Neural Networks for Pattern Recognition
... They can be employed to address various classification or regression challenges. The concept and theory of neural networks have been extensively discussed in numerous works [27][28][29][30]. Neural networks possess the capacity to process data with the aim of uncovering concealed patterns. ...
... Ensembles (also called committees) of neural networks (EoNN) have their origins in the realm of ensemble learning systems. The foundational principles of this approach can be traced back to earlier referenced works that comprehensively delve into neural network concepts [28,30], as well as works dedicated to the study of ensembles [40]. EoNN consists of individual trained artificial neural networks (ANN), each providing predictions that are subsequently aggregated, with the aim of reducing errors in comparison to standalone neural networks. ...
... The utilization of EoNN as the core of a cost estimation model relies on combining a set of trained neural networks to form an ensemble. As outlined in [28], this set might encompass various types of networks or similar networks trained to different local minima. The two methods built on this premise, which were applied during research presented herein, are (1) ensemble averaging and (2) generalized averaging. ...
This paper presents research results on the development of an original cost prediction model for construction costs in sewerage projects. The focus is placed on fast cost estimates applicable in the early stages of a project, based on fundamental information available during the initial design phase of sanitary sewers prior to the detailed design. The originality and novelty of this research lie in the application of artificial neural network ensembles, which include a combination of several individual neural networks and the use of simple averaging and generalized averaging approaches. The research resulted in the development of two ensemble-based models, including five neural networks that were trained and tested using data collected from 125 sewerage projects completed in the Czech Republic between 2018 and 2022. The data included information relevant to various aspects of projects and contract costs, updated to account for changes in costs over time. The developed models present satisfactory predictive performance, especially the ensemble model based on simple averaging, which offers prediction accuracy within the range of ±30% (in terms of percentage errors) for over 90% of the training and testing samples. The developed models, based on the ensembles of neural networks, outperformed the benchmark model based on the classical approach and the use of multiple linear regression.
... They can handle both linear and non-linear relationships using different kernel functions. Also based on optimization, neural networks, inspired by the structure and function of biological neural networks, are versatile models capable of learning complex patterns and relationships from data [38]. They consist of interconnected layers of artificial neurons that process information. ...
... Curve fitting is related to overfitting [19], which occurs when a model excessively performs well on the training data but fails to generalize to unseen data. On the other hand, when the model is too simple and lacks the flexibility to capture the underlying patterns in the data, we have an underfitting [38]. The goal of machine learning algorithms is to find the balance between underfitting (high bias) and overfitting (high variance) in modelling data [35]. ...
... A neural network is a computational model composed of interconnected nodes, called neurons, organized into layers. Each neuron processes incoming information and produces an output, which becomes the input for other neurons in subsequent layers [38]. Through training, neural networks learn to adjust the weights of connections between neurons to recognize patterns and relationships in complex data effectively. ...
Machine learning is a rapidly growing field with the potential to revolutionize many areas of science, including physics. This review provides a brief overview of machine learning in physics, covering the main concepts of supervised, unsupervised, and reinforcement learning, as well as more specialized topics such as causal inference, symbolic regression, and deep learning. We present some of the principal applications of machine learning in physics and discuss the associated challenges and perspectives.
... In step (3), we train and test our proposed end-to-end Siamese neural network to build an efficient binary function embedding model required to generate an embedding for every binary function in our repository. In Step (4), given a new binary function , (e.g., a newly discovered vulnerability), we initially extract its related features obtained at step (2). Then, we generate its embedding using our trained model obtained at learning step (3). ...
... Our Siamese neural network architecture is depicted in Figure 3. It is composed of two identical three-layered multi-layer perceptron neural networks [2]. Our designed NN is suitable for our problem. ...
... For each query binary function in Dataset-I, we look up all its similar functions among 116508 functions, and record the BinFinder performance. From Figure 4(a), we see that BinFinder precision is above 80% for ∈ [1][2][3][4][5] and furthermore, it is above 70% for ∈ [6][7][8][9][10], and above 50% for ∈ [10][11][12][13][14][15][16][17][18][19][20]. We also observe from Figure 4(b) that, BinFinder nDCG values are above 80% for ∈ [1][2][3][4][5][6][7], and above 70% for ∈[8-20]. ...
