RAP-Optimizer: Resource-Aware Predictive Model for Cost Optimization of Cloud AIaaS Applications

Published: Oct. 8, 2024

AI-driven applications are rapidly growing, and more joining the market competition. As a result, AI-as-a-Service (AIaaS) model is experiencing rapid growth. Many of these AIaas-based not properly optimized initially. Once they start large volume traffic, different challenges revealing themselves. One maintaining profit margin for sustainability AIaaS application-based business model, which depends on proper utilization computing resources. This paper introduces Resource Award Predictive (RAP) cost optimization called RAP-Optimizer. It developed by combining Deep Neural Network (DNN) with simulated annealing algorithm. designed to reduce resource underutilization minimize number active hosts in cloud environments. dynamically allocates resources handles API requests efficiently. The RAP-Optimizer reduces physical an average 5 per day, leading 45% decrease server costs. impact has been observed over 12-month period. observational data show significant improvement utilization. effectively operational costs from $2,600 $1,250 month. Furthermore, increases 179%, $600 $1,675 inclusion Dynamic Dropout Control (DDC) algorithm DNN training process mitigates overfitting, achieving 97.48% validation accuracy loss 2.82%. These results indicate that enhances management cost-efficiency application, making it valuable solution modern

Language: Английский

---

LW-MorphCNN: A lightweight morphological attention-based subtype classification network for lung cancer

Measurement Science and Technology, Journal Year: 2024, Volume and Issue: 36(1), P. 015703 - 015703

Published: Oct. 23, 2024

Abstract Lung cancer is generally considered one of the most deadly cancers globally. If it can be identified early and diagnosed correctly, survival probability patients significantly improved. In this process, histopathological examination a commonly used method for diagnosing detecting lung cancer. It crucial to accurately identify subtypes from images, as helps doctors formulate effective treatment plans. However, visual inspection in diagnosis requires large amount time also depends on subjective perception clinicians. Therefore, paper proposes lightweight subtype classification network based morphological attention (LW-MorphCNN), which automatically classify images benign tumors, ADC (adenocarcinoma), SCC (squamous cell carcinoma) provided public dataset LC25000 (Lung Colon). This takes input conducts comparative analysis with classic networks such VGG16, VGG19, DenseNet121, ResNet50, well existing methods proposed same work. The superior other terms parameter quantity performance, an accuracy rate F1 - score reaching 99.47% 99.44% respectively. Clinicians install LW-MorphCNN hospital confirm results.

Language: Английский

---

Deep Learning and Handcrafted Features for Thyroid Nodule Classification

International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(6)

Published: Nov. 1, 2024

ABSTRACT In this research, we present a refined image‐based computer‐aided diagnosis (CAD) system for thyroid cancer detection using ultrasound imagery. This integrates specialized convolutional neural network (CNN) architecture designed to address the unique aspects of image datasets. Additionally, it incorporates novel statistical model that utilizes two‐dimensional random coefficient autoregressive (2D‐RCA) method precisely analyze textural characteristics images, thereby capturing essential texture‐related information. The classification framework relies on composite feature vector combines deep learning features from CNN and handcrafted 2D‐RCA model, processed through support machine (SVM) algorithm. Our evaluation methodology is structured in three phases: initial assessment features, analysis CNN‐derived final their combined effect performance. Comparative analyses with well‐known networks such as VGG16, VGG19, ResNet50, AlexNet highlight superior performance our approach. outcomes indicate significant enhancement diagnostic accuracy, achieving accuracy 97.2%, sensitivity 84.42%, specificity 95.23%. These results not only demonstrate notable advancement nodules but also establish new standard efficiency CAD systems.

Language: Английский

---

RAP-Optimizer: Resource-Aware Predictive Model for Cost Optimization of Cloud AIaaS Applications

Electronics, Journal Year: 2024, Volume and Issue: 13(22), P. 4462 - 4462

Published: Nov. 14, 2024

Artificial Intelligence (AI) applications are rapidly growing, and more joining the market competition. As a result, AI-as-a-service (AIaaS) model is experiencing rapid growth. Many of these AIaaS-based not properly optimized initially. Once they start large volume traffic, different challenges revealing themselves. One maintaining profit margin for sustainability AIaaS application-based business model, which depends on proper utilization computing resources. This paper introduces resource award predictive (RAP) cost optimization called RAP-Optimizer. It developed by combining deep neural network (DNN) with simulated annealing algorithm. designed to reduce underutilization minimize number active hosts in cloud environments. dynamically allocates resources handles API requests efficiently. The RAP-Optimizer reduces physical an average 5 per day, leading 45% decrease server costs. impact was observed over 12-month period. observational data show significant improvement utilization. effectively operational costs from USD 2600 1250 month. Furthermore, increases 179%, 600 1675 inclusion dynamic dropout control (DDC) algorithm DNN training process mitigates overfitting, achieving 97.48% validation accuracy loss 2.82%. These results indicate that enhances management cost-efficiency applications, making it valuable solution modern

Language: Английский

---

Enhanced Lung Cancer Prediction via Integrated Multi‐Space Feature Adaptation, Collaborative Alignment and Disentanglement Learning

Engineering Reports, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 2, 2024

ABSTRACT Lung cancer, marked by the rapid and uncontrolled proliferation of abnormal cells in lungs, continues to be one leading causes cancer‐related deaths globally. Early accurate diagnosis is critical for improving patient outcomes. This research presents an enhanced lung cancer prediction model integrating Adaptation Multiple Spaces Feature L1‐norm Regularization (AMSF‐L1ELM) with Primitive Generation Collaborative Relationship Alignment Disentanglement Learning (PADing). Initially, AMSF‐L1ELM was employed address challenges feature alignment multi‐domain adaptation, achieving a baseline performance test accuracy 83.20%, precision 83.43%, recall 83.74%, F1‐score 83.07%. After incorporating PADing, exhibited significant improvements, increasing 98.07%, 98.11%, 98.05%, 98.06%, ROC‐AUC 100%. Cross‐validation results further validated model's robustness, average 99.73%, 99.55%, 99.64%, 99.64% across five folds. The study utilized four distinct datasets covering range imaging modalities diagnostic labels: Chest CT‐Scan dataset from Kaggle, NSCLC‐Radiomics‐Interobserver1 TCIA, LungCT‐Diagnosis IQ‐OTH/NCCD Kaggle. In total, 4085 images were selected, distributed between source target domains. These demonstrate effectiveness PADing enhancing multiple domains complex medical data.

Language: Английский

---