A survey of uncertainty in deep neural networks

Artificial Intelligence Review, Journal Year: 2023, Volume and Issue: 56(S1), P. 1513 - 1589

Published: July 29, 2023

Abstract Over the last decade, neural networks have reached almost every field of science and become a crucial part various real world applications. Due to increasing spread, confidence in network predictions has more important. However, basic do not deliver certainty estimates or suffer from over- under-confidence, i.e. are badly calibrated. To overcome this, many researchers been working on understanding quantifying uncertainty network’s prediction. As result, different types sources identified approaches measure quantify proposed. This work gives comprehensive overview estimation networks, reviews recent advances field, highlights current challenges, identifies potential research opportunities. It is intended give anyone interested broad introduction, without presupposing prior knowledge this field. For that, introduction most given their separation into reducible model irreducible data presented. The modeling these uncertainties based deterministic Bayesian (BNNs), ensemble test-time augmentation introduced branches fields as well latest developments discussed. practical application, we discuss measures uncertainty, for calibrating an existing baselines available implementations. Different examples wide spectrum challenges medical image analysis, robotics, earth observation idea needs regarding applications networks. Additionally, limitations quantification methods mission- safety-critical discussed outlook next steps towards broader usage such given.

Language: Английский

---

Virtual Screening Algorithms in Drug Discovery: A Review Focused on Machine and Deep Learning Methods

Drugs and Drug Candidates, Journal Year: 2023, Volume and Issue: 2(2), P. 311 - 334

Published: May 5, 2023

Drug discovery and repositioning are important processes for the pharmaceutical industry. These demand a high investment in resources time-consuming. Several strategies have been used to address this problem, including computer-aided drug design (CADD). Among CADD approaches, it is essential highlight virtual screening (VS), an silico approach based on computer simulation that can select organic molecules toward therapeutic targets of interest. The techniques applied by VS structure ligands (LBVS), receptors (SBVS), or fragments (FBVS). Regardless type be applied, they divided into categories depending algorithms: similarity-based, quantitative, machine learning, meta-heuristics, other algorithms. Each category has its objectives, advantages, disadvantages. This review presents overview algorithms VS, describing them showing their use contribution development process.

Language: Английский

---

A systematic review of data fusion techniques for optimized structural health monitoring

Information Fusion, Journal Year: 2023, Volume and Issue: 103, P. 102136 - 102136

Published: Nov. 10, 2023

Advancements in structural health monitoring (SHM) techniques have spiked the past few decades due to rapid evolution of novel sensing and data transfer technologies. This development has facilitated simultaneous recording a wide range data, which could contain abundant damage-related features. Concurrently, age omnipresent started with massive amounts SHM collected from large-size heterogeneous sensor networks. The abundance information diverse sources needs be aggregated enable robust decision-making strategies. Data fusion is process integrating various produce more useful, accurate, reliable about system behavior. paper reviews recent developments applied systems. theoretical concepts, applications, benefits, limitations current methods challenges are presented, future trends discussed. Furthermore, set criteria proposed evaluate contents original review papers this field, road map provided discussing possible work.

Language: Английский

---

Machine learning and deep learning-based approach in smart healthcare: Recent advances, applications, challenges and opportunities

AIMS Public Health, Journal Year: 2024, Volume and Issue: 11(1), P. 58 - 109

Published: Jan. 1, 2024

<abstract> <p>In recent years, machine learning (ML) and deep (DL) have been the leading approaches to solving various challenges, such as disease predictions, drug discovery, medical image analysis, etc., in intelligent healthcare applications. Further, given current progress fields of ML DL, there exists promising potential for both provide support realm healthcare. This study offered an exhaustive survey on DL system, concentrating vital state art features, integration benefits, applications, prospects future guidelines. To conduct research, we found most prominent journal conference databases using distinct keywords discover scholarly consequences. First, furnished along with cutting-edge ML-DL-based analysis smart a compendious manner. Next, integrated advancement services including ML-healthcare, DL-healthcare, ML-DL-healthcare. We then DL-based applications industry. Eventually, emphasized research disputes recommendations further studies based our observations.</p> </abstract>

Language: Английский

---

Uncertainty-aware diabetic retinopathy detection using deep learning enhanced by Bayesian approaches

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 8, 2025

Abstract Deep learning-based medical image analysis has shown strong potential in disease categorization, segmentation, detection, and even prediction. However, high-stakes complex domains like healthcare, the opaque nature of these models makes it challenging to trust predictions, particularly uncertain cases. This sort uncertainty can be crucial analysis; diabetic retinopathy is an example where slight errors without indication confidence have adverse impacts. Traditional deep learning rely on single-point limiting their ability provide measures essential for robust clinical decision-making. To solve this issue, Bayesian approximation approaches evolved are gaining market traction. In work, we implemented a transfer approach, building upon DenseNet-121 convolutional neural network detect retinopathy, followed by extensions trained model. techniques, including Monte Carlo Dropout, Mean Field Variational Inference, Deterministic were applied represent posterior predictive distribution, allowing us evaluate model predictions. Our experiments combined dataset (APTOS 2019 + DDR) with pre-processed images showed that Bayesian-augmented outperforms state-of-the-art test accuracy, achieving 97.68% Dropout model, 94.23% 91.44% We also measure how certain predictions are, using entropy standard deviation metric each approach. evaluated both AUC accuracy scores at multiple data retention levels. addition overall performance boosts, results highlight does not only improve classification detection but reveals beneficial insights about estimation help build more trustworthy decision-making solutions.

