Search results for: naive Bayes classifier

Authors: Yulan Wu

Abstract:

The spread of fake news on social media has posed significant societal harm to the public and the nation, with its threats spanning various domains, including politics, economics, health, and more. News on social media often covers multiple domains, and existing models studied by researchers and relevant organizations often perform well on datasets from a single domain. However, when these methods are applied to social platforms with news spanning multiple domains, their performance significantly deteriorates. Existing research has attempted to enhance the detection performance of multi-domain datasets by adding single-domain labels to the data. However, these methods overlook the fact that a news article typically belongs to multiple domains, leading to the loss of domain knowledge information contained within the news text. To address this issue, research has found that news records in different domains often use different vocabularies to describe their content. In this paper, we propose a fake news detection framework that combines domain knowledge and expert knowledge. Firstly, it utilizes an unsupervised domain discovery module to generate a low-dimensional vector for each news article, representing domain embeddings, which can retain multi-domain knowledge of the news content. Then, a feature extraction module uses the domain embeddings discovered through unsupervised domain knowledge to guide multiple experts in extracting news knowledge for the total feature representation. Finally, a classifier is used to determine whether the news is fake or not. Experiments show that this approach can improve multi-domain fake news detection performance while reducing the cost of manually labeling domain labels.

Keywords: fake news, deep learning, natural language processing, multiple domains

Procedia PDF Downloads 39

Authors: Jigg L. Pelayo, Ricardo G. Villar, Einstine M. Opiso

Abstract:

The advances of aerial laser scanning in the Philippines has open-up entire fields of research in remote sensing and machine vision aspire to provide accurate timely information for the government and the public. Rapid mapping of polygonal roads and roof boundaries is one of its utilization offering application to disaster risk reduction, mitigation and development. The study uses low density LiDAR data and high resolution aerial imagery through object-oriented approach considering the theoretical concept of data analysis subjected to machine learning algorithm in minimizing the constraints of feature extraction. Since separating one class from another in distinct regions of a multi-dimensional feature-space, non-trivial computing for fitting distribution were implemented to formulate the learned ideal hyperplane. Generating customized hybrid feature which were then used in improving the classifier findings. Supplemental algorithms for filtering and reshaping object features are develop in the rule set for enhancing the final product. Several advantages in terms of simplicity, applicability, and process transferability is noticeable in the methodology. The algorithm was tested in the different random locations of Misamis Oriental province in the Philippines demonstrating robust performance in the overall accuracy with greater than 89% and potential to semi-automation. The extracted results will become a vital requirement for decision makers, urban planners and even the commercial sector in various assessment processes.

Keywords: feature extraction, machine learning, OBIA, remote sensing

Procedia PDF Downloads 338

Authors: Gong Zhilin, Jing Yang, Jian Yin

Abstract:

The purpose of this paper is to develop a novel system for credit card fraud detection based on sequential modeling of data using hybrid deep learning models. The projected model encapsulates five major phases are pre-processing, imbalance-data handling, feature extraction, optimal feature selection, and fraud detection with an ensemble classifier. The collected raw data (input) is pre-processed to enhance the quality of the data through alleviation of the missing data, noisy data as well as null values. The pre-processed data are class imbalanced in nature, and therefore they are handled effectively with the K-means clustering-based SMOTE model. From the balanced class data, the most relevant features like improved Principal Component Analysis (PCA), statistical features (mean, median, standard deviation) and higher-order statistical features (skewness and kurtosis). Among the extracted features, the most optimal features are selected with the Self-improved Arithmetic Optimization Algorithm (SI-AOA). This SI-AOA model is the conceptual improvement of the standard Arithmetic Optimization Algorithm. The deep learning models like Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and optimized Quantum Deep Neural Network (QDNN). The LSTM and CNN are trained with the extracted optimal features. The outcomes from LSTM and CNN will enter as input to optimized QDNN that provides the final detection outcome. Since the QDNN is the ultimate detector, its weight function is fine-tuned with the Self-improved Arithmetic Optimization Algorithm (SI-AOA).

Keywords: credit card, data mining, fraud detection, money transactions

Procedia PDF Downloads 104

Authors: Keerti Kembhavi, M. R. Archana, V. Anjaneyappa

Abstract:

Most asphalt pavement condition evaluation use rating frameworks in which asphalt pavement distress is estimated by type, extent, and severity. Rating is carried out by the pavement condition rating (PCR), which is tedious and expensive. This paper presents the development of a low-cost technique for image pavement distress analysis that permits the identification of pothole and cracks. The paper explores the application of image processing tools for the detection of potholes and cracks. Longitudinal cracking and pothole are detected using Fuzzy-C- Means (FCM) and proceeded with the Spectral Theory algorithm. The framework comprises three phases, including image acquisition, processing, and extraction of features. A digital camera (Gopro) with the holder is used to capture pavement distress images on a moving vehicle. FCM classifier and Spectral Theory algorithms are used to compute features and classify the longitudinal cracking and pothole. The Matlab2016Ra Image preparing tool kit utilizes performance analysis to identify the viability of pavement distress on selected urban stretches of Bengaluru city, India. The outcomes of image evaluation with the utilization semi-computerized image handling framework represented the features of longitudinal crack and pothole with an accuracy of about 80%. Further, the detected images are validated with the actual dimensions, and it is seen that dimension variability is about 0.46. The linear regression model y=1.171x-0.155 is obtained using the existing and experimental / image processing area. The R2 correlation square obtained from the best fit line is 0.807, which is considered in the linear regression model to be ‘large positive linear association’.

Keywords: crack detection, pothole detection, spectral clustering, fuzzy-c-means

Procedia PDF Downloads 154

Authors: Libor Zachoval, Daire O Broin, Oisin Cawley

Abstract:

E-learning platforms, such as Blackboard have two major shortcomings: limited data capture as a result of the limitations of SCORM (Shareable Content Object Reference Model), and lack of incorporation of Artificial Intelligence (AI) and machine learning algorithms which could lead to better course adaptations. With the recent development of Experience Application Programming Interface (xAPI), a large amount of additional types of data can be captured and that opens a window of possibilities from which online education can benefit. In a corporate setting, where companies invest billions on the learning and development of their employees, some learner behaviours can be troublesome for they can hinder the knowledge development of a learner. Behaviours that hinder the knowledge development also raise ambiguity about learner’s knowledge mastery, specifically those related to gaming the system. Furthermore, a company receives little benefit from their investment if employees are passing courses without possessing the required knowledge and potential compliance risks may arise. Using xAPI and rules derived from a state-of-the-art review, we identified three learner behaviours, primarily related to guessing, in a corporate compliance course. The identified behaviours are: trying each option for a question, specifically for multiple-choice questions; selecting a single option for all the questions on the test; and continuously repeating tests upon failing as opposed to going over the learning material. These behaviours were detected on learners who repeated the test at least 4 times before passing the course. These findings suggest that gauging the mastery of a learner from multiple-choice questions test scores alone is a naive approach. Thus, next steps will consider the incorporation of additional data points, knowledge estimation models to model knowledge mastery of a learner more accurately, and analysis of the data for correlations between knowledge development and identified learner behaviours. Additional work could explore how learner behaviours could be utilised to make changes to a course. For example, course content may require modifications (certain sections of learning material may be shown to not be helpful to many learners to master the learning outcomes aimed at) or course design (such as the type and duration of feedback).

