Search

Although deep neural networks (DNNs) have performed well on many perceptual tasks, they are vulnerable to adversarial examples that are generated by adding slight but maliciously crafted perturbations to benign images. Adversarial detection is an important technique for identifying adversarial examples before they are entered into target DNNs. Previous studies that were performed to detect adversarial examples either targeted specific attacks or required expensive computation. Designing a lightweight unsupervised detector is still a challenging problem. In this paper, we propose an A uto E ncoder-based A dversarial E xamples (AEAE) detector that can guard DNN models by detecting adversarial examples with low computation in an unsupervised manner. The AEAE includes only a shallow autoencoder that performs two roles. First, a well-trained autoencoder has learned the manifold of benign examples. This autoencoder can produce a large reconstruction error for adversarial images with large perturbations, so we can detect significantly perturbed adversarial examples based on the reconstruction error. Second, the autoencoder can filter out small noises and change the DNN's prediction on adversarial examples with small perturbations. It helps to detect slightly perturbed adversarial examples based on the prediction distance. To cover these two cases, we utilize the reconstruction error and prediction distance from benign images to construct a two-tuple feature set and train an adversarial detector using the isolation forest algorithm. We show empirically that AEAE is an unsupervised and inexpensive detector against most state-of-the-art attacks. Through the detection in these two cases, there is nowhere to hide adversarial examples. • We observe that adversarial detection is sensitive to the perturbation level. • We train a shallow autoencoder to find two key features from adversarial examples. • We propose a lightweight and unsupervised adversarial detector. [ABSTRACT FROM AUTHOR]

In this paper, we present two novel physics-infused neural network (NN) architectures that satisfy various invariance conditions for constructing efficient and robust Deep Learning (DL)-based sub-grid scale (SGS) turbulence models for use in Large Eddy Simulation (LES) procedures widely used in various fluid engineering applications. The first architecture, called tensor basis neural networks (TBNN) , recently proposed in the context of Reynolds-averaged Navier–Stokes (RANS) modeling, and introduced herein for SGS modeling of wall-bounded turbulence, embeds the analytical expansion of the SGS turbulence stresses into integrity basis tensors composed of the symmetric and anti-symmetric parts of the resolved velocity gradient tensor, thus incorporating the Galilean, rotational and reflectional invariances. In our second approach, a relatively simple yet powerful architecture, called Galilean Invariance embedded Neural Network (GINN) incorporates the Galilean invariance, and takes in as inputs the independent components of the integrity basis tensors in addition to the invariant inputs in a single input layer. Explicit filtering of data from direct simulations of the canonical channel flow at two friction Reynolds numbers R e τ ≈ 395 and 590 provided accurate data for training and testing these models. Both sets of models are used to predict the SGS stresses for feature datasets generated with different filter widths, and at different Reynolds numbers. The GINN model yields less prediction error on test datasets (mean squared error ∼ 0. 4) compared to the TBNN model (mean squared error ∼ 0. 5). Upon comparison, it is revealed that GINN has better feature extraction capacity owing to its ability to establish relations between and extract information from cross-components of the integrity basis tensors and the SGS stresses. Based on their predictive performance, both models proposed herein have shown great promise for use in a posteriori actual LESs. The present work illustrates the significance of physics infusion as well as invariance embedding into NN architecture for constructing efficient and robust DL-based SGS turbulent models in achieving superior predictive efficacy. [ABSTRACT FROM AUTHOR]

