arXiv subjects

Jihye Kim

Publications and source records attributed to Jihye Kim.

12 recordsLinked to original sources

arXiv2026

Synchronization Pathways and Resilience in Power Grids

Ensuring a sustainable energy supply requires maintaining power-grid stability. Rotor dynamics are governed by the swing equation, which takes the form of a second-order Kuramoto model with a correlation between power and total coupling strength. Yet the microscopic mechanisms that nucleate and propagate synchronized clusters remain poorly understood. Using a minimal model motivated by empirical grid data and the \Adhoc potential method, we reveal two distinct seed cluster types and propagation pathways: a population-driven seed cluster at the center of the power distribution propagating to its tails by rotor accretion, and, for a symmetric distribution, coupling-driven seed clusters at its tails propagating inward to the center by cluster merger. These differences generate distinct order-parameter patterns, while inertia controls whether the seed clusters persist with distinct angular velocities. The same propagation pathways also govern recovery following external disturbances. We further confirm that the same selection--persistence rule holds in data-derived annealed representations of European power grids. Therefore, our results can inform strategies for sustaining stable power-grid operation. More generally, our pathway-based framework reframes synchronization by emphasizing the dynamics of cluster formation and recovery rather than relying solely on static criteria.

nlin.AO↗

arXiv2026

Physics-constrained machine learning for decoding multi-nanobubble configurations in graphene

Identifying multiple graphene nanobubbles from electronic spectra is challenging because their strain-induced features overlap. We develop a physics-constrained machine-learning framework that decodes nanobubble configurations from density-of-states (DOS) spectra. For spatially separated nanobubbles, previous full quantum-transport calculations established that the multi-bubble DOS is numerically equivalent to the normalized sum of the constituent single-bubble spectra. We encode this validated additive relation in a compact neural decomposition model. For each target spectrum, the basis coefficients are optimized independently, and the resulting weights directly identify the constituent geometries. The method accurately reconstructs configurations of increasing complexity and remains robust to repeated constituents, incomplete basis dictionaries, and simulated measurement noise. The framework provides an interpretable route for characterizing strain-engineered graphene nanostructures and may extend to other quantum materials with additive spectral responses.

cond-mat.mes-hall↗

arXiv2026

Right or Wrong, Models Comply: Directional Blindness in LLM Moral Judgment

As language models take integrated roles across many domains, the response of LLMs to user pushback becomes a critical alignment property. Yet many existing evaluations treat compliance as unidirectional, measuring whether models resist pressure but not whether they resist it selectively. We introduce Compliance Asymmetry (A = BCR/HCR), a bidirectional diagnostic that compares beneficial output change under helpful nudges with harmful change under misleading nudges. Across 9 models and 972,000 nudge-condition responses, we find that this selectivity differs in factual and moral judgments: models follow helpful nudges more than harmful ones on factual questions (A = 1.58), but follow both directions at nearly identical rates on moral questions (A = 1.04). This phenomenon persists across model families, capability levels, and nudging types. Interestingly, we also find that chain-of-thought prompting amplifies helpful and harmful compliance together, while identity-based prompting suppresses both by nearly identical margins. These results identify direction-blind moral compliance as a distinct failure mode in current LLMs and suggest that alignment should target directionally calibrated updating rather than lower compliance alone.

cs.CL↗

arXiv2026

MIRAGE: A Polarity-Flipping Encoding Subspace in LLM Agents

When LLM agents are coerced into covertly encoding sensitive data (Base64, ROT13, acrostic, synonym chains, and beyond), the resulting outputs evade output-side detection but the underlying computation does not. Across nine encoding families and eight models from five architecture families, that computation is supported by a shared low-dimensional encoding subspace in the residual stream. A logistic-regression probe trained on eight encoding families recovers the held-out ninth at AUC 0.975-1.000, reading the computation rather than surface features. The same direction exhibits a second mechanistic signature at the planning token, flipping polarity to activate positively when the model will simulate the encoding inline and negatively when it will outsource it to a tool call, distinguishing two execution strategies before the encoded text exists. We build MIRAGE (Model-Internal Readout of Agentic Generation Exfiltration), a two-channel real-time monitor exploiting both signals. On 126 agentic exfiltration scenarios, it reaches AUC = 0.918, substantially outperforming output-only detection (AUC = 0.518). Monitor performance is fundamentally a property of the host model's geometry: benign-encoding false-positive rate ranges from 0% on Qwen-7B to 100% on Phi-3.5, revealing that the probe faithfully reads whether a model's geometry separates covert from overt encoding. Across all tested adversarial budgets, every attack suppressing the subspace also destroyed encoding fidelity, reported as an empirical regularity on the evaluated budgets, not a structural impossibility claim.

