Increased Training Time Research Articles

TPCNet: Representation learning for H i mapping

Abstract We introduce TPCNet, a neural network predictor that combines Convolutional and Transformer architectures with Positional encodings, for neutral atomic hydrogen (H i) spectral analysis. Trained on synthetic datasets, our models predict cold neutral gas fraction (fCNM) and H i opacity correction factor ($\mathcal {R_{\mathrm{H{\small I}}}}$) from emission spectra based on the learned relationships between the desired output parameters and observables (optically-thin column density and peak brightness). As a follow-up to Murray et al. (2020)’s shallow Convolutional Neural Network (CNN), we construct deep CNN models and compare them to TPCNet models. TPCNet outperforms deep CNNs, achieving a 10percnt average increase in testing accuracy, algorithmic (training) stability, and convergence speed. Our findings highlight the robustness of the proposed model with sinusoidal positional encoding applied directly to the spectral input, addressing perturbations in training dataset shuffling and convolutional network weight initializations. Higher spectral resolutions with increased spectral channels offer advantages, albeit with increased training time. Diverse synthetic datasets enhance model performance and generalization, as demonstrated by producing fCNM and $\mathcal {R_{\mathrm{H{\small I}}}}$ values consistent with evaluation ground truths. Applications of TPCNet to observed emission data reveal strong agreement between the predictions and Gaussian decomposition-based estimates (from emission and absorption surveys), emphasizing its potential in H i spectral analysis.

Exploring Synergy of Denoising and Distillation: Novel Method for Efficient Adversarial Defense

Escalating advancements in artificial intelligence (AI) has prompted significant security concerns, especially with its increasing commercialization. This necessitates research on safety measures to securely utilize AI models. Existing AI models are vulnerable to adversarial attacks, which are a specific form of assault methodology. Although various countermeasures have been explored, practical defense models are scarce. Current adversarial defense methods suffer from reduced accuracy, increased training time, and incomplete defense against adversarial attacks, indicating performance limitations and a lack of robustness. To address these limitations, we propose a composite defense model, the knowledge Distillation and deNoising Network (DiNo-Net), which integrates knowledge distillation and feature denoising techniques. Furthermore, we analyzed a correlation between the loss surface of adversarial perturbations and denoising techniques. Using DiNo-Net, we confirmed that increasing the temperature during the knowledge distillation process effectively amplifies the loss surface around the ground truth. Consequently, this enables more efficient denoising of the adversarial perturbations. It achieved a defense success rate of 72.7%, which is a remarkable improvement over the 41.0% success rate of models with only denoising defense mechanisms. Furthermore, DiNo-Net reduced the training time and maintained higher accuracy, confirming its efficient defense performance. We hope that this relationship will spur the development of fundamental defense strategies.

Federated Learning Using Multi-Modal Sensors with Heterogeneous Privacy Sensitivity Levels

Data from multi-modal sensors, such as Red-Green-Blue (RGB) cameras, thermal cameras, microphones, and mmWave radars, have gradually been adopted in various classification problems for better accuracy. Some sensors, like RGB cameras and microphones, however, capture privacy-invasive data, which are less likely to be used in centralized learning. Although the Federated Learning (FL) paradigm frees clients from sharing their sensor data, doing so results in reduced classification accuracy and increased training time. In this article, we introduce a novel Heterogeneous Privacy Federated Learning (HPFL) paradigm to better capitalize on the less privacy-invasive sensor data, such as thermal images and mmWave point clouds, by uploading them to the server for closing the performance gap between FL and centralized learning. HPFL not only allows clients to keep the more privacy-invasive sensor data private, such as RGB images and human voices, but also gives each client total freedom to define the levels of their privacy concern on individual sensor modalities. For example, more sensitive users may prefer to keep their thermal images private, while others do not mind sharing these images. We carry out extensive experiments to evaluate the HPFL paradigm using two representative classification problems: semantic segmentation and emotion recognition. Several key findings demonstrate the merits of HPFL: (i) compared to FedAvg, it improves foreground accuracy by 18.20% in semantic segmentation and boosts the F1-score by 4.20% in emotion recognition, (ii) with heterogeneous privacy concern levels, it achieves an even larger F1-score improvement of 6.17–16.05% in emotion recognition, and (iii) it also outperforms the state-of-the-art FL approaches by 12.04–17.70% in foreground accuracy and 2.54–4.10% in F1-score.

