arXiv subjects

Dong-Hee Kim

Publications and source records attributed to Dong-Hee Kim.

At least 19 recordsLinked to original sources

arXiv2026

Spectator-transition crosstalk in a spin-3/2 silicon-vacancy qudit in silicon carbide revealed by broadband Ramsey interferometry

Color-center spins in 4H-SiC offer a rare combination of wafer-scale materials maturity with long spin coherence and chip-level photonics, making them promising building blocks for scalable quantum technologies. In particular, the silicon vacancy hosts an S=3/2 ground state, a native qudit that enables compact encodings and subspace-selective control but also introduces spectator transitions: short, detuned pulses can coherently drive nonaddressed level pairs and create crosstalk. Here we use broadband Ramsey interferometry to reveal and quantify such spectator-transition crosstalk. Experimentally, the Ramsey Fourier spectra display multiple lines beyond the addressed single-quantum transition. Analytically, we map each line to a pairwise energy difference between qudit levels of the rotating-frame Hamiltonian and assign its weight via compact amplitudes set by the prepared state and the microwave pulse parameters, predicting a deterministic six-branch structure. Numerical time-domain propagation with the experimental sampling reproduces the detuning map, and the measured peak positions coincide with the analytic branch lines without frequency fitting. Together these results provide a practical, spectator-aware framework for multilevel control in the silicon-vacancy qudit. The approach offers clear guidance to suppress crosstalk or, conversely, to exploit spectator lines, e.g., as additional constraints for in situ pulse calibration and for phase-sensitive quantum state and process estimation.

quant-ph↗

arXiv2026

Grounding Confusable Movie Moments

Characters, locations, and actions recur throughout a movie, so many of its moments differ only in fine-grained details. Grounding such a moment in a full-length movie requires resolving that detail rather than matching the attributes it shares with other moments, yet existing temporal movie grounding benchmarks rarely test this. We introduce CoM$^2$-Bench, a benchmark of confusable query pairs, where the two queries in each pair share most attributes but differ in a single distinguishing detail within a full movie, posing a challenge for recent temporal grounding approaches. To locate where grounding breaks down, we analyze a retrieve-then-localize pipeline, in which a retriever first selects a short candidate chunk and a localizer then performs temporal grounding within it. Separating the two stages lets us attribute each failure to either retrieval or localization. We find that retrieval is the major bottleneck: state-of-the-art retrievers often select a moment's confusable counterpart, because their choices follow the content the two moments share rather than the detail that tells them apart. To address these failures, we propose a simple, training-free remedy, Agentic Window Search (AWS): an MLLM agent $\textit{Decomposes}$ the query into attribute-level sub-queries, then $\textit{Inspects}$ each candidate window against them and $\textit{Expands}$ the window when evidence is cut off at its edge. AWS works with existing retrievers and localizers without additional training, raising mIoU from 18.5 to 26.8.

cs.CV↗

arXiv2026

Beyond Visual Boundaries: Rethinking Scene Segmentation for Movie RAG

Understanding long-form video remains a fundamental challenge for multimodal large language models (MLLMs). Sparse frame sampling fails to capture fine-grained visual details, while dense sampling quickly exceeds context length limits. Retrieval-augmented generation (RAG) offers a promising middle ground by selectively retrieving relevant video segments for grounded generation, yet its effectiveness critically depends on the quality of the video segments used as retrieval units. In this paper, we investigate RAG for movie understanding, which demands story-level reasoning over characters, events, and narrative arcs spanning hours of content. Scene segmentation, a long-studied problem that partitions movies into semantically coherent units, is a natural candidate for defining such retrieval units. We reexamine whether existing methods actually serve this role through comprehensive evaluation on downstream movie understanding tasks, and find that they consistently fail to outperform naive uniform temporal chunking. Our audit of the most standard scene segmentation benchmarks reveals why: current annotations prioritize visually salient transitions over narrative event structure. Motivated by this mismatch, we introduce NarraScene, a narrative-centric scene segmentation dataset annotated with a three-level cognitive taxonomy spanning physical, character, and narrative change, where every valid boundary requires a narrative-level shift. When used as retrieval units, these narrative-grounded segments outperform uniform chunking on downstream movie understanding tasks, suggesting that the central challenge for scene segmentation in movie RAG is not detecting boundaries, but identifying the narrative event units that matter for movie understanding.