Keywords: artificial intelligence, corporate e-learning environment, knowledge maintenance, xAPI

Procedia PDF Downloads 90

Authors: Faiza Belbachir, Thibault Schienhinski

Abstract:

The massive development of online social networks allows users to post and share their opinions on various topics. With this huge volume of opinion, it is interesting to extract and interpret these information for different domains, e.g., product and service benchmarking, politic, system of recommendation. This is why opinion detection is one of the most important research tasks. It consists on differentiating between opinion data and factual data. The difficulty of this task is to determine an approach which returns opinionated document. Generally, there are two approaches used for opinion detection i.e. Lexical based approaches and Machine Learning based approaches. In Lexical based approaches, a dictionary of sentimental words is used, words are associated with weights. The opinion score of document is derived by the occurrence of words from this dictionary. In Machine learning approaches, usually a classifier is trained using a set of annotated document containing sentiment, and features such as n-grams of words, part-of-speech tags, and logical forms. Majority of these works are based on documents text to determine opinion score but dont take into account if these texts are really correct. Thus, it is interesting to exploit other information to improve opinion detection. In our work, we will develop a new way to consider the opinion score. We introduce the notion of trust score. We determine opinionated documents but also if these opinions are really trustable information in relation with topics. For that we use lexical SentiWordNet to calculate opinion and trust scores, we compute different features about users like (numbers of their comments, numbers of their useful comments, Average useful review). After that, we combine opinion score and trust score to obtain a final score. We applied our method to detect trust opinions in TRIPADVISOR collection. Our experimental results report that the combination between opinion score and trust score improves opinion detection.

Keywords: Tripadvisor, opinion detection, SentiWordNet, trust score

Procedia PDF Downloads 168

Authors: Adesuyi Ayodeji Steve, Zahn Munch

Abstract:

This study investigates the use of MODIS NDVI to identify agricultural land cover change areas on an annual time step (2007 - 2012) and characterize the trend in the study area. An ISODATA classification was performed on the MODIS imagery to select only the agricultural class producing 3 class groups namely: agriculture, agriculture/semi-natural, and semi-natural. NDVI signatures were created for the time series to identify areas dominated by cereals and vineyards with the aid of ancillary, pictometry and field sample data. The NDVI signature curve and training samples aided in creating a decision tree model in WEKA 3.6.9. From the training samples two classification models were built in WEKA using decision tree classifier (J48) algorithm; Model 1 included ISODATA classification and Model 2 without, both having accuracies of 90.7% and 88.3% respectively. The two models were used to classify the whole study area, thus producing two land cover maps with Model 1 and 2 having classification accuracies of 77% and 80% respectively. Model 2 was used to create change detection maps for all the other years. Subtle changes and areas of consistency (unchanged) were observed in the agricultural classes and crop practices over the years as predicted by the land cover classification. 41% of the catchment comprises of cereals with 35% possibly following a crop rotation system. Vineyard largely remained constant over the years, with some conversion to vineyard (1%) from other land cover classes. Some of the changes might be as a result of misclassification and crop rotation system.

Keywords: change detection, land cover, modis, NDVI

Procedia PDF Downloads 374

Authors: Syed Asif Hassan, Syed Atif Hassan

Abstract:

Multidrug-resistant Tuberculosis (MDR-TB) is an infection caused by the resistant strains of Mycobacterium tuberculosis that do not respond either to isoniazid or rifampicin, which are the most important anti-TB drugs. The increase in the occurrence of a drug-resistance strain of MTB calls for an intensive search of novel target-based therapeutics. In this context LprG (Rv1411c) a lipoprotein from MTB plays a pivotal role in the immune evasion of Mtb leading to survival and propagation of the bacterium within the host cell. Therefore, a machine learning method will be developed for generating a computational model that could predict for a potential anti LprG activity of the novel antituberculosis compound. The present study will utilize dataset from PubChem database maintained by National Center for Biotechnology Information (NCBI). The dataset involves compounds screened against MTB were categorized as active and inactive based upon PubChem activity score. PowerMV, a molecular descriptor generator, and visualization tool will be used to generate the 2D molecular descriptors for the actives and inactive compounds present in the dataset. The 2D molecular descriptors generated from PowerMV will be used as features. We feed these features into three different classifiers, namely, random forest, a deep neural network, and a recurring neural network, to build separate predictive models and choosing the best performing model based on the accuracy of predicting novel antituberculosis compound with an anti LprG activity. Additionally, the efficacy of predicted active compounds will be screened using SMARTS filter to choose molecule with drug-like features.

Keywords: antituberculosis drug, classifier, machine learning, molecular descriptors, prediction

Procedia PDF Downloads 359

Authors: Annalakshmi G., Sakthivel Murugan S.

Abstract:

This paper presents a local directional encoded derivative binary pattern (LDEDBP) feature extraction method that can be applied for the classification of submarine coral reef images. The classification of coral reef images using texture features is difficult due to the dissimilarities in class samples. In coral reef image classification, texture features are extracted using the proposed method called local directional encoded derivative binary pattern (LDEDBP). The proposed approach extracts the complete structural arrangement of the local region using local binary batten (LBP) and also extracts the edge information using local directional pattern (LDP) from the edge response available in a particular region, thereby achieving extra discriminative feature value. Typically the LDP extracts the edge details in all eight directions. The process of integrating edge responses along with the local binary pattern achieves a more robust texture descriptor than the other descriptors used in texture feature extraction methods. Finally, the proposed technique is applied to an extreme learning machine (ELM) method with a meta-heuristic algorithm known as weighted distance grey wolf optimizer (GWO) to optimize the input weight and biases of single-hidden-layer feed-forward neural networks (SLFN). In the empirical results, ELM-WDGWO demonstrated their better performance in terms of accuracy on all coral datasets, namely RSMAS, EILAT, EILAT2, and MLC, compared with other state-of-the-art algorithms. The proposed method achieves the highest overall classification accuracy of 94% compared to the other state of art methods.