Comparison of Elbow Flexion in Youth Baseball Pitchers With and Without Throwing-Arm Pain.

More than half of all youth baseball pitchers report throwing-related pain in their throwing arm throughout a season. The purpose of this study was to investigate differences in elbow flexion throughout the pitching cycle between youth baseball pitchers with and without throwing-arm pain. It was hypothesized that pitchers with throwing-arm pain would have decreased elbow flexion throughout the pitching cycle compared with those who were pain-free. Controlled laboratory study. A total of 38 youth baseball pitchers (mean age, 13.3 ± 1.7 years; height, 164.4 ± 12.9 cm; weight, 57.1 ± 14 kg) were retrospectively selected from a database. Based on responses to a health history questionnaire, the pitchers were placed into a pain group if they indicated they were experiencing throwing-arm pain. Pitchers who indicated they were not experiencing throwing-arm pain were matched according to age, height, and weight to the pain group. All pitchers threw 3 fastballs to a catcher at the regulation distance. The mean elbow flexion of the 3 trials was used during analysis to investigate peak elbow flexion and time-normalized (0%-100%) elbow flexion across the pitch cycle (stride-foot contact to ball release). Elbow flexion was compared between the pain and pain-free groups using 1-dimensional statistical nonparametric mapping, and the mean peak elbow flexion between groups was compared using the Mann-Whitney U test. No significant differences were observed between the groups in elbow flexion throughout the pitching cycle (P > .05) and no group differences in peak elbow flexion (U = 122; P = .09). Study findings indicated no significant differences in elbow flexion between youth baseball pitchers with versus without throwing-arm pain, unlike previous research reporting that pitchers with a history of medial elbow pain had altered elbow flexion and higher pitch velocities compared with those without a history of pain. Clinicians should consider other potential factors related to throwing-arm pain beyond elbow flexion. Moreover, it is advisable to focus on evidence-based modifiable factors shown to increase the risk of pain and injury in youth pitchers, such as exceeding pitch counts, number of innings pitched, increased training time, range-of-motion, and strength deficits.

A temporal–spectral fusion transformer with subject-specific adapter for enhancing RSVP-BCI decoding

The Rapid Serial Visual Presentation (RSVP)-based Brain–Computer Interface (BCI) is an efficient technology for target retrieval using electroencephalography (EEG) signals. The performance improvement of traditional decoding methods relies on a substantial amount of training data from new test subjects, which increases preparation time for BCI systems. Several studies introduce data from existing subjects to reduce the dependence of performance improvement on data from new subjects, but their optimization strategy based on adversarial learning with extensive data increases training time during the preparation procedure. Moreover, most previous methods only focus on the single-view information of EEG signals, but ignore the information from other views which may further improve performance. To enhance decoding performance while reducing preparation time, we propose a Temporal-Spectral fusion transformer with Subject-specific Adapter (TSformer-SA). Specifically, a cross-view interaction module is proposed to facilitate information transfer and extract common representations across two-view features extracted from EEG temporal signals and spectrogram images. Then, an attention-based fusion module fuses the features of two views to obtain comprehensive discriminative features for classification. Furthermore, a multi-view consistency loss is proposed to maximize the feature similarity between two views of the same EEG signal. Finally, we propose a subject-specific adapter to rapidly transfer the knowledge of the model trained on data from existing subjects to decode data from new subjects. Experimental results show that TSformer-SA significantly outperforms comparison methods and achieves outstanding performance with limited training data from new subjects. This facilitates efficient decoding and rapid deployment of BCI systems in practical use.

