arXiv subjects

Sang Min Lee

Publications and source records attributed to Sang Min Lee.

At least 19 recordsLinked to original sources

arXiv2025

MedErr-CT: A Visual Question Answering Benchmark for Identifying and Correcting Errors in CT Reports

Computed Tomography (CT) plays a crucial role in clinical diagnosis, but the growing demand for CT examinations has raised concerns about diagnostic errors. While Multimodal Large Language Models (MLLMs) demonstrate promising comprehension of medical knowledge, their tendency to produce inaccurate information highlights the need for rigorous validation. However, existing medical visual question answering (VQA) benchmarks primarily focus on simple visual recognition tasks, lacking clinical relevance and failing to assess expert-level knowledge. We introduce MedErr-CT, a novel benchmark for evaluating medical MLLMs' ability to identify and correct errors in CT reports through a VQA framework. The benchmark includes six error categories - four vision-centric errors (Omission, Insertion, Direction, Size) and two lexical error types (Unit, Typo) - and is organized into three task levels: classification, detection, and correction. Using this benchmark, we quantitatively assess the performance of state-of-the-art 3D medical MLLMs, revealing substantial variation in their capabilities across different error types. Our benchmark contributes to the development of more reliable and clinically applicable MLLMs, ultimately helping reduce diagnostic errors and improve accuracy in clinical practice. The code and datasets are available at https://github.com/babbu3682/MedErr-CT.

cs.CV↗

arXiv2024

Harmful Suicide Content Detection

Harmful suicide content on the Internet is a significant risk factor inducing suicidal thoughts and behaviors among vulnerable populations. Despite global efforts, existing resources are insufficient, specifically in high-risk regions like the Republic of Korea. Current research mainly focuses on understanding negative effects of such content or suicide risk in individuals, rather than on automatically detecting the harmfulness of content. To fill this gap, we introduce a harmful suicide content detection task for classifying online suicide content into five harmfulness levels. We develop a multi-modal benchmark and a task description document in collaboration with medical professionals, and leverage large language models (LLMs) to explore efficient methods for moderating such content. Our contributions include proposing a novel detection task, a multi-modal Korean benchmark with expert annotations, and suggesting strategies using LLMs to detect illegal and harmful content. Owing to the potential harm involved, we publicize our implementations and benchmark, incorporating an ethical verification process.

cs.CY↗

arXiv2024

Estimation of photon number distribution and derivative characteristics of photon-pair sources

The evaluation of a photon-pair source employs characteristic metrics like the photon-pair generation rate, heralding efficiency, and second-order correlation function, all of which are determined by the photon number distribution of the source. The photon number distribution, however, can be altered due to spectral or spatial filtering and optical losses, leading to changes in the above characteristics. In this paper, we theoretically describe the effects of different filterings, losses, and noise counts on the photon number distribution and related characteristics. From the theoretical description, an analytic expression for the effective mode number of the joint spectral density is also derived. Compared with previous methods for estimating the photon number distribution and characteristics, an improved methodology is introduced along with a suitable metric of accuracy for estimating the photon number distribution, focusing on photon-pair sources. We discuss the accuracy of the calculated characteristics from the estimated (or reconstructed) photon number distribution through repeated simulations and bootstrapped experimental data.

quant-ph↗

arXiv2023

Quantum Optical Induced-Coherence Tomography by a Hybrid Interferometer

Quantum interferometry based on induced-coherence phenomena has demonstrated the possibility of undetected-photon measurements. Perturbation in the optical path of probe photons can be detected by interference signals generated by quantum mechanically correlated twin photons propagating through a different path, possibly at a different wavelength. To the best of our knowledge, this work demonstrates for the first time a hybrid-type induced-coherence interferometer that incorporates a Mach-Zehnder-type interferometer for visible photons and a Michelson-type interferometer for infrared photons, based on double-pass pumped spontaneous parametric down-conversion. This configuration enables infrared optical measurements via the detection of near-visible photons and provides methods for characterizing the quality of measurements by identifying photon pairs of different origins. The results verify that the induced-coherence interference visibility is approximately the same as the heralding efficiencies between twin photons along the relevant spatial modes. Applications to both time-domain and frequency-domain quantum-optical induced-coherence tomography for three-dimensional test structures are demonstrated. The results prove the feasibility of practical undetected-photon sensing and imaging techniques based on the presented structure.