Keywords: feature extraction, local directional pattern, ELM classifier, GWO optimization

Procedia PDF Downloads 139

Authors: Kaiwen Zheng, Wanting Zhou, Nan Tang, Lei Li, Yuanhang He

Abstract:

The integrated circuit industry has become a cornerstone of the information society, finding widespread application in areas such as industry, communication, medicine, and aerospace. However, with the increasing complexity of integrated circuits, Hardware Trojans (HTs) implanted by attackers have become a significant threat to their security. In this paper, we proposed a hardware trojan detection method for large-scale circuits. As HTs introduce physical characteristic changes such as structure, area, and power consumption as additional redundant circuits, we proposed a machine-learning-based hardware trojan detection method based on the physical characteristics of gate-level netlists. This method transforms the hardware trojan detection problem into a machine-learning binary classification problem based on physical characteristics, greatly improving detection speed. To address the problem of imbalanced data, where the number of pure circuit samples is far less than that of HTs circuit samples, we used the SMOTETomek algorithm to expand the dataset and further improve the performance of the classifier. We used three machine learning algorithms, K-Nearest Neighbors, Random Forest, and Support Vector Machine, to train and validate benchmark circuits on Trust-Hub, and all achieved good results. In our case studies based on AES encryption circuits provided by trust-hub, the test results showed the effectiveness of the proposed method. To further validate the method’s effectiveness for detecting variant HTs, we designed variant HTs using open-source HTs. The proposed method can guarantee robust detection accuracy in the millisecond level detection time for IC, and FPGA design flows and has good detection performance for library variant HTs.

Keywords: hardware trojans, physical properties, machine learning, hardware security

Procedia PDF Downloads 111

Authors: Yulan Wu

Abstract:

With the rapid development of social media, the issue of fake news has gained considerable prominence, drawing the attention of both the public and governments. The widespread dissemination of false information poses a tangible threat across multiple domains of society, including politics, economy, and health. However, much research has concentrated on supervised training models within specific domains, their effectiveness diminishes when applied to identify fake news across multiple domains. To solve this problem, some approaches based on domain labels have been proposed. By segmenting news to their specific area in advance, judges in the corresponding field may be more accurate on fake news. However, these approaches disregard the fact that news records can pertain to multiple domains, resulting in a significant loss of valuable information. In addition, the datasets used for training must all be domain-labeled, which creates unnecessary complexity. To solve these problems, an unsupervised domain knowledge discovery framework for fake news detection is proposed. Firstly, to effectively retain the multidomain knowledge of the text, a low-dimensional vector for each news text to capture domain embeddings is generated. Subsequently, a feature extraction module utilizing the unsupervisedly discovered domain embeddings is used to extract the comprehensive features of news. Finally, a classifier is employed to determine the authenticity of the news. To verify the proposed framework, a test is conducted on the existing widely used datasets, and the experimental results demonstrate that this method is able to improve the detection performance for fake news across multiple domains. Moreover, even in datasets that lack domain labels, this method can still effectively transfer domain knowledge, which can educe the time consumed by tagging without sacrificing the detection accuracy.

Keywords: fake news, deep learning, natural language processing, multiple domains

Procedia PDF Downloads 55

Authors: V. Veeraprathap, G. S. Harish, G. Narendra Kumar

Abstract:

Uncontrolled growth of abnormal cells in the lung in the form of tumor can be either benign (non-cancerous) or malignant (cancerous). Patients with Lung Cancer (LC) have an average of five years life span expectancy provided diagnosis, detection and prediction, which reduces many treatment options to risk of invasive surgery increasing survival rate. Computed Tomography (CT), Positron Emission Tomography (PET), and Magnetic Resonance Imaging (MRI) for earlier detection of cancer are common. Gaussian filter along with median filter used for smoothing and noise removal, Histogram Equalization (HE) for image enhancement gives the best results without inviting further opinions. Lung cavities are extracted and the background portion other than two lung cavities is completely removed with right and left lungs segmented separately. Region properties measurements area, perimeter, diameter, centroid and eccentricity measured for the tumor segmented image, while texture is characterized by Gray-Level Co-occurrence Matrix (GLCM) functions, feature extraction provides Region of Interest (ROI) given as input to classifier. Two levels of classifications, K-Nearest Neighbor (KNN) is used for determining patient condition as normal or abnormal, while Artificial Neural Networks (ANN) is used for identifying the cancer stage is employed. Discrete Wavelet Transform (DWT) algorithm is used for the main feature extraction leading to best efficiency. The developed technology finds encouraging results for real time information and on line detection for future research.

Keywords: artificial neural networks, ANN, discrete wavelet transform, DWT, gray-level co-occurrence matrix, GLCM, k-nearest neighbor, KNN, region of interest, ROI

Procedia PDF Downloads 125

Authors: Sangita Pokhrel, Nalinda Somasiri, Rebecca Jeyavadhanam, Swathi Ganesan

Abstract:

Tourism is a booming industry with huge future potential for global wealth and employment. There are countless data generated over social media sites every day, creating numerous opportunities to bring more insights to decision-makers. The integration of Big Data Technology into the tourism industry will allow companies to conclude where their customers have been and what they like. This information can then be used by businesses, such as those in charge of managing visitor centers or hotels, etc., and the tourist can get a clear idea of places before visiting. The technical perspective of natural language is processed by analysing the sentiment features of online reviews from tourists, and we then supply an enhanced long short-term memory (LSTM) framework for sentiment feature extraction of travel reviews. We have constructed a web review database using a crawler and web scraping technique for experimental validation to evaluate the effectiveness of our methodology. The text form of sentences was first classified through Vader and Roberta model to get the polarity of the reviews. In this paper, we have conducted study methods for feature extraction, such as Count Vectorization and TFIDF Vectorization, and implemented Convolutional Neural Network (CNN) classifier algorithm for the sentiment analysis to decide the tourist’s attitude towards the destinations is positive, negative, or simply neutral based on the review text that they posted online. The results demonstrated that from the CNN algorithm, after pre-processing and cleaning the dataset, we received an accuracy of 96.12% for the positive and negative sentiment analysis.

