arXiv subjects

Hyunjin Kim

Publications and source records attributed to Hyunjin Kim.

At least 19 recordsLinked to original sources

arXiv2026

Vitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers

We propose the first compression approach for image-to-shape Diffusion Transformers (DiTs) that substantially reduces model size while preserving geometric fidelity. Despite remarkable progress in 3D shape generation, large DiT-based models remain computationally prohibitive in resource-constrained settings. Furthermore, it is difficult to directly transfer existing diffusion model compression strategies developed for different domains to 3D generation, and prior 3D efficiency approaches focus primarily on inference speed rather than backbone compression. To address this limitation, we build a geometry-aware compression framework tailored to image-to-shape DiTs. Guided by the observation that 3D DiT layers exhibit non-uniform importance for geometry synthesis, we introduce a vitality-guided framework integrating structured pruning, adaptive quantization, and targeted fine-tuning. Our method achieves up to 66% model-size reduction across state-of-the-art image-to-3D models while maintaining synthesis fidelity comparable to full-sized counterparts. This highlights the potential of our framework as a plug-and-play solution for efficient 3D shape generation across diverse models.

cs.CV↗

arXiv2026

EgoPhys: Learning Generalizable Physics Models of Deformable Objects from Egocentric Video

Humans naturally understand object physics through everyday interactions, but faithfully predicting complex deformable dynamics, such as elastic materials and fabrics, remains a major challenge for computer vision and robotics. We present EgoPhys, a framework that constructs deformable physical digital twins from egocentric RGB-only video using generalizable priors. EgoPhys overcomes the limitations of existing methods to enable controllable deformable digital twin generation from egocentric videos by distilling per-object inverse-physics solutions into a compact codebook, enabling prediction of dense spring stiffness fields for unseen objects without per-spring test-time optimization. Trained with generalizable priors from diverse egocentric interactions, EgoPhys outperforms baselines in reconstruction, future prediction, and zero-shot generalization. To support training and evaluation, we curate an egocentric interaction dataset covering diverse deformable objects, scenes, and manipulation styles. We deploy EgoPhys on a real xArm6 robot, demonstrating that a digital twin initialized from a single egocentric human play video can serve as an internal world representation to aid in deformable-object planning, highlighting egocentric RGB observations as a scalable path toward real-to-sim pipelines.

cs.CV↗

arXiv2026

Accelerating Diffusion via Hybrid Data-Pipeline Parallelism Based on Conditional Guidance Scheduling

Diffusion models have achieved remarkable progress in high-fidelity image, video, and audio generation, yet inference remains computationally expensive. Nevertheless, current diffusion acceleration methods based on distributed parallelism suffer from noticeable generation artifacts and fail to achieve substantial acceleration proportional to the number of GPUs. Therefore, we propose a hybrid parallelism framework that combines a novel data parallel strategy, condition-based partitioning, with an optimal pipeline scheduling method, adaptive parallelism switching, to reduce generation latency and achieve high generation quality in conditional diffusion models. The key ideas are to (i) leverage the conditional and unconditional denoising paths as a new data-partitioning perspective and (ii) adaptively enable optimal pipeline parallelism according to the denoising discrepancy between these two paths. Our framework achieves $2.31\times$ and $2.07\times$ latency reductions on SDXL and SD3, respectively, using two NVIDIA RTX~3090 GPUs, while preserving image quality. This result confirms the generality of our approach across U-Net-based diffusion models and DiT-based flow-matching architectures. Our approach also outperforms existing methods in acceleration under high-resolution synthesis settings. Code is available at https://github.com/kaist-dmlab/Hybridiff.

cs.CV↗

arXiv2026

TimelyFreeze: Adaptive Parameter Freezing Mechanism for Pipeline Parallelism

Pipeline parallelism enables training models that exceed single-device memory, but practical throughput remains limited by pipeline bubbles. Although parameter freezing can improve training throughput by adaptively skipping backward computation, existing methods often over-freeze parameters, resulting in unnecessary accuracy degradation. To address this issue, we propose TimelyFreeze, which models the pipeline schedule as a directed acyclic graph and solves a linear program to compute optimal freeze ratios that minimize batch execution time under accuracy constraints. Experiments show that TimelyFreeze achieves up to 40% training throughput improvement on LLaMA-8B with comparable accuracy. Overall, it enables faster large-scale model training without compromising convergence and generalizes across diverse pipeline-parallel settings.