quant-ph↗

arXiv2022

AI can evolve without labels: self-evolving vision transformer for chest X-ray diagnosis through knowledge distillation

Although deep learning-based computer-aided diagnosis systems have recently achieved expert-level performance, developing a robust deep learning model requires large, high-quality data with manual annotation, which is expensive to obtain. This situation poses the problem that the chest x-rays collected annually in hospitals cannot be used due to the lack of manual labeling by experts, especially in deprived areas. To address this, here we present a novel deep learning framework that uses knowledge distillation through self-supervised learning and self-training, which shows that the performance of the original model trained with a small number of labels can be gradually improved with more unlabeled data. Experimental results show that the proposed framework maintains impressive robustness against a real-world environment and has general applicability to several diagnostic tasks such as tuberculosis, pneumothorax, and COVID-19. Notably, we demonstrated that our model performs even better than those trained with the same amount of labeled data. The proposed framework has a great potential for medical imaging, where plenty of data is accumulated every year, but ground truth annotations are expensive to obtain.

eess.IV↗

arXiv2021

Vision Transformer using Low-level Chest X-ray Feature Corpus for COVID-19 Diagnosis and Severity Quantification

Developing a robust algorithm to diagnose and quantify the severity of COVID-19 using Chest X-ray (CXR) requires a large number of well-curated COVID-19 datasets, which is difficult to collect under the global COVID-19 pandemic. On the other hand, CXR data with other findings are abundant. This situation is ideally suited for the Vision Transformer (ViT) architecture, where a lot of unlabeled data can be used through structural modeling by the self-attention mechanism. However, the use of existing ViT is not optimal, since feature embedding through direct patch flattening or ResNet backbone in the standard ViT is not intended for CXR. To address this problem, here we propose a novel Vision Transformer that utilizes low-level CXR feature corpus obtained from a backbone network that extracts common CXR findings. Specifically, the backbone network is first trained with large public datasets to detect common abnormal findings such as consolidation, opacity, edema, etc. Then, the embedded features from the backbone network are used as corpora for a Transformer model for the diagnosis and the severity quantification of COVID-19. We evaluate our model on various external test datasets from totally different institutions to evaluate the generalization capability. The experimental results confirm that our model can achieve the state-of-the-art performance in both diagnosis and severity quantification tasks with superior generalization capability, which are sine qua non of widespread deployment.

eess.IV↗

arXiv2021

Vision Transformer for COVID-19 CXR Diagnosis using Chest X-ray Feature Corpus

Under the global COVID-19 crisis, developing robust diagnosis algorithm for COVID-19 using CXR is hampered by the lack of the well-curated COVID-19 data set, although CXR data with other disease are abundant. This situation is suitable for vision transformer architecture that can exploit the abundant unlabeled data using pre-training. However, the direct use of existing vision transformer that uses the corpus generated by the ResNet is not optimal for correct feature embedding. To mitigate this problem, we propose a novel vision Transformer by using the low-level CXR feature corpus that are obtained to extract the abnormal CXR features. Specifically, the backbone network is trained using large public datasets to obtain the abnormal features in routine diagnosis such as consolidation, glass-grass opacity (GGO), etc. Then, the embedded features from the backbone network are used as corpus for vision transformer training. We examine our model on various external test datasets acquired from totally different institutions to assess the generalization ability. Our experiments demonstrate that our method achieved the state-of-art performance and has better generalization capability, which are crucial for a widespread deployment.

eess.IV↗

arXiv2021

Severity Quantification and Lesion Localization of COVID-19 on CXR using Vision Transformer

Under the global pandemic of COVID-19, building an automated framework that quantifies the severity of COVID-19 and localizes the relevant lesion on chest X-ray images has become increasingly important. Although pixel-level lesion severity labels, e.g. lesion segmentation, can be the most excellent target to build a robust model, collecting enough data with such labels is difficult due to time and labor-intensive annotation tasks. Instead, array-based severity labeling that assigns integer scores on six subdivisions of lungs can be an alternative choice enabling the quick labeling. Several groups proposed deep learning algorithms that quantify the severity of COVID-19 using the array-based COVID-19 labels and localize the lesions with explainability maps. To further improve the accuracy and interpretability, here we propose a novel Vision Transformer tailored for both quantification of the severity and clinically applicable localization of the COVID-19 related lesions. Our model is trained in a weakly-supervised manner to generate the full probability maps from weak array-based labels. Furthermore, a novel progressive self-training method enables us to build a model with a small labeled dataset. The quantitative and qualitative analysis on the external testset demonstrates that our method shows comparable performance with radiologists for both tasks with stability in a real-world application.

