Legacy Models Research Articles

ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Training.

Negative flips are errors introduced in a classification system when a legacy model is updated. Existing methods to reduce the negative flip rate (NFR) either do so at the expense of overall accuracy by forcing a new model to imitate the old models, or use ensembles, which multiply inference cost prohibitively. We analyze the role of ensembles in reducing NFR and observe that they remove negative flips that are typically not close to the decision boundary, but often exhibit large deviations in the distance among their logits. Based on the observation, we present a method, called Ensemble Logit Difference Inhibition (ELODI), to train a classification system that achieves paragon performance in both error rate and NFR, at the inference cost of a single model. The method distills a homogeneous ensemble to a single student model which is used to update the classification system. ELODI also introduces a generalized distillation objective, Logit Difference Inhibition (LDI), which only penalizes the logit difference of a subset of classes with the highest logit values. On multiple image classification benchmarks, model updates with ELODI demonstrate superior accuracy retention and NFR reduction.

A novel characterization methodology for vapor-injected compressors: A comparative analysis with existing black-box models

In regions characterized by high temperature gradients, vapor compression systems often necessitate operation at very high pressure ratios resulting in a reduction in system capacity. Economized vapor injection compressors are used to avoid these issues, yet a precise predictive map for various compressor technologies with minimal data and relatively better performance remains unclear. This paper establishes a black-box compressor model to accurately predict compressor power, injection mass ratio, and evaporator mass flow rate in compressors with a single vapor injection port. This model is compared against three legacy models from literature and the ANN model, for reference. All five models are evaluated based on their ability to predict the aforementioned metrics. The proposed black-box model can predict the relevant metrics all within 5 % Mean Absolute Percentage Error (MAPE). Additionally, a refrigerant sensitivity analysis is performed with the black-box models. The model is trained using data from R410A and then used the coefficients to predict the performance of the same compressor when using R454B, and vice versa. It can estimate the evaporator mass flow rate with an accuracy within 3 %, the power within 2 %, and the injection mass ratio with MAPE less than 3 %.

Extended metric validation of a semi-physical Space Weather Modeling Framework conductance model on field-aligned current estimations

In this study, a detailed metric survey on the “Galaxy 15” (April 2010) space weather event is conducted to validate MAGNetosphere–Ionosphere–Thermosphere (MAGNIT), a semi-physical auroral ionospheric conductance model characterizing four precipitation sources, against AMPERE measurements via field-aligned current (FAC) characteristics. As part of this study, the comparative performance of three ionosphere electrodynamic specifications involving auroral conductance models, MAGNIT, Ridley Legacy Model (RLM) (empirical), and Conductance Model for Extreme Events (CMEE) (empirical), within the Space Weather Modeling Framework (SWMF), is demonstrated. Overall, MAGNIT exhibits marginally improved predictions; root mean square error values in upward and downward FACs of MAGNIT predictions compared to AMPERE data are smaller than those of CMEE and Ridley Ionosphere Model (RIM) by ∼12.7% and ∼6.24% before the storm, ∼4.52% and ∼2.13% better during the main phase, ∼1.98% and ∼1.27% worse during the second minimum, and better by ∼1.84% and ∼1.49% by the beginning of the recovery, respectively. In all three model configurations, the dusk and night magnetic local time (MLT) sectors over-predict throughout the storm, while the day and dawn MLT sectors under-predict in response to interplanetary magnetic field (IMF) conditions. In addition to accuracy and bias, similar results and conclusions are drawn from additional metrics, including in the categories of correlation, precision, extremes, and skill, and recommendations are made for the best-performing model configuration in each metric category. Visual data–model comparisons conducted by studying the FAC location and latitude/MLT spread throughout various phases of the storm suggest that the spatial extent of the FACs is captured relatively well in the night-side auroral oval, unlike in the day-side oval. The spread in latitude of the FACs matches that in the previous literature on other model performances. This information on auroral precipitation sources and their weight on FACs, along with metrics from model–data comparisons, can be used to modify MAGNIT settings to optimize SWMF model performance.