Keywords: counter vectorization, convolutional neural network, crawler, data technology, long short-term memory, web scraping, sentiment analysis

Procedia PDF Downloads 57

Authors: Levent Basayigit, Sinan Demir, Yusuf Ucar, Burhan Kara

Abstract:

Many research projects require accurate delineation of the different land cover type of the agricultural area. Especially it is critically important for the definition of specific plants like cannabis. However, the complexity of vegetation stands structure, abundant vegetation species, and the smooth transition between different seconder section stages make vegetation classification difficult when using traditional approaches such as the maximum likelihood classifier. Most of the time, classification distinguishes only between trees/annual or grain. It has been difficult to accurately determine the cannabis mixed with other plants. In this paper, a mixed distribution models approach is applied to classify pure and mix cannabis parcels using Worldview-2 imagery in the Lakes region of Turkey. Five different land use types (i.e. sunflower, maize, bare soil, and cannabis) were identified in the image. A constrained Gaussian mixture discriminant analysis (GMDA) was used to unmix the image. In the study, 255 reflectance ratios derived from spectral signatures of seven bands (Blue-Green-Yellow-Red-Rededge-NIR1-NIR2) were randomly arranged as 80% for training and 20% for test data. Gaussian mixed distribution model approach is proved to be an effective and convenient way to combine very high spatial resolution imagery for distinguishing cannabis vegetation. Based on the overall accuracies of the classification, the Gaussian mixed distribution model was found to be very successful to achieve image classification tasks. This approach is sensitive to capture the illegal cannabis planting areas in the large plain. This approach can also be used for monitoring and determination with spectral reflections in illegal cannabis planting areas.

Keywords: Gaussian mixture discriminant analysis, spectral mixture model, Worldview-2, land parcels

Procedia PDF Downloads 165

Authors: Animish Jain

Abstract:

This study deals with using Multiplicative Weight Update within artificial intelligence and machine learning to create models that can diagnose skin cancer using microscopic images of cancer samples. In this study, the multiplicative weight update method is used to take the predictions of multiple models to try and acquire more accurate results. Logistic Regression, Convolutional Neural Network (CNN), and Support Vector Machine Classifier (SVMC) models are employed within the Multiplicative Weight Update system. These models are trained on pictures of skin cancer from the ISIC-Archive, to look for patterns to label unseen scans as either benign or malignant. These models are utilized in a multiplicative weight update algorithm which takes into account the precision and accuracy of each model through each successive guess to apply weights to their guess. These guesses and weights are then analyzed together to try and obtain the correct predictions. The research hypothesis for this study stated that there would be a significant difference in the accuracy of the three models and the Multiplicative Weight Update system. The SVMC model had an accuracy of 77.88%. The CNN model had an accuracy of 85.30%. The Logistic Regression model had an accuracy of 79.09%. Using Multiplicative Weight Update, the algorithm received an accuracy of 72.27%. The final conclusion that was drawn was that there was a significant difference in the accuracy of the three models and the Multiplicative Weight Update system. The conclusion was made that using a CNN model would be the best option for this problem rather than a Multiplicative Weight Update system. This is due to the possibility that Multiplicative Weight Update is not effective in a binary setting where there are only two possible classifications. In a categorical setting with multiple classes and groupings, a Multiplicative Weight Update system might become more proficient as it takes into account the strengths of multiple different models to classify images into multiple categories rather than only two categories, as shown in this study. This experimentation and computer science project can help to create better algorithms and models for the future of artificial intelligence in the medical imaging field.

Keywords: artificial intelligence, machine learning, multiplicative weight update, skin cancer

Procedia PDF Downloads 48

Authors: Tamar Grdzelishvili, Zaza Sinauridze

Abstract:

Introduction: Preeclampsia is a complication of the second trimester of pregnancy, which is characterized by high morbidity and multiorgan damage. Many complex pathogenic mechanisms are now implicated to be responsible for this disease (1). Preeclampsia is one of the leading causes of maternal mortality worldwide. Statistics are enough to convince you of the seriousness of this pathology: about 100,000 women die of preeclampsia every year. It occurs in 3-14% (varies significantly depending on racial origin or ethnicity and geographical region) of pregnant women, in 75% of cases - in a mild form, and in 25% - in a severe form. During severe pre-eclampsia-eclampsia, perinatal mortality increases by 5 times and stillbirth by 9.6 times. Considering that the only way to treat the disease is to end the pregnancy, the main thing is timely diagnosis and prevention of the disease. Identification of high-risk pregnant women for PE and giving prophylaxis would reduce the incidence of preterm PE. First-trimester screening model developed by the Fetal Medicine Foundation (FMF), which uses the Bayes-theorem to combine maternal characteristics and medical history together with measurements of mean arterial pressure, uterine artery pulsatility index, and serum placental growth factor, has been proven to be effective and have superior screening performance to that of traditional risk factor-based approach for the prediction of PE (2) Methods: Retrospective single center screening study. The study population consisted of women from the Tbilisi maternity hospital “Pineo medical ecosystem” who met the following criteria: they spoke Georgian, English, or Russian and agreed to participate in the study after discussing informed consent and answering questions. Prior to the study, the informed consent forms approved by the Institutional Review Board were obtained from the study subjects. Early assessment of preeclampsia was performed between 11-13 weeks of pregnancy. The following were evaluated: anamnesis, dopplerography of the uterine artery, mean arterial blood pressure, and biochemical parameter: Pregnancy-associated plasma protein A (PAPP-A). Individual risk assessment was performed with performed by Fast Screen 3.0 software ThermoFisher scientific. Results: A total of 513 women were recruited and through the study, 51 women were diagnosed with preeclampsia (34.5% in the pregnant women with high risk, 6.5% in the pregnant women with low risk; P<0.000 1). Conclusions: First-trimester screening combining maternal factors with uterine artery Doppler, blood pressure, and pregnancy-associated plasma protein-A is useful to predict PE in a routine care setting. More patient studies are needed for final conclusions. The research is still ongoing.