Structural adhesives have seen a surge in applications for bonding Fibre reinforced polymers to strengthen structures. However, being polymeric materials, they can exhibit viscoelasticity, leading to creep, which can adversely affect bonded strengthening systems. This paper delves into the nonlinear shear creep behaviour of a typical structural adhesive using a machine learning approach. A deep neural networks (DNN) model was trained using a dataset of 7176 data points obtained from 25 sets of butt joint shear creep tests performed at various stress levels and temperatures. The DNN model successfully captures the complex relationship between temperature, stress, time, and shear creep with a performance value of R2 larger than 0.99. The creep compliance curves predicted by the model are in close agreement with the experimental results, highlighting the ability of the DNN model to effectively model the complex nonlinear shear creep behaviour, which is difficult to achieve with traditional physics-based viscoelastic models. The SHapley Additive exPlanations analysis further identified temperature as the most significant factor influencing the creep, followed by stress and time. The constructed DNN model and its corresponding analyses can provide a valuable reference for the structural adhesive to limit the negative effects of creep in practical applications. • Butt joint shear creep tests performed at various stress levels and temperatures. • Modelling of nonlinear creep of the structural adhesive using the ML method. • A DNN predictive model constructed from the test results achieves high accuracy. • SHAP approach used to interpret and explain the developed DNN model. • The significance of temperature, stress, and time on the nonlinear creep results. [ABSTRACT FROM AUTHOR]

This paper proposes a novel Deep Belief Network (DBN) architecture, the 'Interval Type-2 Fuzzy DBN (IT2FDBN)', which models the weights and biases with IT2 FSs. Thus, it introduces a novel algorithm for augmented deep leaning, which has the capability to address all the limitations of the classical DBN (CDBN) and T1 fuzzy DBN (T1FDBN). We comparatively evaluate the performance of the IT2FDBN by conducting experiments using the popular MNIST handwritten digit recognition datasets. Additionally, to demonstrate its robustness and generalization capabilities, we also conduct experiments taking two noisy variants of MNIST dataset, viz. the MNIST with AWGN (additive white Gaussian noise) and the MNIST with motion blur. We conduct extensive simulations by considering different combinations of nodes in the hidden layers of the DBN for better model selection. We thoroughly compare the results using well-known performance measures such as root mean square error (RMSE) and Error rate. We show that, in terms of RMSE values and error rates, the proposed IT2FDBN outperforms both T1FDBN and CDBN across all the three datasets. Further, we also provide the results of convergence, runtime-based comparison, and statistical analysis in support of our proposal. [ABSTRACT FROM AUTHOR]

Phishing websites distribute unsolicited content and are frequently used to commit email and internet fraud. Detecting them before any user information is submitted is critical. Several efforts have been made to detect these phishing websites in recent years. Most existing approaches use hand-crafted lexical and statistical features from a website's textual content to train classification models to detect phishing web pages. However, these phishing detection approaches have limitations, including (1) the tediousness of extracting hand-crafted features, which require specialized domain knowledge to determine which features are useful for a particular platform; and (2) the difficulties encountered by models built on hand-crafted features to capture the semantic patterns in words and characters in URL and HTML content. To address these challenges, this paper proposes WebPhish, an end-to-end deep neural network trained using embedded raw URLs and HTML content to detect website phishing attacks. First, the proposed model automatically employs an embedding technique to extract the corresponding characters into homologous dense vectors. Then, the concatenation layer merges the URL and HTML embedding matrices. Following that, Convolutional layers are used to model its semantic dependencies. Extensive experiments were conducted with real-world phishing data, which yielded an accuracy of 98.1%, showing that WebPhish outperforms baseline detection approaches in identifying phishing pages. • WebPhish employs both raw URL and HTML content to detect phishing web pages. • WebPhish uses character-level embeddings to enable the feature vectors to generalize to new data. • Extensive experiments conducted on a real-world dataset yields 98.1% accuracy. • Its application on the airline Twitter dataset shows the proposed model's flexibility. [ABSTRACT FROM AUTHOR]

