Search Results (104)

Accurate and concise temperature prediction models have important applications in meteorological science, agriculture, energy, and electricity. This study aims to compare the performance of simple models and deep learning models in temperature prediction and explore whether simple models can replace deep learning models in specific scenarios to save computing resources. Based on 37 years of daily temperature time series data from 10 cities from 1987 to 2024, the Simple Moving Average (SMA), Seasonal Average Method with Lookback Years (SAM-Lookback), and Long Short-Term Memory (LSTM) models are fitted to evaluate the accuracy of simple models and deep learning models in temperature prediction. The performance of different models is intuitively compared by calculating the RMSE and Percentage Error of each city. The results show that LSTM has higher accuracy in most cities, but the prediction results of SMA and LSTM are similar and perform equally well, while SAM-Lookback is relatively weak. Full article

The large-scale dynamics of the universe is generally described in terms of the time-dependent scale factor $a\left(t\right)$. To make contact with observational data, the $a\left(t\right)$ function needs to be related to the observable $z\left(r\right)$ function, redshift versus distance. Model fitting of data has shown that the equation that governs $z\left(r\right)$ needs to contain a constant term, which has been identified as Einstein’s cosmological constant. Here, it is shown that the required constant term is not a cosmological constant but is due to an overlooked geometric difference between proper time t and look-back time $t_{\mathrm{lb}}$ along lines of sight, which fan out isotropically in all directions of the 3D (3-dimensional) space that constitutes the observable universe. The constant term is needed to satisfy the requirement of spatial isotropy in the local limit. Its magnitude is independent of the epoch in which the observer lives and agrees with the value found by model fitting of observational data. Two of the observational consequences of this explanation are examined: an increase in the age of the universe from 13.8 Gyr to 15.4 Gyr, and a resolution of the $H_0$ tension, which restores consistency to cosmological theory. Full article

Transformer-based time series models are being increasingly employed for time series data analysis. However, their training remains memory intensive, especially with high-dimensional data and extended look-back windows, while model-level memory optimizations are well studied, the batch formation process remains an underexplored factor to performance inefficiency. This paper introduces a memory-efficient batching framework based on view-based sliding windows operating directly on GPU-resident tensors. This approach eliminates redundant data materialization caused by tensor stacking and reduces data transfer volumes without modifying model architectures. We present two variants of our solution: (1) per-batch optimization for datasets exceeding GPU memory, and (2) dataset-wise optimization for in-memory workloads. We evaluate our proposed batching framework systematically using peak GPU memory consumption and epoch runtime as efficiency metrics across varying batch sizes, sequence lengths, feature dimensions, and model architectures. Results show consistent memory savings, averaging 90% and runtime improvements of up to 33% across multiple transformer-based models (Informer, Autoformer, Transformer, and PatchTST) and a linear baseline (DLinear) without compromising model accuracy. We extensively validate our method using synthetic and standard real-world benchmarks, demonstrating accuracy preservation and practical scalability in distributed GPU environments. The proposed method highlights batch formation process as a critical component for improving training efficiency. Full article

Automated monitoring of the coolant temperature can enable autonomous operation of generation IV reactors (GenIV), thus reducing their operating and maintenance costs. Automation can be accomplished with machine learning (ML) models trained on historical sensor data. However, the performance of ML usually depends on the availability of large amount of training data, which is difficult to obtain for GenIV, as this technology is still under development. We propose the use of transfer learning (TL), which involves utilizing knowledge across different domains, to compensate for this lack of training data. TL can be used to create pre-trained ML models with data from small-scale research facilities, which can then be fine-tuned to monitor GenIV reactors. In this work, we develop pre-trained Transformer and long short-term memory (LSTM) networks by training them on temperature measurements from thermal hydraulic flow loops operating with water and Galinstan fluids at room temperature at Argonne National Laboratory. The pre-trained models are then fine-tuned and re-trained with minimal additional data to perform predictions of the time series of high temperature measurements obtained from the Engineering Test Unit (ETU) at Kairos Power. The performance of the LSTM and Transformer networks is investigated by varying the size of the lookback window and forecast horizon. The results of this study show that LSTM networks have lower prediction errors than Transformers, but LSTM errors increase more rapidly with increasing lookback window size and forecast horizon compared to the Transformer errors. Full article

