arXiv subjects

Hsin-Yu Wu

Publications and source records attributed to Hsin-Yu Wu.

10 recordsLinked to original sources

arXiv2025

Med-RewardBench: Benchmarking Reward Models and Judges for Medical Multimodal Large Language Models

Multimodal large language models (MLLMs) hold significant potential in medical applications, including disease diagnosis and clinical decision-making. However, these tasks require highly accurate, context-sensitive, and professionally aligned responses, making reliable reward models and judges critical. Despite their importance, medical reward models (MRMs) and judges remain underexplored, with no dedicated benchmarks addressing clinical requirements. Existing benchmarks focus on general MLLM capabilities or evaluate models as solvers, neglecting essential evaluation dimensions like diagnostic accuracy and clinical relevance. To address this, we introduce Med-RewardBench, the first benchmark specifically designed to evaluate MRMs and judges in medical scenarios. Med-RewardBench features a multimodal dataset spanning 13 organ systems and 8 clinical departments, with 1,026 expert-annotated cases. A rigorous three-step process ensures high-quality evaluation data across six clinically critical dimensions. We evaluate 32 state-of-the-art MLLMs, including open-source, proprietary, and medical-specific models, revealing substantial challenges in aligning outputs with expert judgment. Additionally, we develop baseline models that demonstrate substantial performance improvements through fine-tuning.

cs.CL↗

arXiv2025

Monitoring Single DNA Docking Site Activity With Sequential Modes of an Optoplasmonic Whispering-Gallery Mode Biosensor

In recent years, there has been rapid advancement in single-molecule techniques, driven by their unparalleled precision in studying molecules whose sizes are beyond the diffraction limit. Among these techniques, optoplasmonic whispering gallery mode sensing has demonstrated great potential in label-free single-molecule characterization. It combines the principles of localized surface plasmon resonance (LSPR) and whispering gallery mode (WGM) sensing, offering exceptional sensing capabilities, even at the level of single ions. However, current optoplasmonic WGM sensing operates in a multiplexed channel, making it challenging to focus on individual binding sites of analyte molecules. In this article, we characterize different binding sites of DNA analyte molecules hybridizing to docking strands on the optoplasmonic WGM sensor, using the ratio of the resonance shift between sequential polar WGM modes. We identify specific docking sites that undergo transient interactions and eventually hybridize with the complementary analyte strands permanently.

physics.bio-ph↗

arXiv2025

Quantum algorithm for solving nonlinear differential equations based on physics-informed effective Hamiltonians

We propose a distinct approach to solving linear and nonlinear differential equations (DEs) on quantum computers by encoding the problem into ground states of effective Hamiltonian operators. Our algorithm relies on constructing such operators in the Chebyshev space, where an effective Hamiltonian is a sum of global differential and data constraints. Once the effective Hamiltonian is formed, solutions of differential equations can be obtained using the ground state preparation techniques (e.g. imaginary-time evolution and quantum singular value transformation), bypassing variational search. Unlike approaches based on discrete grids, the algorithm enables evaluation of solutions beyond fixed grid points and implements constraints in the physics-informed way. Our proposal inherits the best traits from quantum machine learning-based DE solving (compact basis representation, automatic differentiation, nonlinearity) and quantum linear algebra-based approaches (fine-grid encoding, provable speed-up for state preparation), offering a robust strategy for quantum scientific computing in the early fault-tolerant era.

quant-ph↗

arXiv2025

Quantum Chebyshev Probabilistic Models for Fragmentation Functions

Quantum generative modeling is emerging as a powerful tool for advancing data analysis in high-energy physics, where complex multivariate distributions are common. However, efficiently learning and sampling these distributions remains challenging. We propose a quantum protocol for a bivariate probabilistic model based on shifted Chebyshev polynomials, trained as a circuit-based representation of two correlated variables, with sampling performed via quantum Chebyshev transforms. As a key application we study fragmentation functions (FFs) of charged pions and kaons from single-inclusive hadron production in electron-positron annihilation. We learn the joint distribution of momentum fraction $z$ and energy scale $Q$, and infer their correlations from the entanglement structure. Building on the generalization capabilities of the quantum model and extended register architecture, we perform fine-grid multivariate sampling for FF dataset augmentation. Our results highlight the growing potential of quantum generative modeling to advance data analysis and scientific discovery in high-energy physics.

quant-ph↗

arXiv2024

Towards a Holistic Framework for Multimodal Large Language Models in Three-dimensional Brain CT Report Generation

Multi-modal large language models (MLLMs) have been given free rein to explore exciting medical applications with a primary focus on radiology report generation. Nevertheless, the preliminary success in 2D radiology captioning is incompetent to reflect the real-world diagnostic challenge in the volumetric 3D anatomy. To mitigate three crucial limitation aspects in the existing literature, including (1) data complexity, (2) model capacity, and (3) evaluation metric fidelity, we collected an 18,885 text-scan pairs 3D-BrainCT dataset and applied clinical visual instruction tuning (CVIT) to train BrainGPT models to generate radiology-adherent 3D brain CT reports. Statistically, our BrainGPT scored BLEU-1 = 44.35, BLEU-4 = 20.38, METEOR = 30.13, ROUGE-L = 47.6, and CIDEr-R = 211.77 during internal testing and demonstrated an accuracy of 0.91 in captioning midline shifts on the external validation CQ500 dataset. By further inspecting the captioned report, we reported that the traditional metrics appeared to measure only the surface text similarity and failed to gauge the information density of the diagnostic purpose. To close this gap, we proposed a novel Feature-Oriented Radiology Task Evaluation (FORTE) to estimate the report's clinical relevance (lesion feature and landmarks). Notably, the BrainGPT model scored an average FORTE F1-score of 0.71 (degree=0.661; landmark=0.706; feature=0.693; impression=0.779). To demonstrate that BrainGPT models possess objective readiness to generate human-like radiology reports, we conducted a Turing test that enrolled 11 physician evaluators, and around 74% of the BrainGPT-generated captions were indistinguishable from those written by humans. Our work embodies a holistic framework that showcased the first-hand experience of curating a 3D brain CT dataset, fine-tuning anatomy-sensible language models, and proposing robust radiology evaluation metrics.