Exploring mid‐infrared spectral transfer functions for the prediction of multiple soil properties using a global dataset

AbstractInfrared spectroscopy is increasingly being adopted as a technology for soil analysis. However, laboratories worldwide are equipped with different infrared spectrometers, leading to variations that hinder the global application of soil spectroscopy. This study evaluates the transferability of soil spectra from a global dataset collected using four mid‐infrared spectrometers. To evaluate the efficacy of five spectral transfer functions (direct standardization, piecewise direct standardization, spectral space transformation [SST], principal components‐canonical correlation analysis [PC‐CCA], and domain‐invariant partial least square [DIPLS] regression), two datasets were used: dataset A (n = 224; standardized samples) was scanned using one primary spectrometer and three secondary spectrometers; dataset B (n = 1904; legacy samples) was scanned only using the primary spectrometer. The first set of chemometrics models was developed using dataset A to compare the performance of different spectrometers. The second set of models was developed using dataset B to evaluate the effectiveness of spectral transfer functions. Both models were developed using partial least squares regression. Spectral transfer functions developed using dataset A indicate that the PC‐CCA method was the best in converging spectra collected from four instruments into a similar space projected using Uniform Manifold Approximation and Projection. Spectral transfer did not result in consistent improvement in the prediction of soil properties compared to the direct use of spectra collected from different spectrometers. These findings carry significant implications for the utilization of legacy models, enabling laboratories to concentrate on acquiring new samples and spectral measurements using established protocols without the need for spectral transfer.

Flexible Peripheral Nerve Interfacing Electrode for Joint Position Control in Closed-Loop Neuromuscular Stimulation.

Addressing peripheral nerve disorders with electronic medicine poses significant challenges, especially in replicating the dynamic mechanical properties of nerves and understanding their functionality. In the field of electronic medicine, it is crucial to design a system that thoroughly understands the functions of the nervous system and ensures a stable interface with nervous tissue, facilitating autonomous neural adaptation. Herein, we present a novel neural interface platform that modulates the peripheral nervous system using flexible nerve electrodes and advanced neuromodulation techniques. Specifically, we have developed a surface-based inverse recruitment model for effective joint position control via direct electrical nerve stimulation. Utilizing barycentric coordinates, this model constructs a three-dimensional framework that accurately interpolates inverse isometric recruitment values across various joint positions, thereby enhancing control stability during stimulation. Experimental results from rabbit ankle joint control trials demonstrate our model's effectiveness. In combination with a proportional-integral-derivative (PID) controller, it shows superior performance by achieving reduced settling time (less than 1.63 s), faster rising time (less than 0.39 s), and smaller steady-state error (less than 3 degrees) compared to the legacy model. Moreover, the model's compatibility with recent advances in flexible interfacing technologies and its integration into a closed-loop controlled functional neuromuscular stimulation (FNS) system highlight its potential for precise neuroprosthetic applications in joint position control. This approach marks a significant advancement in the management of neurological disorders with advanced neuroprosthetic solutions.

An artificial intelligence-generated model predicts 90-day survival in alcohol-associated hepatitis: A global cohort study.