cs.CV↗

arXiv2026

Crossover and universality breaking in the dilute Baxter-Wu model

The critical behavior of the Baxter-Wu model belongs to the universality class of the four-state Potts model. While the introduction of annealed vacancies does not alter the criticality of the four-state Potts model, the dilute Baxter-Wu model has remained the subject of several competing scenarios. Here we investigate the phase diagram of the spin-$1$ Baxter-Wu model in the presence of a crystal field using transfer-matrix calculations and large-scale Monte Carlo simulations. Our results reveal a systematic evolution of the effective critical behavior with increasing crystal field, accompanied by increasingly strong finite-size corrections near the crossover to the first-order regime. Along the line of continuous transitions, the central charge remains close to $c=1$, while the scaling dimensions systematically deviate from the spin-$1/2$ limit as the crystal field increases, consistent with either continuously varying effective critical exponents or a slow crossover between competing critical behaviors. The first-order regime is independently characterized through multicanonical simulations, which confirm the expected finite-size scaling and interfacial behavior. Taken together, our results provide a unified picture of the dilute spin-$1$ Baxter-Wu model, substantially narrowing the range of possible scenarios for the crossover between continuous and first-order phase transitions.

cond-mat.stat-mech↗

arXiv2026

Diversity Over Frequency: Rethinking Tool Use in Visual Chain-of-Thought Agents

Visual agents employ external visual tools within visual chains of thought to incorporate fine-grained evidence. While prior work has mainly studied these tools in visual search tasks, their role in more complex visual reasoning remains underexplored. In this paper, we move beyond simple visual search tasks to investigate more challenging tasks, including 3D spatial reasoning and medical visual question answering, where agents must integrate tool-acquired local evidence with the global context. We identify a {tool-use collapse phenomenon: models progressively stop using tools while still achieving higher task accuracy. Moreover, we observe a clear asymmetry: (i) completely eliminating tool use degrades performance, whereas (ii) incentivizing tool use yields only marginal gains despite substantially increasing usage. We find that vanilla training and tool-use encouragement both reduce rollout diversity, explaining why higher tool use does not yield stronger reasoning performance. Motivated by these findings, we add an entropy regularization term to encourage diverse rollout exploration, achieving the best performance despite gradually declining tool usage. Overall, our findings suggest a training-time view of tools as scaffolding, where broader exploration over language generation and visual tool invocation improves reasoning despite tool-use collapse. Project page: https://scaffolded-exploration.github.io

cs.CV↗

arXiv2026

Cluster Dynamics Stay Fast-Until Tricriticality

Cluster Monte Carlo algorithms are widely regarded as the most effective route to overcoming critical slowing down in lattice spin systems. Whether this acceleration persists in the presence of vacancies and multicritical fluctuations, however, remains unresolved. We address this question through a systematic dynamic-scaling study of hybrid cluster-local update schemes in the two-dimensional Blume-Capel model, which exhibits a line of continuous Ising-like transitions terminating at a tricritical point. Along the entire critical line, hybrid dynamics retain the near-optimal efficiency of pure cluster updates despite the presence of annealed vacancies. Strikingly, this acceleration collapses precisely at tricriticality, where the dynamic critical exponent reverts to the local-update value. We trace this breakdown to the correlated percolation of vacancies, whose emergent system-spanning geometry obstructs nonlocal relaxation in the spin sector. Our results identify a fundamental geometric limitation of cluster acceleration at tricriticality and establish vacancy percolation as the mechanism controlling dynamic universality in hybrid Monte Carlo dynamics.