cs.DC↗

arXiv2025

GOATex: Geometry & Occlusion-Aware Texturing

We present GOATex, a diffusion-based method for 3D mesh texturing that generates high-quality textures for both exterior and interior surfaces. While existing methods perform well on visible regions, they inherently lack mechanisms to handle occluded interiors, resulting in incomplete textures and visible seams. To address this, we introduce an occlusion-aware texturing framework based on the concept of hit levels, which quantify the relative depth of mesh faces via multi-view ray casting. This allows us to partition mesh faces into ordered visibility layers, from outermost to innermost. We then apply a two-stage visibility control strategy that progressively reveals interior regions with structural coherence, followed by texturing each layer using a pretrained diffusion model. To seamlessly merge textures obtained across layers, we propose a soft UV-space blending technique that weighs each texture's contribution based on view-dependent visibility confidence. Empirical results demonstrate that GOATex consistently outperforms existing methods, producing seamless, high-fidelity textures across both visible and occluded surfaces. Unlike prior works, GOATex operates entirely without costly fine-tuning of a pretrained diffusion model and allows separate prompting for exterior and interior mesh regions, enabling fine-grained control over layered appearances. For more qualitative results, please visit our project page: https://goatex3d.github.io/.

cs.CV↗

arXiv2025

Investigating the Degradation of LATP Solid Electrolyte in High Alkaline Li-$O_2$ Batteries

In this study, we address the challenge of electrolyte degradation in all-solid-state humidified Li-$O_2$ batteries, which offer high theoretical energy density and potential cost advantages over conventional lithium-ion batteries. Combining STEM-EELS, and XPS characterizations with DFT calculations, we reveal the leaching of $(PO_4)^{3+}$ and $Al^{3+}$ ions from the $Li_{1.3}Al_{0.3}Ti_{1.7}(PO_4)_3$ (LATP) solid electrolyte upon battery discharge, caused by the highly alkaline environment. Upon charging, the leached ions precipitate as $Li_3PO_4$ and $AlPO_4$, which accumulate on the LATP surface and contribute to battery degradation. A Ti-rich layer is observed at the surface after a few cycles due to depletion of other cations. Our findings suggest that the degradation products are formed through repeated dissolution and precipitation in the discharge-charge cycles. Furthermore, our results indicate that the Ti-rich layer on the LATP surface can potentially reduce parasitic reactions. Our study provides mechanistic understanding of LATP solid electrolyte degradation in humidified Li-$O_2$ cell, paving the way for designing more durable and efficient Li-$O_2$ batteries.

cond-mat.mtrl-sci↗

arXiv2025

Metropolis-Hastings Sampling for 3D Gaussian Reconstruction

We propose an adaptive sampling framework for 3D Gaussian Splatting (3DGS) that leverages comprehensive multi-view photometric error signals within a unified Metropolis-Hastings approach. Vanilla 3DGS heavily relies on heuristic-based density-control mechanisms (e.g., cloning, splitting, and pruning), which can lead to redundant computations or premature removal of beneficial Gaussians. Our framework overcomes these limitations by reformulating densification and pruning as a probabilistic sampling process, dynamically inserting and relocating Gaussians based on aggregated multi-view errors and opacity scores. Guided by Bayesian acceptance tests derived from these error-based importance scores, our method substantially reduces reliance on heuristics, offers greater flexibility, and adaptively infers Gaussian distributions without requiring predefined scene complexity. Experiments on benchmark datasets, including Mip-NeRF360, Tanks and Temples and Deep Blending, show that our approach reduces the number of Gaussians needed, achieving faster convergence while matching or modestly surpassing the view-synthesis quality of state-of-the-art models.

