arXiv subjects

Yujin Park

Publications and source records attributed to Yujin Park.

16 recordsLinked to original sources

arXiv2026

Expected performance of a water Cherenkov detector for reactor antineutrino--electron scattering

We evaluate the expected performance of a 170-ton water Cherenkov detector for reactor antineutrino--electron elastic scattering at the RENO near-detector site. The reactor-spectrum normalization is extracted from the reconstructed recoil-electron directional distribution using a constrained profile-likelihood fit for an exposure of 365.25 days. For the minimum prompt-hit multiplicity $N_{\mathrm{prompt}}\geq8$ scenario, which neglects PMT radioactivity, the total uncertainty on the reactor-spectrum normalization is ${}^{+1.08\%}_{-1.07\%}$. For the more restrictive $N_{\mathrm{prompt}}\geq20$ scenario, which includes the simulated PMT-radioactivity components, the total uncertainty is ${}^{+1.49\%}_{-1.47\%}$. These results demonstrate that directional information from a water Cherenkov detector can provide percent-level sensitivity to the reactor elastic-scattering normalization.

hep-ex↗

arXiv2026

Surprise-Guided MergeSort: Budget-Efficient Human-in-the-Loop Ranking via Adaptive Comparison Scheduling

Pairwise comparison is the gold standard for subjective ranking tasks; however, exhaustive annotation requires a massive number of human comparisons ($O(n^2)$). While sorting-based methods have reduced this burden to $O(n\log n)$, they still require expensive human judgment for every single comparison. To further improve annotation efficiency, we propose leveraging a Vision-Language Model (VLM) not as an annotator replacement, but as a \emph{question prioritizer} to identify which comparisons genuinely require human judgment. The proposed \textbf{Surprise-Guided MergeSort (SGS)} framework achieves this through three integrated components: (1) a bottom-up MergeSort scheduler that structures comparisons and exploits transitivity, (2) a composite Surprise Scorer -- combining position-bias-cancelled VLM confidence, Elo gap, and vote entropy -- to quantify comparison ambiguity, and (3) an adaptive budget allocator that routes high-surprise pairs to humans while automating low-surprise pairs via transitivity inference. Validation was conducted on six diverse benchmarks spanning text similarity (STS-B, BIOSSES, SICKR-STS) and image quality assessment (KonIQ-10k, TID2013, LIVE Challenge). SGS effectively identified and skipped up to 535 non-informative comparisons per session. Consequently, it achieved Kendall's $\tau{\times}100$ improvements of $+6$ to $+12$ over Active Elo under the same total budget. These results demonstrate that combining VLM-guided surprise metrics with algorithmic sorting provides a generally consistent accuracy-efficiency trade-off across diverse domains.

cs.LG↗

arXiv2026

MetaRanker: Human-in-the-loop Active Ranking for Metalens Image Quality

Image quality in modern imaging systems emerges from the coupled effects of the sensor, optics, and computational reconstruction. Ultra-thin metalenses offer a path toward substantial miniaturization of optical modules, but practical designs often exhibit pronounced chromatic and field-dependent aberrations that necessitate computational reconstruction. In current metalens pipelines, reconstruction models are commonly trained and selected using distortion-based fidelity objectives, such as PSNR, yet these proxies can be weakly correlated with human preference and downstream utility, reflecting the well-known perception--distortion trade-off. We introduce MetaRanker, a human-in-the-loop active ranking framework that formalizes metalens image quality in terms of semantic interpretability, defined as the degree to which humans can reliably recognize objects and structures in the presence of optical artifacts. MetaRanker combines a probabilistic preference model with uncertainty-aware query selection, and leverages vision--language models to provide lightweight semantic priors. Importantly, these priors are used only to guide the sampling of informative comparisons; human judgments remain the primary supervision signal throughout. Across real-world and synthetic metalens datasets with distinct degradation profiles, MetaRanker produces rankings that align most closely with human assessments, while reducing the number of pairwise annotations required by approximately 80% relative to exhaustive pairwise evaluation. Finally, we show that standard image quality assessment metrics exhibit limited alignment with human interpretability in the metalens domain, positioning MetaRanker as a practical step toward perceptually grounded metalens evaluation and co-design.