cs.CL↗

arXiv2026

Optimality in group-driven social dynamics on hypergraphs

We explore the role of intrinsic structural properties of hypergraphs in governing group-driven social dynamics with social reinforcement. First, we analyze simplicial contagion dynamics on random hypergraphs in which the level of hyperedge nestedness is systematically controlled. By developing the facet-based approximate master equation (FAME) method, we demonstrate that hyperedge nestedness induces a non-monotonic change in the outbreak threshold for simplicial contagion, displaying the lowest threshold at an intermediate level of hyperedge nestedness due to competition between simple and higher-order contagion processes. Next, we formulate the group-driven voter model (GVM) and investigate the consensus time for the GVM on hypergraphs with N nodes. Focusing on a representative case of the GVM, we show that the consensus time scales logarithmically with the system size as A ln N, where the prefactor A displays the fastest consensus formation at an intermediate level of social reinforcement due to competition between group-constraint and nonlinearity factors. Taken together, our results highlight the importance of competing effects arising from higher-order interactions in shaping optimality in group-driven social dynamical processes.

physics.soc-ph↗

arXiv2026

Effect of higher-order interactions on noisy majority-rule dynamics with random group sizes

We study opinion dynamics with higher-order interactions, motivated by the fact that social influence often takes place in groups rather than only through pairwise contacts. We introduce a noisy majority-rule model on annealed hypergraphs with heterogeneous group sizes and investigate how the distribution of interaction sizes affects collective ordering and relaxation. Using analytical theory and Monte Carlo simulations, we show that group-size heterogeneity strongly shapes both the transition between ordered and mixed states and the associated time scales. In particular, broader and heavier-tailed distributions make ordering more robust by enhancing the effect of rare large-group events. They also modify the finite-size scaling of relaxation, producing a crossover from the standard logarithmic behavior to faster ordering in sufficiently broad ensembles. In the pure majority-rule limit, we further show that the exit probability near coexistence obeys a universal error-function scaling form controlled by a single structural parameter. Our results demonstrate that the full distribution of group sizes is a key determinant of nonequilibrium ordering in higher-order opinion dynamics.

physics.soc-ph↗

arXiv2025

Ordering-disordering dynamics of the $q$-voter model under random external bias

We investigate a variant of the two-state $q$-voter model in which agents update their states under a random external field (which points upward with probability $s$ and downward with probability $1-s$) with probability $p$ or adopt the unanimous opinion of $q$ randomly selected neighbors with probability $ 1-p$. Using mean-field analysis and Monte Carlo simulations, we identify an order-disorder transition at $p_c$ when $s=\tfrac{1}{2}$. Notably, in the regime of $p>p_c$, we estimate the time for systems to reach disordered state from consensus state and find the logarithmic scaling $T_{\text{dis}} \sim \mathcal{B}\ln N$, with $\mathcal{B} = 1/(2p)$ for $q = 1$, while for $q > 1$, $\mathcal{B}$ depends on both $p > p_c$ and $q$. We observe that disordering dynamics slow down significantly for nonlinear strengths $q$ between $2$ and $3$, independent of the probability $p$. On the other hand, when $s=0$ or $s=1$, the system is bound to reach consensus, with the consensus time scaling logarithmically with system size as $T_{\text{con}} \sim \mathcal{B}\ln N$, where $\mathcal{B} = 1/p$ for $q = 1$ and $\mathcal{B} = 1$ for $q > 1$. Furthermore, in the limit of $p = 0$, we derive a closed-form exit probability valid for arbitrary values of $q$ and demonstrate a finite-size scaling collapse. These results clarify how external cues and peer conformity jointly control ordering and disordering in binary opinion dynamics.

physics.soc-ph↗

arXiv2024

Competition between group interactions and nonlinearity in voter dynamics on hypergraphs