The non-stationary nature of energy data is a serious challenge for energy forecasting methods. Frequent model updates are necessary to adapt to distribution shifts and avoid performance degradation. However, retraining regression models with lookback windows large enough to capture energy patterns is computationally expensive, as increasing the number of features leads to longer training times. To address this problem, we propose an approach that guarantees fast convergence through dimensionality reduction. Using a synthetic neighborhood dataset, we first validate three deep learning models—an artificial neural network (ANN), a 1D convolutional neural network (1D-CNN), and a long short-term memory (LSTM) network. Then, in order to mitigate the long training time, we apply principal component analysis (PCA) and a variational autoencoder (VAE) for feature reduction. As a way to ensure the suitability of the proposed models for a residential context, we also explore the trade-off between low error and training speed by considering three test scenarios: a global model, a local model for each building, and a global model that is fine-tuned for each building. Our results demonstrate that by selecting the optimal dimensionality reduction method and model architecture, it is possible to decrease the mean squared error (MSE) by up to 63% and accelerate training by up to 80%. Full article

Trajectory prediction is critical for ensuring the safety, reliability, and scalability of Unmanned Aerial Vehicle (UAV) in urban environments. Despite advances in deep learning, existing methods often struggle with dynamic UAV conditions, such as rapid directional changes and limited forecasting horizons, while lacking comprehensive real-time validation and generalization capabilities. This study addresses these challenges by proposing a gated recurrent unit (GRU)-based deep learning framework optimized through Look_Back and Forward_Length labeling to capture complex temporal patterns. The model demonstrated state-of-the-art performance, surpassing existing unmanned aerial vehicles (UAV) and aircraft trajectory prediction approaches, including FlightBERT++, in terms of both accuracy and robustness. It achieved reliable long-range predictions up to 4 s, and its real-time feasibility was validated due to its efficient resource utilization. The model’s generalization capability was confirmed through evaluations on two independent UAV datasets, where it consistently predicted unseen trajectories with high accuracy. These findings highlight the model’s ability to handle rapid maneuvers, extend prediction horizons, and generalize across platforms. This work establishes a robust trajectory prediction framework with practical applications in collision avoidance, mission planning, and anti-drone systems, paving the way for safer and more scalable UAV operations. Full article

This paper studies a p-layers deep learning artificial neural network (DLANN) for European multi-asset options. Firstly, a p-layers DLANN is constructed with undetermined weights and bias. Secondly, according to the terminal values of the partial differential equation (PDE) and the points that satisfy the PDE of multi-asset options, some discrete data are fed into the p-layers DLANN. Thirdly, using the least square error as the objective function, the weights and bias of the DLANN are trained well. In order to optimize the objective function, the partial derivatives for the weights and bias of DLANN are carefully derived. Moreover, to improve the computational efficiency, a time-segment DLANN is proposed. Numerical examples are presented to confirm the accuracy, efficiency, and stability of the proposed p-layers DLANN. Computational examples show that the DLANN’s relative error is less than $0.5\%$ for different numbers of assets $d=1,2,3,4$. In the future, the p-layers DLANN can be extended into American options, Asian options, Lookback options, and so on. Full article

Background: Breast cancer patients who develop brain metastases have a high mortality rate and a massive decrease in quality of life. Approximately 10–15% of all patients with breast cancer (BC) and 5–40% of all patients with metastatic BC develop brain metastasis (BM) during the course of the disease. However, there is only limited knowledge about prognostic factors in the treatment of patients with brain metastases in breast cancer (BMBC). Therefore, we retrospectively analyzed data of BMBC patients from the University Hospital of Würzburg for treatment patterns to find characteristics associated with a better or worse prognosis. These findings should help to treat the ever-increasing collective of patients with BMBC better in the future. Methods: The clinical data of 337 patients with cerebral metastatic breast cancer (date of death between 2004 and 2021) treated at the Department of Gynecology and Obstetrics of the University Hospital Würzburg were retrospectively analyzed, with a focus on patients’ survival. Results: The involvement of regional lymph nodes at initial diagnosis, the immunohistochemical subtype of TNBC at the onset of BMBC, and extracranial metastases at the time of BM diagnosis (bone, liver, lung metastases) were associated with a worse prognosis. In contrast, the immunohistochemical subtype of HER2/neu, the sole occurrence of a singular BM, the local surgical removal of BMs, and radiotherapy (especially stereotactic radiotherapy) were associated with prolonged survival. The number of therapies before the diagnosis of BMs also had a prognostic influence. Conclusions: Looking back at data is crucial for pinpointing risk elements affecting survival after a BM diagnosis. In our investigation, along with established factors like immunohistologic subtype, BM count, surgical excision, stereotactic irradiation, and type of extracranial metastasis, we also found that the number of therapies before BM diagnosis and the initial lymph node status were associated with patients’ survival. Potentially, these factors could be included in prospective prognostic scores for evaluating brain metastasis survival rates, thereby aiding in making appropriate treatment suggestions for impacted patients. Full article