MAR-YOLOv9: A multi-dataset object detection method for agricultural fields based on YOLOv9

With the development of deep learning technology, object detection has been widely applied in various fields. However, in cross-dataset object detection, conventional deep learning models often face performance degradation issues. This is particularly true in the agricultural field, where there is a multitude of crop types and a complex and variable environment. Existing technologies still face performance bottlenecks when dealing with diverse scenarios. To address these issues, this study proposes a lightweight, cross-dataset enhanced object detection method for the agricultural domain based on YOLOv9, named Multi-Adapt Recognition-YOLOv9 (MAR-YOLOv9). The traditional 32x downsampling Backbone network has been optimized, and a 16x downsampling Backbone network has been innovatively designed. A more streamlined and lightweight Main Neck structure has been introduced, along with innovative methods for feature extraction, up-sampling, and Concat connection. The hybrid connection strategy allows the model to flexibly utilize features from different levels. This solves the issues of increased training time and redundant weights caused by the detection neck and auxiliary branch structures in traditional YOLOv9, enabling MAR-YOLOv9 to maintain high performance while reducing the model’s computational complexity and improving detection speed, making it more suitable for real-time detection tasks. In comparative experiments on four plant datasets, MAR-YOLOv9 improved the mAP@0.5 accuracy by 39.18% compared to seven mainstream object detection algorithms, and by 1.28% compared to the YOLOv9 model. At the same time, the model size was reduced by 9.3%, and the number of model layers was decreased, reducing computational costs and storage requirements. Additionally, MAR-YOLOv9 demonstrated significant advantages in detecting complex agricultural images, providing an efficient, lightweight, and adaptable solution for object detection tasks in the agricultural field. The curated data and code can be accessed at the following link: https://github.com/YangxuWangamI/MAR-YOLOv9.

Enhancing trainee performance in obstetric ultrasound through an artificial intelligence system: randomized controlled trial.

Performing obstetric ultrasound scans is challenging for inexperienced operators; therefore, the prenatal screening artificial intelligence system (PSAIS) software was developed to provide real-time feedback and guidance for trainees during their scanning procedures. The aim of this study was to investigate the potential benefits of utilizing such an artificial intelligence system to enhance the efficiency of obstetric ultrasound training in acquiring and interpreting standard basic views. A prospective, single-center randomized controlled study was conducted at The First Affiliated Hospital of Sun Yat-sen University. From September 2022 to April 2023, residents with no prior obstetric ultrasound experience were recruited and assigned randomly to either a PSAIS-assisted training group or a conventional training group. Each trainee underwent a four-cycle practical scan training program, performing 20 scans in each cycle on pregnant volunteers at 18-32 gestational weeks, focusing on acquiring and interpreting standard basic views. At the end of each cycle, a test scan evaluated trainees' ability to obtain standard ultrasound views without PSAIS assistance, and image quality was rated by both the trainees themselves and an expert (in a blinded manner). The primary outcome was the number of training cycles required for each trainee to meet a certain standard of proficiency (i.e. end-of-cycle test scored by the expert at ≥ 80%). Secondary outcomes included the expert ratings of the image quality in each trainee's end-of-cycle test and the discordance between ratings by trainees and the expert. In total, 32 residents and 1809 pregnant women (2720 scans) were recruited for the study. The PSAIS-assisted trainee group required significantly fewer training cycles compared with the non-PSAIS-assisted group to meet quality requirements (P = 0.037). Based on the expert ratings of image quality, the PSAIS-assisted training group exhibited superior ability in acquiring standard imaging views compared with the conventional training group in the third (P = 0.012) and fourth (P < 0.001) cycles. In both groups, the discordance between trainees' ratings of the quality of their own images and the expert's ratings decreased with increasing training time. A statistically significant difference in overall trainee-expert rating discordance between the two groups emerged at the end of the first training cycle and remained at every cycle thereafter (P < 0.013). By assisting inexperienced trainees in obtaining and interpreting standard basic obstetric scanning views, the use of artificial intelligence-assisted systems has the potential to improve training effectiveness. © 2024 International Society of Ultrasound in Obstetrics and Gynecology.