Inference centers need more data to have a more comprehensive and beneficial learning model, and for this purpose, they need to collect data from data providers. On the other hand, data providers are cautious about delivering their datasets to inference centers in terms of privacy considerations. In this paper, by modifying the structure of the autoencoder, we present a method that manages the utility-privacy trade-off well. To be more precise, the data is first compressed using the encoder, then confidential and non-confidential features are separated and uncorrelated using the classifier. The confidential feature is appropriately combined with noise, and the non-confidential feature is enhanced, and at the end, data with the original data format is produced by the decoder. The suggested architecture additionally enables data providers to modify the degree of privacy needed for private features and the level of utility for non-private features. The proposed method has been examined for both image and categorical databases, and the results show a significant performance improvement compared to previous methods. [ABSTRACT FROM AUTHOR]

This paper addresses how physical knowledge can improve machine learning in process control. A data-driven tracking control framework using physics-informed neural networks (PINNs) and deep reinforcement learning (DRL) is proposed for dynamical systems, which is particularly important when iterations or repetitions of data collection experiments are limited or even not allowed. An indirect control approach is followed to accomplish this. First, a new process model is identified using PINNs trained from a model representing the experimental data collected from the plant. Then, the DRL-based control agent utilizes the identified PINN model to deliver an offline policy. We evaluate our framework for data-driven tracking control of nonisothermal CSTR, considering measurement noise and stochastic dynamics for closed-loop control. Our approach performed comparably to an NMPC without requiring a model to predict the process dynamics. • Physical knowledge was implemented to improve machine learning. • A indirect data-driven control approach was developed. • Physics-informed neural networks and deep reinforcement learning are combined. • The developed framework was tested against nonlinear model predictive control. [ABSTRACT FROM AUTHOR]

Visible light communication (VLC) is an emerging technology showing promising signs of being incorporated into 6G. VLC has challenges to overcome, mainly the nonlinear distortion from light-emitting diodes (LEDs), attenuation at high frequencies, and the presence of ambient light. This paper presents the hardware implementation of VLC with a deep neural network (DNN) based demodulator that can mitigate the nonlinear distortions and improve the received signal quality, especially when high-powered LEDs are used. Off-the-shelf available LEDs have different properties from different manufacturers, and developing a model of all the combined effects is intricate. Few machine learning (ML) based solutions have been presented to mitigate these hindrances like Autoencoders, but the difficulty is with their pre-training. The presented method uses DNN that can be used robustly without pre-training. It is tested with experimentally obtained data using high-powered, off-the-shelf available LEDs. The DNN is trained using the received training and data symbols that are transmitted simultaneously. The DNN has improved the bit error rate (BER) by recovering the data symbols that had undergone distortion. It is observed that the DNN can make up for decreased efficiency due to training symbols by improving the BER for higher-order modulations. It is also observed that the required BER performance can be achieved with fewer training symbols in the transmitted packet. • A DNN based method is presented to mitigate nonlinear distortions in VLC systems. • The signal will be distorted by high-powered LEDs, unknown channel and Photodetector. • The DNN is trained online, so pretraining and mathematical modeling is not necessary. • The proposed system is evaluated with experimentally obtained data. • Experiments are conducted with unknown LED effects and in presence of ambient light. [ABSTRACT FROM AUTHOR]

This paper presents a data-driven approach to approximate the dynamics of a nonlinear time-varying system (NTVS) by a linear time-varying system (LTVS), which results from the Koopman operator and deep neural networks. Analysis of the approximation error between states of the NTVS and the resulting LTVS is presented. Simulations on a representative NTVS show that the proposed method achieves small approximation errors, even when the system changes rapidly. Furthermore, simulations in an example of quadcopters demonstrate the computational efficiency of the proposed approach. [ABSTRACT FROM AUTHOR]

