The shape-dependent aero- and hydro-dynamics found in nature have been adopted in a wide range of areas spanning from daily transportation to forefront biomedical research. Here, we report DNA-linked nanoparticle films exhibiting shape-dependent magnetic locomotion, controlled by DNA sequences. Fabricated through a DNA-directed layer-by-layer assembly of iron oxide and gold nanoparticles, the multifunctional films exhibit rotational and translational motions under magnetic fields, along with reversible shape morphing via DNA strand exchange reactions. Notably, the shape of the film significantly influences its magnetic responsiveness, attributable to shape-dependent drag forces acting on mesoscopic films. The distinctive shape dependence combined with the shape-changing capability offers an approach to regulate magnetic locomotion within a constant magnetic field, as demonstrated here through the go and stop motion of nanoparticle films without altering the magnetic field.

해양 사고는 인명과 환경에 심각한 피해를 초래할 수 있다 이러한 사고에 신속하고 효과적으로 대응하기 위해서는 부유체의 이동 경로를 정확하게 예측하는 것이 필수적이다 특히, 해양 구조 및 오염 방지를 위한 즉각적인 대응이 요구되는 상황에서는 바람 데이터의 정확한 예측이 중요한 역할을 한다. 본 연구에서는 부유체 경로 예측에 중요한 바람 데이터를 예측하기 위해 딥러닝 기반의 Maximum Absolute Scaler를 적용한 일차원 합성곱 신경망 (1D-CNN) 모델의 활용을 제안하였다. 이를 위해 한반도 인근 7개 관측소에서 수집된 데이터를 전처리하여 모델 학습에 사용하였다. 실험 결과 제안된 모델은 기존 수치 모델 대비 RMSE에서 약 15.8%, MAE에서 12.22%, Adjusted R² 값에서 18.25% 향상된 성능을 보였다. 이는 제안한 모델이 실시간 해양 사고 대응 시스템의 정확도를 높이는 데 기여할 수 있음을 시사한다.

A genetic algorithm (GA) evolves the population of candidate solutions to a particular problem through crossover and mutation operators. Since a newly generated solution may not satisfy the constraints of the given problem, it is common to penalize infeasible solutions or to repair them to feasible solutions. For the minimum vertex cover (MVC) problem, we propose an adaptive greedy repair operator. Our repair operator first repairs an infeasible solution into a feasible one using a randomized greedy algorithm, and then removes unnecessary vertices from the feasible vertex cover, making it a minimal vertex cover. During the repair process, when adding or removing vertices, the degree of exploration and exploitation in the randomized greedy algorithm is adaptively adjusted based on the convergence level of the population. When the population lacks high-quality solutions, the operator strives to generate superior solutions greedily. However, when the solution set has enough high-quality solutions, it explores unexplored choices to break through the limitations of the existing solution set. We compared our GA with a deterministic greedy algorithm, a randomized greedy algorithm, and GAs using various repair operators. Experimental results on benchmark graphs from the Benchmarks with Hidden Optimum Solutions Library (BHOSLIB) demonstrated that the proposed repair operator improved the performance of the GA for the MVC problem.

본 연구는 국립현대미술관의 소장품 데이터를 활용하여 도메인 특화 질의응답(QA)에서의 파인튜닝(Fine-Tuning)과 검색 증강 생성(Retrieval-Augmented Generation, RAG) 기법을 비교했다. Gemma 3 언어모델을 대상으로 저랭크 적응(Low-Rank Adaptation, LoRA)을 기반으로 한 매개변수 효율적 파인튜닝(Parameter-Efficient Fine-Tuning, PEFT)을 수행하고, 이와 비교하기 위해 벡터 데이터베이스 기반의 RAG 시스템을 구축하여 50개의 QA쌍으로 각 모델들을 평가했다. 평가 지표로는 응답 관련성(Response Relevance), 사실적 정확성(Factual Correctness), 의미적 유사성(Semantic Similarity), 답변 정확성(Answer Correctness)을 사용했다. 실험 결과 RAG 시스템은 응답 관련성을 제외한 모든 지표에서 파인튜닝 및 기본 모델보다 우수했으며, 특히 사실적 정확성과 답변 정확성에서 큰 성능향상을 보여줬다. 이는 RAG 시스템이 도메인 지식을 활용한 QA 시스템을 구축하기에 적합한 방식임을 시사한다.

Maritime accidents can cause significant damage to both human life and the environment. To respond quickly and effectively, accurate wind prediction is crucial. This study focuses on optimizing the weights of an ensemble voting model for wind data prediction using genetic algorithms (GAs). The GA-based approach significantly improved wind prediction accuracy by optimizing the weights of various models. Data from seven observation stations near South Korea was used, and the GA-based model outperformed the ECMWF High-Resolution model, achieving a 24.6% improvement in RMSE. This highlights the potential of GA in wind prediction models and its significant application in maritime accident response systems.

본 논문은 예술 작품 데이터에 특화된 언어모델 개발을 목표로 한다. 기존 언어모델의 도메인 한계를 극복하고 구조화된 출력 형식을 지원하기 위해, 파라미터 효율적 미세조정(Parameter-Efficient Fine-Tuning, PEFT) 기법을 적용하여 데이터 전처리 및 모델 튜닝을 수행하였다. 또한, 적절한 기반 모델을 선정하기 위해 다양한 상용 및 오픈소스 언어모델의 성능을 비교하고, 훈련 데이터셋의 형식이 출력의 품질에 미치는 영향을 분석하였다. 실험 결과, Llama 3.1 기반의 Instruct 모델에 마크다운 형식의 데이터셋을 적용하여 훈련했을 때 가장 우수한 성능을 보였지만 파인튜닝만으로는 도메인 특화 언어모델 개발에 한계가 있음을 확인할 수 있었다.

Air pollution poses major risks to public health, driving the adoption of low-cost sensor (LCS) networks for fine-grained and real-time monitoring. However, the variable accuracy of LCS data compared with reference instruments necessitates robust quality control (QC) frameworks. Over the past decade, machine learning (ML) has emerged as a powerful tool to calibrate sensors, detect anomalies, and mitigate drift in large-scale deployment. This survey reviews advances in three methodological categories: traditional ML models, deep learning architectures, and hybrid or unsupervised methods. We also examine spatiotemporal QC frameworks that exploit redundancies across time and space, as well as real-time implementations based on edge–cloud architectures. Applications include personal exposure monitoring, integration with atmospheric simulations, and support for policy decision making. Despite these achievements, several challenges remain. Traditional models are lightweight but often fail to generalize across contexts, while deep learning models achieve higher accuracy but demand large datasets and remain difficult to interpret. Spatiotemporal approaches improve robustness but face scalability constraints, and real-time systems must balance computational efficiency with accuracy. Broader adoption will also require clear standards, reliable uncertainty quantification, and sustained trust in corrected data. In summary, ML-based QC shows strong potential but is still constrained by data quality, transferability, and governance gaps. Future work should integrate physical knowledge with ML, leverage federated learning for scalability, and establish regulatory benchmarks. Addressing these challenges will enable ML-driven QC to deliver reliable, high-resolution data that directly support science-based policy and public health.