Three-dimensional hydro-acoustic imaging is a research hot spot in the underwater acoustic signal processing field, which has a wide range of application prospects in marine environmental resource surveying, seabed topography and geomorphological mapping, and underwater early warning and monitoring. To solve the problem that the resolution of the current imaging sonar reduces rapidly with increase in distance and a scanning gap exists in side-scan sonar, we designed a down-looking 3D-imaging sonar with a linear array structure. The imaging scheme adopts a time-domain spatial beam-forming method with the Back Projection (BP) algorithm as the core, and the formation of a virtual plane array can effectively improve the along-track resolution. To cope with the interference of the carrier motion error on the imaging, we proposed a high-precision sub-wavelength motion compensation method based on a real-time acoustic calibration system. Simulation and real data experiments show that the motion compensation method can effectively eliminate the influence of motion error and make the imaging energy more focused, leading to higher-quality acoustic images. Under equal average energy, the maximum superimposed sound intensity values in the imaging results increased by 20.75 dB and 6.57 dB, respectively, for simulation and real data. After motion compensation, the resolution of this imaging system reached 3 cm × 3 cm × 2.5 cm @ Depth = 17 m, TBP = 30 s · Hz. Full article

We analyzed the properties of a sample of edge-on low-surface brightness galaxies, which are referred to as Cao23 ELSBGs. Cao23 ELSBGs exhibit a wide range of luminosities (−22 < $M_r$ < −13) with a mean scale length of 3.19 ± 1.48 kpc. Compared to HI-rich dwarf ELSBGs, Cao23 ELSBGs display more extended disk structures and redder (g-r) colors. They are also, on average, more massive than HI-rich dwarf ELSBGs. Star formation rates (SFRs) were calculated using WISE 12 $\mathsf{\mu }$m luminosity conversions and spectral energy distribution (SED) fitting methods, respectively. Cao23 ELSBGs fall below the main sequence with specific star formation rates (sSFRs) primarily in the range of 0.01–0.1 Gyr⁻¹. More massive Cao23 LSBGs tend to have lower sSFRs. Additionally, we derived the non-parametric star formation histories (SFHs) of Cao23 ELSBGs by SED fitting, dividing the SFHs into seven look back time bins with constant SFRs assumed for each bin. Our analysis indicates that high-mass ($M_{\ast }$ > 10^9.0$M\odot$) Cao23 ELSBGs assembled their mass earlier than their lower-mass counterparts, supporting a downsizing trend for LSBGs. Full article

This review paper focuses on the application of neural networks in semiconductor packaging, particularly examining how the Back Propagation Neural Network (BPNN) model predicts the work-in-process (WIP) arrival rates at various stages of semiconductor packaging processes. Our study demonstrates that BPNN models effectively forecast WIP quantities at each processing step, aiding production planners in optimizing machine allocation and thus reducing product manufacturing cycles. This paper further explores the potential applications of neural networks in enhancing production efficiency, forecasting capabilities, and process optimization within the semiconductor industry. We discuss the integration of real-time data from manufacturing systems with neural network models to enable more accurate and dynamic production planning. Looking ahead, this paper outlines prospective advancements in neural network applications for semiconductor packaging, emphasizing their role in addressing the challenges of rapidly changing market demands and technological innovations. This review not only underscores the practical implementations of neural networks but also highlights future directions for leveraging these technologies to maintain competitiveness in the fast-evolving semiconductor industry. Full article

Introduction: Nailfold video-capillaroscopy (NVC) is a non-invasive cost-effective technique involving the microscopic examination of small blood vessels of the distal nailfold with a magnification device. It provides valuable information regarding the microcirculation including anomalies such as tortuous or dilated capillaries, hemorrhages, and avascular areas, which can characterize connective tissue diseases. The utility of NVC in the diagnosis and monitoring of systemic sclerosis (SSc) has been investigated in numerous studies allowing the distinction of the specific microvascular pattern of scleroderma from different conditions other than scleroderma (non-scleroderma pattern). Sarcoidosis (SA) is a systemic inflammatory disease that can affect various organs, including the lungs, skin, and lymph nodes. The purpose of our review was to evaluate the current state of the art in the use of NVC in the diagnosis of SA, to understand the indications for its use and any consequent advantages in the management of the disease in different settings in terms of benefits for patients. Materials and Methods: We searched for the key terms “sarcoidosis” and “video-capillaroscopy” in a computerized search of Pub-Med, extending the search back in time without setting limits. We provided a critical overview of the literature, based on a precise evaluation. After our analysis, we examined the six yielded works looking for answers to our questions. Results: Few studies have evaluated that microcirculation is often compromised in SA, with alterations in blood flow and consequent tissue damage. Discussion: Basing on highlighted findings, NVC appears to be a useful tool in the initial evaluation of sarcoidosis patients. Furthermore, capillaroscopy is useful in the evaluation of the coexistence of sarcoidosis and scleroderma spectrum disorder or overlap syndromes. Conclusions: In conclusions, no specific pattern has been described for sarcoidosis, and further re-search is needed to fully understand the implications of nailfold capillaroscopy find-ings in this disease and to establish standardized guidelines for its use in clinical practice. Full article

