Stack Architecture Research Articles

Assessing stacked physics-informed machine learning models for co-located wind–solar power forecasting

Increasingly advanced stochastic energy management systems are employed to facilitate the integration of wind and solar PV in worldwide power grids. In this context, forecasting is a key tool limiting the success of said control actions. This paper explores the suitability of stacked machine learning based models to predict wind and solar power available in the same site using a physics informed approach. The method recombines basic meteorological metrics widely available to compute new physics informed ones facilitating the learning procedure, while other are weak ML-models themselves. Further, to facilitate the integration of the point forecasters in the stochastic optimization field, we propose a simple unsupervised estimation of the error distribution. In this way, scenarios can be easily and homogeneously characterized for different resolutions and horizons. A study case is presented employing the Open Access dataset SOLETE, to facilitate benchmarking and replication of results. The results show accuracy improvements over the previously reported work over the same dataset.

Machine learning and discriminant function analysis in the formulation of generic models for sex prediction using patella measurements

Sex prediction from bone measurements that display sexual dimorphism is one of the most important aspects of forensic anthropology. Some bones like the skull and pelvis display distinct morphological traits that are based on shape. These morphological traits which are sexually dimorphic across different population groups have been shown to provide an acceptably high degree of accuracy in the prediction of sex. A sample of 100 patella of Mixed Ancestry South Africans (MASA) was collected from the Dart collection. Six parameters: maximum height (maxh), maximum breadth (maxw), maximum thickness (maxt), the height of articular facet (haf), lateral articular facet breadth (lafb), and medial articular facet breath (mafb) were used in this study. Stepwise and direct discriminant function analyses were performed for measurements that exhibited significant differences between male and female mean measurements, and the “leave-one-out” approach was used for validation. Moreover, we have used eight classical machine learning techniques along with feature ranking techniques to identify the best feature combinations for sex prediction. A stacking machine learning technique was trained and validated to classify the sex of the subject. Here, we have used the top performing three ML classifiers as base learners and the predictions of these models were used as inputs to different machine learning classifiers as meta learners to make the final decision. The measurements of the patella of South Africans are sexually dimorphic and this observation is consistent with previous studies on the patella of different countries. The range of average accuracies obtained for pooled multivariate discriminant function equations is 81.9–84.2%, while the stacking ML technique provides 90.8% accuracy which compares well with those presented for previous studies in other parts of the world. In conclusion, the models proposed in this study from measurements of the patella of different population groups in South Africa are useful resent with reasonably high average accuracies.

Towards Execution-Efficient LSTMs via Hardware-Guided Grow-and-Prune Paradigm

Long short-term memory (LSTM) applications need fast yet compact models. Neural network compression approaches, e.g., the grow-and-prune paradigm, have proved to be promising for cutting down network complexity by skipping insignificant weights. However, current compression strategies remain mostly hardware-agnostic and network complexity reduction does not always translate to execution efficiency. In this work, we propose a hardware-guided symbiotic training methodology for compact, accurate, yet execution-efficient inference models. It is based on our observation that hardware may introduce substantial non-monotonic behavior, which we call the latency hysteresis effect, when evaluating network size versus inference latency. This observation raises question about the mainstream smaller-dimension-is-better compression strategy, which often leads to a sub-optimal model architecture. Leveraging the hardware-impacted hysteresis effect and sparsity, we enable a symbiosis of model compactness and accuracy with execution efficiency, thus reducing LSTM latency while increasing its accuracy. We have evaluated our approach on language modeling and speech recognition applications. Relative to the traditional stacked LSTM architecture obtained for the Penn Treebank dataset, we reduce the number of parameters by 18.0× (30.5×) and measured run-time latency by up to 2.4× (5.2×) on Nvidia GPUs (Intel Xeon CPUs) without any accuracy degradation. For the DeepSpeech2 architecture obtained for the AN4 dataset, we reduce the model size by 7.0× (19.4×), word error rate from 12.9% to 9.9% (10.4%), and measured run-time latency by up to 1.7× (2.4×) on Nvidia GPUs (Intel Xeon CPUs). Our method consistently outperforms prior art for both applications, with compact, accurate, and execution-efficient inference models.