Alcohol-associated hepatitis (AH) poses significant short-term mortality. Existing prognostic models lack precision for 90-day mortality. Utilizing artificial intelligence in a global cohort, we sought to derive and validate an enhanced prognostic model. The Global AlcHep initiative, a retrospective study across 23 centers in 12 countries, enrolled patients with AH per National Institute for Alcohol Abuse and Alcoholism criteria. Centers were partitioned into derivation (11 centers, 860 patients) and validation cohorts (12 centers, 859 patients). Focusing on 30 and 90-day postadmission mortality, 3 artificial intelligence algorithms (Random Forest, Gradient Boosting Machines, and eXtreme Gradient Boosting) informed an ensemble model, subsequently refined through Bayesian updating, integrating the derivation cohort's average 90-day mortality with each center's approximate mortality rate to produce posttest probabilities. The ALCoholic Hepatitis Artificial INtelligence Ensemble score integrated age, gender, cirrhosis, and 9 laboratory values, with center-specific mortality rates. Mortality was 18.7% (30d) and 27.9% (90d) in the derivation cohort versus 21.7% and 32.5% in the validation cohort. Validation cohort 30 and 90-day AUCs were 0.811 (0.779-0.844) and 0.799 (0.769-0.830), significantly surpassing legacy models like Maddrey's Discriminant Function, Model for End-Stage Liver Disease variations, age-serum bilirubin-international normalized ratio-serum Creatinine score, Glasgow, and modified Glasgow Scores ( p < 0.001). ALCoholic Hepatitis Artificial INtelligence Ensemble score also showcased superior calibration against MELD and its variants. Steroid use improved 30-day survival for those with an ALCoholic Hepatitis Artificial INtelligence Ensemble score > 0.20 in both derivation and validation cohorts. Harnessing artificial intelligence within a global consortium, we pioneered a scoring system excelling over traditional models for 30 and 90-day AH mortality predictions. Beneficial for clinical trials, steroid therapy, and transplant indications, it's accessible at: https://aihepatology.shinyapps.io/ALCHAIN/ .

Abstract 6423: KRAS mutation detection by Sanger sequencing using the SeqStudio Flex capillary electrophoresis instrument

Abstract The detection of mutations in the KRAS gene highly associated with cancer is one of the most widely used Sanger sequencing capillary electrophoresis (CE) applications in molecular diagnostics. In addition to their traditional role in cancer diagnostic, KRAS mutations have recently gained additional relevance as pharmacogenomics (PGx) markers for a significant number of new oncology therapeutics. KRAS mutation detection assays are used as companion diagnostics (CDx) for these cancer treatments. Analysis by Sanger Sequencing of KRAS mutations is cost effective and lends itself for use in resource limited healthcare settings. While there are currently no commercial kits for targeted sequencing of the KRAS mutation hotspots available, the online Applied Biosystems Primer Designer™ Tool (thermofisher.com) allows standardization of the PCR primer design between labs and repeat orders of the KRAS assays by using these commercially available assay ID numbers. Here we demonstrate the use of Primer Designer KRAS mutation detections assays for mutation in codon 12, 13 and 61 totaling to 8 allele variants. The data was generated on the SeqStudio Flex CE instrument using reference standard DNA as well as DNA extracted from FFPE solid tumor samples. We show accuracy of variant call (OPA 100% (64/64)) and concordance (OPA 98.3% (59/60)) in a side-by-side comparison of the SeqStudio Flex with the legacy Model 3500 Genetic Analyzer instrument using a set of commercially available (Horizon Discovery) KRAS reference DNA samples. Primary solid tumor samples are processed to Fresh Frozen Paraffin Embedded samples (FFPE) for further disease marker evaluation. These samples are stable for long term at ambient temperatures and provide the opportunity to evaluate prospective biomarkers at large scale to understand their potential utility. Discovery Life Sciences has generated a large-scale biorepository of human biospecimen, including FFPE, plasma/sera, buffy coat, and PBMCs. Archival FFPE samples from oncology indications associated with KRAS mutations, such as non-small cell lung cancer and colorectal cancer, were analyzed and identified samples with KRAS mutations, demonstrating the potential diagnostic utility of these assays. Finally, cell-free (cf) DNA from double spun plasma from relevant indications was also successfully evaluated for KRAS mutations, highlighting the potential for these assays in liquid biopsy studies. The study demonstrates the robustness of the Primer Designer KRAS mutation detection sequencing assays using reference standard DNA as well as DNA derived from FFPE DNA and cfDNA from patient derived tumor samples. For Research Use Only. Not for use in diagnostic procedures. Citation Format: Achim Karger, Ami Patel, Kaylee Tseng, Camila Ornelas, Mary Napier, Rod Beck, Kerri Collwell, Shawn Fahl, Carole Bornarth. KRAS mutation detection by Sanger sequencing using the SeqStudio Flex capillary electrophoresis instrument [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6423.