eess.IV↗

arXiv2020

Experimental Learning of Pure Quantum States using Sequential Single-Shot Measurement Outcomes

We experimentally implement a machine-learning method for accurately identifying unknown pure quantum states. The method, called single-shot measurement learning, achieves the theoretical optimal accuracy for $ε= O(N^{-1})$ in state learning and reproduction, where $ε$ and $N$ denote the infidelity and number of state copies, without employing computationally demanding tomographic methods. This merit results from the inclusion of weighted randomness in the learning rule governing the exploration of diverse learning routes. We experimentally verify the advantages of our scheme by using a linear-optics setup to prepare and measure single-photon polarization qubits. The experimental results show highly accurate state learning and reproduction exhibiting infidelity of $O(N^{-0.983})$ down to $10^{-5}$, without estimation of the experimental parameters.

quant-ph↗

arXiv2018

Quantum teleportation of shared quantum secret

Quantum teleportation allows to transfer unknown quantum states between distant parties. It is not only a primitive of quantum communications but also an essential task in realization of the quantum networks for promising applications such as quantum cryptography and distributed quantum computation. Despite recent substantial progresses in the realization of quantum communications, teleportation of shared quantum information between multiple senders and receivers is still missing. Here we propose and experimentally demonstrate quantum teleportation among spatially separate parties in a quantum network. The protocol can be jointly performed by distributed participants, while none of them can fully access the transferred information. It can be generally extended to quantum communications incorporating an arbitrary number of parties without a trusted central node. Our work opens a route to the realization of distributed quantum communications and computations in quantum networks

quant-ph↗

arXiv2018

Simultaneous measurement of HOM-like and N00N-like two-photon wavepackets in a Mach-Zehnder interferometer

We propose and experimentally demonstrate two-photon interference effects in a Mach-Zehnder interferometer in which two different kinds of two-photon states are prepared by introducing a time delay between the two input photons. The two-photon states are simultaneously prepared in a symmetrically superposed state with two temporally separated photons in two different spatial modes or in the same spatial mode within the interferometer. We observe two-photon interference fringes involving both the Hong-Ou-Mandel interference effect and the interference of path-entangled two-photon states simultaneously in a single interferometric setup. The observed two-photon interference effect can provide a simultaneous observation of the interferometric properties of the single-photon and two-photon wavepackets

quant-ph↗

arXiv2018

CHSH inequality test via disturbance-free measurement

We propose disturbance-free measurement using a "weak-value" scheme, in which a weakly measured quantum system is post-selected (to the initial state) to confirm that there is no disturbance. The probability of obtaining the non-disturbed state is asymptotically close to unity. We theoretically show that outcomes of the disturbance-free measurement for a two-qubit state satisfy the Clauser--Horne--Shimony--Holt inequality. We experimentally demonstrate the test for a typical (maximally entangled) two-qubit state based on a linear optical system. In experiments, polarization-entangled photon-pairs generated by the spontaneous parametric down-conversion process are measured by instruments such as strength-variable polarization-measurement apparatuses and a fiber-based Bell state analyzer.

quant-ph↗

arXiv2018

Learning unknown pure quantum states

We propose a learning method for estimating unknown pure quantum states. The basic idea of our method is to learn a unitary operation $\hat{U}$ that transforms a given unknown state $|ψ_τ\rangle$ to a known fiducial state $|f\rangle$. Then, after completion of the learning process, we can estimate and reproduce $|ψ_τ\rangle$ based on the learned $\hat{U}$ and $|f\rangle$. To realize this idea, we cast a random-based learning algorithm, called `single-shot measurement learning,' in which the learning rule is based on an intuitive and reasonable criterion: the greater the number of success (or failure), the less (or more) changes are imposed. Remarkably, the learning process occurs by means of a single-shot measurement outcome. We demonstrate that our method works effectively, i.e., the learning is completed with a {\em finite} number, say $N$, of unknown-state copies. Most surprisingly, our method allows the maximum statistical accuracy to be achieved for large $N$, namely $\simeq O(N^{-1})$ scales of average infidelity. This result is comparable to those yielded from the standard quantum tomographic method in the case where additional information is available. It highlights a non-trivial message, that is, a random-based adaptive strategy can potentially be as accurate as other standard statistical approaches.