Fuzzy adaptive jellyfish search-optimized stacking machine learning for engineering planning and design

This paper presents a novel fuzzy adaptive jellyfish search-optimized stacking system (FAJS-SS) that integrates the jellyfish search (JS) optimizer, the fuzzy adaptive (FA) logic controller, and stacking ensemble machine learning. First, FA logic is incorporated into JS optimizer to construct an efficient metaheuristic algorithm for global optimization. The proposed algorithm is benchmarked against various well-known optimizers using mathematical functions. The FAJS optimizer is then used to optimize the hyperparameters of the stacking system (SS). Cases that involve construction productivity, the compressive strength of a masonry structure, the shear capacity of reinforced deep beams, the axial strength of steel tube-confined concrete, and the resilient modulus of subgrade soils were investigated. Results of analyses reveal that the FAJS-SS predicts more accurately than the other machine learning systems in the literature. Accordingly, the proposed fuzzy adaptive metaheuristic-optimized stacking system is effective for providing engineering informatics in the planning and design phase.

Accelerated functional brain aging in major depressive disorder: evidence from a large scale fMRI analysis of Chinese participants

Major depressive disorder (MDD) is one of the most common mental health conditions that has been intensively investigated for its association with brain atrophy and mortality. Recent studies suggest that the deviation between the predicted and the chronological age can be a marker of accelerated brain aging to characterize MDD. However, current conclusions are usually drawn based on structural MRI information collected from Caucasian participants. The universality of this biomarker needs to be further validated by subjects with different ethnic/racial backgrounds and by different types of data. Here we make use of the REST-meta-MDD, a large scale resting-state fMRI dataset collected from multiple cohort participants in China. We develop a stacking machine learning model based on 1101 healthy controls, which estimates a subject’s chronological age from fMRI with promising accuracy. The trained model is then applied to 1276 MDD patients from 24 sites. We observe that MDD patients exhibit a +4.43 years (p < 0.0001, Cohen’s d = 0.31, 95% CI: 2.23–3.88) higher brain-predicted age difference (brain-PAD) compared to controls. In the MDD subgroup, we observe a statistically significant +2.09 years (p < 0.05, Cohen’s d = 0.134525) brain-PAD in antidepressant users compared to medication-free patients. The statistical relationship observed is further checked by three different machine learning algorithms. The positive brain-PAD observed in participants in China confirms the presence of accelerated brain aging in MDD patients. The utilization of functional brain connectivity for age estimation verifies existing findings from a new dimension.

Advanced Prognosis methodology based on behavioral indicators and Chained Sequential Memory Neural Networks with a diesel engine application

This paper presents a novel methodology in the field of Prognosis and predictive Maintenance (PdM) of industrial components. It has been designed as an inclusive PdM approach grounded on flexible strategies capable of characterizing the behavior of an industrial system regardless of its nature in terms of its physics, dynamics, or evolution in time. The proposed method includes two behavioral indicators computed through a robust method based on Behavior Patterns. These two indicators (Deviation and Similarity) provide a precise characterization of the behaviors of an industrial system. The prognosis of both indicators is carried out through three different Neural Network (NN) architectures: a Multilayer Perceptron (MLP) and two types of Long–Short Term Memory (LSTM) NNs with two different configurations. Among these configurations, this study proposes a novel LSTM architecture characterized by its Chained Sequential Memory (CSM) architecture based on Peephole Connections. The three architectures are studied and compared in detail in order to determine which one achieves better results in prognosis. The originality of this approach lies in the prognosis of behaviors by applying indicators to enhance and make more intuitive the characterization and prognosis of the state of the system. The proposed LSTM CSM architecture reduces the forecast error by around 50% in comparison to MLP and Stacked LSTM architectures. This study includes an application to a real case in which the new methodology is implemented for the prognosis of the cooling system of a power plant diesel generator. The results obtained prove the advantages and possibilities that the proposed methodology has for industrial applications.

