024

Discontinuity-preserving Normal Integration with Auxiliary Edges

Many surface reconstruction methods incorporate normal integration, which is a process to obtain a depth map from surface gradients. In this process, the input may represent a surface with discontinuities, e.g., due to self-occlusion. To reconstruct an accurate depth map from the input normal map, hidden surface gradients occurring from the jumps must be handled. To model these jumps correctly, we design a novel discretization scheme for the domain of normal integration. Our key idea is to introduce auxiliary edges, which bridge between piecewise-smooth patches in the domain so that the magnitude of hidden jumps can be explicitly expressed. Using the auxiliary edges, we design a novel algorithm to optimize the discontinuity and the depth map from the input normal map. Our method optimizes discontinuities by using a combination of iterative re-weighted least squares and iterative filtering of the jump magnitudes on auxiliary edges to provide strong sparsity regularization. Compared to previous discontinuity-preserving normal integration methods, which model the magnitudes of jumps only implicitly, our method reconstructs subtle discontinuities accurately thanks to our explicit representation of jumps allowing for strong sparsity regularization.

cs.CV↗

arXiv2024

WIMP constraints from black hole low-mass X-ray binaries

The abnormally fast orbital decay observed in the black hole (BH) Low-Mass X-ray binaries (BH-LMXB) A0620-00 and XTE J1118+480 can be explained by the dynamical friction between Dark Matter (DM) and the companion star orbiting around the low-mass BH (of a few $M_\odot$) of the system. In this case the value of the index $γ_{\rm sp}$ of the DM spike surrounding the BH can be pinned down with an accuracy of a few percent, way better than that for much bigger systems such as the super massive BHs (SMBHs) in the Galactic Center or in M87. We have used data from XTE J1118+480 to put bounds on the WIMP annihilation cross section times velocity $\langle σv\rangle$, assuming that DM annihilation is driven by the $b\bar{b}$ annihilation channel and that it proceeds in $s$-wave. The bounds are driven by the radio synchrotron signal produced by $e^\pm$ final states propagating in the magnetic field near the BH. For DM masses $m_χ$ up to the TeV scale XTE J1118+480 allows to constrain $\langle σv\rangle$ well below $\langleσv\rangle_{\rm thermal}$, corresponding to the observed DM relic density in the Universe for a thermal WIMP. On the other hand, for $m_χ\gtrsim$ 15 GeV, the bounds from the SMBHs in the GC or in M87 do not reach $\langleσv\rangle_{\rm thermal}$ when the very large uncertainties on the corresponding spike indices are taken into account, in spite of potentially producing much larger DM densities compared to XTE J1118+480. Our bounds for XTE J1118+480 have a mild sensitivity on spatial diffusion, but diffusion enhances the sensitivity of the results upon the intensity of the magnetic field. Taken at face value the bound from XTE J1118+480 on $\langle σv\rangle$ is the most constraining compared to all others for $m_χ\lesssim$ 1 TeV, unless the intensity of the magnetic field is significantly smaller than its equipartition estimation.

hep-ph↗

arXiv2024

Hybrid Neural Representations for Spherical Data

In this paper, we study hybrid neural representations for spherical data, a domain of increasing relevance in scientific research. In particular, our work focuses on weather and climate data as well as comic microwave background (CMB) data. Although previous studies have delved into coordinate-based neural representations for spherical signals, they often fail to capture the intricate details of highly nonlinear signals. To address this limitation, we introduce a novel approach named Hybrid Neural Representations for Spherical data (HNeR-S). Our main idea is to use spherical feature-grids to obtain positional features which are combined with a multilayer perception to predict the target signal. We consider feature-grids with equirectangular and hierarchical equal area isolatitude pixelization structures that align with weather data and CMB data, respectively. We extensively verify the effectiveness of our HNeR-S for regression, super-resolution, temporal interpolation, and compression tasks.

cs.LG↗

arXiv2023

A Machine Learning Approach to Understanding the Physical Properties of Magnetic Flux Ropes in the Solar Wind at 1 AU

Interplanetary magnetic flux ropes (MFRs) are commonly observed structures in the solar wind, categorized as magnetic clouds (MCs) and small-scale MFRs (SMFRs) depending on whether they are associated with coronal mass ejections. We apply machine learning to systematically compare SMFRs, MCs, and ambient solar wind plasma properties. We construct a dataset of 3-minute averaged sequential data points of the solar wind's instantaneous bulk fluid plasma properties using about twenty years of measurements from \emph{Wind}. We label samples by the presence and type of MFRs containing them using a catalog based on Grad-Shafranov (GS) automated detection for SMFRs and NASA's catalog for MCs (with samples in neither labeled non-MFRs). We apply the random forest machine learning algorithm to find which categories can be more easily distinguished and by what features. MCs were distinguished from non-MFRs with an AUC of 94% and SMFRs with an AUC of 89% and had distinctive plasma properties. In contrast, while SMFRs were distinguished from non-MFRs with an AUC of 86%, this appears to rely solely on the $\langle B \rangle$ > 5 nT threshold applied by the GS catalog. The results indicate that SMFRs have virtually the same plasma properties as the ambient solar wind, unlike the distinct plasma regimes of MCs. We interpret our findings as additional evidence that most SMFRs at 1 au are generated within the solar wind, and furthermore, suggesting that they should be considered a salient feature of the solar wind's magnetic structure rather than transient events.

physics.space-ph↗

arXiv2023

A Closer Look at Small-Scale Magnetic Flux Ropes in the Solar Wind at 1 AU: Results from Improved Automated Detection