Binary function clone search is an essential capability that enables multiple applications and use cases, including reverse engineering, patch security inspection, threat analysis, vulnerable function detection, etc. As such, a surge of interest has been expressed in designing and implementing techniques to address function similarity on binary executables and firmware images. Although existing approaches have merit in fingerprinting function clones, they present limitations when the target binary code has been subjected to significant code transformation resulting from obfuscation, compiler optimization, and/or cross-compilation to multiple-CPU architectures. In this regard, we design and implement a system named BinFinder, which employs a neural network to learn binary function embeddings based on a set of extracted features that are resilient to both code obfuscation and compiler optimization techniques. Our experimental evaluation indicates that BinFinder outperforms state-of-the-art approaches for multi-CPU architectures by a large margin, with 46% higher Recall against Gemini, 55% higher Recall against SAFE, and 28% higher Recall against GMN. With respect to obfuscation and compiler optimization clone search approaches, BinFinder outperforms the asm2vec (single CPU architecture approach) with 30% higher Recall and BinMatch (multi-CPU architecture approach) with 10% higher Recall. Finally, our work is the first to provide noteworthy results with respect to binary clone search over the tigress obfuscator, which is a well-established open-source obfuscator.
... Tikhonov regularization is often referred to as ridge regression or L 2 regularization in machine learning. Bishop [30] also • Bishop [30] also highlights that training with noise is a form of regularization in neural network models. His findings and the research by others, such as Shalev-Shwartz and Ben-David [31], suggest that regularization results in stable algorithms. ...
... Tikhonov regularization is often referred to as ridge regression or L 2 regularization in machine learning. Bishop [30] also • Bishop [30] also highlights that training with noise is a form of regularization in neural network models. His findings and the research by others, such as Shalev-Shwartz and Ben-David [31], suggest that regularization results in stable algorithms. ...
In a data-centric era, concerns regarding privacy and ethical data handling grow as machine learning relies more on personal information. This empirical study investigates the privacy, generalization, and stability of deep learning models in the presence of additive noise in federated learning frameworks. Our main objective is to provide strategies to measure the generalization, stability, and privacy-preserving capabilities of these models and further improve them.To this end, five noise infusion mechanisms at varying noise levels within centralized and federated learning settings are explored. As model complexity is a key component of the generalization and stability of deep learning models during training and evaluation, a comparative analysis of three Convolutional Neural Network (CNN) architectures is provided.The paper introduces Signal-to-Noise Ratio (SNR) as a quantitative measure of the trade-off between privacy and training accuracy of noise-infused models, aiming to find the noise level that yields optimal privacy and accuracy. Moreover, the Price of Stability and Price of Anarchy are defined in the context of privacy-preserving deep learning, contributing to the systematic investigation of the noise infusion strategies to enhance privacy without compromising performance. Our research sheds light on the delicate balance between these critical factors, fostering a deeper understanding of the implications of noise-based regularization in machine learning. By leveraging noise as a tool for regularization and privacy enhancement, we aim to contribute to the development of robust, privacy-aware algorithms, ensuring that AI-driven solutions prioritize both utility and privacy.
... The machine learning (ML) algorithm is quite popular for the various application domains. The performance of ML algorithms is highly dependent on quality of input datasets [1]. It contains features and class label. ...
... Mathematically PT approach can be provided as below, where dataset D has three features v =< v 1 , v 2 , v 3 > , two labels c =< c 1 , c 2 > , and four samples < s (1) , s (2) , s (3) , s (4) >. ...
Multi-label classification is used to solve the problem where multiple labels are associated with single sample. Naive Bayes (NB) classifier is widely used for single label classification due to its high performance and simplicity. Therefore it is vital to extend NB for multi-label classification. In single label classification feature weighted NB gives high accuracy by solving the conditional independence assumption of NB. However, NB is not much explored for multi-label classification. This paper proposes correlation dependent feature weighted NB (MLCFWNB) for multi-label classification. The proposed MLCFWNB is tested over eight benchmark datasets. The experimental result suggest that MLCFWNB wins 60% times in case of different multi-label learning evaluation parameters.