A Study on Network Anomaly Detection Using Fast Persistent Contrastive Divergence

As network technology evolves, cyberattacks are not only increasing in frequency but also becoming more sophisticated. To proactively detect and prevent these cyberattacks, researchers are developing intrusion detection systems (IDSs) leveraging machine learning and deep learning techniques. However, a significant challenge with these advanced models is the increased training time as model complexity grows, and the symmetry between performance and training time must be taken into account. To address this issue, this study proposes a fast-persistent-contrastive-divergence-based deep belief network (FPCD-DBN) that offers both high accuracy and rapid training times. This model combines the efficiency of contrastive divergence with the powerful feature extraction capabilities of deep belief networks. While traditional deep belief networks use a contrastive divergence (CD) algorithm, the FPCD algorithm improves the performance of the model by passing the results of each detection layer to the next layer. In addition, the mix of parameter updates using fast weights and continuous chains makes the model fast and accurate. The performance of the proposed FPCD-DBN model was evaluated on several benchmark datasets, including NSL-KDD, UNSW-NB15, and CIC-IDS-2017. As a result, the proposed method proved to be a viable solution as the model performed well with an accuracy of 89.4% and an F1 score of 89.7%. By achieving superior performance across multiple datasets, the approach shows great potential for enhancing network security and providing a robust defense against evolving cyber threats.

Comprehensive survey on the effectiveness of sharpness aware minimization and its progressive variants

ABSTRACT As advancements push for larger and more complex Artificial Intelligence (AI) models to improve performance, preventing the occurrence of overfitting when training overparameterized Deep Neural Networks (DNNs) remains a challenge. Despite the presence of various regularization techniques aimed at mitigating this issue, poor generalization remains a concern, especially when handling diverse and limited data. This paper explores one of the latest and most promising strategies to address this challenge, Sharpness Aware Minimization (SAM), which concurrently minimizes loss value and sharpness-related loss. While this method exhibits substantial effectiveness, it comes with a notable trade-off in increased training time and is founded on several approximations. Consequently, several variants of SAM have emerged to alleviate these limitations and bolster model performance. This survey paper examines the significant advancements achieved by SAM, delves into its constraints, and categorizes recent progressive variants that further enhance current State-of-the-Art results.

ATHENA-FL: Avoiding Statistical Heterogeneity with One-versus-All in Federated Learning

Federated learning (FL) is a distributed approach to train machine learning models without disclosing private data from participating clients to a central server. Nevertheless, FL training struggles to converge when clients have distinct data distributions, which leads to an increased training time and model prediction error. We propose ATHENA-FL, a federated learning system that considers clients with heterogeneous data distributions to generate accurate models in fewer training epochs than state-of-the-art approaches. ATHENA-FL reduces communication costs, providing an additional positive aspect for resource-constrained scenarios. ATHENA-FL mitigates data heterogeneity by introducing a preliminary step before training that clusters clients with similar data distribution. To handle that, we use the weights of a locally trained neural network used as a probe. The proposed system also uses the one-versus-all model to train one binary detector for each class in the cluster. Thus, clients can compose complex models combining multiple detectors. These detectors are shared with all participants through the system's database. We evaluate the clustering procedure using different layers from the neural network and verify that the last layer is sufficient to cluster the clients efficiently. The experiments show that using the last layer as input for the clustering algorithm transmits 99.68% fewer bytes to generate clusters compared to using all the neural network weights. Finally, our results show that ATHENA-FL correctly identifies samples, achieving up to 10.9% higher accuracy than traditional training. Furthermore, ATHENA-FL achieves lower training communication costs compared with MobileNet architecture, reducing the number of transmitted bytes between 25% and 97% across evaluated scenarios.

The feasibility and effectiveness of remote cognitive training on cognitive function and work performance in workers.