cs.CL↗

arXiv2024

Multidimensional Quantum Generative Modeling by Quantum Hartley Transform

We develop an approach for building quantum models based on the exponentially growing orthonormal basis of Hartley kernel functions. First, we design a differentiable Hartley feature map parametrized by real-valued argument that enables quantum models suitable for solving stochastic differential equations and regression problems. Unlike the naturally complex Fourier encoding, the proposed Hartley feature map circuit leads to quantum states with real-valued amplitudes, introducing an inductive bias and natural regularization. Next, we propose a quantum Hartley transform circuit as a map between computational and Hartley basis. We apply the developed paradigm to generative modeling from solutions of stochastic differential equations, and utilize the quantum Hartley transform for fine sampling from parameterized distributions through an extended register. Finally, we present tools for implementing multivariate quantum generative modeling for both correlated and uncorrelated distributions. As a result, the developed quantum Hartley models offer a distinct quantum approach to generative AI at increasing scale.

quant-ph↗

arXiv2024

DeNVeR: Deformable Neural Vessel Representations for Unsupervised Video Vessel Segmentation

This paper presents Deformable Neural Vessel Representations (DeNVeR), an unsupervised approach for vessel segmentation in X-ray angiography videos without annotated ground truth. DeNVeR utilizes optical flow and layer separation techniques, enhancing segmentation accuracy and adaptability through test-time training. Key contributions include a novel layer separation bootstrapping technique, a parallel vessel motion loss, and the integration of Eulerian motion fields for modeling complex vessel dynamics. A significant component of this research is the introduction of the XACV dataset, the first X-ray angiography coronary video dataset with high-quality, manually labeled segmentation ground truth. Extensive evaluations on both XACV and CADICA datasets demonstrate that DeNVeR outperforms current state-of-the-art methods in vessel segmentation accuracy and generalization capability while maintaining temporal coherency.

cs.CV↗

arXiv2023

Quantum Chebyshev Transform: Mapping, Embedding, Learning and Sampling Distributions

We develop a paradigm for building quantum models in the orthonormal space of Chebyshev polynomials. We show how to encode data into quantum states with amplitudes being Chebyshev polynomials with degree growing exponentially in the system size. Similar to the quantum Fourier transform which maps computational basis space into the phase (Fourier) basis, we describe the quantum circuit for the mapping between computational and Chebyshev spaces. We propose an embedding circuit for generating the orthonormal Chebyshev basis of exponential capacity, represented by a continuously-parameterized shallow isometry. This enables automatic quantum model differentiation, and opens a route to solving stochastic differential equations. We apply the developed paradigm to generative modeling from physically- and financially-motivated distributions, and use the quantum Chebyshev transform for efficient sampling of these distributions in extended computational basis.

quant-ph↗

arXiv2022

Anomalous DNA hybridisation kinetics on gold nanorods revealed via a dual single-molecule imaging and optoplasmonic sensing platform

Observing the hybridisation kinetics of DNA probes immobilised on plasmonic nanoparticles is key in plamon enhanced fluorescence detection from weak emitting species, and refractive index based single-molecule detection on optoplasmonic sensors. The role of the local field in providing plasmonic signal enhancements for single-molecule detection has been studied in great detail. Nevertheless, few studies have compared the experimental results in both techniques for single-molecule studies. Here we developed the first optical setup that integrates optoplasmonic and DNA-PAINT based detection of oligonucleotides to compare these sub-platforms and provide complementary insights into single-molecule processes. We record the fluorescence and optoplasmonic sensor signals for individual, transient hybridisation events. The hybridisation events are observed in the same sample cell and over a prolonged time (i.e. towards high binding site occupancies). A decrease in the association rate over the measurement duration is reported. Our dual optoplasmonic sensing and imaging platform offers insight into the observed phenomenon, revealing that irreversible hybridisation events accumulate over detected step signals in optoplasmonic sensing. Our results point to novel physicochemical mechanisms that result in the stabilisation of DNA hybridisation on optically-excited plasmonic nanoparticles.

physics.optics↗

arXiv2021

Comparing transient oligonucleotide hybridization kinetics using DNA-PAINT and optoplasmonic single-molecule sensing on gold nanorods

We report a comparison of two photonic techniques for single-molecule sensing: fluorescence nanoscopy and optoplasmonic sensing. As the test system, oligonucleotides with and without fluorescent labels are transiently hybridized to complementary docking strands attached to gold nanorods. Comparing the measured single-molecule kinetics helps to examine the influence of fluorescent labels as well as factors arising from different sensing geometries. Our results demonstrate that DNA dissociation is not significantly altered by the fluorescent label, while DNA association is affected by geometric factors in the two techniques. These findings open the door to exploiting plasmonic sensing and fluorescence nanoscopy in a complementary fashion, which will aid in building more powerful sensors and uncovering the intricate effects that influence the behavior of single molecules.

physics.optics↗