Searcharxiv⌕ Search

arXiv subjects

Atsushi Yoshida

Publications and source records attributed to Atsushi Yoshida.

16 recordsLinked to original sources

Fine-Grained Japanese Clinical Named Entity Recognition: A Controlled Comparison of BERT and Open-Weight Large Language Models

Background: Generative large language models (LLMs) offer an alternative to encoder-based clinical named entity recognition (NER), but their performance on fine-grained Japanese extraction tasks and sensitivity to generation settings remain uncertain. Methods: We compared two fine-tuned BERT-CRF baselines with nine open-weight LLMs on iCorpus, which defines 113 entity categories. The evaluation used 433 sentences from 36 test documents; 1,672 sentences were used for training. LLMs were evaluated using zero-shot prompting with and without JSON Schema constraints and three-seed QLoRA fine-tuning. Fine-tuned models were assessed under different output-token limits, including 900 and larger model-specific limits of 2,600 or 4,096 tokens. Performance was measured using sentence-level label-surface matching with relaxed and exact-surface F1. Results: The highest zero-shot relaxed F1 was 0.337, compared with 0.836-0.841 for the BERT baselines. At the larger output limits, all eight fine-tuned LLMs with at least 13B total parameters exceeded both BERT baselines (relaxed F1, 0.882-0.914; exact-surface F1, 0.840-0.880). Across three-seed means, increasing the output limit improved relaxed F1 by 0.052-0.085, alongside higher recall and fewer length-limited outputs. JSON Schema constraints changed relaxed F1 by -0.014 to +0.008 among seven models outside the two reasoning-oriented configurations, while medical-general model pairs showed no consistent advantage for medical specialization. Conclusions: On this historical Japanese case-report benchmark, task-specific fine-tuning combined with adequate generation capacity was associated with stronger LLM extraction performance. Output-token limits materially affected comparisons with BERT, whereas enforcing output structure alone did not overcome low zero-shot accuracy.

cs.AI↗

The LLENS Platform for On-Premises Large Language Model Agents in a Japanese University Hospital: Development and Evaluation Study

Large language model agents that use tools could take on part of the electronic health record and desk work that occupies physicians' time, but in many hospitals the medical information systems run inside closed networks, so the model and the agent harness must also be inside the network. We designed, implemented, and deployed LLENS, a clinical support AI agent platform that runs inside the closed network of Hokkaido University Hospital on a single host with 8 NVIDIA H200 GPUs, the open-weight model Kimi K2.6 (~1 trillion parameters), and a web application rebuilt from Open WebUI. The harness has 10 components, including page image reading, full-text attachment input, a path-addressed document tree without a vector index, a network-isolated sandbox, artifact return, self-checking of generated output, subagents, and saved procedures. Kimi K2.6 was chosen from 3 open-weight candidates and scored 480/500 on the 2026 Japanese Medical Licensing Examination; it generated 119 tokens/s for a single user and served 16 concurrent requests at 908 tokens/s in total. On 180 questions over hospital documents (3 seeds each), mean accuracy was 10.7% with no documents, 81.3% with fixed retrieval (top 10 chunks of one lexical search), and 97.2% with interactive access, in which the model reads with document tools and the sandbox (15.9 percentage points above fixed retrieval, 95% CI 10.9-21.3); the difference lay in filtering/counting questions (16.7% vs 87.8%). In an automated test of 13 work scenarios (4 runs each), the expected components were triggered in all runs for 12, and 5 scenarios were run with synthetic data on the deployment host. The platform has run on the hospital network since September 14, 2026.

cs.CL↗

Development of the algorithm for differentiating bone metastases and trauma of the ribs in bone scintigraphy and demonstration of visual evidence of the algorithm -- Using only anterior bone scan view of thorax

Background: Although there are many studies on the application of artificial intelligence (AI) models to medical imaging, there is no report of an AI model that determines the accumulation of ribs in bone metastases and trauma only using the anterior image of thorax of bone scintigraphy. In recent years, a method for visualizing diagnostic grounds called Gradient-weighted Class Activation Mapping (Grad-CAM) has been proposed in the area of diagnostic images using Deep Convolutional Neural Network (DCNN). As far as we have investigated, there are no reports of visualization of the diagnostic basis in bone scintigraphy. Our aim is to visualize the area of interest of DCNN, in addition to developing an algorithm to classify and diagnose whether RI accumulation on the ribs is bone metastasis or trauma using only anterior bone scan view of thorax. Material and Methods: For this retrospective study, we used 838 patients who underwent bone scintigraphy to search for bone metastases at our institution. A frontal chest image of bone scintigraphy was used to create the algorithm. We used 437 cases with bone metastases on the ribs and 401 cases with abnormal RI accumulation due to trauma. Result: AI model was able to detect bone metastasis lesion with a sensitivity of 90.00% and accuracy of 86.5%. And it was possible to visualize the part that the AI model focused on with Grad-CAM.