Language: Английский

---

A Review on Deep Sequential Models for Forecasting Time Series Data

Applied Computational Intelligence and Soft Computing, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 19

Published: June 3, 2022

Deep sequential (DS) models are extensively employed for forecasting time series data since the dawn of deep learning era, and they provide forecasts values required in subsequent steps. DS models, unlike other traditional statistical data, can learn hidden patterns temporal sequences have memorizing from prior points. Given widespread usage several domains, a comprehensive study describing their applications is necessary. This work presents review contemporary performance diverse an investigation that were various applications. Three namely, artificial neural network (ANN), long short-term memory (LSTM), temporal-conventional (TCNN) along with elaborated. We showed comparison between such terms application fields, model structure activation functions, optimizers, implementation, goal more about optimal used. Furthermore, challenges perspectives future development presented discussed. conclude LSTM widely employed, particularly form hybrid model, which most accurate predictions made when shape hybrids used as model.

Language: Английский

---

Identification of kidney stones in KUB X-ray images using VGG16 empowered with explainable artificial intelligence

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: March 14, 2024

Abstract A kidney stone is a solid formation that can lead to failure, severe pain, and reduced quality of life from urinary system blockages. While medical experts interpret kidney-ureter-bladder (KUB) X-ray images, specific images pose challenges for human detection, requiring significant analysis time. Consequently, developing detection becomes crucial accurately classifying KUB images. This article applies transfer learning (TL) model with pre-trained VGG16 empowered explainable artificial intelligence (XAI) establish takes categorizes them as stones or normal cases. The findings demonstrate the achieves testing accuracy 97.41% in identifying X-rays dataset used. delivers highly accurate predictions but lacks fairness explainability their decision-making process. study incorporates Layer-Wise Relevance Propagation (LRP) technique, an enhance transparency effectiveness address this concern. XAI specifically LRP, increases model's transparency, facilitating comprehension predictions. play important role assisting doctors identification stones, thereby execution effective treatment strategies.

Language: Английский

---

Amazigh CNN speech recognition system based on Mel spectrogram feature extraction method

International Journal of Speech Technology, Journal Year: 2024, Volume and Issue: 27(1), P. 287 - 296

Published: March 1, 2024

Language: Английский

---

Intelligent skin disease prediction system using transfer learning and explainable artificial intelligence

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 11, 2025

Abstract Skin diseases impact millions of people around the world and pose a severe risk to public health. These have wide range effects on skin’s structure, functionality, appearance. Identifying predicting skin are laborious processes that require complete physical examination, review patient’s medical history, proper laboratory diagnostic testing. Additionally, it necessitates significant number histological clinical characteristics for examination subsequent treatment. As disease’s complexity quantity features grow, identifying becomes more challenging. This research proposes deep learning (DL) model utilizing transfer (TL) quickly identify like chickenpox, measles, monkeypox. A pre-trained VGG16 is used learning. The can predict by symptom patterns. Images from four classes monkeypox, normal included in dataset. dataset separated into training experimental results performed demonstrate with 93.29% testing accuracy. However, does not explain why how system operates because models black boxes. Deep models’ opacity stands way their widespread application healthcare sector. In order make this valuable health sector, article employs layer-wise relevance propagation (LRP) determine scores each input. identified symptoms provide insights could support timely diagnosis treatment decisions diseases.

Language: Английский

---

A Smart Gas Sensor Using Machine Learning Algorithms: Sensor Types Based on IED Configurations, Fabrication Techniques, Algorithmic Approaches, Challenges, Progress, and Limitations: A Review

IEEE Sensors Journal, Journal Year: 2023, Volume and Issue: 23(11), P. 11336 - 11355

Published: April 24, 2023

Over the past decade, machine learning (ML) and artificial intelligence (AI) have attracted great interest in research various practical applications. Currently, smart, fast, high sensitivity with excellent selectivity are becoming increasingly interesting due to need for environmental safety medical The main challenge is improve sensor selectivity, which requires combination of interdisciplinary areas successfully develop smart gas/chemical sensing devices better performance. In this review, we present a few principles gas based on low-cost interdigital electrodes (IDEs), such as electrochemical, resistive, capacitive, acoustic sensors. addition, most important current methods improving performance, different materials, techniques used fabricate IDE sensors, their advantages limitations presented. comparison between ML AI algorithms pattern recognition classification also discussed. discussion then establishes application cases algorithms, provide efficient data processing methods, design sensors that highly selective. challenges applications critically study shows importance structural optimization sensitive, selective

Language: Английский

---