cs.CV↗

arXiv2026

Dodgersort: Uncertainty-Aware VLM-Guided Human-in-the-Loop Pairwise Ranking

Pairwise comparison labeling is emerging as it yields higher inter-rater reliability than conventional classification labeling, but exhaustive comparisons require quadratic cost. We propose Dodgersort, which leverages CLIP-based hierarchical pre-ordering, a neural ranking head and probabilistic ensemble (Elo, BTL, GP), epistemic--aleatoric uncertainty decomposition, and information-theoretic pair selection. It reduces human comparisons while improving the reliability of the rankings. In visual ranking tasks in medical imaging, historical dating, and aesthetics, Dodgersort achieves a 11--16\% annotation reduction while improving inter-rater reliability. Cross-domain ablations across four datasets show that neural adaptation and ensemble uncertainty are key to this gain. In FG-NET with ground-truth ages, the framework extracts 5--20$\times$ more ranking information per comparison than baselines, yielding Pareto-optimal accuracy--efficiency trade-offs.

cs.CV↗

arXiv2025

Rethinking Retrieval-Augmented Generation for Medicine: A Large-Scale, Systematic Expert Evaluation and Practical Insights

Large language models (LLMs) are transforming the landscape of medicine, yet two fundamental challenges persist: keeping up with rapidly evolving medical knowledge and providing verifiable, evidence-grounded reasoning. Retrieval-augmented generation (RAG) has been widely adopted to address these limitations by supplementing model outputs with retrieved evidence. However, whether RAG reliably achieves these goals remains unclear. Here, we present the most comprehensive expert evaluation of RAG in medicine to date. Eighteen medical experts contributed a total of 80,502 annotations, assessing 800 model outputs generated by GPT-4o and Llama-3.1-8B across 200 real-world patient and USMLE-style queries. We systematically decomposed the RAG pipeline into three components: (i) evidence retrieval (relevance of retrieved passages), (ii) evidence selection (accuracy of evidence usage), and (iii) response generation (factuality and completeness of outputs). Contrary to expectation, standard RAG often degraded performance: only 22% of top-16 passages were relevant, evidence selection remained weak (precision 41-43%, recall 27-49%), and factuality and completeness dropped by up to 6% and 5%, respectively, compared with non-RAG variants. Retrieval and evidence selection remain key failure points for the model, contributing to the overall performance drop. We further show that simple yet effective strategies, including evidence filtering and query reformulation, substantially mitigate these issues, improving performance on MedMCQA and MedXpertQA by up to 12% and 8.2%, respectively. These findings call for re-examining RAG's role in medicine and highlight the importance of stage-aware evaluation and deliberate system design for reliable medical LLM applications.

cs.CL↗

arXiv2025

EZ-Sort: Efficient Pairwise Comparison via Zero-Shot CLIP-Based Pre-Ordering and Human-in-the-Loop Sorting

Pairwise comparison is often favored over absolute rating or ordinal classification in subjective or difficult annotation tasks due to its improved reliability. However, exhaustive comparisons require a massive number of annotations (O(n^2)). Recent work has greatly reduced the annotation burden (O(n log n)) by actively sampling pairwise comparisons using a sorting algorithm. We further improve annotation efficiency by (1) roughly pre-ordering items using the Contrastive Language-Image Pre-training (CLIP) model hierarchically without training, and (2) replacing easy, obvious human comparisons with automated comparisons. The proposed EZ-Sort first produces a CLIP-based zero-shot pre-ordering, then initializes bucket-aware Elo scores, and finally runs an uncertainty-guided human-in-the-loop MergeSort. Validation was conducted using various datasets: face-age estimation (FGNET), historical image chronology (DHCI), and retinal image quality assessment (EyePACS). It showed that EZ-Sort reduced human annotation cost by 90.5% compared to exhaustive pairwise comparisons and by 19.8% compared to prior work (when n = 100), while improving or maintaining inter-rater reliability. These results demonstrate that combining CLIP-based priors with uncertainty-aware sampling yields an efficient and scalable solution for pairwise ranking.