cs.CV↗

arXiv2025

Resolving Intervalley Gaps and Many-Body Resonances in Moir\'e Superconductor

Magic-angle twisted multilayer graphene stands out as a highly tunable class of moir\'e materials that exhibit strong electronic correlations and robust superconductivity. However, understanding the relations between the low-temperature superconducting phase and the preceding correlated phases established at higher temperatures remains a challenge. Here, we employ scanning tunneling microscopy and spectroscopy to track the formation sequence of correlated phases established by the interplay of dynamic correlations, intervalley coherence, and superconductivity in magic-angle twisted trilayer graphene (MATTG). We discover the existence of two well-resolved gaps pinned at the Fermi level within the superconducting doping range. While the outer gap, previously associated with pseudogap phase, persists at high temperatures and magnetic fields, the newly revealed inner gap is more fragile in line with superconductivity MATTG transport experiments. Andreev reflection spectroscopy taken at the same location confirms a clear trend that closely follows the doping behaviour of the inner gap, and not the outer one. Moreover, spectroscopy taken at nanoscale domain boundaries further corroborates the contrasting behavior of the two gaps, with the inner gap remaining resilient to structural variations, as expected from the finite superconducting coherence length. By comparing our findings with recent topological heavy-fermion models, we identify that the outer gap originates from the splitting of the Abrikosov-Suhl-Kondo resonance due to the breaking of the valley symmetry arising from correlation-driven effects. Our results suggest an intricate but tractable hierarchy of correlated phases in twisted multilayer graphene.

cond-mat.supr-con↗

arXiv2025

Visual Acoustic Fields

Objects produce different sounds when hit, and humans can intuitively infer how an object might sound based on its appearance and material properties. Inspired by this intuition, we propose Visual Acoustic Fields, a framework that bridges hitting sounds and visual signals within a 3D space using 3D Gaussian Splatting (3DGS). Our approach features two key modules: sound generation and sound localization. The sound generation module leverages a conditional diffusion model, which takes multiscale features rendered from a feature-augmented 3DGS to generate realistic hitting sounds. Meanwhile, the sound localization module enables querying the 3D scene, represented by the feature-augmented 3DGS, to localize hitting positions based on the sound sources. To support this framework, we introduce a novel pipeline for collecting scene-level visual-sound sample pairs, achieving alignment between captured images, impact locations, and corresponding sounds. To the best of our knowledge, this is the first dataset to connect visual and acoustic signals in a 3D context. Extensive experiments on our dataset demonstrate the effectiveness of Visual Acoustic Fields in generating plausible impact sounds and accurately localizing impact sources. Our project page is at https://yuelei0428.github.io/projects/Visual-Acoustic-Fields/.

cs.CV↗

arXiv2024

Talking with Tables for Better LLM Factual Data Interactions

Large Language Models (LLMs) often struggle with requests related to information retrieval and data manipulation that frequently arise in real-world scenarios under multiple conditions. In this paper, we demonstrate that leveraging tabular structures in LLM interactions, is more effective than utilizing other structures for handling prevalent requests that operate over factual data. Through comprehensive evaluations across various scenarios and request types, we show that providing tabular structures yields a 40.29\% average performance gain along with better robustness and token efficiency. Through attention-value analysis, we discover that tables help LLMs better locate relevant information, explaining these improvements. Beyond tables and text, we evaluate whether (1) blending structuredness within text, such as providing templates or fixing the order of attributes, and (2) other representative structures, such as knowledge graphs and JSON are helpful. We observe that utilizing tables offers the best balance between efficiency and effectiveness. The method remains robust to task complexity and adapts to unstructured sources through text-to-table conversion. Overall, we highlight the untapped potential of tabular representations for future LLM applications.

cs.AI↗

arXiv2024

Artificial Intelligence and Strategic Decision-Making: Evidence from Entrepreneurs and Investors

This paper explores how artificial intelligence (AI) may impact the strategic decision-making (SDM) process in firms. We illustrate how AI could augment existing SDM tools and provide empirical evidence from a leading accelerator program and a startup competition that current Large Language Models (LLMs) can generate and evaluate strategies at a level comparable to entrepreneurs and investors. We then examine implications for key cognitive processes underlying SDM -- search, representation, and aggregation. Our analysis suggests AI has the potential to enhance the speed, quality, and scale of strategic analysis, while also enabling new approaches like virtual strategy simulations. However, the ultimate impact on firm performance will depend on competitive dynamics as AI capabilities progress. We propose a framework connecting AI use in SDM to firm outcomes and discuss how AI may reshape sources of competitive advantage. We conclude by considering how AI could both support and challenge core tenets of the theory-based view of strategy. Overall, our work maps out an emerging research frontier at the intersection of AI and strategy.