Roman Urdu Fake Reviews Detection Using Stacked LSTM Architecture

Roman Urdu Fake Reviews Detection Using Stacked LSTM Architecture

Prognostic Model of ICU Admission Risk in Patients with COVID-19 Infection Using Machine Learning.

With the onset of the COVID-19 pandemic, the number of critically sick patients in intensive care units (ICUs) has increased worldwide, putting a burden on ICUs. Early prediction of ICU requirement is crucial for efficient resource management and distribution. Early-prediction scoring systems for critically ill patients using mathematical models are available, but are not generalized for COVID-19 and Non-COVID patients. This study aims to develop a generalized and reliable prognostic model for ICU admission for both COVID-19 and non-COVID-19 patients using best feature combination from the patient data at admission. A retrospective cohort study was conducted on a dataset collected from the pulmonology department of Moscow City State Hospital between 20 April 2020 and 5 June 2020. The dataset contains ten clinical features for 231 patients, of whom 100 patients were transferred to ICU and 131 were stable (non-ICU) patients. There were 156 COVID positive patients and 75 non-COVID patients. Different feature selection techniques were investigated, and a stacking machine learning model was proposed and compared with eight different classification algorithms to detect risk of need for ICU admission for both COVID-19 and non-COVID patients combined and COVID patients alone. C-reactive protein (CRP), chest computed tomography (CT), lung tissue affected (%), age, admission to hospital, and fibrinogen parameters at hospital admission were found to be important features for ICU-requirement risk prediction. The best performance was produced by the stacking approach, with weighted precision, sensitivity, F1-score, specificity, and overall accuracy of 84.45%, 84.48%, 83.64%, 84.47%, and 84.48%, respectively, for both types of patients, and 85.34%, 85.35%, 85.11%, 85.34%, and 85.35%, respectively, for COVID-19 patients only. The proposed work can help doctors to improve management through early prediction of the risk of need for ICU admission of patients during the COVID-19 pandemic, as the model can be used for both types of patients.

Constraint-based type inference for FreezeML

FreezeML is a new approach to first-class polymorphic type inference that employs term annotations to control when and how polymorphic types are instantiated and generalised. It conservatively extends Hindley-Milner type inference and was first presented as an extension to Algorithm W. More modern type inference techniques such as HM(X) and OutsideIn(X) employ constraints to support features such as type classes, type families, rows, and other extensions. We take the first step towards modernising FreezeML by presenting a constraint-based type inference algorithm. We introduce a new constraint language, inspired by the Pottier/Rémy presentation of HM(X), in order to allow FreezeML type inference problems to be expressed as constraints. We present a deterministic stack machine for solving FreezeML constraints and prove its termination and correctness.

5G NOMA Defense Application Environment and Stacked LSTM Network Architectures

In 5G and beyond 5G wireless communication networks, the NOMA scheme is widely considered a major non-orthogonal access technique for improving system capacity and data rates. The main challenges in current NOMA systems are limited channel feedback and the difficulty of integrating it with advanced adaptive coding and modulation algorithms. This study analyses S-LSTM-based DL NOMA receivers in i.i.d. Nakagami-m fading channel circumstances as opposed to previously presented solutions. The LSTM has the advantage of responding dynamically to changing channel conditions. When compared to a typical NOMA system, a typical NOMA system has a 12% lower outage probability, a 39% increase in net throughput, and a maximum SER reduction of 48%. Complex modulated M-ary PSK and M-ary QAM data symbols are employed in D/L NOMA transmission. Classic receivers such as LS and MMSE are outperformed by the S-LSTM-based DL-NOMA receiver. The CP and non-linear clipping noise simulation curves compare the performance of the MMSE and LS receivers with that of the DL NOMA receiver in real-time propagation circumstances. The DL-based detector outperforms the MMSE for SNRs greater than 15 dB because the S-LSTM method is more robust than the clipping noise.