Performance enhancement of short-term wind speed forecasting model using Realtime data.

The ever-increasing demand for electricity has presented a grave threat to traditional energy sources, which are finite, rapidly depleting, and have a detrimental environmental impact. These shortcomings of conventional energy resources have caused the globe to switch from traditional to renewable energy sources. Wind power significantly contributes to carbon-free energy because it is widely accessible, inexpensive, and produces no harmful emissions. Better and more efficient renewable wind power production relies on accurate wind speed predictions. Accurate short-term wind speed forecasting is essential for effectively handling unsteady wind power generation and ensuring that wind turbines operate safely. The significant stochastic nature of the wind speed and its dynamic unpredictability makes it difficult to forecast. This paper develops a hybrid model, L-LG-S, for precise short-term wind speed forecasting to address problems in wind speed forecasting. In this research, state-of-the-art machine learning and deep learning algorithms employed in wind speed forecasting are compared with the proposed approach. The effectiveness of the proposed hybrid model is tested using real-world wind speed data from a wind turbine located in the city of Karachi, Pakistan. Moreover, the mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) are used as accuracy evaluation indices. Experimental results show that the proposed model outperforms the state-of-the-art legacy models in terms of accuracy for short-term wind speed in training, validation and test predictions by 98% respectively.

Evaluating the Performance of ChatGPT in Ophthalmology: An Analysis of Its Successes and Shortcomings

Foundation models are a novel type of artificial intelligence algorithms, in which models are pretrained at scale on unannotated data and fine-tuned for a myriad of downstream tasks, such as generating text. This study assessed the accuracy of ChatGPT, a large language model (LLM), in the ophthalmology question-answering space. Evaluation of diagnostic test or technology. ChatGPT is a publicly available LLM. We tested 2 versions of ChatGPT (January 9 "legacy" and ChatGPT Plus) on 2 popular multiple choice question banks commonly used to prepare for the high-stakes Ophthalmic Knowledge Assessment Program (OKAP) examination. We generated two 260-question simulated exams from the Basic and Clinical Science Course (BCSC) Self-Assessment Program and the OphthoQuestions online question bank. We carried out logistic regression to determine the effect of the examination section, cognitive level, and difficulty index on answer accuracy. We also performed a post hoc analysis using Tukey's test to decide if there were meaningful differences between the tested subspecialties. We reported the accuracy of ChatGPT for each examination section in percentage correct by comparing ChatGPT's outputs with the answer key provided by the question banks. We presented logistic regression results with a likelihood ratio (LR) chi-square. We considered differences between examination sections statistically significant at a P value of < 0.05. The legacy model achieved 55.8% accuracy on the BCSC set and 42.7% on the OphthoQuestions set. With ChatGPT Plus, accuracy increased to 59.4% ± 0.6% and 49.2% ± 1.0%, respectively. Accuracy improved with easier questions when controlling for the examination section and cognitive level. Logistic regression analysis of the legacy model showed that the examination section (LR, 27.57; P= 0.006) followed by question difficulty (LR, 24.05; P < 0.001) were most predictive of ChatGPT's answer accuracy. Although the legacy model performed best in general medicine and worst in neuro-ophthalmology (P < 0.001) and ocular pathology (P= 0.029), similar post hoc findings were not seen with ChatGPT Plus, suggesting more consistent results across examination sections. ChatGPT has encouraging performance on a simulated OKAP examination. Specializing LLMs through domain-specific pretraining may be necessary to improve their performance in ophthalmic subspecialties. Proprietary or commercial disclosure may be found after the references.