Keywords: first-trimester, preeclampsia, screening, pregnancy-associated plasma protein

Procedia PDF Downloads 48

Authors: Basant Singh Sikarwar, Mukesh Roy, Ayush Goyal, Priya Ranjan

Abstract:

This work combines two bodies of knowledge which includes biomedical basis of blood stain formation and fluid communities’ wisdom that such formation of blood stain depends heavily on physical properties. Moreover biomedical research tells that different patterns in stains of blood are robust indicator of blood donor’s health or lack thereof. Based on these valuable insights an innovative screening tool is proposed which can act as an aide in the diagnosis of diseases such Anemia, Hyperlipidaemia, Tuberculosis, Blood cancer, Leukemia, Malaria etc., with enhanced confidence in the proposed analysis. To realize this powerful technique, simple, robust and low-cost micro-fluidic devices, a micro-capillary viscometer and a pendant drop tensiometer are designed and proposed to be fabricated to measure the viscosity, surface tension and wettability of various blood samples. Once prognosis and diagnosis data has been generated, automated linear and nonlinear classifiers have been applied into the automated reasoning and presentation of results. A support vector machine (SVM) classifies data on a linear fashion. Discriminant analysis and nonlinear embedding’s are coupled with nonlinear manifold detection in data and detected decisions are made accordingly. In this way, physical properties can be used, using linear and non-linear classification techniques, for screening of various diseases in humans and cattle. Experiments are carried out to validate the physical properties measurement devices. This framework can be further developed towards a real life portable disease screening cum diagnostics tool. Small-scale production of screening cum diagnostic devices is proposed to carry out independent test.

Keywords: blood, physical properties, diagnostic, nonlinear, classifier, device, surface tension, viscosity, wettability

Procedia PDF Downloads 355

Authors: Souvik Phadikar, Nidul Sinha, Rajdeep Ghosh

Abstract:

In brainwaves research using electroencephalogram (EEG) signals, finding the most relevant and effective feature set for identification of activities in the human brain is a big challenge till today because of the random nature of the signals. The feature extraction method is a key issue to solve this problem. Finding those features that prove to give distinctive pictures for different activities and similar for the same activities is very difficult, especially for the number of activities. The performance of a classifier accuracy depends on this quality of feature set. Further, more number of features result in high computational complexity and less number of features compromise with the lower performance. In this paper, a novel idea of the selection of optimal feature set using a heuristically optimized deep autoencoder is presented. Using various feature extraction methods, a vast number of features are extracted from the EEG signals and fed to the autoencoder deep neural network. The autoencoder encodes the input features into a small set of codes. To avoid the gradient vanish problem and normalization of the dataset, a meta-heuristic search algorithm is used to minimize the mean square error (MSE) between encoder input and decoder output. To reduce the feature set into a smaller one, 4 hidden layers are considered in the autoencoder network; hence it is called Heuristically Optimized Deep Autoencoder (HO-DAE). In this method, no features are rejected; all the features are combined into the response of responses of the hidden layer. The results reveal that higher accuracy can be achieved using optimal reduced features. The proposed HO-DAE is also compared with the regular autoencoder to test the performance of both. The performance of the proposed method is validated and compared with the other two methods recently reported in the literature, which reveals that the proposed method is far better than the other two methods in terms of classification accuracy.

Keywords: autoencoder, brainwave signal analysis, electroencephalogram, feature extraction, feature selection, optimization

Procedia PDF Downloads 94

Authors: Michael Hast

Abstract:

Within their first year, infants can differentiate between objects based on their weight. By at least 5 years children hold consistent weight-related misconceptions about the physical world, such as that heavy things fall faster than lighter ones because of their weight. Such misconceptions are seen as a challenge for science education since they are often highly resistant to change through instruction. Understanding the time point of emergence of such ideas could, therefore, be crucial for early science pedagogy. The paper thus discusses two studies that jointly address the issue by examining young children’s search behaviour in hidden displacement tasks under consideration of relative object weight. In both studies, they were tested with a heavy or a light ball, and they either had information about one of the balls only or both. In Study 1, 88 toddlers aged 2 to 3½ years watched a ball being dropped into a curved tube and were then allowed to search for the ball in three locations – one straight beneath the tube entrance, one where the curved tube lead to, and one that corresponded to neither of the previous outcomes. Success and failure at the task were not impacted by weight of the balls alone in any particular way. However, from around 3 years onwards, relative lightness, gained through having tactile experience of both balls beforehand, enhanced search success. Conversely, relative heaviness increased search errors such that children increasingly searched in the location immediately beneath the tube entry – known as the gravity bias. In Study 2, 60 toddlers aged 2, 2½ and 3 years watched a ball roll down a ramp and behind a screen with four doors, with a barrier placed along the ramp after one of four doors. Toddlers were allowed to open the doors to find the ball. While search accuracy generally increased with age, relative weight did not play a role in 2-year-olds’ search behaviour. Relative lightness improved 2½-year-olds’ searches. At 3 years, both relative lightness and relative heaviness had a significant impact, with the former improving search accuracy and the latter reducing it. Taken together, both studies suggest that between 2 and 3 years of age, relative object weight is increasingly taken into consideration in navigating naïve physical concepts. In particular, it appears to contribute to the early emergence of misconceptions relating to object weight. This insight from developmental psychology research may have consequences for early science education and related pedagogy towards early conceptual change.

Keywords: conceptual development, early science education, intuitive physics, misconceptions, object weight

Procedia PDF Downloads 174

Authors: Gaelle Candel, David Naccache

Abstract:

t-SNE is an embedding method that the data science community has widely used. It helps two main tasks: to display results by coloring items according to the item class or feature value; and for forensic, giving a first overview of the dataset distribution. Two interesting characteristics of t-SNE are the structure preservation property and the answer to the crowding problem, where all neighbors in high dimensional space cannot be represented correctly in low dimensional space. t-SNE preserves the local neighborhood, and similar items are nicely spaced by adjusting to the local density. These two characteristics produce a meaningful representation, where the cluster area is proportional to its size in number, and relationships between clusters are materialized by closeness on the embedding. This algorithm is non-parametric. The transformation from a high to low dimensional space is described but not learned. Two initializations of the algorithm would lead to two different embeddings. In a forensic approach, analysts would like to compare two or more datasets using their embedding. A naive approach would be to embed all datasets together. However, this process is costly as the complexity of t-SNE is quadratic and would be infeasible for too many datasets. Another approach would be to learn a parametric model over an embedding built with a subset of data. While this approach is highly scalable, points could be mapped at the same exact position, making them indistinguishable. This type of model would be unable to adapt to new outliers nor concept drift. This paper presents a methodology to reuse an embedding to create a new one, where cluster positions are preserved. The optimization process minimizes two costs, one relative to the embedding shape and the second relative to the support embedding’ match. The embedding with the support process can be repeated more than once, with the newly obtained embedding. The successive embedding can be used to study the impact of one variable over the dataset distribution or monitor changes over time. This method has the same complexity as t-SNE per embedding, and memory requirements are only doubled. For a dataset of n elements sorted and split into k subsets, the total embedding complexity would be reduced from O(n²) to O(n²=k), and the memory requirement from n² to 2(n=k)², which enables computation on recent laptops. The method showed promising results on a real-world dataset, allowing to observe the birth, evolution, and death of clusters. The proposed approach facilitates identifying significant trends and changes, which empowers the monitoring high dimensional datasets’ dynamics.