Planar heterojunction perovskite solar cell with graded energy band architecture via fast-drying spray deposition

The fast-drying spray deposition (FDSD) technique for perovskite solar cells (PSCs) is developed to enable the stacking of perovskite absorbers with different work functions, which allows the creation of an additional built-in electric field at the interface during the fermi level realignment process upon contact. FDSD is functional under high relative humidity (RH) ambiance and by design, deposits dry film without the need for post-deposition annealing treatment. Based on a spray coating process, FDSD is also highly scalable. Leveraging FDSD’s multilayer deposition capability, this work explores the implementation of graded energy band architectures to achieve PSCs with enhanced carrier extraction and photovoltaic performances. To demonstrate the potential benefit of this approach, two triple cation mixed halide perovskite formulas are chosen. The two formulas, when stacked together in correct order, produce a heterojunction PSC device with an extra built-in electric field, which helps drift charge carriers towards desired electrodes. The architecture with the proper energy band alignment therefore exhibits enhanced carrier extraction efficiency and, despite being subjected to over 60–80% RH during fabrication, reaches the mean power conversion efficiency (PCE) of 7.4%, with the maximum value of 9.5%. The average PCE translates to over 9.9% and 10.3% improvements over the devices based on the two constituent formulas individually. FDSD demonstrates great flexibility i.e., in-humid-air fabrication process and requiring no post annealing treatments, thereby enabling extremely robust and scalable stacked architecture PSCs with low cost and good performance.

Viscoelastic properties of bioinspired asymmetric helicoidal CFRP composites

A dynamic mechanical thermal analyser (DMTA) was used to develop insight into the dynamic mechanical properties of bioinspired asymmetric helicoidal carbon fibre reinforced plastic (CFRP) composites as a function of fibre architecture using inter-ply stacking angles of 0° (UD), 0/90° (cross-ply), 5°, 15°, 10°, 20°, 25° and 30°. Here, we show that the dynamic mechanical properties of asymmetric helicoidal CFRP composites are linearly correlated to their oriented ply fractions between 0° and 45° off the loading axis. We furthermore provide evidence from the tan-δ curves that asymmetric helicoidal CFRP composites are heterogeneous materials with separate viscoelastic phases and glass transition temperatures, resulting from the stacking architectures of these composites. Inter-ply stacking angles are finally noted as critical factors affecting the extent of macromolecular mobility within helicoidally stacked continuous fibre CFRP composites.Graphical abstract

Quantifying Nonlocality: How Outperforming Local Quantum Codes Is Expensive.

Quantum low-density parity-check (LDPC) codes are a promising avenue to reduce the cost of constructing scalable quantum circuits. However, it is unclear how to implement these codes in practice. Seminal results of Bravyi etal. [Phys. Rev. Lett. 104, 050503 (2010)PRLTAO0031-900710.1103/PhysRevLett.104.050503] have shown that quantum LDPC codes implemented through local interactions obey restrictions on their dimension k and distance d. Here we address the complementary question of how many long-range interactions are required to implement a quantum LDPC code with parameters k and d. In particular, in 2D we show that a quantum LDPC code with distance d∝n^{1/2+ϵ} requires Ω(n^{1/2+ϵ}) interactions of length Ω[over ˜](n^{ϵ}). Further, a code satisfying k∝n with distance d∝n^{α} requires Ω[over ˜](n) interactions of length Ω[over ˜](n^{α/2}). As an application of these results, we consider a model called a stacked architecture, which has previously been considered as a potential way to implement quantum LDPC codes. In this model, although most interactions are local, a few of them are allowed to be very long. We prove that limited long-range connectivity implies quantitative bounds on the distance and code dimension.

