GIST Scholar
Not a member yet
30271 research outputs found
Sort by
AMPED: Adaptive Multi-objective Projection for balancing Exploration and skill Diversification
No full textSkill-based reinforcement learning (SBRL) enables rapid adaptation in environments with sparse rewards by pretraining a skill-conditioned policy. Effective skill learning requires jointly maximizing both exploration and skill diversity. However, existing methods often face challenges in simultaneously optimizing for these two conflicting objectives. In this work, we propose a new method, Adaptive Multi-objective Projection for balancing Exploration and skill Diversification (AMPED), which explicitly addresses both: during pre-training, a gradient-surgery projection balances the exploration and diversity gradients, and during fine-tuning, a skill selector exploits the learned diversity by choosing skills suited to downstream tasks. Our approach achieves performance that surpasses SBRL baselines across various benchmarks. Through an extensive ablation study, we identify the role of each component and demonstrate that each element in AMPED is contributing to performance. We further provide theoretical and empirical evidence that, with a greedy skill selector, greater skill diversity reduces fine-tuning sample complexity. These results highlight the importance of explicitly harmonizing exploration and diversity and demonstrate the effectiveness of AMPED in enabling robust and generalizable skill learning
Design of Virtual Driving Test Environment for Collecting and Validating Bad Weather SiLS Data Based on Multi-Source Images Using DCU with V2X-Car Edge Cloud
No full textIn real-world autonomous driving tests, unexpected events such as pedestrians or wild animals suddenly entering the driving path can occur. Conducting actual test drives under various weather conditions may also lead to dangerous situations. Furthermore, autonomous vehicles may operate abnormally in bad weather due to limitations of their sensors and GPS. Driving simulators, which replicate driving conditions nearly identical to those in the real world, can drastically reduce the time and cost required for market entry validation; consequently, they have become widely used. In this paper, we design a virtual driving test environment capable of collecting and verifying SiLS data under adverse weather conditions using multi-source images. The proposed method generates a virtual testing environment that incorporates various events, including weather, time of day, and moving objects, that cannot be easily verified in real-world autonomous driving tests. By setting up scenario-based virtual environment events, multi-source image analysis and verification using real-world DCUs (Data Concentrator Units) with V2X-Car edge cloud can effectively address risk factors that may arise in real-world situations. We tested and validated the proposed method with scenarios employing V2X communication and multi-source image analysis. © © 2025 The Authors.TRUEsciescopu
Prolonged daytime presence and oxidative impact of nitryl chloride, ClNO2, in winter urban environment☆
No full textNitryl chloride (ClNO2) is widely recognized as a nocturnally formed species that influences next-day air quality through early-morning photolysis. However, its formation mechanisms and diurnal behavior remain poorly constrained, particularly its persistence beyond the morning. Here, we present the first wintertime observations of ClNO2 in South Korea, revealing its sustained presence and photochemical impacts under urban conditions. Observed ClNO2 concentration averaged 208 pptv in the morning and 27 pptv in the afternoon, with a campaign maximum of 2.25 ppbv (1-min resolution). These patterns, along with supporting chemical and meteorological parameters, suggest that elevated morning ClNO2 resulted from reduced photolytic loss under weak solar radiation and continued N2O5 uptake. In contrast, observational evidence indicates a possible linkage between particulate NO3- photolysis and additional afternoon formation, particularly in aerosols enriched with anthropogenic chloride. This persistent ClNO2 shifted the diurnal peak of Cl radical production to late morning, with similar to 33 % of daily production occurring in the afternoon-surpassing the morning contribution (22 %). Observation-constrained box modeling further showed that ClNO2-driven ozone (O-3) production was comparable in both morning and afternoon periods, each contributing similar to 38 % to the total ClNO2-related O-3 enhancement. This demonstrates that substantial Cl-initiated oxidation can persist well into the afternoon, even under moderate NO3- levels (6.4 +/- 6.9 mu g/m(3), average +/- 1 sigma). These findings underscore the importance of considering ClNO2-driven oxidation throughout the day and highlight the need for further observations across seasons and urban environments to better constrain its atmospheric role.FALSEsciescopu
Overcoming imaging plate saturation with a multi-scan reconstruction technique for high-flux deuteron diagnostics