eess.IV↗

Investigation of coronal properties of X-ray bright G-dwarf stars based on the solar surface magnetic field -- corona relation

We investigated the coronal properties of G-dwarf stars including the Sun over a wide range of X-ray luminosity $L_{\rm X}$ ($3\times 10^{26}$ to $2\times 10^{30}~{\rm erg~s^{-1}}$). We analyzed the archival data of ten X-ray bright ($L_{\it X}>10^{28}~{\rm erg~s^{-1}}$) G-dwarf stars to derive their emission measure (EM) and the coronal temperature ($T$) during the periods when no prominent stellar flares were observed. We attempted to explain the relation on the basis of our understanding of the present Sun: a steady corona model based on the so-called RTV scaling laws and the observed power-law distribution function of surface magnetic features. We derived a theoretical scaling law of the EM--$T$ relation for a star with multiple active regions, and applied it to the observations combined with data in literature. We found that with the solar parameters, our scaling law seems to be consistent with the data of slowly-rotating stars. However, more X-ray bright stars are located well above the scaling law based on the solar parameter. The scaling law may explain the observations if those stars show a power-law distribution function of active regions with the same power-law index but a 10-100 times larger coefficient. This suggests that X-ray bright stars show more active regions for a given size than the Sun. Since our samples include rapidly-rotating stars, we infer that the offset of the X-ray bright stars from the present-Sun-based scaling law is due to the enhancement of the surface magnetic field generation by their rapid rotation.

astro-ph.SR↗

Active gas features in three HSC-SSP CAMIRA clusters revealed by high angular resolution analysis of MUSTANG-2 SZE and XXL X-ray observations

We present results from simultaneous modeling of high angular resolution GBT/MUSTANG-2 90 GHz Sunyaev-Zel'dovich effect (SZE) measurements and XMM-XXL X-ray images of three rich galaxy clusters selected from the HSC-SSP Survey. The combination of high angular resolution SZE and X-ray imaging enables a spatially resolved multi-component analysis, which is crucial to understand complex distributions of cluster gas properties. The targeted clusters have similar optical richnesses and redshifts, but exhibit different dynamical states in their member galaxy distributions: a single-peaked cluster, a double-peaked cluster, and a cluster belonging to a supercluster. A large-scale residual pattern in both regular Compton-parameter $y$ and X-ray surface brightness distributions is found in the single-peaked cluster, indicating a sloshing mode. The double-peaked cluster shows an X-ray remnant cool core between two SZE peaks associated with galaxy concentrations. The temperatures of the two peaks reach $\sim20-30$ keV in contrast to the cool core component of $\sim2$ keV, indicating a violent merger. The main SZE signal for the supercluster is elongated along a direction perpendicular to the major axis of the X-ray core, suggesting a minor merger before core passage. The $S_X$ and $y$ distributions are thus perturbed at some level, regardless of the optical properties. We find that the integrated Compton $y$ parameter and the temperature for the major merger are boosted from those expected by the weak-lensing mass and those for the other two clusters show no significant deviations, which is consistent with predictions of numerical simulations.

astro-ph.CO↗

Effects of asymmetric splayed columnar defects on the anomalous peak effect in Ba$_{0.6}$K$_{0.4}$Fe$_{2}$As$_{2}$

In order to enhance the critical current density ($J_c$) of superconductors, introduction of columnar defects (CDs) through swift-particle irradiation is effective. By dispersing the direction of CDs (splayed CDs), not only further enhancement of $J_c$ has been confirmed but also an anomalous peak effect (APE) in $J_c$ at a certain magnetic field determined by the irradiation dose was observed. It has been proposed that the APE arises from the suppression of kink motion and/or the effects of vortex entanglement in systems with splayed CDs. In this study, we measure $J_c$ properties of optimally K-doped Ba-122-type iron-based superconductor Ba$_{0.6}$K$_{0.4}$Fe$_{2}$As$_{2}$ single crystals with splayed CDs that are introduced asymmetrically with respect to the $c$-axis by irradiating 2.6 GeV U ions. We discuss the significance of the average direction of splayed CDs to the APE. We also discuss the relationship between the APE and the entanglement of vortices.