Characterization of Non-Neutral Urban Canopy Wind Profile Using CFD Simulations—A Data-Driven Approach

Horizontally averaged wind profiles inside the urban canopy are used in many studies and numerical models. The existing analytical models are only applicable to a small range of aspect ratios and mostly neutral atmospheric conditions due to their underlying assumptions. In this study, a surrogate model for predicting horizontally averaged wind profiles in the street canyons of an idealized urban canopy for a wide range of urban morphologies and thermal forcing scenarios is developed with the help of machine learning techniques and computational fluid dynamics (CFD) simulation data. The influence of urban morphological parameters, atmospheric stability and wind conditions on the urban canopy wind flow is modeled using machine learning algorithms applied on CFD simulation results. The numerical model is validated using wind-tunnel data. Steady-state Reynolds averaged Navier–Stokes (RANS) simulations with a standard k-ϵ turbulence model for 252 different simulation conditions are performed on an idealized building geometry that consists of a regular array of 4 × 4 cubes. The simulation results are averaged horizontally to obtain the mean velocity and temperature profiles. Surrogate models are developed using the simulation outputs as training examples and the best model is chosen by comparing the performance of different machine learning models. The surrogate artificial neural network (ANN) model of this study outperforms the current state-of-the-art models in the prediction of horizontally averaged mean wind profile inside the urban canopy. The mean error (ME) and root-mean-square error (RMSE) of the discrete point prediction ANN model of this study are 0.016 m/s and 0.060 m/s, respectively, which is significantly lower compared to the best of the legacy models for which the errors are 0.048 m/s and 0.387 m/s, respectively.

Empirical Path Loss Modelling for Selected LTE Networks in FUTA Campus, Ondo State, Nigeria

Deployed Long Term Evolution (LTE) networks in Nigeria can barely meet the desired 100 Mbps downlink throughput leading to unsatisfactory quality of experience by mobile users. Typically, mobile network operators (MNOs) rely on network planning tools designed for generalized environments. These tools employ legacy propagation models that may not be suited to the operational environments under consideration. As such, the efficiency of such legacy path loss models suffers when they are used in environments different from those for which they have been designed, and this poses a major challenge to the MNOs. This is because the Nigerian geographical areas and topographical features vary widely from the areas where the legacy models were developed. Several studies in Nigeria and other African countries have shown that the legacy path loss models perform unsatisfactorily when compared with field measurement data. To address this challenge and enable accurate path loss prediction for an urban campus environment, extensive measurements at 2600 MHz were carried out in the main campus of the Federal University of Technology Akure (FUTA), Ondo State, Nigeria. The measurement results were compared with the path loss predictions from the commonly-used legacy propagation models (Free space and 3GPP TR 36.873). The results show that the legacy path loss models under-predict the path loss averagely by 20-40 dB, and up to 88 dB in some cases, for the considered environment. Root mean square error (RMSE) values in the range of 1.895 and 9.159 were also observed along the routes. The measurement results will enable the MNOs to adjust the path losses in order to deliver improved quality of service.

E2E Service Class Mapping in Heterogeneous IOUT: SDN-Based Architecture

Over the past few years, the Internet-of-Underwater-Things (IoUT) became a prominent technology for investigating the underwater conditions. Moreover, the communication networks became heterogeneous over the past few years due to deployment of various technologies between the source and target nodes on the end-to-end (E2E) path. In heterogeneous IoUT networks, the nodes on an E2E path may consist of different traffic classes (TCs). Hence, the legacy model of networking and the default software-defined networking (SDN) will not be effective with inter-domain communication and E2E mapping of TCs. In this study, we propose a state-of-the art methodology utilizing SDN to fulfill the E2E service needs among domains with various TCs on the E2E path for the heterogeneous IoUT networks. We propose a module in the controller based upon a multi-criteria decision-making strategy i.e. simple additive weighting module (SAWM) to select a significant service class from the TCs. Moreover, we map the TCs in the global programmable controller (GPC) to meet an E2E quality-of-service (QoS) that meets application demands. Finally, we validate the supremacy of our proposed scheme through comparisons for E2E TCs mapping in terms of network performance metrics. The proof-of-concept experiments validate the effectiveness of the proposed strategy.