Observation Method in the Control of Stacker Capacity Under Landslide Hazard – A Case Study

Abstract The article presents both an application and the purpose of the observation method in the control of stacker capacity. It lists the types of the measured (observed) quantities, which serve as a basis for the observation method. It also describes the procedure of the method and discusses its individual steps. It further provides examples of applying the method in defining the capacity levels of a stacking machine ZGOT-11500, based on the recorded surface and subsurface soil displacement values. The article also offers the increment values and speeds for the individual parameters, which serve as a warning against deterioration of the geotechnical condition of the soil. Knowledge of the relationships between the parameters that describe soil deformation and the required defined stacker capacity may serve as a basis for further research and experiments on the observation method, which may increase the safety of stacking operations. The analysis was based on the results of geotechnical and geodetic measurements, as well as on the operating parameters of the stacker, acquired over a period of 5 months.

UAV Multispectral Image-Based Urban River Water Quality Monitoring Using Stacked Ensemble Machine Learning Algorithms—A Case Study of the Zhanghe River, China

Timely monitoring of inland water quality using unmanned aerial vehicle (UAV) remote sensing is critical for water environmental conservation and management. In this study, two UAV flights were conducted (one in February and the other in December 2021) to acquire images of the Zhanghe River (China), and a total of 45 water samples were collected concurrently with the image acquisition. Machine learning (ML) methods comprising Multiple Linear Regression, the Least Absolute Shrinkage and Selection Operator, a Backpropagation Neural Network (BP), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were applied to retrieve four water quality parameters: chlorophyll-a (Chl-a), total nitrogen (TN), total phosphors (TP), and permanganate index (CODMn). Then, ML models based on the stacking approach were developed. Results show that stacked ML models could achieve higher accuracy than a single ML model; the optimal methods for Chl-a, TN, TP, and CODMn were RF-XGB, BP-RF, RF, and BP-RF, respectively. For the testing dataset, the R2 values of the best inversion models for Chl-a, TN, TP, and CODMn were 0.504, 0.839, 0.432, and 0.272, the root mean square errors were 1.770 μg L−1, 0.189 mg L−1, 0.053 mg L−1, and 0.767 mg L−1, and the mean absolute errors were 1.272 μg L−1, 0.632 mg L−1, 0.045 mg L−1, and 0.674 mg L−1, respectively. This study demonstrated the great potential of combined UAV remote sensing and stacked ML algorithms for water quality monitoring.

Predicting hospital emergency department visits with deep learning approaches

Overcrowding in emergency department (ED) causes lengthy waiting times, reduces adequate emergency care and increases rate of mortality. Accurate prediction of daily ED visits and allocating resources in advance is one of the solutions to ED overcrowding problem. In this paper, a deep stacked architecture is being proposed and applied to the daily ED visits prediction problem with deep components such as Long Short Term Memory (LSTM), Gated Recurrent Units (GRU) and simple Recurrent Neural Network (RNN). The proposed architecture achieves very high mean accuracy level (94.28–94.59%) in daily ED visits predictions. We have also compared the performance of this architecture with non-stacked deep models and traditional prediction models. The results indicate that deep stacked models outperform (4–7%) the traditional prediction models and other non-stacked deep learning models (1–2%) in our prediction tasks. The application of deep neural network in ED visits prediction is novel as this is one of the first studies to apply a deep stacked architecture in this field. Importantly, our models have achieved better prediction accuracy (in one case comparable) than the state-of-the-art in the literature.

Graded multilayer triple cation perovskites for high speed and detectivity self-powered photodetector via scalable spray coating process

Rapid advancements in perovskite materials have led to potential applications in various optoelectronic devices, such as solar cells, light-emitting diodes, and photodetectors. Due to good photoelectric properties, perovskite enables low-cost and comparable performance in terms of responsivity, detectivity, and speed to those of the silicon counterpart. In this work, we utilized triple cation perovskite, well known for its high performance, stability, and wide absorption range, which is crucial for broadband photodetector applications. To achieve improved detectivity and faster response time, graded multilayer perovskite absorbers were our focus. Sequential spray deposition, which allows stacked perovskite architecture without disturbing lower perovskite layers, was used to generate single, double, and triple-layer perovskite photodetectors with proper energy band alignment. In this work, we achieved a record on self-powered perovskite photodetector fabricated from a scalable spray process in terms of EQE and responsivity of 65.30% and 0.30 A W-1. The multilayer devices showed faster response speed than those of single-layer perovskite photodetectors with the champion device reaching 70 µs and 88 µs for rising and falling times. The graded band structure and the internal electric field generated from perovskite heterojunction also increase specific detectivity about one magnitude higher in comparison to the single-layer with the champion device achieving 6.82 × 1012 cmHz1/2 W−1.