cs.CV↗

arXiv2025

On the Energy Distribution of the Galactic Center Excess' Sources

The Galactic Center Excess (GCE) may yet herald the discovery of annihilating dark matter. Weighing against that conclusion are analyses showing evidence for dim point sources within the spatial structure of the emission. Due to technical limitations these analyses are purely spatial with all spectral information that could disentangle the excess from astrophysical backgrounds discarded. Here, we demonstrate that a neural network simulation-based inference approach can jointly analyze the spatial and spectra data. The addition is profound: energy information drives the putative point sources to be significantly dimmer, indicating either the GCE is truly diffuse in nature or made of an exceptionally large number of sources. Quantitatively, for our best fit background model, the excess is essentially consistent with Poisson emission as predicted by dark matter. If due to point sources, our median prediction is $\mathcal{O}(10^5)$ sources, or more than 35,000 at 90\% confidence, both orders of magnitude larger than the hundreds preferred by earlier point-source analyses of the GCE, although variations allowed by background systematics could reduce the required number of sources by roughly an order of magnitude.

astro-ph.HE↗

arXiv2025

A comparative study of physics capabilities of a liquid argon and a water based liquid scintillator at DUNE

We present a comprehensive comparison of the physics sensitivities of a Liquid Argon Time Projection Chamber (LArTPC) and a Water-based Liquid Scintillator (WbLS) detector, considering their potential deployment as the fourth far detector module in the DUNE facility. Using GLoBES-based simulations, we evaluate their performance in measuring standard neutrino oscillation parameters ($\theta_{23}, \delta_{13}$ and $\Delta m^{2}_{31}$), both in standard 3-neutrino case, as well as in presence of new physics scenarios involving light sterile neutrinos and neutral-current non-standard interactions (NC NSI). Our findings show that THEIA (a WbLS-based detector) significantly outperforms LArTPC in resolving the CP phase $\delta_{13}$,- especially near maximal CP violation, and in lifting the octant degeneracy of $\theta_{23}$ due to its superior energy resolution and ability to clearly identify the second oscillation maximum. Furthermore, THEIA offers competitive reconstruction precision even with relatively moderate energy resolutions ($7-10\%/\sqrt{E}$) and demonstrates enhanced robustness under new physics scenarios. These results support the physics-driven case for a hybrid DUNE configuration utilizing both LArTPC and WbLS technologies for optimized sensitivity across the full spectrum of neutrino oscillation and physics beyond the standard model.

hep-ph↗

arXiv2024

Time-delayed gamma-ray signatures of heavy axions from core-collapse supernovae

Heavy axions that couple to both quantum electrodynamics and quantum chromodynamics with masses on the order of MeV - GeV and high-scale decay constants in excess of $\sim$$10^8$ GeV may arise generically in e.g. axiverse constructions. In this work we provide the most sensitive search to-date for the existence of such heavy axions using Fermi-LAT data towards four recent supernovae (SN): Cas A, SN1987A, SN2023ixf, and SN2024ggi. We account for heavy axion production in the proto-neutron-star cores through nuclear and electromagnetic processes and then the subsequent decay of the axions into photons. While previous works have searched for gamma-rays from SN1987A using the Solar Maximum Mission that observed SN1987A during the SN itself, we show that using Fermi Large Area Telescope data provides an approximately five orders of magnitude improvement in flux sensitivity for axions with lifetimes larger than around 10 years. We find no evidence for heavy axions and exclude large regions of previously-unexplored parameter space.