Information and Communication Technology (ICT) such as smartphones has been attracting attention to prevent elderly people from becoming isolated. For this reason, recent research has proposed training methods for acquiring smartphone functions. However, since the types of smartphone functions required vary from person to person, a one-size-fits-all approach fails to engage all individuals adequately, leading to limited outcomes. On the other hand, with a view to social implementation, it is necessary to clarify a method that is effective in both cost and time. Previous research suggests that self-efficacy and social capital are the keys to acquiring smartphone skills among elderly people. Therefore, in this review, while looking back at previous research, we propose a study to demonstrate that by providing careful individual instruction by an experienced instructor to elderly people with little experience in smartphones and then having them take turns teaching other participants after the instruction, their self-efficacy and social capital can be increased, and a positive spiral effect can be achieved to maximize the improvement of smartphone skills widely. Full article

Modules based on c-Si cells account for more than 90% of the photovoltaic capacity installed worldwide, which is why the analysis in this paper focusses on this cell type. This study provides an overview of the current state of silicon-based photovoltaic technology, the direction of further development and some market trends to help interested stakeholders make decisions about investing in PV technologies, and it can be an excellent incentive for young scientists interested in this field to find a narrower field of research. This analysis covers all process steps, from the production of metallurgical silicon from raw material quartz to the production of cells and modules, and it includes technical, economic and environmental aspects. The economic aspect calls for more economical production. The ecological aspect looks for ways to minimise the negative impact of cell production on the environment by reducing emissions and using environmentally friendly materials. The technical aspect refers to the state of development of production technologies that contribute to achieving the goals of the economic, environmental and sustainability-related aspects. This involves ways to reduce energy consumption in all process steps, cutting ingots into wafers with the smallest possible cutting width (less material waste), producing thin cells with the greatest possible dimensional accuracy, using cheaper materials and more efficient production. An extremely important goal is to achieve the highest possible efficiency of PV cells, which is achieved by reducing cell losses (optical, electrical, degradation). New technologies in this context are Tunnel Oxide Passivated Contact (TOPcon), Interdigitated Back Contact Cells (IBCs), Heterojunction Cells (HJTs), Passivated Emitter Rear Totally Diffused cells (PERTs), silicon heterojunction cells (SHJs), Multi-Bush, High-Density Cell Interconnection, Shingled Cells, Split Cells, Bifacial Cells and others. The trend is also to increase the cell size and thus increase the output power of the module but also to reduce the weight of the module per kW of power. Research is also focused to maximise the service life of PV cells and minimise the degradation of their operating properties over time. The influence of shade and the increase in cell temperature on the operating properties should preferably be minimised. In this context, half-cut and third-cut cell technology, covering the cell surface with a layer that reduces soiling and doping with gallium instead of boron are newer technologies that are being applied. All of this leads to greater sustainability in PV technology, and solar energy becomes more affordable and necessary in the transition to a “green” economy. Full article

Since the new guidelines for endometrial cancer risk classification have been published, many reviews have proposed a critical re-evaluation. In this review, we look back to how the molecular classification system was built and its evolution in time to highlight the major flaws, particularly the biases stemming from the inherent limitations of the cohorts involved in the discoveries. A significant drawback in some cohorts is the inclusion criteria, as well as the retrospective nature and the notably sparse numbers, especially in the POLEmut (nonsynonymous mutation in EDM domain of POLE) risk groups, all of which impact the reliability of outcomes. Additionally, a disregard for variations in follow-up duration leads to a non-negligible bias, which raises a substantial concern in data interpretation and guideline applicability. Finally, according to the results that we obtained through a re-analysis of the confirmation cohort, the p53abn (IHC positive for p53 protein) subgroup, which is predominant in non-endometrioid histology (73–80%), loses its predictivity power in the endometrioid cohort of patients. The exclusion of non-endometrioid subtypes from the cohort led to a complete overlap of three molecular subgroups (all except POLEmut) for both overall and progression-free survival outcomes, suggesting the need for a more histotype-specific approach. In conclusion, this review challenges the current ESGO/ESTRO/ESP guidelines on endometrial cancer risk classification and highlights the limitations that must be addressed to better guide the clinical decision-making process. Full article