Timber structures have been widely utilized in the field of construction. Nevertheless, these structures are susceptible to environmental factors that can cause internal damage, including cavities, cracks, and corrosion, which can significantly compromise their structural integrity and safety. However, the detection of such internal damages is often challenging due to their concealed nature. Existing methods for detecting internal damage in timber structures suffer from low detection efficiency and recognition accuracy. This study proposes an end-to-end damage detection framework that integrates emphasized channel attention, propagation, and aggregation-time-delay neural network (ECAPA-TDNN) with percussion techniques. This framework is designed to evaluate the internal damage status of timber structures. The paper also introduces a method for creating internal voids within timber structures and fabricates timber specimens containing various sizes of internal damage. Vibration responses at different locations on the specimens are obtained using percussion techniques and serve as input data for the network. The results indicated that the ECAPA-TDNN architecture could effectively identifies damage within the timber structure and accurately distinguish between different sizes of damage, furnishing superlative recognition performance transcending prevailing benchmark models. This study elucidates the potential of percussion-based techniques and deep learning methods in the damage detection field, as well as the possibility for the proposed method to be extended to other applications, including transformer inspection, vehicle vibration monitoring, and medical treatment. [ABSTRACT FROM AUTHOR]

The easy and accurate identification of varieties is fundamental in agriculture, especially in the olive sector, where more than 1200 olive varieties are currently known worldwide. Varietal misidentification leads to many potential problems for all actors in the sector: farmers and nursery workers may establish the incorrect variety, leading to maladaptation in the field; olive oil and table olive producers may label and sell a nonauthentic product; consumers may be misled; and breeders may commit errors during targeted crossings between different varieties. To date, the standard for varietal identification and certification consists of two methods: morphological classification and genetic analysis. The morphological classification consists of the visual pairwise comparison of different organs of the olive tree, where the most important organ is considered to be the endocarp. In contrast, different methods for genetic classification exist (RAPDs, SSR, and SNP). Both classification methods present advantages and disadvantages. Visual morphological classification requires highly specialized personnel and is prone to human error. Genetic identification methods are more accurate but incur a high cost and are difficult to implement. This paper introduces OliVaR, a novel approach to olive varietal identification. OliVaR used a teacher–student deep learning architecture to learn the defining characteristics of the endocarp of each specific olive variety and perform varietal classification. We construct what is, to the best of our knowledge, the largest olive variety dataset to date, comprising image data for 131 varieties from the Mediterranean basin. We thoroughly test OliVaR on this dataset and show that it correctly predicts olive varieties with over 86% accuracy. • Olive endocarp images are sufficient for variety recognition using deep learning. • Applying texture-enhancing to olive endocarps highlights morphological traits. • Knowledge-driven DNN is more suitable than prior approaches for olive recognition. • Olivar reduces cost analysis and time and ensures quality by complementing DNA techniques. • OliVaR can identify and classify 131 olive varieties with 86% accuracy. [ABSTRACT FROM AUTHOR]

• Present a novel POI recommendation method using BERT. • Consider a personalized travel route for each tourist based on Topic. • Present an improved measurement method of users' preferences. Route schema is challenging for tourists because they must choose Points of Interest (POIs) in unknown areas that meet their preferences and limitations. Historically, sequential methods were utilized to generate recommendations based on previous user interactions. Despite their efficacy, however, such left-to-right unidirectional models are suboptimal due to the following factors: a) user behavior sequences are restricted in their ability to utilize hidden representations in unidirectional architectures; b) a rigidly ordered sequence is frequently assumed but is not always possible. This paper proposes a novel personalized sequential recommendation model, termed BERTSeqHybrid, which utilizes Bidirectional Encoder Representations from Transformers (BERT) to circumvent these limitations. In addition to contextual data from POIs, asymmetric schemas, and topic modeling are employed to improve the user-user similarity model. Furthermore, a novel method for evaluating user preferences is proposed utilizing explicit demographic data to mitigate the cold start problem. In the experimental evaluation, the developed methodology, which was applied to two different datasets (Yelp and Flickr), produced superior root mean square error RMSE, F-Score, mean average precision (MAP), and normalized discounted cumulative gain (NDCG) indexes. [ABSTRACT FROM AUTHOR]