... Secondly, we include in our model a pruning mechanism based on the global magnitude algorithm [2,17]. However, we modify this algorithm in such a way to decide not only how much (i.e., how many synapses) to prune, but also when prune. ...
... This process is performed by removing connections based on a given strategy. In this work, we use the global magnitude pruning algorithm [2,17], that simply consists in removing all the connections whose weights are smaller, in absolute value, than a threshold that is defined as the -th percentile, where is the desired pruning rate (i.e., the percentage of connections to remove). ...
... An issue unrollings often face (within GSL and beyond) is their tendency to produce layers with preactivation outputs that are nonlinear functions of the parameters; e.g., parameter products and parameters in denominators (Monga et al., 2021). This prevents the deep network from being a true neural network (NN), i.e., a function composed of layers, where each layer consists of affine transformations of data or intermediate activations followed by non-linear functions applied pointwise (Bishop, 1995). To address this issue, network designers may opt for reparameterization, but at the expense of parameter interpretability and often degraded empirical performance (Monga et al., 2021;Shrivastava et al., 2020). ...
Graphs serve as generic tools to encode the underlying relational structure of data. Often this graph is not given, and so the task of inferring it from nodal observations becomes important. Traditional approaches formulate a convex inverse problem with a smoothness promoting objective and rely on iterative methods to obtain a solution. In supervised settings where graph labels are available, one can unroll and truncate these iterations into a deep network that is trained end-to-end. Such a network is parameter efficient and inherits inductive bias from the optimization formulation, an appealing aspect for data constrained settings in, e.g., medicine, finance, and the natural sciences. But typically such settings care equally about uncertainty over edge predictions, not just point estimates. Here we introduce novel iterations with independently interpretable parameters, i.e., parameters whose values - independent of other parameters' settings - proportionally influence characteristics of the estimated graph, such as edge sparsity. After unrolling these iterations, prior knowledge over such graph characteristics shape prior distributions over these independently interpretable network parameters to yield a Bayesian neural network (BNN) capable of graph structure learning (GSL) from smooth signal observations. Fast execution and parameter efficiency allow for high-fidelity posterior approximation via Markov Chain Monte Carlo (MCMC) and thus uncertainty quantification on edge predictions. Synthetic and real data experiments corroborate this model's ability to provide well-calibrated estimates of uncertainty, in test cases that include unveiling economic sector modular structure from S$\&$P$500$ data and recovering pairwise digit similarities from MNIST images. Overall, this framework enables GSL in modest-scale applications where uncertainty on the data structure is paramount.
... Les chercheurs McCulloch et al. [76] ont publié dans les années 1940 les premiers travaux dans ce domaine. Au point de vue de la classication, Bishop [77] a proposé une modélisation discrimante du réseau. En eet, une fonction discriminante linéaire h est dénie dans l'espace des entrées E. Elle représente une combinaison linéaire du vecteur des poids ...
En raison de l’augmentation considérable des images dans la vie quotidienne, de nombreuses
applications nécessitent une étude sur leur similarité. La Carte des Dissimilarités Locales (CDL)
est une mesure, construite autour de la distance de Hausdorff, qui est très efficace pour localiser et
quantifier les différences de structures entre les images. Cette mesure a été proposée par Baudrier
et al. [1]. Avant cela, aucune solution spécifiquement locale n’a été proposée par la communauté
scientifique. À partir d’une CDL, il est cependant difficile d’interpréter et de prendre une décision
sur la similarité entre deux images. De plus, la mesure est mise en échec sur des images contenant
à la fois des structures et des textures et le comportement statistique des valeurs de la CDL n’a
jamais été étudié. Tout cela limitait ses domaines d’application. Cette thèse propose d’abord
une distribution statistique pour modéliser les valeurs des niveaux de gris des CDL des images
structurelles. Les deux paramètres de la distribution sont pertinents pour discriminer les paires
d’images en classes similaires et dissimilaires. Des modèles d’apprentissage automatique et des
tests statistiques sont utilisés pour classer les paires d’images. Mais, avant d’aborder les tests, une
extension de l’approche au problème de classification d’images multi-classes est proposée. Ensuite,
les mesures d’informations telles que l’Information Mutuelle (IM) et l’Information Disjointe (ID)
sont utilisées pour adapter la CDL sur des images avec un mélange de structures et de textures.