cond-mat.stat-mech↗

arXiv2026

Spatially Grounded Long-Horizon Task Planning in the Wild

Recent advances in robot manipulation increasingly leverage Vision-Language Models (VLMs) for high-level reasoning, such as decomposing task instructions into sequential action plans expressed in natural language that guide downstream low-level motor execution. However, current benchmarks do not assess whether these plans are spatially executable, particularly in specifying the exact spatial locations where the robot should interact to execute the plan, limiting evaluation of real-world manipulation capability. To bridge this gap, we define a novel task of grounded planning and introduce GroundedPlanBench, a newly curated benchmark for spatially grounded long-horizon action planning in the wild. GroundedPlanBench jointly evaluates hierarchical sub-action planning and spatial action grounding (where to act), enabling systematic assessment of whether generated sub-actions are spatially executable for robot manipulation. We further introduce Video-to-Spatially Grounded Planning (V2GP), an automated data generation framework that leverages real-world robot video demonstrations to improve spatially grounded long-horizon planning. Our evaluations reveal that spatially grounded long-horizon planning remains a major bottleneck for current VLMs. Our results demonstrate that V2GP provides a promising approach for improving both action planning and spatial grounding performance, validated on our benchmark as well as through real-world robot manipulation experiments, advancing progress toward spatially actionable planning.

cs.RO↗

arXiv2026

Superconductivity in the kagome Hubbard model under the flat-band-preserving disorder

We investigate the disordered flat-band superconductivity within the attractive Hubbard model on the kagome lattice by contrasting the flat-band-preserving disorder [Phys. Rev. B 98, 235109 (2018)] with the random hopping disorder that breaks the flat-band degeneracy. Through Bogoliubov-de Gennes mean-field calculations, we find that the superfluid weight is much more robust under the flat-band-preserving disorder, while the system eventually undergoes a transition to an insulator as disorder becomes strong enough. The almost linear interaction-dependence of the superfluid weight in the weak coupling limit found with the flat-band-preserving disorder confirms the persistent flat-band signature, whereas the exponential behavior of a dispersive-band character arises with the random hopping counterpart. In addition, in the exact diagonalization of the one-particle density matrix, we identify an occupation spectrum structure attributed to the flat-band states, demonstrating the connection between the resilient flat band and the enhanced robustness of superconductivity.

cond-mat.supr-con↗

arXiv2025

Transfer-matrix approach to the Blume-Capel model on the triangular lattice

We investigate the spin-$1$ Blume-Capel model on an infinite strip of the triangular lattice using the transfer-matrix method combined with a sparse-matrix factorization technique. Through finite-size scaling analysis of numerically exact spectra for strip widths up to $L = 19$, we accurately locate the tricritical point improving upon recent Monte Carlo estimates. In the first-order regime, we observe exponential scaling of the spectral gap, reflecting the linear growth of interfacial tension as the temperature decreases below the tricritical point. Finally, we validate our tricritical point estimate through precise agreement with conformal field theory predictions for the tricritical Ising universality class. Our results underscore the continued utility of the transfer-matrix approach for studying phase transitions in complex lattice models.

cond-mat.stat-mech↗

arXiv2025

SynRES: Towards Referring Expression Segmentation in the Wild via Synthetic Data

Despite the advances in Referring Expression Segmentation (RES) benchmarks, their evaluation protocols remain constrained, primarily focusing on either single targets with short queries (containing minimal attributes) or multiple targets from distinctly different queries on a single domain. This limitation significantly hinders the assessment of more complex reasoning capabilities in RES models. We introduce WildRES, a novel benchmark that incorporates long queries with diverse attributes and non-distinctive queries for multiple targets. This benchmark spans diverse application domains, including autonomous driving environments and robotic manipulation scenarios, thus enabling more rigorous evaluation of complex reasoning capabilities in real-world settings. Our analysis reveals that current RES models demonstrate substantial performance deterioration when evaluated on WildRES. To address this challenge, we introduce SynRES, an automated pipeline generating densely paired compositional synthetic training data through three innovations: (1) a dense caption-driven synthesis for attribute-rich image-mask-expression triplets, (2) reliable semantic alignment mechanisms rectifying caption-pseudo mask inconsistencies via Image-Text Aligned Grouping, and (3) domain-aware augmentations incorporating mosaic composition and superclass replacement to emphasize generalization ability and distinguishing attributes over object categories. Experimental results demonstrate that models trained with SynRES achieve state-of-the-art performance, improving gIoU by 2.0% on WildRES-ID and 3.8% on WildRES-DS. Code and datasets are available at https://github.com/UTLLab/SynRES.