quant-ph↗

arXiv2017

Two-photon interference of polarization-entangled photons in a Franson interferometer

We present two-photon interference experiments with polarization-entangled photon pairs in a polarization-based Franson-type interferometer. Although the two photons do not meet at a common beamsplitter, a phase-insensitive Hong-Ou-Mandel type two-photon interference peak and dip fringes are observed, resulting from the two-photon interference effect between two indistinguishable two-photon probability amplitudes leading to a coincidence detection. A spatial quantum beating fringe is also measured for nondegenerate photon pairs in the same interferometer, although the two-photon states have no frequency entanglement. When unentangled polarization-correlated photons are used as an input state, the polarization entanglement is successfully recovered through the interferometer via delayed compensation.

quant-ph↗

arXiv2017

Observation of two-photon interference effect with single non-photon-number resolving detector

Multiphoton interference effects can be measured with a single detector when two input photons are temporally well separated when compared with the dead time of the single-photon avalanche detector. Here we experimentally demonstrate that the Hong-Ou-Mandel interference effect can be observed with a single non-photon-number resolving detector via a time-delayed coincidence measurement of successive electrical signals from the detector. The two-photon interference experiment is performed by utilizing temporally well-separated pairwise weak coherent pulses and the interference fringes are successfully measured with a high visibility in the range of the limited upper bound for the weak coherent photon source.

quant-ph↗

arXiv2016

Two-photon interference of temporally separated photons

We present experimental demonstrations of two-photon interference involving temporally separated photons within two types of interferometers: a Mach-Zehnder interferometer and a polarization-based Michelson interferometer. The two-photon states are probabilistically prepared in a symmetrically superposed state within the two interferometer arms by introducing a large time delay between two input photons; this state is composed of two temporally separated photons, which are in two different or the same spatial modes. We then observe two-photon interference fringes involving both the Hong-Ou-Mandel interference effect and the interference of path-entangled two-photon states simultaneously in a single interferometric setup. The observed two-photon interference fringes provide simultaneous observation of the interferometric properties of the single-photon and two-photon wavepackets. The observations can also facilitate a more comprehensive understanding of the origins of the interference phenomena arising from spatially bunched/anti-bunched two-photon states comprised of two temporally separated photons within the interferometer arms.

quant-ph↗

arXiv2016

Highly bright photon-pair generation in Doppler-broadened ladder-type atomic system

We report a bright photon-pair source with a coincidence counting rate per input power (cps/mW) of tens of thousands, obtained via spontaneous four-wave mixing from a Doppler-broadened atomic ensemble of the 5S1/2-5P3/2-5D5/2 transition of 87Rb. The photon-pair generation rate is enhanced by the two-photon coherence contributions from almost all the atomic velocity groups in the Doppler-broadened ladder-type atomic system. We obtained the violation of the Cauchy-Schwarz inequality by a factor of 2370(150). We believe that our scheme for highly bright paired photons is important as a useful quantum light source for quantum entanglement swapping between completely autonomous sources.

quant-ph↗

arXiv2016

Superradiant Emission of Ultra-Bright Photon Pairs in Doppler-Broadened Atomic Ensemble

With a recent rising interest of single photon superradiance due to its potential usefulness for efficient collection of single photon from an atomic ensemble, bright and narrow photon pair source is a key component in realization of quantum communication and quantum computer based on coherent interaction between light and atomic ensemble. We report the superradiant emission of ultra-bright photon pairs with a coincidence counting rate per input power of 64,600 cps/mW via spontaneous four-wave mixing in a thermal vapour cell. The photon-pair generation rate is enhanced by the contribution of two-photon coherence of almost atomic velocity groups in the Doppler-broadened ladder-type atomic system. The quadratic proportionality of the probability of detecting a heralded single photon as a function of the optical depth clarifies that the ultra-brightness results from the superradiance. In addition, a single photon superradiant beating at a high optical depth of the atomic ensemble is observed for the first time.

quant-ph↗