hep-ph↗

arXiv2024

Supernova axions convert to gamma-rays in magnetic fields of progenitor stars

It has long been established that axions could have been produced within the nascent proto-neutron-star formed following the type II supernova SN1987A, escaped the star due to their weak interactions, and then converted to gamma-rays in the Galactic magnetic fields; the non-observation of a gamma-ray flash coincident with the neutrino burst leads to strong constraints on the axion-photon coupling for axion masses $m_a \lesssim 10^{-10}$ eV. In this work we use SN1987A to constrain higher mass axions, all the way to $m_a \sim 10^{-3}$ eV, by accounting for axion production from the Primakoff process, nucleon bremsstrahlung, and pion conversion along with axion-photon conversion on the still-intact magnetic fields of the progenitor star. Moreover, we show that gamma-ray observations of the next Galactic supernova, leveraging the magnetic fields of the progenitor star, could detect quantum chromodynamics axions for masses above roughly 50 $\mu$eV, depending on the supernova. We propose a new full-sky gamma-ray satellite constellation that we call the GALactic AXion Instrument for Supernova (GALAXIS) to search for such future signals along with related signals from extragalactic neutron star mergers.

hep-ph↗

arXiv2023

Was There a 3.5 keV Line?

The 3.5 keV line is a purported emission line observed in galaxies, galaxy clusters, and the Milky Way whose origin is inconsistent with known atomic transitions and has previously been suggested to arise from dark matter decay. We systematically re-examine the bulk of the evidence for the 3.5 keV line, attempting to reproduce six previous analyses that found evidence for the line. Surprisingly, we only reproduce one of the analyses; in the other five we find no significant evidence for a 3.5 keV line when following the described analysis procedures on the original data sets. For example, previous results claimed 4$σ$ evidence for a 3.5 keV line from the Perseus cluster; we dispute this claim, finding no evidence for a 3.5 keV line. We find evidence for background mismodeling in multiple analyses. We show that analyzing these data in narrower energy windows diminishes the effects of mismodeling but returns no evidence for a 3.5 keV line. We conclude that there is little robust evidence for the existence of the 3.5 keV line. Some of the discrepancy of our results from those of the original works may be due to the earlier reliance on local optimizers, which we demonstrate can lead to incorrect results. For ease of reproducibility, all code and data are publicly available.

astro-ph.CO↗

arXiv2023

Signatures of Primordial Energy Injection from Axion Strings

Axion strings are horizon-size topological defects that may be produced in the early Universe. Ultra-light axion-like particles may form strings that persist to temperatures below that of big bang nucleosynthesis. Such strings have been considered previously as sources of gravitational waves and cosmic microwave background (CMB) polarization rotation. In this work we show, through analytic arguments and dedicated adaptive mesh refinement cosmological simulations, that axion strings deposit a sub-dominant fraction of their energy into high-energy Standard Model (SM) final states, for example, by the direct production of heavy radial modes that subsequently decay to SM particles. This high-energy SM radiation is absorbed by the primordial plasma, leading to novel signatures in precision big bang nucleosynthesis, the CMB power spectrum, and gamma-ray surveys. In particular, we show that CMB power spectrum data constrains axion strings with decay constants $f_a \lesssim 10^{12}$ GeV, up to model dependence on the ultraviolet completion, for axion masses $m_a \lesssim 10^{-29}$ eV; future CMB surveys could find striking evidence of axion strings with lower decay constants.

hep-ph↗

arXiv2022

A Search for Dark Matter Lines at the Galactic Center with 14 Years of Fermi Data

Dark matter (DM) in the Milky Way halo may annihilate or decay to photons, producing monochromatic gamma rays. We search for DM-induced spectral lines using 14 years of data from the Large Area Telescope onboard the Fermi Gamma-ray Space Telescope ($\textit{Fermi}$-LAT) between $10\,\mathrm{GeV}$ and $2\,\mathrm{TeV}$ in the inner Milky Way leveraging both the spatial and spectral morphology of an expected signal. We present new constraints as strong as $\langle σv \rangle \lesssim 6\times 10^{-30}\, \mathrm{cm}^3/\mathrm{s}$ for the two-to-two annihilations and $τ\gtrsim 10^{30}\,\mathrm{s}$ for one-to-two decays, representing leading sensitivity between $10\,\mathrm{GeV}$ and $\sim$$500\,\mathrm{GeV}$. We consider the implications of our line-constraints on the Galactic Center Excess (GCE), which is a previously-observed excess of continuum $\sim$GeV gamma-rays that may be explained by DM annihilation. The Higgs portal and neutralino-like DM scenarios, which have been extensively discussed as possible origins of the GCE, are constrained by our work because of the lack of observed one-loop decays to two photons. More generally, we interpret our null results in a variety of annihilating and decaying DM models, such as neutralinos, gravitinos, and glueballs, showing that in many cases the line search is more powerful than the continuum, despite the continuum annihilation being at tree level.