cond-mat.supr-con↗

Anisotropic physical properties and large critical current density in KCa${_2}$Fe${_4}$As${_4}$F${_2}$ single crystal

We present a systematic study of electrical resistivity, Hall coefficient, magneto-optical imaging, magnetization, and STEM analyses of KCa${_2}$Fe${_4}$As${_4}$F${_2}$ single crystals. Sharp diamagnetic transition and magneto-optical imaging reveal homogeneity of single crystal and prominent Bean-like penetrations of vortices. Large anisotropy of electrical resistivity, with ${ρ_c / ρ_{ab}}$ > 100, and semiconductor-like ${ρ_c}$ suggest that the electronic state is quasi two-dimensional. Hall effect measurements indicate that KCa${_2}$Fe${_4}$As${_4}$F${_2}$ is a multiband system with holes as main carriers. Magnetization measurements reveal significantly larger J$_c$ compared with that in other iron-based superconductors with different values of J$_c$ depending on the direction of magnetic field. Origin of these J$_c$ characteristics is discussed based on microstructural observations using STEM. In addition, further enhancement of J$_c$ in KCa${_2}$Fe${_4}$As${_4}$F${_2}$ for future application is demonstrated in terms of heavy-ion irradiation.

cond-mat.supr-con↗

X-ray properties of high-richness CAMIRA clusters in the Hyper Suprime-Cam Subaru Strategic Program field

We present the first results of a pilot X-ray study of 37 rich galaxy clusters at $0.1<z<1.1$ in the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP) field. Diffuse X-ray emissions from these clusters were serendipitously detected in the XMM-Newton fields of view. We systematically analyze X-ray images of 37 clusters and emission spectra of a subsample of 17 clusters with high photon statistics by using the XMM-Newton archive data. The frequency distribution of the offset between the X-ray centroid or peak and the position of the brightest cluster galaxy was derived for the optical cluster sample. The fraction of relaxed clusters estimated from the X-ray peak offsets in 17 clusters is $29\pm11(\pm13)$\%, which is smaller than that of the X-ray cluster samples such as HIFLUGCS. Since the optical cluster search is immune to the physical state of X-ray-emitting gas, it is likely to cover a larger range of the cluster morphology. We also derived the luminosity-temperature relation and found that the slope is marginally shallower than those of X-ray-selected samples and consistent with the self-similar model prediction of 2. Accordingly, our results show that the X-ray properties of the optical clusters are marginally different from those observed in the X-ray samples. The implication of the results and future prospects are briefly discussed.

astro-ph.CO↗

Field-driven transition in the Ba$_{1-x}$K$_x$Fe$_2$As$_2$ superconductor with splayed columnar defects

Through 2.6 GeV U irradiations, we have induced bimodal splayed columnar defects in Ba$_{1-x}$K$_x$Fe$_2$As$_2$ single crystals with splay angles, $\pm 5 ^\circ$, $\pm 10 ^\circ$, $\pm 15 ^\circ$, and $\pm 20 ^\circ$. Critical current densities through magnetization measurements were carefully evaluated, where a splay angle of $\pm 5 ^\circ$ brought about the highest $J_\mathrm{c}$. Mageto-optical images close to $T_\mathrm{c}$ indicates highly anisotropic discontinuity lines in the remnant state, and with anisotropy increasing with greater splay angles. Moreover, amongst those with splayed columnar defects, anomalous non-monotonic field dependences of $J_\mathrm{c}$ and $S$ with an extrema at some fraction of the matching field are observed. We discuss that such $J_\mathrm{c}$ enhancement arises from a field-driven coupling transition in which intervortex interactions reorganize the vortex structure to be accommodated into columnar defects, thereby increasing pinning at higher fields.

cond-mat.supr-con↗

The intrinsic abundance ratio and X-factor of CO isotopologues in L1551 shielded from FUV photodissociation

L1551 is chosen because it is relatively isolated in the Taurus molecular cloud shielded from FUV photons, providing an ideal environment for studying the target properties. Our observations cover ~40'x40' with resolution ~30", which are the maps with highest spatial dynamical range to date. We derive the X(13CO)/X(C18O) value on the sub-parsec scales in the range of ~3-27 with a mean value of 8.0+-2.8. Comparing to the visual extinction map derived from the Herschel observations, we found that the abundance ratio reaches its maximum at low Av (i.e., Av ~ 1-4mag), and decreases to the typical solar system value of 5.5 inside L1551 MC. The high X(13CO)/X(C18O) value at the boundary of the cloud is most likely due to the selective FUV photodissociation of C18O. This is in contrast with Orion-A where its internal OB stars keep the abundance ratio at a high level greater than ~10. In addition, we explore the variation of the X-factor, because it is an uncertain but widely used quantity in extragalactic studies. We found that X-factor is proportional to N(H2)^1.0 which is consistent with previous simulations. Excluding the high density region, the average X-factor is similar to the Milky Way average value.