Social dynamics are often driven by both pairwise (i.e., dyadic) relationships and higher-order (i.e., polyadic) group relationships, which one can describe using hypergraphs. To gain insight into the impact of polyadic relationships on dynamical processes on networks, we formulate and study a polyadic voter process, which we call the group-driven voter model (GVM), that incorporates the effect of group interactions by nonlinear interactions that are subject to a group (i.e., hyperedge) constraint. By examining the competition between nonlinearity and group sizes, we show that the GVM achieves consensus faster than standard voter-model dynamics, with an optimal minimizing exit time. We substantiate this finding by using mean-field theory on annealed uniform hypergraphs with $N$ nodes, for which the exit time scales as ${\cal A}\ln N$, where the prefactor ${\cal A}$ depends both on the nonlinearity and on group-constraint factors. Our results reveal how competition between group interactions and nonlinearity shapes GVM dynamics. We thereby highlight the importance of such competing effects in complex systems with polyadic interactions.

physics.soc-ph↗

arXiv2023

Contagion dynamics on hypergraphs with nested hyperedges

In complex social systems encoded as hypergraphs, higher-order (i.e., group) interactions taking place among more than two individuals are represented by hyperedges. One of the higher-order correlation structures native to hypergraphs is the nestedness: Some hyperedges can be entirely contained (that is, nested) within another larger hyperedge, which itself can also be nested further in a hierarchical manner. Yet the effect of such hierarchical structure of hyperedges on the dynamics has remained unexplored. In this context, here we propose a random nested-hypergraph model with a tunable level of nestedness and investigate the effects of nestedness on a higher-order susceptible-infected-susceptible process. By developing an analytic framework called the facet approximation, we obtain the steady-state fraction of infected nodes on the random nested-hypergraph model more accurately than existing methods. Our results show that the hyperedge-nestedness affects the phase diagram significantly. Monte Carlo simulations support the analytical results.

physics.soc-ph↗

arXiv2023

CrossSplit: Mitigating Label Noise Memorization through Data Splitting

We approach the problem of improving robustness of deep learning algorithms in the presence of label noise. Building upon existing label correction and co-teaching methods, we propose a novel training procedure to mitigate the memorization of noisy labels, called CrossSplit, which uses a pair of neural networks trained on two disjoint parts of the labelled dataset. CrossSplit combines two main ingredients: (i) Cross-split label correction. The idea is that, since the model trained on one part of the data cannot memorize example-label pairs from the other part, the training labels presented to each network can be smoothly adjusted by using the predictions of its peer network; (ii) Cross-split semi-supervised training. A network trained on one part of the data also uses the unlabeled inputs of the other part. Extensive experiments on CIFAR-10, CIFAR-100, Tiny-ImageNet and mini-WebVision datasets demonstrate that our method can outperform the current state-of-the-art in a wide range of noise ratios.

cs.CV↗

arXiv2021

Quantum readout error mitigation via deep learning

Quantum computing devices are inevitably subject to errors. To leverage quantum technologies for computational benefits in practical applications, quantum algorithms and protocols must be implemented reliably under noise and imperfections. Since noise and imperfections limit the size of quantum circuits that can be realized on a quantum device, developing quantum error mitigation techniques that do not require extra qubits and gates is of critical importance. In this work, we present a deep learning-based protocol for reducing readout errors on quantum hardware. Our technique is based on training an artificial neural network with the measurement results obtained from experiments with simple quantum circuits consisting of singe-qubit gates only. With the neural network and deep learning, non-linear noise can be corrected, which is not possible with the existing linear inversion methods. The advantage of our method against the existing methods is demonstrated through quantum readout error mitigation experiments performed on IBM five-qubit quantum devices.

quant-ph↗

arXiv2019

Development of high-sensitivity chip calorimeters for cellular metabolic heat sensing

Cellular metabolic rate is a good indicator of the physiological state of cells and its changes, which can be measured by total heat flux accompanying metabolism. Chip calorimeters can provide label-free and high throughput measurements of cellular metabolic rate, however, lack of high power resolution and microfluidic sample handling capability has been preventing their wide applications. We report high-resolution chip calorimeters integrated with thin-film parylene microfluidics, which can reliably measure metabolic heat from mammalian cells with controlled stimuli. The molding and bonding technique allowed fast and reliable parylene microfluidic channel fabrications and highly sensitive vanadium oxide thermistor enabled temperature resolution as small as ~ 15 μK, which led to a three-orders-of-magnitude improvement in volume specific power resolutions. Measurements of metabolic heat were successfully demonstrated with adherent and nonadherent cells. We expect the chip calorimeter will provide a universal platform for fundamental cell-biology studies and biomedical applications including cell-based assay for drug discovery.

physics.bio-ph↗