No full textImaging plates are widely used for charged-particle diagnostics, but their performance at high flux is limited by scanner saturation, leaving calibration gaps at low energies. In this work, we introduce a multi-scan reconstruction technique that recovers pre-saturation photostimulated luminescence (PSL) signals through repeated scans. This approach enables reliable extension of the dynamic range of imaging plates without hardware modification. As a proof of concept, we present the absolute calibration of BAS-TR imaging plates for monoenergetic deuterons in the 5–200 keV range, a regime where no prior data were available. The reconstructed PSL yield per deuteron shows good agreement with energy deposition simulations above 8 keV, with deviations at 5 keV attributed to surface effects. Overall, this method provides a practical solution to overcome saturation in imaging plate measurements and offers valuable reference data for ion diagnostics in nuclear fusion, plasma, and accelerator experiments. © 2025 Elsevier Ltd.FALSEsciescopu
A review of perovskite/Si tandem solar cells: internal and external components toward high efficiency, long-term durability, and commercialization
No full textThe rapid evolution of photovoltaic (PV) technology has made solar modules a key solution to meet growing global energy demands. In this context, achieving higher PV efficiency and reducing energy costs have become paramount objectives. Tandem solar cells, in which perovskite subcells are integrated with silicon (Si) subcells, represent a viable solution to surpass the Shockley-Queisser (S-Q) limit that constrains the efficiency of single-junction solar cells. These tandem configurations have demonstrated remarkable efficiency, reaching up to 34.85%, and are at the forefront of current PV research. This review focuses on recent studies aimed at enhancing the efficiency, stability, and scalability of tandem solar cells, including categorizing key areas of development in tandem solar cells into internal components (e.g., Si and perovskite subcells and interconnecting layers) and external components (e.g., encapsulation and busbars). Additionally, we address the fabrication process and levelized cost of energy (LCOE) of perovskite/Si tandem solar cells for cost-effective mass production. Moreover, we provide an outlook on the technological advancements required for the successful commercialization of tandem solar cells.FALSEsciescopu
An EEMD-based LSTM method for reconstructing the attenuated interference signals in a laser doppler vibrometry system
No full textSpeckle-induced signal dropouts are a persistent challenge in Laser Doppler Vibrometry (LDV), degrading the accuracy of velocity measurements. This paper proposes a framework with an ensemble empirical mode decomposition (EEMD)-based long short-term memory (LSTM) to address this issue by reconstructing the attenuated interference signals that cause these dropouts. The core of our framework is a two-stage process. First, the attenuated interference signal is decomposed into intrinsic mode functions (IMFs) using EEMD, selectively weighted, and then recombined to generate a non-attenuated interference signal. Second, an LSTM network is trained to learn this entire transformation. It acts as a computationally efficient model that maps the attenuated input signal directly to the reconstructed output signal. The performance of the proposed EEMD-based LSTM method was rigorously verified using both a software simulator and a hardware-in-the-loop experiment with an analog circuit. The results confirm that our method effectively reduces the velocity signal dropouts in both simulation and experimental settings, demonstrating its significant potential for improving the reliability of LDV systems.FALSEsciescopu
Inhibition of de novo ceramide synthesis mitigates alpha-synuclein pathology in a Parkinson’s disease mouse model
No full textParkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by the loss of dopaminergic neurons and the accumulation of α-synuclein aggregates. Ceramide metabolism is increasingly implicated in protein aggregation and mitochondrial dysfunction, both of which are prevalent in neurodegenerative disorders. While prior studies using cell lines have hinted at ceramide’s role in PD, the in vivo relevance and therapeutic efficacy of inhibiting its synthesis remained largely unexplored. We aimed to evaluate the therapeutic potential of inhibiting ceramide synthesis in various models of PD, including the A53T α-synuclein transgenic mouse model, primary neurons from patients with PD, and patient-derived midbrain organoids. We found that inhibiting de novo ceramide biosynthesis decreases α-synuclein aggregation and improves motor and cognitive function in A53T α-synuclein transgenic mice. Treatment with myriocin, a serine palmitoyltransferase inhibitor, restored mitochondrial morphology, enhanced mitophagy, and reduced neuroinflammation. Single-nucleus transcriptomic analysis revealed that myriocin normalized gene networks related to synaptic transmission, mitochondrial homeostasis, and inflammation. Additionally, human midbrain organoids derived from PD patient-induced pluripotent stem cells exhibited reduced α-synuclein aggregation and preserved dopaminergic neurons following myriocin treatment. Together, these results suggest that targeting ceramide synthesis is a promising strategy for addressing protein aggregation and neuronal death in PD.TRUEsciescopu