Nous proposons, enfin, d’appliquer la mesure au problème de détection de changements sur des
séries d’images. Nous savons aussi que, de nos jours, de nombreuses images numériques sont
falsifiées pour de la propagande ou pour cacher des informations importantes. La détection de
ces falsifications intéresse donc de nombreux acteurs majeurs de la sécurité. Dans cette thèse,
nous nous intéressons uniquement à la détection de falsifications par copier-coller. Toutes nos
approches sont basées uniquement sur la CDL et essentiellement sur les deux paramètres de la
distribution proposée. Elles sont pertinentes et certaines méthodes sont même comparées avec
des approches d’apprentissage profond de l’état de l’art.
... Normalisation techniques are conventionally considered as a technique to preprocess the data set. Data preprocessing is one of the vital stages in the development of a solution, and the choice of the preprocessing technique can significantly affect the performance of algorithms in ANN [93]. Recently, the collected data have become more complex to save memory and computational cost, and the feature's relevance is based on normalisation techniques [94]. ...
This research study aims to understand the application of Artificial Neural Networks (ANNs) to forecast the Self-Compacting Recycled Coarse Aggregate Concrete (SCRCAC) compressive strength. From different literature, 602 available data sets from SCRCAC mix designs are collected, and the data are rearranged, reconstructed, trained and tested for the ANN model development. The models were established using seven input variables: the mass of cementitious content, water, natural coarse aggregate content, natural fine aggregate content, recycled coarse aggregate content, chemical admixture and mineral admixture used in the SCRCAC mix designs. Two normalization techniques are used for data normalization to visualize the data distribution. For each normalization technique, three transfer functions are used for modelling. In total, six different types of models were run in MATLAB and used to estimate the 28th day SCRCAC compressive strength. Normalization technique 2 performs better than 1 and TANSING is the best transfer function. The best k-fold cross-validation fold is k = 7. The coefficient of determination for predicted and actual compressive strength is 0.78 for training and 0.86 for testing. The impact of the number of neurons and layers on the model was performed. Inputs from standards are used to forecast the 28th day compressive strength. Apart from ANN, Machine Learning (ML) techniques like random forest, extra trees, extreme boosting and light gradient boosting techniques are adopted to predict the 28th day compressive strength of SCRCAC. Compared to ML, ANN prediction shows better results in terms of sensitive analysis. The study also extended to determine 28th day compressive strength from experimental work and compared it with 28th day compressive strength from ANN best model. Standard and ANN mix designs have similar fresh and hardened properties. The average compressive strength from ANN model and experimental results are 39.067 and 38.36 MPa, respectively with correlation coefficient is 1. It appears that ANN can validly predict the compressive strength of concrete.
... Simply put, these are design choices which force the algorithm to learn one specific pattern over another. There are many ways to encode such biases in an artificial neural network, among which regularization strategies [21,290,170], architectural restrictions [118,210,347], parameter sharing [127,249], training recipes [125,81,72,51], or invariance or equivariance to known transformations [184,264,353,169,87]. While beneficial for practical use cases, the reliance on inductive biases tailored to specific tasks and modalities does not align with the quest for a unified deep learning algorithm. ...