econ.GN↗

arXiv2024

PartSTAD: 2D-to-3D Part Segmentation Task Adaptation

We introduce PartSTAD, a method designed for the task adaptation of 2D-to-3D segmentation lifting. Recent studies have highlighted the advantages of utilizing 2D segmentation models to achieve high-quality 3D segmentation through few-shot adaptation. However, previous approaches have focused on adapting 2D segmentation models for domain shift to rendered images and synthetic text descriptions, rather than optimizing the model specifically for 3D segmentation. Our proposed task adaptation method finetunes a 2D bounding box prediction model with an objective function for 3D segmentation. We introduce weights for 2D bounding boxes for adaptive merging and learn the weights using a small additional neural network. Additionally, we incorporate SAM, a foreground segmentation model on a bounding box, to improve the boundaries of 2D segments and consequently those of 3D segmentation. Our experiments on the PartNet-Mobility dataset show significant improvements with our task adaptation approach, achieving a 7.0%p increase in mIoU and a 5.2%p improvement in mAP@50 for semantic and instance segmentation compared to the SotA few-shot 3D segmentation model.

cs.CV↗

arXiv2024

Observation of Cooper-pair density modulation state

Superconducting states that break space-group symmetries of the underlying crystal can exhibit nontrivial spatial modulation of the order parameter. Previously, such remarkable states were intimately associated with the breaking of translational symmetry, giving rise to the density-wave orders, with wavelengths spanning several unit cells. However, a related basic concept has been long overlooked: when only intra-unit-cell symmetries of the space group are broken, the superconducting states can display a distinct type of nontrivial modulation preserving long-range lattice translation. Here, we refer to this new concept as the pair density modulation (PDM), and report the first observation of a PDM state in exfoliated thin flakes of iron-based superconductor FeTe$_{\text{0.55}}$Se$_{\text{0.45}}$. Using scanning tunneling microscopy, we discover robust superconducting gap modulation with the wavelength corresponding to the lattice periodicity and the amplitude exceeding 30% of the gap average. Importantly, we find that the observed modulation originates from the large difference in superconducting gaps on the two nominally equivalent iron sublattices. The experimental findings, backed up by model calculations, suggest that in contrast to the density-wave orders, the PDM state is driven by the interplay of sublattice symmetry breaking and a peculiar nematic distortion specific to the thin flakes. Our results establish new frontiers for exploring the intertwined orders in strong-correlated electronic systems and open a new chapter for iron-based superconductors.

cond-mat.supr-con↗

arXiv2024

Can Separators Improve Chain-of-Thought Prompting?

Chain-of-thought (CoT) prompting is a simple and effective method for improving the reasoning capabilities of Large Language Models (LLMs). The basic idea of CoT is to let LLMs break down their thought processes step-by-step by putting exemplars in the input prompt. However, the densely structured prompt exemplars of CoT may cause the cognitive overload of LLMs. Inspired by human cognition, we introduce COT-SEP, a method that strategically employs separators at the end of each exemplar in CoT prompting. These separators are designed to help the LLMs understand their thought processes better while reasoning. Interestingly, it turns out that COT-SEP significantly improves the LLMs' performances on complex reasoning tasks (e.g., GSM8K, AQuA, CSQA), compared with the vanilla CoT, which does not use separators. We also study the effects of the type and the location of separators tested on multiple LLMs, including GPT-3.5-Turbo, GPT-4, and LLaMA-2 7B.

cs.CL↗

arXiv2023

Toward Reinforcement Learning-based Rectilinear Macro Placement Under Human Constraints

Macro placement is a critical phase in chip design, which becomes more intricate when involving general rectilinear macros and layout areas. Furthermore, macro placement that incorporates human-like constraints, such as design hierarchy and peripheral bias, has the potential to significantly reduce the amount of additional manual labor required from designers. This study proposes a methodology that leverages an approach suggested by Google's Circuit Training (G-CT) to provide a learning-based macro placer that not only supports placing rectilinear cases, but also adheres to crucial human-like design principles. Our experimental results demonstrate the effectiveness of our framework in achieving power-performance-area (PPA) metrics and in obtaining placements of high quality, comparable to those produced with human intervention. Additionally, our methodology shows potential as a generalized model to address diverse macro shapes and layout areas.