hep-ph↗

arXiv2022

Higgsino Dark Matter Confronts 14 years of Fermi Gamma Ray Data

Thermal higgsino dark matter (DM), with mass around 1 TeV, is a well-motivated, minimal DM scenario that arises in supersymmetric extensions of the Standard Model. Higgsinos may naturally be the lightest superpartners in Split-supersymmetry models that decouple the scalar superpartners while keeping higgsinos and gauginos close to the TeV scale. Higgsino DM may annihilate today to give continuum gamma-ray emission at energies less than a TeV in addition to a line-like signature at energies equal to the mass. Previous searches for higgsino DM, for example with the H.E.S.S. gamma-ray telescope, have not reached the necessary sensitivity to probe the higgsino annihilation cross-section. In this work we make use of 14 years of $\textit{Fermi}$ gamma-ray data at energies above $\sim$10 GeV to search for the continuum emission near the Galactic Center from higgsino annihilation. We interpret our results using DM profiles from Milky Way analogue galaxies in the FIRE-2 hydrodynamic cosmological simulations. We set the strongest constraints to-date on higgsino-like DM. Our results show a mild, $\sim$2$σ$ preference for higgsino DM with a mass near the thermal higgsino mass and, depending on the DM density profile, the expected cross-section.

hep-ph↗

arXiv2022

Personalization Trade-offs in Designing a Dialogue-based Information System for Support-Seeking of Sexual Violence Survivors

The lack of reliable, personalized information often complicates sexual violence survivors' support-seeking. Recently, there is an emerging approach to conversational information systems for support-seeking of sexual violence survivors, featuring personalization with wide availability and anonymity. However, a single best solution might not exist as sexual violence survivors have different needs and purposes in seeking support channels. To better envision conversational support-seeking systems for sexual violence survivors, we explore personalization trade-offs in designing such information systems. We implement a high-fidelity prototype dialogue-based information system through four design workshop sessions with three professional caregivers and interviewed with four self-identified survivors using our prototype. We then identify two forms of personalization trade-offs for conversational support-seeking systems: (1) specificity and sensitivity in understanding users and (2) relevancy and inclusiveness in providing information. To handle these trade-offs, we propose a reversed approach that starts from designing information and inclusive tailoring that considers unspecified needs, respectively.

cs.HC↗

arXiv2021

A deep search for decaying dark matter with XMM-Newton blank-sky observations

Sterile neutrinos with masses in the keV range are well-motivated extensions to the Standard Model that could explain the observed neutrino masses while also making up the dark matter (DM) of the Universe. If sterile neutrinos are DM then they may slowly decay into active neutrinos and photons, giving rise to the possibility of their detection through narrow spectral features in astrophysical X-ray data sets. In this work, we perform the most sensitive search to date for this and other decaying DM scenarios across the mass range from 5 to 16 keV using archival XMM-Newton data. We reduce 547 Ms of data from both the MOS and PN instruments using observations taken across the full sky and then use this data to search for evidence of DM decay in the ambient halo of the Milky Way. We determine the instrumental and astrophysical baselines with data taken far away from the Galactic Center, and use Gaussian Process modeling to capture additional continuum background contributions. No evidence is found for unassociated X-ray lines, leading us to produce the strongest constraints to date on decaying DM in this mass range.

astro-ph.CO↗