Over the past decade, deep learning has advanced the analysis of text, image, audio, and video. More recently, transformers and self-supervised learning have triggered a global competition to train gigantic models on Internet-scale datasets, with massive computational resources. This thesis deals with large-scale 3D point cloud analysis and adopts a different approach focused on efficiency. We introduce methods which improve several aspects of the state-of-the-art: faster training, fewer parameters, smaller compute or memory footprint, and better utilization of realistically-available data. In doing so, we strive to devise solutions towards a more frugal and accessible Artificial Intelligence (AI). We first introduce a 3D semantic segmentation model that combines the efficiency of superpoint-based methods with the expressivity of transformers. We build a hierarchical data representation which allows us to drastically accelerate the parsing of large 3D point clouds. Our network proves to match or even surpass state-of-the-art approaches on a range of sensors and acquisition environments, while boasting orders of magnitude fewer parameters, with faster training and inference. We then build on this framework to tackle panoptic segmentation of large-scale 3D point clouds. Existing instance and panoptic segmentation methods do not scale well to large scene with numerous objects because the computation of their loss function implies a costly matching step between true and predicted instances. Instead, we frame this task as a scalable graph clustering problem, which a small network is trained to address from local objectives only, without computing the actual object instances at train time. Our lightweight model can process ten-million-point scenes at once on a single GPU in a few seconds, opening the door to 3D panoptic segmentation at unprecedented scales. Finally, we propose to exploit the complementarity of image and point cloud modalities to enhance 3D scene understanding. We place ourselves in a realistic acquisition setting where multiple arbitrarily-located images observe the same scene, with potential occlusions. Unlike previous 2D-3D fusion approaches, we learn to select information from various views of the same object based on their respective observation conditions: camera-to-object distance, occlusion rate, optical distortion, etc. Our efficient implementation achieves state-of-the-art results both in indoor and outdoor settings, with minimal requirements: raw point clouds, arbitrarily-positioned images, and their cameras poses. Overall, this thesis upholds the principle that for settings with limited data availability, exploiting the structure of the problem unlocks both efficient and performant architectures.
... MLPs are trained using the backpropagation algorithm, a supervised learning technique that adjusts the weights and biases of the neurons based on the discrepancy between the predicted output and the actual output (Goodfellow et al. 2016). The backpropagation algorithm computes the gradient of the error function with respect to the network's weights and biases, subsequently updating these parameters using a gradient descent optimization algorithm (Bishop 1995). This iterative process continues until the error is minimized, signifying the completion of MLP training. ...
The quality of the sample plays a vital role in developing accurate models using machine learning. This aspect is equally important when evaluating regional landslide susceptibility using machine learning. Previous studies have mostly employed random generation methods to select samples, which often fail to select representative samples. Therefore, this study proposes the KK-sampling method, which uses K-means and KNN algorithms to analyze relevant attributes of the study area and select samples. To evaluate the effectiveness of the proposed method, this study employed MLP, RF, and XGBoost models in conjunction with the KK-sampling method, with Zhong County, Chongqing serving as a case study. The results indicate that the KK-sampling method significantly improves the stability and accuracy of the model. Additionally, this study analyzed the importance of landslide factors in Zhong County using SHAP values. The findings provide a reference for establishing a reasonable and effective landslide susceptibility model in the region.
... For all other experiments, the plasticity rules were parametrized using multi-layer perceptrons [MLPs,33], with different inputs from the rate network, depending on the chosen model: ...
Synaptic plasticity is thought to be critical for building and maintaining brain circuits. Models of plasticity, or plasticity rules, are typically designed by hand, and evaluated based on their ability to elicit similar neuron or circuit properties to ground truth. While this approach has provided crucial insights into plasticity mechanisms, it is limited in its scope by human intuition and cannot identify all plasticity mechanisms that are consistent with the empirical data of interest. In other words, focusing on individual hand-crafted rules ignores the potential degeneracy of plasticity mechanisms that explain the same empirical data, and may thus lead to inaccurate experimental predictions. Here, we use an unsupervised, adversarial approach to infer plasticity rules directly from neural activity recordings. We show that even in a simple, idealised network model, many mechanistically different plasticity rules are equally compatible with empirical data. Our results suggest the need for a shift in the study of plasticity rules, considering as many degenerate plasticity mechanisms consistent with data as possible, before formulating experimental predictions.
... Sun et al. (2015b) utilized Granger causality for selecting machine learning features in two-dimensional space. This approach outperformed the traditional feature selection techniques like Principal Component Analysis (PCA) (Abdi & Williams (2010)), Functional Connectome (FC) (Bishop et al. (1995)), and Recursive Feature Elimination (RFE) (Guyon et al. (2002)) due to the ability of Granger causality to identify the causal connection between the input variable and the chosen time series. Nogueira et al. (2021) published a survey paper that mainly focused on the applications of causal discovery in machine learning. ...