cs.LG↗

arXiv2025

Tricriticality and finite-size scaling in the triangular Blume-Capel ferromagnet

We report on numerical simulations of the two-dimensional spin-$1$ Blume-Capel ferromagnet embedded in a triangular lattice. Utilizing a range of Monte Carlo and finite-size scaling techniques, we explore several critical aspects along the crystal field--temperature ($Δ, T$) transition line. Wang-Landau simulations measuring the joint density of states in combination with the method of field mixing allow us to probe the phase coexistence curve in high resolution, determining the tricritical point $(Δ_{\rm t}, T_{\rm t})$ with improved accuracy and verifying the tricritical exponents. Extensive multicanonical simulations identifying transition points across the phase diagram characterize the Ising universality class for $Δ< Δ_{\rm t}$ with precise determination of thermal and magnetic critical exponents expected in the second-order regime. On the other hand, for $Δ> Δ_{\rm t}$, a finite-size scaling analysis is dedicated to revealing the first-order signature in the surface tension that linearly increases upon lowering the temperature deeper into the first-order transition regime. Finally, a comprehensive picture of the phase diagram for the model is presented, collecting transition points obtained from the combined numerical approach in this study and previous estimates in the literature.

cond-mat.stat-mech↗

arXiv2024

CNG-SFDA:Clean-and-Noisy Region Guided Online-Offline Source-Free Domain Adaptation

Domain shift occurs when training (source) and test (target) data diverge in their distribution. Source-Free Domain Adaptation (SFDA) addresses this domain shift problem, aiming to adopt a trained model on the source domain to the target domain in a scenario where only a well-trained source model and unlabeled target data are available. In this scenario, handling false labels in the target domain is crucial because they negatively impact the model performance. To deal with this problem, we propose to update cluster prototypes (i.e., centroid of each sample cluster) and their structure in the target domain formulated by the source model in online manners. In the feature space, samples in different regions have different pseudo-label distribution characteristics affected by the cluster prototypes, and we adopt distinct training strategies for these samples by defining clean and noisy regions: we selectively train the target with clean pseudo-labels in the clean region, whereas we introduce mix-up inputs representing intermediate features between clean and noisy regions to increase the compactness of the cluster. We conducted extensive experiments on multiple datasets in online/offline SFDA settings, whose results demonstrate that our method, CNG-SFDA, achieves state-of-the-art for most cases. Code is available at https://github.com/hyeonwoocho7/CNG-SFDA.

cs.CV↗

arXiv2024

Joint-Embedding Predictive Architecture for Self-Supervised Learning of Mask Classification Architecture

In this work, we introduce Mask-JEPA, a self-supervised learning framework tailored for mask classification architectures (MCA), to overcome the traditional constraints associated with training segmentation models. Mask-JEPA combines a Joint Embedding Predictive Architecture with MCA to adeptly capture intricate semantics and precise object boundaries. Our approach addresses two critical challenges in self-supervised learning: 1) extracting comprehensive representations for universal image segmentation from a pixel decoder, and 2) effectively training the transformer decoder. The use of the transformer decoder as a predictor within the JEPA framework allows proficient training in universal image segmentation tasks. Through rigorous evaluations on datasets such as ADE20K, Cityscapes and COCO, Mask-JEPA demonstrates not only competitive results but also exceptional adaptability and robustness across various training scenarios. The architecture-agnostic nature of Mask-JEPA further underscores its versatility, allowing seamless adaptation to various mask classification family.

cs.CV↗

arXiv2024

Neural-network quantum state study of the long-range antiferromagnetic Ising chain