Model integration methods for hydro-model platform under cloud computing environments

Computing platforms providing cloud simulation services have raised new challenges on the model integration. Unlike calls to the model programs (components) in traditional simulation software, here the models should be dynamically integrated in the “plug and play” mode regardless of the differences in model type and developer. To this end two integration methods have been proposed, i.e., coarse-grained EXE integration and interactive integration. In an EXE integration method, the simulation program is directly called and thus only a data conversion interface is needed while rewriting of the model source code is not required. In contrast, an interactive integration method wraps the model components using the standard wrapper with communication interfaces, and therefore, it can communicate and exchange data with the platform in a real time. The first method is suitable for the integration of legacy models, while the second one can control the progress of simulation schemes and facilitate the scheduling of computing resources. Examples of the model integration and platform application have been presented in hydraulics/hydrodynamics to demonstrate the effectiveness of the integration method and the cloud computing platform.

Use of Brain Biomechanical Models for Monitoring Impact Exposure in Contact Sports

Head acceleration measurement sensors are now widely deployed in the field to monitor head kinematic exposure in contact sports. The wealth of impact kinematics data provides valuable, yet challenging, opportunities to study the biomechanical basis of mild traumatic brain injury (mTBI) and subconcussive kinematic exposure. Head impact kinematics are translated into brain mechanical responses through physics-based computational simulations using validated brain models to study the mechanisms of injury. First, this article reviews representative legacy and contemporary brain biomechanical models primarily used for blunt impact simulation. Then, it summarizes perspectives regarding the development and validation of these models, and discusses how simulation results can be interpreted to facilitate injury risk assessment and head acceleration exposure monitoring in the context of contact sports. Recommendations and consensus statements are presented on the use of validated brain models in conjunction with kinematic sensor data to understand the biomechanics of mTBI and subconcussion. Mainly, there is general consensus that validated brain models have strong potential to improve injury prediction and interpretation of subconcussive kinematic exposure over global head kinematics alone. Nevertheless, a major roadblock to this capability is the lack of sufficient data encompassing different sports, sex, age and other factors. The authors recommend further integration of sensor data and simulations with modern data science techniques to generate large datasets of exposures and predicted brain responses along with associated clinical findings. These efforts are anticipated to help better understand the biomechanical basis of mTBI and improve the effectiveness in monitoring kinematic exposure in contact sports for risk and injury mitigation purposes.

Fungible Citizenship

Fungible Citizenship

A reinvestigation into Munson’s model for room closure in bedded rock salt

Accurate predictions of room closure are important for hazardous waste repositories in rock salt formations, such as the Waste Isolation Pilot Plant (WIPP). When Munson and co-workers simulated several room closure experiments conducted at the WIPP during the 1980’s and 1990’s, their simulated closure curves closely agreed with the closure measurements. A careful review of their work, however, raised concerns and prompted the reinvestigation in this paper. To begin the reinvestigation, Munson’s legacy Room D closure simulation was reasonably recreated in a current-day finite element code. Next, special care was taken to obtain numerically converged results, re-introduce the anhydrite strata intermittently ignored by Munson, and calibrate the Munson–Dawson (M–D) constitutive model for salt as much as possible from laboratory test measurements. When this new model was used to simulate Room D’s closure, it under-predicted the horizontal and vertical closure rates by 2 . 34 × and 3 . 10 × , respectively, at 5.7 years after room excavation. As a result, the M–D model was extended to capture the newly established creep behavior at low equivalent stresses ( < 8 MPa ) and replace the Tresca with the Hosford equivalent stress. Simulations using the new M–D model over-predicted the horizontal closure rate by 1 . 15 × and under-predicted the vertical closure rate by 1 . 08 × at 5.7 years, averaged over three room closure experiments. Although further improvements could be made, the new model has a stronger scientific foundation than Munson’s legacy model and appears ready for careful engineering use.