We aimed to determine whether remote cognitive training (CT) is feasible and has the potential to improve cognitive function and work performance in Japanese workers. From June to September 2020, this intervention time series study enrolled workers aged 18-65 years from 10 companies located in a metropolitan area of Japan. Cognitive function tests and self-administered questionnaires were completed by the participants three times: at baseline, after 12 weeks of CT, and after a further 12 weeks following cessation. We measured work performance with the question: "How would you rate your performance (compared with your optimum performance) over the past 4 weeks?" Responses were made via a visual analog scale (0-100). Cognitive function was assessed using the THINC-integrated tool, which is a brief, objective computerized cognitive assessment battery. For our computerized remote CT intervention, BrainHQ was used on the basis of our scientific rationale and the empirical literature. We recommended three 20-min sessions of BrainHQ per week and sent participants three reminders. In total, 119 participants were recruited to this study. Only 22.7% of the subjects achieved the recommended training time of 720 min over 12 weeks. The median training time was used to divide participants into long and short- training groups. The long-training group showed a greater improvement in attention and executive function than the short-training group but there was no significant improvement in work performance after CT compared to baseline. Our results suggest that although remote CT was not feasible enough, the effects on cognitive function can be expected by increasing training time and motivation.

Identification of Spinal Disorders Based on Spine X-ray Dataset Processing Using the LBP and CNN Algorithms

This research will use deep learning in conducting spinal x-ray image analysis but computational time problems are a problem of this study. Computations on deep learning across multiple nodes can increase training time and longer computation time compared to machine learning models. Based on experimental results, the best spine x-ray image classification results when using the CNN model with accuracy at the training stage, evaluation stage and test stage were 69.00%, 83.33% and 81.16% respectively. CNN models optimized with LBP get the lowest accuracy, with results at the training stage of 62.64%, validation stage of 75.00% and testing stage of 65.22%. LBP feature extraction turns out to have several drawbacks when combined with the CNN model, one major drawback is its inability to process global spatial information while retaining local texture information which causes LBP to be unable to capture the entire structure or context of the image, focusing only on local patterns so that many features of the image are lost. Another issue is the sensitivity of CNNs to image data, which can affect classification accuracy.

A Review on Large-Scale Data Processing with Parallel and Distributed Randomized Extreme Learning Machine Neural Networks

The randomization-based feedforward neural network has raised great interest in the scientific community due to its simplicity, training speed, and accuracy comparable to traditional learning algorithms. The basic algorithm consists of randomly determining the weights and biases of the hidden layer and analytically calculating the weights of the output layer by solving a linear overdetermined system using the Moore–Penrose generalized inverse. When processing large volumes of data, randomization-based feedforward neural network models consume large amounts of memory and drastically increase training time. To efficiently solve the above problems, parallel and distributed models have recently been proposed. Previous reviews of randomization-based feedforward neural network models have mainly focused on categorizing and describing the evolution of the algorithms presented in the literature. The main contribution of this paper is to approach the topic from the perspective of the handling of large volumes of data. In this sense, we present a current and extensive review of the parallel and distributed models of randomized feedforward neural networks, focusing on extreme learning machine. In particular, we review the mathematical foundations (Moore–Penrose generalized inverse and solution of linear systems using parallel and distributed methods) and hardware and software technologies considered in current implementations.

A comparative study of an on premise AutoML solution for medical image classification

Automated machine learning (AutoML) allows for the simplified application of machine learning to real-world problems, by the implicit handling of necessary steps such as data pre-processing, feature engineering, model selection and hyperparameter optimization. This has encouraged its use in medical applications such as imaging. However, the impact of common parameter choices such as the number of trials allowed, and the resolution of the input images, has not been comprehensively explored in existing literature. We therefore benchmark AutoKeras (AK), an open-source AutoML framework, against several bespoke deep learning architectures, on five public medical datasets representing a wide range of imaging modalities. It was found that AK could outperform the bespoke models in general, although at the cost of increased training time. Moreover, our experiments suggest that a large number of trials and higher resolutions may not be necessary for optimal performance to be achieved.

