Searcharxiv⌕ Search

arXiv subjects

Jihye Kang

Publications and source records attributed to Jihye Kang.

7 recordsLinked to original sources

StyleComposer: Training-Free Multi-Reference Style Composition

The color, texture, and structure of an image can come from different references, but reference-guided generators transfer each reference as a single style signal. We find that in a diffusion transformer, every pathway that brings in a reference transfers multiple attributes, yet each attribute has one representation in which it is most directly controllable. Guided by this finding, we design StyleComposer to route color through the chromatic coordinates of the latent, texture through the attention keys and values, and structure through the queries, and to limit what else each route admits from its reference. Without training or inversion, StyleComposer follows the references and the prompt together more closely than stylization methods and multi-image generators, and it sets the strength of each attribute separately.

cs.CV↗

How Well Is My Artwork Protected? Predicting and Tailoring Protection against Style Mimicry by Personalization

Protection tools add an imperceptible perturbation to each artwork so that a model personalized on it fails to imitate its style, and they apply one recipe, a protection method at a fixed strength, to every artwork. We find that the same recipe protects artworks very differently, well beyond the noise of personalization seeds, and that cheap signals predict this protection before any personalization. Because protection rises and falls together across recipes, our method, Patina, predicts the protection of every recipe from one short trial protection and gives each artwork the weakest recipe whose predicted protection reaches a chosen level. At the same level of visibility, Patina lets more artworks pass the protection threshold than any single recipe and warns the artist about artworks for which no recipe is predicted to pass it.

cs.CR↗

Fast and Strong Image Protection against Diffusion Personalization

Personalized text-to-image diffusion models can reproduce a person from a handful of photographs. Image protection perturbs each image before publication so that fine-tuning on it fails, but it optimizes every perturbation from scratch, so its cost grows with the number of images. We present the protective diffusion model (PDM), which is trained once and samples the perturbation of a new image over many denoising steps. PDM distills an iterative teacher and is then post-trained with the target residual projection (TRP) loss as a reward, a stronger objective than existing ones. PDM outperforms nine baselines on faces and artworks and nearly matches its teacher at a tenth of the cost.

cs.CV↗

Mean Accuracy Is the Wrong Estimand for a Release Decision

A release decision about a fine-tuned model is an operational commitment about the \emph{next} rerun of its training recipe. Yet standard benchmark reporting gives only the mean accuracy of past seeds. These target fundamentally different questions: release decisions depend on catastrophic tail risk, whereas standard evaluation reports only central tendency. Because fine-tuning induces bimodal mode collapse, high average accuracy frequently conceals severe failure rates. Auditing $2\,124$ fine-tuned models across eleven language models, we show that shipping models based on mean accuracy violates a $10\%$ deployment failure budget $48.0\%$ of the time. Common heuristics like $\bar b - 2σ$ prove equally dangerous: by underestimating variance on small samples, they become most permissive when evidence is weakest. We introduce \textbf{recipe cards}, an evaluation framework that directly certifies rerun tail failure rates using exact finite-sample binomial bounds. Recipe cards reduce held-out budget violations to $0.6\%$, enforce a mathematical sample floor ($N \ge 22$) that prevents certification on thin evidence, and explicitly refuse unverified claims under domain and task shifts.

cs.LG↗

Beyond User-Side Defenses: Image Protection in Open-Weight Diffusion Models

Model providers can support image protection, but open-weight deployment introduces practical constraints. We propose four requirements: protection effectiveness, clean fidelity, compatibility, and tamper resistance. We incorporate long-horizon attack simulations to strengthen resistance to protection removal. Experiments with a single protection tool show improved resistance to re-alignment at a cost to clean fidelity, motivating further research toward practical model-side protection.

cs.CR↗

Onset Mechanism of M6.5 Solar Flare Observed in Active Region 12371

We studied a flare onset process in terms of stability of a three-dimensional (3D) magnetic field in active region 12371 producing an eruptive M6.5 flare in 2015 June 22. In order to reveal the 3D magnetic structure, we first extrapolated the 3D coronal magnetic fields based on time series of the photospheric vector magnetic fields under a nonlinear force-free field (NLFFF) approximation. The NLFFFs nicely reproduced the observed sigmoidal structure which is widely considered to be preeruptive magnetic configuration. In particular, we found that the sigmoid is composed of two branches of sheared arcade loops. On the basis of the NLFFFs, we investigated the sheared arcade loops to explore the onset process of the eruptive flare using three representative magnetohydrodynamic instabilities: the kink, torus, and double arc instabilities (DAI). The DAI, recently proposed by Ishiguro & Kusano, is a double arc loop that can be more easily destabilized than a torus loop. Consequently, the NLFFFs are found to be quite stable against the kink and torus instabilities. However, the sheared arcade loops formed prior to the flare possibly become unstable against the DAI. As a possible scenario for the onset process of the M6.5 flare, we suggest a three-step process: (1) double arc loops are formed by the sheared arcade loops through the tether-cutting reconnection during an early phase of the flare, (2) the DAI contributes to the expansion of destabilized double arc loops, and (3) finally, the torus instability makes the full eruption.

astro-ph.SR↗

Investigation of Force-Freeness of Solar Emerging Magnetic Field via Application of the Virial Theorem to MHD Simulations

Force-freeness of a solar magnetic field is a key to reconstructing invisible coronal magnetic structure of an emerging flux region on the Sun where active phenomena such as flares and coronal mass ejections frequently occur. We have performed magnetohydrodynamic (MHD) simulations which are adjusted to investigate force-freeness of an emerging magnetic field by using the virial theorem. Our focus is on how the force-free range of an emerging flux region develops and how it depends on the twist of a pre-emerged magnetic field. As an emerging flux region evolves, the upper limit of the force-free range continuously increases while the lower limit is asymptotically reduced to the order of a photospheric pressure scale height above the solar surface. As the twist becomes small the lower limit increases and then seems to be saturated. We also discuss the applicability of the virial theorem to an evolving magnetic structure on the Sun.

astro-ph.SR↗