Optimal management of green hydrogen production in renewable energy systems using deep reinforcement learning methods
No full textThis research focuses on developing a deep reinforcement learning (DRL) framework to optimize green hydrogen production within renewable energy systems. By integrating a DRL-based model, the study aims to enhance real-time management of energy supply, storage, and distribution, involving an electrolyzer and balancing energy flows from photovoltaic (PV) sources, an energy storage system (ESS) and grid power. Utilizing real-world data, the DRL model adapts dynamically to fluctuations in renewable energy output and market prices, thereby optimizing operational efficiency. The study compares various DRL algorithms, including proximal policy optimization (PPO), soft actor-critic (SAC), and advantage actor-critic (A2C), assessing their performance in maximizing predefined reward functions. The findings demonstrate the robustness of the PPO algorithm, demonstrating significant reward accumulation and adaptability in managing dynamic environments. This validation is supported by empirical data and learning curves, confirming the DRL model’s proficiency in optimizing energy use and enhancing operational performance in green hydrogen systems. The integration of DRL with the framework for green hydrogen and renewable energy suggests a comprehensive solution that improves energy efficiency, operational costs, and sustainability initiatives. The research highlights the potential of advanced machine learning techniques for enhanced operational efficiency of renewable energy systems. © 2025 Elsevier Ltd.FALSEsciescopu
Motor-intent decoding from synthetic EEG data using denoising diffusion probabilistic models
No full textDecoding motor-intent directly from electroencephalogram (EEG) signals presents significant opportunities for advancing bio-inspired rehabilitation strategies and developing sophisticated human-computer interfaces. Despite the historical dominance of discriminative deep learning decoders, limitations in data availability and the development of effective decoding pipelines remain key obstacles to realizing this potential. This study introduces a novel framework predicated on electromyogram (EMG)-prompted diffusion models for the direct decoding of motor-intent from EMG and EEG signals. We demonstrate that this approach reduces classification error by 12.70% relative to recent discriminative decoders. Furthermore, our results surpassed conventional positive pair augmentation techniques, such as jittering, exhibiting a 3.41% improvement in performance. These findings underscore the transformative potential of generative models for generating synthetic training data and optimizing decoding pipelines in neuro-signal processing. We anticipate that this work will stimulate further investigation into the application of these techniques to improve the efficacy of rehabilitation interventions and facilitate more intuitive human-computer interactions, ultimately contributing to advancements in neuro-assistive device development and personalized rehabilitation strategies.FALSEsciescopu
Unravel key factors in α,β-unsaturated carboxylic acid salts one-pot synthesis from CO2 and alkenes: parameterization of bidentate ligands in transition-metal complexes
No full textTransition metal-catalyzed CO2 conversion to value-added chemicals faces the challenge of balancing selectivity and activity. Here, we integrate multivariate linear regression (MLR) with density functional theory (DFT) to establish a predictive framework to optimize catalytic systems for CO2/C2H4 coupling. The classical steric descriptors percent buried volume (%Vbur) and bite angle were shown to be non-equivalent, and their interplay is explained by an “interaction term transformation strategy”. By rationally tuning the electronic, steric and geometric properties of the catalyst, we design a high performance PPh2-ImPy-Ni(0) catalyst for CO2/C2H4 coupling, achieving a record TON of 570 with 82 % yield. A novel buried volume of octants (VBO) descriptor is proposed that can be extended to identify thresholds separating high-active/low-active regions in the CO2/alkene coupling, analogous to previous ligand classification systems. Most importantly, our stepwise energy decomposition approach – from DFT-computed Gibbs free energy (ΔG) to electronic energy (ΔE), interaction energy (ΔEint), and ultimately to the PIO-based bond index (PBI) – provides a generalizable framework for mechanistically interpreting reaction reactivity. These studies illuminate the broad applicability of hypotheses concerning the structural impact of various classes of bidentate ligands on reaction mechanisms, offering a robust methodology likely to be instrumental in advancing future ligand research. © 2025FALSEscopu