Predicting the Kijang Emas Bullion Price using LSTM Networks

This study investigates the potential of Deep Learning techniques, specifically LSTM networks, in forecasting Kijang Emas future value over a long period. Six LSTM models comprising of Simple LSTM, Bidirectional LSTM, and Stacked LSTM architecture were built and trained against a 15-year historical price data for Kijang Emas. The models’ performance was then measured against ARIMA (5,1,0) as a baseline reference and evaluated against the RAE, MSE and RMSE metric. The results revealed that LSTM networks models performed well in forecasting Kijang Emas price based on the test dataset where the average RMSE was between 49.9 to 50.3 while the Bidirectional LSTM was found to exhibit better performance as compared to the other LSTM models.

Lateral spectrum splitting system with perovskite photovoltaic cells

We examine the potential of a multijunction spectrum-splitting photovoltaic (PV) solar energy system with perovskite PV cells. Spectrum splitting allows combinations of different energy band gap PV cells that are laterally separated and avoids the complications of fabricating tandem stack architectures. Volume holographic optical elements have been shown to be effective for the spectrum-splitting operation and can be incorporated into compact module packages. However, one of the remaining issues for spectrum splitting systems is the availability of low-cost wide band gap and intermediate band gap cells that are required for realizing high overall conversion efficiency. Perovskite PV cells have been fabricated with a wide range of band gap energies that potentially satisfy the requirements for multijunction spectrum-splitting systems. A spectrum-splitting system is evaluated for a combination of perovskite PV cells with energy band gaps of 2.30, 1.63, and 1.25 eV and with conversion efficiencies of 10.4%, 21.6%, and 20.4%, respectively, which have been demonstrated experimentally in the literature. First, the design of a cascaded volume holographic lens for spectral separation in three spectral bands is presented. Second, a rigorous coupled wave model is developed for computing the diffraction efficiency of a cascaded hologram. The model accounts for cross-coupling between higher diffraction orders in the upper and lower holograms, which previous models have not accounted for but is included here with the experimental verification. Lastly, the optical losses in the system are analyzed and the hypothetical power conversion efficiency is calculated to be 26.7%.

Residual Stresses in Ultrafine‐Grained Laminated Metal Composites Analyzed by X‐ray Diffraction and the Hole‐Drilling Method

Accumulative roll bonding is an advanced manufacturing process, which is capable of simultaneously refining the grain size into the nanometer regime and bonding different metallic sheet materials. Herein, homogenous aluminum/aluminum as well as heterogeneous aluminum/steel laminated metal composite (LMCs) are fabricated. The residual stresses are experimentally determined by X‐ray diffraction and the hole‐drilling method. Generally, a complex residual stress profile is found in all LMCs. The level of residual stress strongly depends on the bonded materials. Compressive residual stresses are induced in all sheets in the near surface area. These stresses range from −5 MPa in aluminum to −240 MPa in steel. In the homogenous aluminum/aluminum LMCs, compressive stresses up to −26 MPa in the softer layers and tensile stresses up to 30 MPa in the stronger layers are built up. This is different to heterogeneous aluminum/steel LMCs, where tensile stresses up to 40 MPa in the softer aluminum layers and compressive stresses up to −72 MPa in the inner harder steel layers are present. Based on the results obtained it is possible to directly design the material combination or stacking architecture of ultrafine‐grained LMCs to tailor the residual stress profile.