Small-scale interplanetary magnetic flux ropes (SMFRs) are similar to ICMEs in magnetic structure, but are smaller and do not exhibit ICME plasma signatures. We present a computationally efficient and GPU-powered version of the single-spacecraft automated SMFR detection algorithm based on the Grad-Shafranov (GS) technique. Our algorithm is capable of processing higher resolution data, eliminates selection bias caused by a fixed $\avg{B}$ threshold, has improved detection criteria demonstrated to have better results on an MHD simulation, and recovers full 2.5D cross sections using GS reconstruction. We used it to detect 512,152 SMFRs from 27 years (1996 to 2022) of 3-second cadence \emph{Wind} measurements. Our novel findings are: (1) the radial density of SMFRs at 1 au (${\sim}1$ per $\si{10^6\kilo\meter}$) and filling factor (${\sim}$35\%) are independent of solar activity, distance to the heliospheric current sheet (HCS), and solar wind plasma type, although the minority of SMFRs with diameters greater than ${\sim}$0.01 au have a strong solar activity dependence; (2) SMFR diameters follow a log-normal distribution that peaks below the resolved range ($\gtrsim 10^4$ km), although the filling factor is dominated by SMFRs between $10^5$ to $10^6$ km; (3) most SMFRs at 1 au have strong field-aligned flows like those from PSP measurements; (4) in terms of diameter $d$, SMFR poloidal flux $\propto d^{1.2}$, axial flux $\propto d^{2.0}$, average twist number $\propto d^{-0.8}$, current density $\propto d^{-0.8}$, and helicity $\propto d^{3.2}$. Implications for the origin of SMFRs and switchbacks are briefly discussed. The new algorithm and SMFR dataset are made freely available.

physics.space-ph↗

arXiv2023

Mesh Density Adaptation for Template-based Shape Reconstruction

In 3D shape reconstruction based on template mesh deformation, a regularization, such as smoothness energy, is employed to guide the reconstruction into a desirable direction. In this paper, we highlight an often overlooked property in the regularization: the vertex density in the mesh. Without careful control on the density, the reconstruction may suffer from under-sampling of vertices near shape details. We propose a novel mesh density adaptation method to resolve the under-sampling problem. Our mesh density adaptation energy increases the density of vertices near complex structures via deformation to help reconstruction of shape details. We demonstrate the usability and performance of mesh density adaptation with two tasks, inverse rendering and non-rigid surface registration. Our method produces more accurate reconstruction results compared to the cases without mesh density adaptation.

cs.GR↗

arXiv2023

LaplacianFusion: Detailed 3D Clothed-Human Body Reconstruction

We propose LaplacianFusion, a novel approach that reconstructs detailed and controllable 3D clothed-human body shapes from an input depth or 3D point cloud sequence. The key idea of our approach is to use Laplacian coordinates, well-known differential coordinates that have been used for mesh editing, for representing the local structures contained in the input scans, instead of implicit 3D functions or vertex displacements used previously. Our approach reconstructs a controllable base mesh using SMPL, and learns a surface function that predicts Laplacian coordinates representing surface details on the base mesh. For a given pose, we first build and subdivide a base mesh, which is a deformed SMPL template, and then estimate Laplacian coordinates for the mesh vertices using the surface function. The final reconstruction for the pose is obtained by integrating the estimated Laplacian coordinates as a whole. Experimental results show that our approach based on Laplacian coordinates successfully reconstructs more visually pleasing shape details than previous methods. The approach also enables various surface detail manipulations, such as detail transfer and enhancement.

cs.GR↗

arXiv2022

Improved White Dwarves Constraints on Inelastic Dark Matter and Left-Right Symmetric Models

WIMPs can be captured in compact stars such as white dwarves (WDs) leading to an increase in the star luminosity through their annihilation process. We show that when the WIMP interacts with the nuclear targets within the WD through inelastic scattering and its mass exceeds a few tens GeV the data on low-temperature large-mass WDs in the Messier 4 globular cluster can probe values of the mass splitting $δ\lesssim$ 40 MeV. Such value largely exceeds those ensuing from direct detection and from solar neutrino searches. We apply such improved constraint to the specific DM scenario of a self-conjugate bi-doublet in the Left-Right Symmetric Model (LRSM), where the standard $SU(2)_L$ group with coupling $g_L$ is extended by an additional $SU(2)_R$ with coupling $g_R$. We show that bounds from WDs significantly reduce the cosmologically viable parameter space of such scenario, in particular requiring $g_R>g_L$. For instance, for $g_R/g_L$ = 1.8 we find the two viable mass ranges 1.2 TeV $\lesssim m_χ\lesssim$ 3 TeV and 5 TeV $\lesssim m_χ\lesssim$ 10 TeV, when the charged $SU(2)_R$ gauge boson mass $M_{W_2}$ is lighter than $\simeq$ 12 TeV. We also discuss the ultraviolet completion of the LRSM model, when the latter is embedded in a Grand Unified Theory. We show that such low-energy parameter space and compatibility to proton-decay bounds require a non-trivial extension of the particle content of the minimal model. We provide a specific example where $M_{W_2}\lesssim$ 10 TeV is achieved by extending the LRSM at high energy with color triplets that are singlets under all other groups, and $g_R/g_L>$1 is obtained by introducing $SU(2)_L$ triplets with no $SU(2)_R$ counterparts, i.e. by breaking the symmetry between the multiplets of $SU(2)_L$ and $SU(2)_R$.

hep-ph↗