Keywords: concept drift, data visualization, dimension reduction, embedding, monitoring, reusability, t-SNE, unsupervised learning

Procedia PDF Downloads 117

Authors: Abdulaziz Alsadhan, Naveed Khan

Abstract:

In recent years intrusions on computer network are the major security threat. Hence, it is important to impede such intrusions. The hindrance of such intrusions entirely relies on its detection, which is primary concern of any security tool like Intrusion Detection System (IDS). Therefore, it is imperative to accurately detect network attack. Numerous intrusion detection techniques are available but the main issue is their performance. The performance of IDS can be improved by increasing the accurate detection rate and reducing false positive. The existing intrusion detection techniques have the limitation of usage of raw data set for classification. The classifier may get jumble due to redundancy, which results incorrect classification. To minimize this problem, Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Local Binary Pattern (LBP) can be applied to transform raw features into principle features space and select the features based on their sensitivity. Eigen values can be used to determine the sensitivity. To further classify, the selected features greedy search, back elimination, and Particle Swarm Optimization (PSO) can be used to obtain a subset of features with optimal sensitivity and highest discriminatory power. These optimal feature subset used to perform classification. For classification purpose, Support Vector Machine (SVM) and Multilayer Perceptron (MLP) used due to its proven ability in classification. The Knowledge Discovery and Data mining (KDD’99) cup dataset was considered as a benchmark for evaluating security detection mechanisms. The proposed approach can provide an optimal intrusion detection mechanism that outperforms the existing approaches and has the capability to minimize the number of features and maximize the detection rates.

Keywords: Particle Swarm Optimization (PSO), Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), Local Binary Pattern (LBP), Support Vector Machine (SVM), Multilayer Perceptron (MLP)

Procedia PDF Downloads 342

Authors: Zachary Pan

Abstract:

Our general mission is to solve the problem of classifying images into different body part types and deciding if each of them is healthy or not. However, for now, we will determine healthiness for only one-sixth of the body parts, specifically the chest. We will detect pneumonia in X-ray scans of those chest images. With this type of AI, doctors can use it as a second opinion when they are taking CT or X-ray scans of their patients. Another ad-vantage of using this machine learning classifier is that it has no human weaknesses like fatigue. The overall ap-proach to this problem is to split the problem into two parts: first, classify the image, then determine if it is healthy. In order to classify the image into a specific body part class, the body parts dataset must be split into test and training sets. We can then use many models, like neural networks or logistic regression models, and fit them using the training set. Now, using the test set, we can obtain a realistic accuracy the models will have on images in the real world since these testing images have never been seen by the models before. In order to increase this testing accuracy, we can also apply many complex algorithms to the models, like multiplicative weight update. For the second part of the problem, to determine if the body part is healthy, we can have another dataset consisting of healthy and non-healthy images of the specific body part and once again split that into the test and training sets. We then use another neural network to train on those training set images and use the testing set to figure out its accuracy. We will do this process only for the chest images. A major conclusion reached is that convolutional neural networks are the most reliable and accurate at image classification. In classifying the images, the logistic regression model, the neural network, neural networks with multiplicative weight update, neural networks with the black box algorithm, and the convolutional neural network achieved 96.83 percent accuracy, 97.33 percent accuracy, 97.83 percent accuracy, 96.67 percent accuracy, and 98.83 percent accuracy, respectively. On the other hand, the overall accuracy of the model that de-termines if the images are healthy or not is around 78.37 percent accuracy.

Keywords: body part, healthcare, machine learning, neural networks

Procedia PDF Downloads 72

Authors: Sridhar A. Malkaram, Tamer E. Fandy

Abstract:

Purpose: 5-Azacytidine (5AC) and decitabine (DC) are DNA hypomethylating agents. We recently demonstrated that both drugs increase the enzymatic activity of the histone deacetylase enzyme SIRT6. Accordingly, we are comparing the changes H3K9 acetylation changes in the whole genome induced by both drugs using leukemia cells. Description of Methods & Materials: Mononuclear cells from the bone marrow of six de-identified naive acute myeloid leukemia (AML) patients were cultured with either 500 nM of DC or 5AC for 72 h followed by ChIP-Seq analysis using a ChIP-validated acetylated-H3K9 (H3K9ac) antibody. Chip-Seq libraries were prepared from treated and untreated cells using SMARTer ThruPLEX DNA- seq kit (Takara Bio, USA) according to the manufacturer’s instructions. Libraries were purified and size-selected with AMPure XP beads at 1:1 (v/v) ratio. All libraries were pooled prior to sequencing on an Illumina HiSeq 1500. The dual-indexed single-read Rapid Run was performed with 1x120 cycles at 5 pM final concentration of the library pool. Sequence reads with average Phred quality < 20, with length < 35bp, PCR duplicates, and those aligning to blacklisted regions of the genome were filtered out using Trim Galore v0.4.4 and cutadapt v1.18. Reads were aligned to the reference human genome (hg38) using Bowtie v2.3.4.1 in end-to-end alignment mode. H3K9ac enriched (peak) regions were identified using diffReps v1.55.4 software using input samples for background correction. The statistical significance of differential peak counts was assessed using a negative binomial test using all individuals as replicates. Data & Results: The data from the six patients showed significant (Padj<0.05) acetylation changes at 925 loci after 5AC treatment versus 182 loci after DC treatment. Both drugs induced H3K9 acetylation changes at different chromosomal regions, including promoters, coding exons, introns, and distal intergenic regions. Ten common genes showed H3K9 acetylation changes by both drugs. Approximately 84% of the genes showed an H3K9 acetylation decrease by 5AC versus 54% only by DC. Figures 1 and 2 show the heatmaps for the top 100 genes and the 99 genes showing H3K9 acetylation decrease after 5AC treatment and DC treatment, respectively. Conclusion: Despite the similarity in hypomethylating activity and chemical structure, the effect of both drugs on H3K9 acetylation change was significantly different. More changes in H3K9 acetylation were observed after 5 AC treatments compared to DC. The impact of these changes on gene expression and the clinical efficacy of these drugs requires further investigation.