astro-ph.GA↗

Aharonov-Bohm phase for an electromagnetic wave background

The canonical Aharonov-Bohm effect is usually studied with time-independent potentials. In this work, we investigate the Aharonov-Bohm phase acquired by a charged particle moving in {\it time-dependent} potentials . In particular, we focus on the case of a charged particle moving in the time varying field of a plane electromagnetic wave. We work out the Aharonov-Bohm phase using both the potential ({\it i.e.} $\oint A_μdx ^μ$) and field ({\it i.e.} $\frac{1}{2}\int F_{μν} d σ^{μν}$) forms of the Aharanov-Bohm phase. We give conditions in terms of the parameters of the system (frequency of the electromagnetic wave, the size of the space-time loop, amplitude of the electromagnetic wave) under which the time varying Aharonov-Bohm effect could be observed.

hep-th↗

New Result in the Production and Decay of an Isotope, 278[113], of the 113th Element

An isotope of the 113th element, i.e., 278[113], was produced in a nuclear reaction with a 70Zn beam on a 209Bi target. We observed six consecutive α-decays following the implantation of a heavy particle in nearly the same position in the semiconductor detector under an extremely low background condition. The fifth and sixth decays are fully consistent with the sequential decays of 262Db and 258Lr in both decay energies and decay times. This indicates that the present decay chain consisted of 278[113], 274Rg (Z=111), 270Mt (Z=109), 266Bh (Z=107), 262Db (Z=105), and 258Lr (Z=103) with firm connections. This result, together with previously reported results from 2004 and 2007, conclusively leads to the unambiguous production and identification of the isotope 278[113] of the 113th element.

nucl-ex↗

Identification of 45 New Neutron-Rich Isotopes Produced by In-Flight Fission of a 238U Beam at 345 MeV/nucleon

A search for new isotopes using in-flight fission of a 345 MeV/nucleon 238U beam has been carried out at the RI Beam Factory at the RIKEN Nishina Center. Fission fragments were analyzed and identified by using the superconducting in-flight separator BigRIPS. We observed 45 new neutron-rich isotopes: 71Mn, 73,74Fe, 76Co, 79Ni, 81,82Cu, 84,85Zn, 87Ga, 90Ge, 95Se, 98Br, 101Kr, 103Rb, 106,107Sr, 108,109Y, 111,112Zr, 114,115Nb, 115,116,117Mo, 119,120Tc, 121,122,123,124Ru, 123,124,125,126Rh, 127,128Pd, 133Cd, 138Sn, 140Sb, 143Te, 145I, 148Xe, and 152Ba.

nucl-ex↗

Ultra-Relativistic Hamiltonian with Various Singular Potentials

It is shown from a simple scaling invariance that the ultra-relativistic Hamiltonian (mu=0) does not have bound states when the potential is Coulombic. This supplements the application of the relativistic virial theorem derived by Lucha and Schoeberl which shows that bound states do not exist for potentials more singular than the Coulomb potential.

hep-th↗

Increasing Potentials in Non-Abelian and Abelian Gauge Theories

An exact solution for an SU(2) Yang-Mills field coupled to a scalar field is given. This solution has potentials with a linear and Coulomb part. This may have some physical importance since many phenomenological QCD studies assume a linear plus Coulomb potential. Usually the linear potential is motivated with lattice gauge theory arguments. Here the linear potential is an exact result of the field equations. We also show that in the Nielsen-Olesen Abelian model there is an exact solution in the BPS limit which has a Coulomb-like electromagnetic field and a logarithmically rising scalar field. Both of these solutions must be cut off from above to avoid infinite field energy.

hep-th↗

A General Relativistic Model for Confinement in SU(2) Yang-Mills Theory

In this paper we present a model of confinement based on an analogy with the confinement mechanism of the Schwarzschild solution of general relativity. Using recently discovered exact, Schwarzschild-like solutions of the SU(2) Yang-Mills-Higgs equations we study the behaviour of a scalar, SU(2) charged test particle placed in the gauge fields of this solution. We find that this test particle is indeed confined inside the color event horizon of our solution. Additionally it is found that this system is a composite fermion even though there are no fundamental fermions in the original Lagrangian.

hep-th↗