arXiv subjects

Sungjun Lee

Publications and source records attributed to Sungjun Lee.

5 recordsLinked to original sources

arXiv2025

Formation of bosonic $^{23}$Na$^{41}$K Feshbach molecules

Ultracold Feshbach molecules are a crucial intermediate step for the creation of quantum degenerate gases of strongly dipolar molecules. After coherent transfer to the rovibrational ground state, these dimers can realize stable dipolar gases with strong, tunable long-range interactions. Here, we report the creation of bosonic $^{23}$Na$^{41}$K Feshbach molecules by radio-frequency (RF) association. An RF pulse applied on the molecular side of an interspecies Feshbach resonance at 73.6(1)~G associates up to $1.1(1)\times10^4$ molecules from a thermal mixture of $^{23}$Na and $^{41}$K atoms. Measurements of the binding energy reveal a broad resonance width of 5.1(2)~G, facilitating robust control over interspecies interactions. The molecule lifetime in the presence of background atoms exceeds 2~ms, extending to 7~ms after removal of $^{23}$Na. These results constitute a key step toward the production of ultracold $^{23}$Na$^{41}$K ground state molecules for the exploration of novel many-body phenomena in strongly dipolar Bose gases.

cond-mat.quant-gas↗

arXiv2025

Short-Window Sliding Learning for Real-Time Violence Detection via LLM-based Auto-Labeling

This paper proposes a Short-Window Sliding Learning framework for real-time violence detection in CCTV footages. Unlike conventional long-video training approaches, the proposed method divides videos into 1-2 second clips and applies Large Language Model (LLM)-based auto-caption labeling to construct fine-grained datasets. Each short clip fully utilizes all frames to preserve temporal continuity, enabling precise recognition of rapid violent events. Experiments demonstrate that the proposed method achieves 95.25\% accuracy on RWF-2000 and significantly improves performance on long videos (UCF-Crime: 83.25\%), confirming its strong generalization and real-time applicability in intelligent surveillance systems.

cs.CV↗

arXiv2023

Dual-species Bose-Einstein condensates of $^{23}$Na and $^{41}$K with tunable interactions

We report the creation of dual-species Bose-Einstein condensates (BECs) of $^{23}$Na and $^{41}$K. Favorable background scattering lengths enable efficient sympathetic cooling of $^{41}$K via forced evaporative cooling of $^{23}$Na in a plugged magnetic trap and an optical dipole trap. The $1/e$ lifetime of the thermal mixture in the stretched hyperfine state exceeds 5 s in the presence of background scattering. At the end of evaporation, we create dual BECs in the immiscible phase, with about $3\times10^5$ $^{23}$Na atoms surrounding $5\times10^4$ $^{41}$K atoms. To further enable the tuning of the interspecies interaction strength, we locate multiple Feshbach resonances at magnetic fields up to 100 G. The broadest $s$-wave resonance located at 73.4(3) G features a favorable width of 1.8(2) G. This work sets the stage for the creation of ultracold gases of strongly dipolar bosonic $^{23}$Na$^{41}$K molecules as well as the exploration of many-body physics in bosonic $^{23}$Na-$^{41}$K mixtures.

cond-mat.quant-gas↗

arXiv2022

Noise-aware Learning from Web-crawled Image-Text Data for Image Captioning

Image captioning is one of the straightforward tasks that can take advantage of large-scale web-crawled data which provides rich knowledge about the visual world for a captioning model. However, since web-crawled data contains image-text pairs that are aligned at different levels, the inherent noises (e.g., misaligned pairs) make it difficult to learn a precise captioning model. While the filtering strategy can effectively remove noisy data, it leads to a decrease in learnable knowledge and sometimes brings about a new problem of data deficiency. To take the best of both worlds, we propose a Noise-aware Captioning (NoC) framework, which learns rich knowledge from the whole web-crawled data while being less affected by the noises. This is achieved by the proposed alignment-level-controllable captioner, which is learned using alignment levels of the image-text pairs as a control signal during training. The alignment-level-conditioned training allows the model to generate high-quality captions by simply setting the control signal to the desired alignment level at inference time. An in-depth analysis shows the effectiveness of our framework in handling noise. With two tasks of zero-shot captioning and text-to-image retrieval using generated captions (i.e., self-retrieval), we also demonstrate our model can produce high-quality captions in terms of descriptiveness and distinctiveness. The code is available at \url{https://github.com/kakaobrain/noc}.

cs.CV↗

arXiv2014

High Speed Friction Microscopy and Nanoscale Friction Coefficient Mapping

As mechanical devices in the nano/micro length scale are increasingly employed, it is crucial to understand nanoscale friction and wear especially at technically relevant sliding velocities. Accordingly, a novel technique has been developed for Friction Coefficient Mapping (FCM), leveraging recent advances in high speed AFM. The technique efficiently acquires friction versus force curves based on a sequence of images at a single location, each with incrementally lower loads. As a result, true maps of the coefficient of friction can be uniquely calculated for heterogeneous surfaces. These parameters are determined at a scan velocity as fast as 2 mm/s for microfabricated SiO2 mesas and Au coated pits, yielding results that are identical to traditional speed measurements despite being ~1000 times faster. To demonstrate the upper limit of sliding velocity for the custom setup, the friction properties of mica are reported from 200 μm/sec up to 2 cm/sec. While FCM is applicable to any AFM and scanning speed, quantitative nanotribology investigations of heterogeneous sliding or rolling components are therefore uniquely possible, even at realistic velocities for devices such as MEMS, biological implants, or data storage systems.

cond-mat.mtrl-sci↗