Keywords: DNA methylation, leukemia, decitabine, 5-Azacytidine, epigenetics

Procedia PDF Downloads 120

Authors: Mohammad H. Taheri, David J. Brown, Nasser Sherkat

Abstract:

Engagement is one of the most important factors in determining successful outcomes and deep learning in students. Existing approaches to detect student engagement involve periodic human observations that are subject to inter-rater reliability. Our solution uses real-time multimodal multisensor data labeled by objective performance outcomes to infer the engagement of students. The study involves four students with a combined diagnosis of cerebral palsy and a learning disability who took part in a 3-month trial over 59 sessions. Multimodal multisensor data were collected while they participated in a continuous performance test. Eye gaze, electroencephalogram, body pose, and interaction data were used to create a model of student engagement through objective labeling from the continuous performance test outcomes. In order to achieve this, a type of continuous performance test is introduced, the Seek-X type. Nine features were extracted including high-level handpicked compound features. Using leave-one-out cross-validation, a series of different machine learning approaches were evaluated. Overall, the random forest classification approach achieved the best classification results. Using random forest, 93.3% classification for engagement and 42.9% accuracy for disengagement were achieved. We compared these results to outcomes from different models: AdaBoost, decision tree, k-Nearest Neighbor, naïve Bayes, neural network, and support vector machine. We showed that using a multisensor approach achieved higher accuracy than using features from any reduced set of sensors. We found that using high-level handpicked features can improve the classification accuracy in every sensor mode. Our approach is robust to both sensor fallout and occlusions. The single most important sensor feature to the classification of engagement and distraction was shown to be eye gaze. It has been shown that we can accurately predict the level of engagement of students with learning disabilities in a real-time approach that is not subject to inter-rater reliability, human observation or reliant on a single mode of sensor input. This will help teachers design interventions for a heterogeneous group of students, where teachers cannot possibly attend to each of their individual needs. Our approach can be used to identify those with the greatest learning challenges so that all students are supported to reach their full potential.

Keywords: affective computing in education, affect detection, continuous performance test, engagement, flow, HCI, interaction, learning disabilities, machine learning, multimodal, multisensor, physiological sensors, student engagement

Procedia PDF Downloads 69

Authors: Thiago Spilborghs Bueno Meyer, Plinio Thomaz Aquino Junior

Abstract:

Through speech, which privileges the functional and interactive nature of the text, it is possible to ascertain the spatiotemporal circumstances, the conditions of production and reception of the discourse, the explicit purposes such as informing, explaining, convincing, etc. These conditions allow bringing the interaction between humans closer to the human-robot interaction, making it natural and sensitive to information. However, it is not enough to understand what is said; it is necessary to recognize emotions for the desired interaction. The validity of the use of neural networks for feature selection and emotion recognition was verified. For this purpose, it is proposed the use of neural networks and comparison of models, such as recurrent neural networks and deep neural networks, in order to carry out the classification of emotions through speech signals to verify the quality of recognition. It is expected to enable the implementation of robots in a domestic environment, such as the HERA robot from the RoboFEI@Home team, which focuses on autonomous service robots for the domestic environment. Tests were performed using only the Mel-Frequency Cepstral Coefficients, as well as tests with several characteristics of Delta-MFCC, spectral contrast, and the Mel spectrogram. To carry out the training, validation and testing of the neural networks, the eNTERFACE’05 database was used, which has 42 speakers from 14 different nationalities speaking the English language. The data from the chosen database are videos that, for use in neural networks, were converted into audios. It was found as a result, a classification of 51,969% of correct answers when using the deep neural network, when the use of the recurrent neural network was verified, with the classification with accuracy equal to 44.09%. The results are more accurate when only the Mel-Frequency Cepstral Coefficients are used for the classification, using the classifier with the deep neural network, and in only one case, it is possible to observe a greater accuracy by the recurrent neural network, which occurs in the use of various features and setting 73 for batch size and 100 training epochs.

Keywords: emotion recognition, speech, deep learning, human-robot interaction, neural networks

Procedia PDF Downloads 132

Authors: Reeta Devi, Hitender Kumar Tyagi, Dinesh Kumar

Abstract:

In present scenario, cardiovascular problems are growing challenge for researchers and physiologists. As heart disease have no geographic, gender or socioeconomic specific reasons; detecting cardiac irregularities at early stage followed by quick and correct treatment is very important. Electrocardiogram is the finest tool for continuous monitoring of heart activity. Heart rate variability (HRV) is used to measure naturally occurring oscillations between consecutive cardiac cycles. Analysis of this variability is carried out using time domain, frequency domain and non-linear parameters. This paper presents HRV analysis of the online dataset for normal sinus rhythm (taken as healthy subject) and sudden cardiac death (SCD subject) using all three methods computing values for parameters like standard deviation of node to node intervals (SDNN), square root of mean of the sequences of difference between adjacent RR intervals (RMSSD), mean of R to R intervals (mean RR) in time domain, very low-frequency (VLF), low-frequency (LF), high frequency (HF) and ratio of low to high frequency (LF/HF ratio) in frequency domain and Poincare plot for non linear analysis. To differentiate HRV of healthy subject from subject died with SCD, k –nearest neighbor (k-NN) classifier has been used because of its high accuracy. Results show highly reduced values for all stated parameters for SCD subjects as compared to healthy ones. As the dataset used for SCD patients is recording of their ECG signal one hour prior to their death, it is therefore, verified with an accuracy of 95% that proposed algorithm can identify mortality risk of a patient one hour before its death. The identification of a patient’s mortality risk at such an early stage may prevent him/her meeting sudden death if in-time and right treatment is given by the doctor.

Keywords: early stage prediction, heart rate variability, linear and non-linear analysis, sudden cardiac death

Procedia PDF Downloads 321

Authors: Xiaoyang Zhao, Kaiying Wang

Abstract:

The choice of stocking density in poultry farming is a potential way for determining welfare level of poultry. However, it is difficult to measure stocking densities in poultry farming because of a lot of variables such as species, age and weight, feeding way, house structure and geographical location in different broiler houses. A method was proposed in this paper to measure the differences of young broilers reared under different stocking densities by sound analysis. Vocalisations of broilers were recorded and analysed under different stocking densities to identify the relationship between sounds and stocking densities. Recordings were made continuously for three-week-old chickens in order to evaluate the variation of sounds emitted by the animals at the beginning. The experimental trial was carried out in an indoor reared broiler farm; the audio recording procedures lasted for 5 days. Broilers were divided into 5 groups, stocking density treatments were 8/m², 10/m², 12/m² (96birds/pen), 14/m² and 16/m², all conditions including ventilation and feed conditions were kept same except from stocking densities in every group. The recordings and analysis of sounds of chickens were made noninvasively. Sound recordings were manually analysed and labelled using sound analysis software: GoldWave Digital Audio Editor. After sound acquisition process, the Mel Frequency Cepstrum Coefficients (MFCC) was extracted from sound data, and the Support Vector Machine (SVM) was used as an early detector and classifier. This preliminary study, conducted in an indoor reared broiler farm shows that this method can be used to classify sounds of chickens under different densities economically (only a cheap microphone and recorder can be used), the classification accuracy is 85.7%. This method can predict the optimum stocking density of broilers with the complement of animal welfare indicators, animal productive indicators and so on.