TDLC: Tensor decomposition‐based direct learning‐compression algorithm for DNN model compression

SummaryAs a deep neural networks (DNNs) model compression method, learning‐compression (LC) algorithm based on pre‐trained models and matrix decomposition increases training time and ignores the structural information of models. In this manuscript, a tensor decomposition‐based direct LC (TDLC) algorithm without pre‐trained models is proposed. In TDLC, the pre‐trained model is eliminated, and tensor decomposition is first applied to LC algorithm to preserve the structural features of the model. There are two key steps in TDLC. An optimal rank selection method is first proposed in compression‐step (C‐step) of TDLC to find global optimal ranks of tensor decomposition. Second, TDLC utilizes cyclical learning rate, which is different from traditional monotonically learning rates schedule, to improve the generalization performance of uncompressed models in learning‐step (L‐step). TDLC obtains the optimal compression model by alternately optimizing L‐step and C‐step. TDLC is compared with 16 state‐of‐the‐art compression methods in experiments part. Extensive experimental results show that TDLC produces high‐accuracy compression models with high compression rate. Comparing with TDLC‐pre‐trained, TDLC notably achieves 30% training time shorten and 11% parameter reduction in Resnet32, while improving accuracy by 0.2%.

Evaluation of a web-based surgical training approach and insights from the Distal Bypass Competition 2021 using a simulator kit

ObjectiveThe study aimed to assess a novel web-based surgical training approach among young vascular surgeons and validate web-based evaluation methods for assessing anastomosis skills through the online Distal Bypass Competition (DBC). MethodsThe OSATS was used to assess anastomosis proficiency using a simulator among 42 participants (17 junior and 25 senior surgeons) in the DBC 2021. An initial evaluation of basic anastomosis skills was conducted via recorded videos, comparing OSATS scores with instructor-derived impression-based scores (Comprehensive Evaluation Score [CES]) during the nomination round. Real-time evaluations conducted over Zoom focused on assessing technical and composition scores of the anastomoses to determine finalists. The ice-skating method addresses potential variability among judges by excluding the highest and lowest scores for each procedural element from the final evaluation. This method, demonstrated in real time, provides trainees with valuable feedback on which movements are highly rated, enhancing the educational impact. Evaluation criteria included precision in needlework, external appearance, handling of the intima, and internal morphology of the anastomosis. Furthermore, voluntary survey questionnaires were administered to gather participant feedback on their experience with DBC 2021. ResultsThe OSATS was associated with a lower coefficient of variation (CV) than was the CES (0.19 ± 0.07 vs 0.24 ± 0.09) (P = .01) with regard to the evaluation of basic anastomotic skills in the nomination round. Interestingly, there was no significant difference in CV between the Procedural Element Score based on OSATS and CES in the senior class (0.16 ± 0.06 vs 0.22 ± 0.09) (P = .13), but a significant difference was particularly evident in the junior class (0.12 ± 0.07 vs 0.27 ± 09) (P = .03). Lower scores were associated with higher CVs in both classes, but the OSATS improved the CV among participants who had inadequate anastomosis skills. In real-time evaluation, composition scores exhibited higher CVs compared with technical scores, even among high-scoring participants. Notably, the internal morphology of the anastomosis had the largest CV in the composition scores. The ice-skating method decreased the CV as compared with the original data evaluated using the OSATS (0.15 ± 0.04 vs 0.09 ± 0.02 for the technical score [P < .01] and .18 ± 1.00 vs 0.07 ± 0.05 for the composition score [P = .03]); accordingly, the ice-skating method was more conducive to objective, real-time evaluation. The results of the post-DBC survey indicated that 89% of participants reported increased training time, and one-half of these participants indicated that their confidence in their clinical skills related to this procedure increased owing to the process of training for the DBC. ConclusionsThe OSATS yielded more detailed evaluations of anastomosis skill with a lower variability compared with the instructor's impression-based evaluation in this web-based training. Real-time evaluation methods, such as the ice-skating method, have proven to be valuable tools for the standardized assessment of web-based surgical education.

An integrated approach for magnification independent breast cancer classification