cs.LG↗

arXiv2023

SyncDiffusion: Coherent Montage via Synchronized Joint Diffusions

The remarkable capabilities of pretrained image diffusion models have been utilized not only for generating fixed-size images but also for creating panoramas. However, naive stitching of multiple images often results in visible seams. Recent techniques have attempted to address this issue by performing joint diffusions in multiple windows and averaging latent features in overlapping regions. However, these approaches, which focus on seamless montage generation, often yield incoherent outputs by blending different scenes within a single image. To overcome this limitation, we propose SyncDiffusion, a plug-and-play module that synchronizes multiple diffusions through gradient descent from a perceptual similarity loss. Specifically, we compute the gradient of the perceptual loss using the predicted denoised images at each denoising step, providing meaningful guidance for achieving coherent montages. Our experimental results demonstrate that our method produces significantly more coherent outputs compared to previous methods (66.35% vs. 33.65% in our user study) while still maintaining fidelity (as assessed by GIQA) and compatibility with the input prompt (as measured by CLIP score). We further demonstrate the versatility of our method across three plug-and-play applications: layout-guided image generation, conditional image generation and 360-degree panorama generation. Our project page is at https://syncdiffusion.github.io.

cs.CV↗

arXiv2023

ExMobileViT: Lightweight Classifier Extension for Mobile Vision Transformer

The paper proposes an efficient structure for enhancing the performance of mobile-friendly vision transformer with small computational overhead. The vision transformer (ViT) is very attractive in that it reaches outperforming results in image classification, compared to conventional convolutional neural networks (CNNs). Due to its need of high computational resources, MobileNet-based ViT models such as MobileViT-S have been developed. However, their performance cannot reach the original ViT model. The proposed structure relieves the above weakness by storing the information from early attention stages and reusing it in the final classifier. This paper is motivated by the idea that the data itself from early attention stages can have important meaning for the final classification. In order to reuse the early information in attention stages, the average pooling results of various scaled features from early attention stages are used to expand channels in the fully-connected layer of the final classifier. It is expected that the inductive bias introduced by the averaged features can enhance the final performance. Because the proposed structure only needs the average pooling of features from the attention stages and channel expansions in the final classifier, its computational and storage overheads are very small, keeping the benefits of low-cost MobileNet-based ViT (MobileViT). Compared with the original MobileViTs on the ImageNet dataset, the proposed ExMobileViT has noticeable accuracy enhancements, having only about 5% additional parameters.

cs.CV↗

arXiv2023

Imaging inter-valley coherent order in magic-angle twisted trilayer graphene

Magic-angle twisted trilayer graphene (MATTG) exhibits a range of strongly correlated electronic phases that spontaneously break its underlying symmetries. The microscopic nature of these phases and their residual symmetries stands as a key outstanding puzzle whose resolution promises to shed light on the origin of superconductivity in twisted materials. Here we investigate correlated phases of MATTG using scanning tunneling microscopy and identify striking signatures of interaction-driven spatial symmetry breaking. In low-strain samples, over a filling range of about 2-3 electrons or holes per moiré unit cell, we observe atomic-scale reconstruction of the graphene lattice that accompanies a correlated gap in the tunneling spectrum. This short-scale restructuring appears as a Kekulé supercell -- implying spontaneous inter-valley coherence between electrons -- and persists in a wide range of magnetic fields and temperatures that coincide with the development of the gap. Large-scale maps covering several moiré unit cells further reveal a slow evolution of the Kekulé pattern, indicating that atomic-scale reconstruction coexists with translation symmetry breaking at the much longer moiré scale. We employ auto-correlation and Fourier analyses to extract the intrinsic periodicity of these phases and find that they are consistent with the theoretically proposed incommensurate Kekulé spiral order. Moreover, we find that the wavelength characterizing moiré-scale modulations monotonically decreases with hole doping away from half-filling of the bands and depends only weakly on the magnetic field. Our results provide essential insights into the nature of MATTG correlated phases in the presence of strain and imply that superconductivity emerges from an inter-valley coherent parent state.

cond-mat.str-el↗