• A novel voice conversion technique for body-conducted unvoiced speech is proposed based on Taylor-Boltzmann Machines. The proposed algorithm was the integration of Taylor Series with Boltzmann machine Deep Neural Networks. • A body conductive microphone referred as the non audible murmur (NAM) microphone is used to capture the non-audible murmur from the neck part behind the ear. • The classifier was trained with the features likes "pitch chroma, spectral centroid, and spectral skewness" that are derived from the incoming body conducted low voice verbal communication signal. • In the initial training phase, a Taylor series approximation DNN model is trained from a collection of stereo data which encompasses the NAM speech of same phonemes represented as log-power spectral features. • Moreover, in the recognition phase, a well trained Taylor-DNN model that is based on Boltzmann Machine intakes the speech features and generates the "log-power-spectra features" and it aids in recognizing the speech waveform. • The paper may gain high interest from the readers and experts of Speech processing, Voice Conversion Techniques, body-conducted unvoiced speech. Communication becomes effectual while the verbal communication signal reaches the destination with profound characteristics. An application based on human voice attracts the researcher's attention. It leads the researcher to proceed towards the speech conversion for low voice humans having murmured voices as their attitude. The researchers introduced a technique for improving the body-conducted unvoiced speech for silent speech communication using Taylor-Boltzmann machine-trained optimized Deep Neural Network (DNN). The projected model includes two main stages: the training stage as well as recognition stage. In the training stage, the following functionalities are undergone: (i) pre-processing, (ii) feature extraction, and (iii) deep learning-based speech enhancement. Initially, the collected input Non-Audible Murmur (NAM) speech signal is subjected to the pre-processing stage, wherein the improved wiener filtering is employed for suppressing the noise in the input signal. From the pre-processed speech signal, the most relevant characteristics such as "spectral skewness, spectral chroma, spectral centroid and improved Mel Frequency Cepstral Coefficients (MFCC)" are extracted, and they are fused together. Then, using these fused features, the Taylor series with optimized DNN and Boltzmann Machine (Taylor-DNN-BM) model is trained. The final enhanced speech is acquired from the Taylor- optimized DNN-BM. The recognition phase encapsulates the feature extraction and deep learning-based speech enhancement stage. The input speech data enters into the feature extraction phase, wherein the characteristics such as "spectral skewness, spectral chroma, spectral centroid and improved (MFCC)" are retrieved. After that, these features are fused and fed as input to the Taylor-optimized DNN-BM model. From Taylor- optimized DNN-BM, the enhanced speech signal is acquired. In DNN, the weight function is fine-tuned using the newly projected Self Improved Flower Pollination Algorithm (SI-FPA) algorithm. This SI-FPA model is the conceptual improvement of the Flower Pollination Algorithm (FPA). Finally, the projected model is compared with the existing models to validate its efficiency in terms of speech enhancement. [ABSTRACT FROM AUTHOR]

Domain adaptation provides the possibility of utilizing the knowledge gained from an auxiliary domain to accomplish the task in another related domain. In this paper, a common feature representation of these domains is learned through a new proposed idea to reduce inter-domain differences more precisely which results in higher accuracy for unsupervised domain adaptation. To decrease these divergences, the proposed method finds a subspace that reduces data distribution discrepancy between domains. To achieve this goal, minimization of the divergence of the class conditional probability distributions is considered. By considering class conditional distributions, more discriminative information of data is preserved compared to applying marginal distributions. Also, an explicit parametric distribution of the source and target domains is considered to reduce the discrepancy between the two domains data which results in higher accuracy compared to the other relevant domain adaptation methods. Experimental studies on benchmark image classification tasks confirmed our assumptions and indicated the significant improvement given by the proposed method compared to the other state-of-the-art methods. • Aligning the feature distribution of the source and target domains explicitly. • Introducing a new loss function to impose the common distribution on the features. • Reducing the class conditional probability distribution distance between the domains. [ABSTRACT FROM AUTHOR]