Defying a legacy or an evolving process? A post-pandemic architectural design pedagogy

The coronavirus disease 2019 (Covid-19) condition has prompted serious questions about the challenges faced by the established two-century-old canons of education in architecture and urbanism. This paper establishes an evolutionary account on how design education in architecture and urbanism has arrived at the pre-Covid-19 condition, explores current challenges and, in the process of encountering the Covid-19 condition, asks the question of what the scope of opportunities is to meet these challenges. A chronological analysis of design pedagogy is undertaken to instigate a debate on its future in a post-pandemic environment. This paper captures the salient characteristics of the legacy model that is inherited from historical schools, demonstrates the influence of and resistance to this model (1960s); identifies the qualities of various alternatives including ten ground-breaking alternative pedagogies (1970s–1990s); highlights strengths of further alternative approaches including critical inquiry, the process-based and learning-by-making pedagogies (2000s) and the social construction-based pedagogies (2010s). Scrutinising the consequences of the Covid-19 condition and the associated ‘transitional emergency model’, the analysis articulates the persisting challenges and examines current adaptations while outlining the scope of future opportunities for a responsive design pedagogy in architecture and urbanism for a post-pandemic world.

Multiexpert Adversarial Regularization for Robust and Data-Efficient Deep Supervised Learning

Deep neural networks (DNNs) can achieve high accuracy when there is abundant training data that has the same distribution as the test data. In practical applications, data deficiency is often a concern. For classification tasks, the lack of enough labeled images in the training set often results in overfitting. Another issue is the mismatch between the training and the test domains, which results in poor model performance. This calls for the need to have robust and data efficient deep learning models. In this work, we propose a deep learning approach called Multi-Expert Adversarial Regularization learning (MEAR) with limited computational overhead to improve the generalization and robustness of deep supervised learning models. The MEAR framework appends multiple classifier heads (experts) to the feature extractor of the legacy model. MEAR aims to learn the feature extractor in an adversarial fashion by leveraging complementary information from the individual experts as well as the ensemble of the experts to be more robust for an unseen test domain. We train state-of-the-art networks with MEAR for two important computer vision tasks, image classification and semantic segmentation. We compare MEAR to a variety of baselines on multiple benchmarks. We show that MEAR is competitive with other methods and more successful at learning robust features.

Validation and Optimisation of Common Clear Sky Models with a use case for North-East Bulgaria

One of the most significant error contributors to preliminary design tools for Photovoltaic power systems is related to the simple parametric Clear Sky models. Therefore, this paper focuses on providing a methodology and a more sophisticated open-source tool for 3 commonly used Clear Sky models. This includes all relevant steps involved in the process - from filtering the raw meteorological data, identification of Clear Sky regions, data redistribution to genetic optimization of selected model parameter, etc.use case is built upon a multiyear dataset obtained from TU Varna meteorological station between 2012-2016. A significantly higher density distribution of Clear sky segments was identified during the summer through the Clear Sky Identification algorithm. To avoid the risk of overfitting the models to purely summer months and poor model fits in winter months, which was found to be the case with the legacy model, the underrepresented clear sky regions (based on θ) were replicated until uniform distribution is attained. Subsequently, a genetic optimization was applied to selected parameters in the Clear Sky algorithms and the updated models showed a significant improvement in low winter months (θ) and even overall performance boost RMSE / MAE /R2. Furthermore, such validations and optimizations are recommended prior to any design or real-time PV-system analysis for the specific location.

Value-Based Leadership in Crowdfunding Digital Startup: A Qualitative Study in iGrow and KITABISA

This study aims to identify the basic values of crowdfunding digital startup leaders at iGrow and kitabisa and also to understand how these values were created. Individual value systems are related to people's attitudes and behavior, especially in the leadership decision-making process. iGrow as an investment company for agriculture products has an orientation toward achieving profits, while kitabisa as a social movement platform has a social orientation toward helping others. This study used a qualitative approach through the in-depth interview from snowball sampling that served as the primary data. The research result proposed nine main values identified from leaders of both startup companies. The main values of iGrow leader were achievement goals, hard work, legacy, togetherness, and role models, while those of Kitabisa leader were rational, learning, hustler, genuine caring, and legacy. From the process in which these values were created, this study proposed a theory on a different path of value creation, called the value creation pathway. These research findings were only limited to the two leaders from iGrow and Kitabisa.