Keywords: broiler, stocking density, poultry farming, sound monitoring, Mel Frequency Cepstrum Coefficients (MFCC), Support Vector Machine (SVM)

Procedia PDF Downloads 123

Authors: Anželika Teresė

Abstract:

In order to solve issues in sign language linguistics, address matters pertaining to maintaining high quality of sign language (SL) translation, contribute to dispelling misconceptions about SL and deaf people, and raise awareness and understanding of the deaf community heritage, this presentation discusses literature in Lithuanian Sign Language (LSL) and inherent metaphors that are created by using the phonological parameter –handshape, location, movement, palm orientation and nonmanual features. The study covered in this presentation is twofold, involving both the micro-level analysis of metaphors in terms of phonological parameters as a sub-lexical feature and the macro-level analysis of the poetic context. Cognitive theories underlie research of metaphors in sign language literature in a range of SL. The study follows this practice. The presentation covers the qualitative analysis of 34 pieces of LSL literature. The analysis employs ELAN software widely used in SL research. The target is to examine how specific types of each phonological parameter are used for the creation of metaphors in LSL literature and what metaphors are created. The results of the study show that LSL literature employs a range of metaphors created by using classifier signs and by modifying the established signs. The study also reveals that LSL literature tends to create reference metaphors indicating status and power. As the study shows, LSL poets metaphorically encode status by encoding another meaning in the same sign, which results in creating double metaphors. The metaphor of identity has been determined. Notably, the poetic context has revealed that the latter metaphor can also be identified as a metaphor for life. The study goes on to note that deaf poets create metaphors related to the importance of various phenomena significance of the lyrical subject. Notably, the study has allowed detecting locations, nonmanual features and etc., never mentioned in previous SL research as used for the creation of metaphors.

Keywords: Lithuanian sign language, sign language literature, sign language metaphor, metaphor at the phonological level, cognitive linguistics

Procedia PDF Downloads 106

Authors: Katie E. Jennings

Abstract:

The interactions between family dynamics and environmental factors with mental health vulnerability in individuals are well known and are a theme for on-going research and debate. The impact upon mental health issues and forensic issues on family dynamics, experience, and emotional wellbeing cannot be over-Emphasised. For forensic patients with diagnosed mental disorders, these relationships and environments may have also been functionally linked to the development and maintenance of those disorders; with significant adverse childhood experiences being a common feature of many Patient’s histories. Mental health hospitals remove the patient from their home environments and provide treatment outside of these relationships and often outside of the home area. There is, therefore, a major focus on Services ensuring that patients are able to build and maintain relationships with family and friends, requiring services to involve families in Patients' care and treatment wherever possible. There are standards set by Government and clinical bodies that require absolute demonstration of the inclusion of family and friends in all aspects of the care and treatment of forensic patients. For some patients and family members, this push to take on a “role” in care can be unhelpful, extremely stressful, and has constant implications for the potential delicate reparation of relationships. Based on work undertaken for over 20 years in forensic mental health settings, this paper explores the positive psychology approach to a dimensional model to family inclusion in mental health care that learns from family court work and allows for the maintenance of relationships to be at both proximal and Distil levels; to prevent the replication of abuse, decrease the need for falsehoods and assist the recovery of all. The model is based on allowing families to choose to not be involved or be involved in different ways if this is seen to be more helpful. It also allows patients to choose the level of potential involvement that they would find helpful, and for this to be reviewed at a timeframe agreed by all parties, rather than when the next survey is due or the patient has a significant care meeting. This paper is significant as there is a lack of research to support services to use a positive psychology approach to work in this area, the assumption that being asked to be involved must be positive for all seems naïve at best for this patient group. Work relating to the psychology of family can significantly contribute to the development of knowledge in this area. The development of a dimensional model will support choice within families and assist in the development of more honest and open relationships.

Keywords: family dynamics, forensic, mental disorder, positive psychology

Procedia PDF Downloads 124

Authors: Massimo Miccoli, Luca Marangoni, Alberto Aniello Scaringi, Alessandro Marceddu, Alessandro Amicone

Abstract:

The susceptibility of deep neural networks (DNNs) to adversarial manipulations is a recognized challenge within the computer vision domain. Adversarial examples, crafted by adding subtle yet malicious alterations to benign images, exploit this vulnerability. Various defense strategies have been proposed to safeguard DNNs against such attacks, stemming from diverse research hypotheses. Building upon prior work, our approach involves the utilization of autoencoder models. Autoencoders, a type of neural network, are trained to learn representations of training data and reconstruct inputs from these representations, typically minimizing reconstruction errors like mean squared error (MSE). Our autoencoder was trained on a dataset of benign examples; learning features specific to them. Consequently, when presented with significantly perturbed adversarial examples, the autoencoder exhibited high reconstruction errors. The architecture of the autoencoder was tailored to the dimensions of the images under evaluation. We considered various image sizes, constructing models differently for 256x256 and 512x512 images. Moreover, the choice of the computer vision model is crucial, as most adversarial attacks are designed with specific AI structures in mind. To mitigate this, we proposed a method to replace image-specific dimensions with a structure independent of both dimensions and neural network models, thereby enhancing robustness. Our multi-modal autoencoder reconstructs the spectral representation of images across the red-green-blue (RGB) color channels. To validate our approach, we conducted experiments using diverse datasets and subjected them to adversarial attacks using models such as ResNet50 and ViT_L_16 from the torch vision library. The autoencoder extracted features used in a classification model, resulting in an MSE (RGB) of 0.014, a classification accuracy of 97.33%, and a precision of 99%.

Keywords: adversarial attacks, malicious images detector, binary classifier, multimodal transformer autoencoder

Procedia PDF Downloads 43