Breast cancer (BC) is the second most common cancer worldwide, and its diagnosis and treatment rely heavily on histopathology image analysis. Moreover, continuous evaluation of deep learning techniques in medical imaging modalities is essential for improving early diagnosis. The Break-His dataset is the de facto standard for such analysis. Researchers have actively explored various approaches to classifying these images into binary, four, and eight based on the magnification-independent (MI) and magnification-dependent (MD). However, existing processes necessitate separate models for each classification based on the magnification factor and entail training multiple models. It leads to increased training time and model complexity. Moreover, model performance in eight-class is relatively low compared to other classifications. This paper introduces a modified Swin-Transformer V2 architecture for multi-labeled and eight-class classification on BC histopathology images. Further, the eight-class output was reconfigured to operate in an integrated approach, allowing the model to predict binary, four, and eight classes simultaneously for both MI and MD. On the Break-His dataset, the proposed model achieved an accuracy of 98.27%, 97.95%, 98.97%, and 99.16% for eight-class, malignant, benign, and binary classification tasks, respectively. Additionally, the multi-labeled approach demonstrated a performance improvement in the eight-class category compared to current leading models, achieving an enhancement of up to 5%, and the model was trained in just 2 h. Overall, the proposed model can improve the prognosis of breast cancer patients by leading to more accurate diagnosis and treatment. The implemented code can be accessed at https://github.com/kbnarayanavit/integrated-approach.

Classification of Malicious URLs Using Machine Learning.

Amid the rapid proliferation of thousands of new websites daily, distinguishing safe ones from potentially harmful ones has become an increasingly complex task. These websites often collect user data, and, without adequate cybersecurity measures such as the efficient detection and classification of malicious URLs, users' sensitive information could be compromised. This study aims to develop models based on machine learning algorithms for the efficient identification and classification of malicious URLs, contributing to enhanced cybersecurity. Within this context, this study leverages support vector machines (SVMs), random forests (RFs), decision trees (DTs), and k-nearest neighbors (KNNs) in combination with Bayesian optimization to accurately classify URLs. To improve computational efficiency, instance selection methods are employed, including data reduction based on locality-sensitive hashing (DRLSH), border point extraction based on locality-sensitive hashing (BPLSH), and random selection. The results show the effectiveness of RFs in delivering high precision, recall, and F1 scores, with SVMs also providing competitive performance at the expense of increased training time. The results also emphasize the substantial impact of the instance selection method on the performance of these models, indicating its significance in the machine learning pipeline for malicious URL classification.

Text Generation using Long Short Term Memory to Generate a LinkedIn Post

LinkedIn is one of the most popular sites out there to advertise oneself to potential employer. This study aims to create a good enough text generation model that it can generate a text as if it were made by someone who posts on LinkedIn. This study will use a Neural Network layer called Long Short Term Memory (LSTM) as the main algorithm and the train data consists of actual posts made by users in LinkedIn. LSTM is an algorithm that is created to reduce vanishing and exploding gradient problem in Neural Network. From the result, final accuracy and loss varies. Increasing learning rate from its default value of 0.001, to 0.01, or even 0.1 creates worse model. Meanwhile, increasing dimensions of LSTM will sometimes increases training time or decreases it while not really increasing model performance. In the end, models chosen at the end are models with around 97% of accuracy. From this study, it can be concluded that it is possible to use LSTM to create a text generation model. However, the result might not be too satisfying. For future work, it is advised to instead use a newer model, such as the Transformer model.

COST-WINNERS: COST reduction WIth Neural NEtworks-based augmented Random Search for simultaneous thermal and electrical energy storage control

The combination of local renewable energy production, dynamic loads, and multiple energy storage systems with different dynamics requires sophisticated control systems to maximize the energy cost efficiency of the combined energy system. Battery and thermal energy storage systems can be combined to increase the local use of on-site renewable energy, reduce peak power demand, and exploit time-of-use energy pricing. In this paper, we focus on how the augmented random search algorithm and artificial neural networks can be used together to solve an energy cost optimization problem involving the control of a battery energy storage system and a thermal energy storage system at the same time in a smart warehouse. As part of this work, a simulated training environment made using the data from the smart warehouse’s operations. In addition to the energy storage systems, the warehouse energy system has integrated a large roof mounted photovoltaic power plant and an industrial-scale cooling system.The developed solution is able to minimize the energy costs by modulating both energy systems, depending on the situation. Additionally, when it is tested against the state-of-the-art solutions, our developed solution at worst matches performance when the alternative algorithm is allowed to increase training time by a factor of nearly three. On average, our presented solution doubles the performance of the benchmark algorithm with much less computational resource expenditure.