韩国科学技术先进研究所

Compressed sensing MRI seeks to accelerate MRI acquisition processes by sampling fewer k-space measurements and then reconstructing the missing data algorithmically. The success of these approaches often relies on strong priors or learned statistical models. While recent diffusion model-based priors have shown great potential, previous methods typically ignore clinically available metadata (e.g. patient demographics, imaging parameters, slice-specific information). In practice, metadata contains meaningful cues about the anatomy and acquisition protocol, suggesting it could further constrain the reconstruction problem. In this work, we propose ContextMRI, a text-conditioned diffusion model for MRI that integrates granular metadata into the reconstruction process. We train a pixel-space diffusion model directly on minimally processed, complex-valued MRI images. During inference, metadata is converted into a structured text prompt and fed to the model via CLIP text embeddings. By conditioning the prior on metadata, we unlock more accurate reconstructions and show consistent gains across multiple datasets, acceleration factors, and undersampling patterns. Our experiments demonstrate that increasing the fidelity of metadata — ranging from slice location and contrast to patient age, sex, and pathology — systematically boosts reconstruction performance. This work highlights the untapped potential of leveraging clinical context for inverse problems and opens a new direction for metadata-driven MRI reconstruction. Code is available at https://github.com/DoHunLee1/ContextMRI.

Vision-language-action (VLA) models translate multimodal observations into low-level robot actions. During robot operation, each control period sets an inference deadline, and overruns leave the robot acting on stale observations, reducing task success. Meeting this deadline motivates on-device or nearby edge execution, where a single robot requires batch-1 inference outside the design point of LLM serving systems. Although VLA architectures combine familiar vision-language, autoregressive, and diffusion-style components, their runtime behavior in this batch-1 control setting remains uncharacterized. We characterize four representative VLA models on an edge GPU server and two onboard SoCs, using single-inference profiling and 43,200 closed-loop episodes. Action tensor dimensionality determines whether a stage is memory- or compute-bound, platform balance can shift that bottleneck, and GPU frequency scaling yields a platform-dependent energy-latency sweet spot. In closed-loop operation, overlapping inference with action execution creates an accuracy-speed-energy tradeoff, and no configuration is Pareto-dominant across deployment SLOs. These results guide joint design of VLA model architectures, hardware, and runtime policies.

Electric vertical take-off and landing vehicles (eVTOLs) can provide fast, potentially lower-carbon trunk services for long-distance urban commuting, but they require coordination with ground feeder modes because they cannot provide door-to-door service. Under fluctuating passenger arrivals, fixed timetables are easy to operate but inflexible, whereas load-based departure rules improve utilization but create unpredictable waiting. This study proposes an adaptive soft-target dispatch strategy that schedules eVTOL departures based on passenger accumulation, stranded demand, and vehicle availability. The integrated land-air dispatch problem is formulated as an event-driven, multi-line decision process and solved using multi-agent reinforcement learning. Numerical experiments show that the proposed strategy reduces average system cost by 22.57% relative to fixed timetables and 0.51% relative to full-match rules, with the largest gains under low-to-medium demand and medium-to-high values of time. A Chengdu airport case study further confirms its cost-saving potential.

Elemental co-doping of titanium dioxide (TiO2) has been recognized as one of the most promising strategies for tailoring its electronic band structure and extending its light absorption into the visible-light range. Understanding the structural modifications induced by doping and their corresponding effects on the electronic and optical properties of TiO2 is crucial for the rational design of efficient photocatalytic materials. Accordingly, this study systematically investigates the structural, electronic, and optical properties of pristine and Cu-N co-doped rutile TiO2 structures using density functional theory (DFT) calculations. For this purpose, the Hubbard corrected generalized gradient approximation(GGA+U) framework was employed using the CASTEP code. This study provides a systematic DFT+U investigation of Cu-N co-doped rutile TiO2, correlating its structural, electronic, and optical properties with a detailed analysis of the synergistic orbital hybridization between Cu-3d and N-2p states. The analysis provides insight into the resulting band edge modifications and the absence of deep trap states. The incorporation of Cu and N induces localized structural relaxation in the TiO2 lattice. Electronic structure calculations reveal a significant reduction in the band gap from approximately 2.95 eV for pristine TiO2 to approximately 2.17 eV for Cu-N co-doped TiO2, whereas the mono-doped systems exhibit intermediate band gaps of 2.49 eV for Cu-doped and 2.61 eV for N-doped TiO2, confirming that the co-doped system achieves the smallest band gap through a synergistic effect. This reduction is primarily attributed to the strong hybridization among the Cu-3d, N-2p, Ti-3d, and O-2p states near the band edges. Furthermore, the optical properties exhibit enhanced visible-light absorption, accompanied by a red shift in the absorption onset and improved dielectric response. Thus, it can be concluded that Cu-N co-doped TiO2 has considerable potential for visible-light-driven photocatalysis and solar-energy applications.

Abstract On-site quality assessment of precast concrete (PC) components is essential to ensure structural integrity, but precamber measurement relies on labor-intensive manual procedures due to time constraints and site congestion. This study proposes an automated precamber estimation framework for PC components using two-dimensional (2D) images. The proposed method integrates a reference-object-free geometric estimation, which utilizes predefined design specifications as ground truth, with vision transformer (ViT)-embedded architecture trained on hybrid synthetic data to robustly capture high-level features. Field validation demonstrated that the framework achieves an average measurement error of 1.91 mm, confirming a reliable performance comparable to manual methods. Notably, the automated process reduced inspection time by 79.82% and labor costs by 91.70% compared to traditional manual procedures. These results demonstrate the practical feasibility of the proposed framework for supporting rapid and efficient on-site quality assessment. Furthermore, the framework holds significant potential to extend into a comprehensive dimensional inspection system for managing various geometric deformations beyond precamber.