We investigate quantum phase transitions in the transverse field Ising chain with algebraically decaying long-range (LR) antiferromagnetic interactions using the variational Monte Carlo method with the restricted Boltzmann machine employed as a trial wave function ansatz. First, we measure the critical exponents and the central charge through the finite-size scaling analysis, verifying the contrasting observations in the previous tensor network studies. The correlation function exponent and the central charge deviate from the short-range (SR) Ising values at a small decay exponent $α_\mathrm{LR}$, while the other critical exponents examined are very close to the SR Ising exponents regardless of $α_\mathrm{LR}$ examined. However, in the further test of the critical Binder ratio, we find that the universal ratio of the SR limit does not hold for $α_\mathrm{LR} < 2$, implying a deviation in the criticality. On the other hand, we find evidence of the conformal invariance breakdown in the conformal field theory (CFT) test of the correlation function. The deviation from the CFT description becomes more pronounced as $α_\mathrm{LR}$ decreases, although a precise breakdown threshold is yet to be determined.

cond-mat.stat-mech↗

arXiv2023

Slowest and Fastest Information Scrambling in the Strongly Disordered XXZ Model

We present a perturbation method to compute the out-of-time-ordered correlator in the strongly disordered Heisenberg XXZ model in the deep many-body localized regime. We characterize the discrete structure of the information propagation across the eigenstates, revealing a highly structured light cone confined by the strictly logarithmic upper and lower bounds representing the slowest and fastest scrambling available in this system. We explain these bounds by deriving the closed-form expression of the effective interaction for the slowest scrambling and by constructing the effective model of a half length for the fastest scrambling. We extend our lowest-order perturbation formulations to the higher dimensions, proposing that the logarithmic upper and lower light cones may persist in a finite two-dimensional system in the limit of strong disorder and weak hopping.

cond-mat.dis-nn↗

arXiv2022

Tensor network calculation of the logarithmic correction exponent in the XY model

We study the logarithmic correction to the scaling of the first Lee-Yang (LY) zero in the classical $XY$ model on square lattices by using tensor renormalization group methods. In comparing the higher-order tensor renormalization group (HOTRG) and the loop-optimized tensor network renormalization (LoopTNR), we find that the entanglement filtering in LoopTNR is crucial to gaining high accuracy for the characterization of the logarithmic correction, while HOTRG still proposes approximate bounds for the zero location associated with two different bond-merging algorithms of the higher-order singular value decomposition and the oblique projectors. Using the LoopTNR data computed up to the system size of $L=1024$ in the $L \times L$ lattices, we estimate the logarithmic correction exponent $r = -0.0643(9)$ from the extrapolation of the finite-size effective exponent, which is comparable to the renormalization group prediction of $r = -1/16$.

cond-mat.stat-mech↗

arXiv2021

Emergence of a finite-size-scaling function in the supervised learning of the Ising phase transition

We investigate the connection between the supervised learning of the binary phase classification in the ferromagnetic Ising model and the standard finite-size-scaling theory of the second-order phase transition. Proposing a minimal one-free-parameter neural network model, we analytically formulate the supervised learning problem for the canonical ensemble being used as a training data set. We show that just one free parameter is capable enough to describe the data-driven emergence of the universal finite-size-scaling function in the network output that is observed in a large neural network, theoretically validating its critical point prediction for unseen test data from different underlying lattices yet in the same universality class of the Ising criticality. We also numerically demonstrate the interpretation with the proposed one-parameter model by providing an example of finding a critical point with the learning of the Landau mean-field free energy being applied to the real data set from the uncorrelated random scale-free graph with a large degree exponent.

cond-mat.stat-mech↗

arXiv2020

A Confidence-Calibrated MOBA Game Winner Predictor

In this paper, we propose a confidence-calibration method for predicting the winner of a famous multiplayer online battle arena (MOBA) game, League of Legends. In MOBA games, the dataset may contain a large amount of input-dependent noise; not all of such noise is observable. Hence, it is desirable to attempt a confidence-calibrated prediction. Unfortunately, most existing confidence calibration methods are pertaining to image and document classification tasks where consideration on uncertainty is not crucial. In this paper, we propose a novel calibration method that takes data uncertainty into consideration. The proposed method achieves an outstanding expected calibration error (ECE) (0.57%) mainly owing to data uncertainty consideration, compared to a conventional temperature scaling method of which ECE value is 1.11%.

cs.LG↗