The ability to understand causality from data is one of the major milestones of human-level intelligence. Causal Discovery (CD) algorithms can identify the cause-effect relationships among the variables of a system from related observational data with certain assumptions. Over the years, several methods have been developed primarily based on the statistical properties of data to uncover the underlying causal mechanism. In this study, we present an extensive discussion on the methods designed to perform causal discovery from both independent and identically distributed (I.I.D.) data and time series data. For this purpose , we first introduce the common terminologies used in causal discovery literature and then provide a comprehensive discussion of the algorithms designed to identify causal relations in different settings. We further discuss some of the benchmark datasets available for evaluating the algorithmic performance, off-the-shelf tools or software packages to perform causal discovery readily, and the common metrics used to evaluate these methods. We also evaluate some widely used causal discovery algorithms on multiple benchmark datasets and compare their performances. Finally, we conclude by discussing the research challenges and the applications of causal discovery algorithms in multiple areas of interest.
... After concatenation, traditional supervised machine learning algorithms can be used for learning. Examples of traditional machine learning methods mentioned in the review article that have been used in multivariate biomarker discovery include Naive Bayes [36], artificial neural networks [37], support vector machine [38], k-nearest neighbors [39], random forest (RF) [40] and least absolute shrinkage operator (LASSO) [41]. In addition, several deep neural networks [42] have been used for concatenated multi-omics data. ...
Advancements in technologies that generate large-scale omics data and the develop- ment of machine learning methods to analyze this data provide new opportunities for the field of medicine, such as improved prevention, diagnosis and treatment of diseases through the application of multivariate biomarkers. Moreover, multi- variate biomarkers offer opportunities for precision medicine where treatments can be tailored to the needs of individual patients. Multivariate biomarker discovery which involves the prediction of clinical outcomes reproducibly using a small set of biomarkers, has emerged as a promising approach. However, from a machine learning perspective, the integration of multi-omics data to discover multi-omics biomarkers remains challenging. In addition, interpretability and explainability are key issues in the translation of models into clinical practice.
Recently proposed group of kernel methods called sparse pre-image kernel machines has an embedded feature selection and offers improved interpretability compared to traditional kernel methods. Another benefit for learning multi-omics biomarkers is that sparse pre-image kernel machines can be extended to multi-view learning. This thesis explores the application of sparse pre-image kernel machines to multivariate biomarker discovery using a multi-omics coronavirus disease 2019 data set. To study whether the stability of feature selection can be improved, this thesis couples a method known as stability selection with sparse pre-image kernel machines. The stability of feature selection and model performance with the selected features are compared to two baseline methods, random forest and logistic regression.
This thesis considers two types of feature selection pipelines for sparse pre-image kernel machines, where the first is a general grid search approach to select a level of regularization, and thus features. In the second pipeline, sparse pre-image kernel machines is combined with stability selection. Results show that stability selection improves the stability of the learned features significantly. In addition, the proposed multi-view approach learns a more balanced set of features compared to other methods in terms of learning features from both views. The findings of this thesis provide insights into the potential application of sparse pre-image kernel machines for the discovery of multi-omics biomarkers in complex diseases.
... In the work presented in [56,28] the authors showed that from the regularization principles and through a solution of a variational problem, the RBF model is a subclass of regularization networks. Particularly important, in the case of the Gaussian RBF (GRBF), is the definition of the shape parameter [11], which is problem dependent and controls the variance (or the width) of the Gaussian basis function. The GRBF is very sensitive to the shape parameter as shown empirically in [45] in case of interpolation, and a usual way to set it is to fix it through cross-validation procedures. ...
Providing a model that achieves a strong predictive performance and at the same time is interpretable by humans is one of the most difficult challenges in machine learning research due to the conflicting nature of these two objectives. To address this challenge, we propose a modification of the Radial Basis Function Neural Network model by equipping its Gaussian kernel with a learnable precision matrix. We show that precious information is contained in the spectrum of the precision matrix that can be extracted once the training of the model is completed. In particular, the eigenvectors explain the directions of maximum sensitivity of the model revealing the active subspace and suggesting potential applications for supervised dimensionality reduction. At the same time, the eigenvectors highlight the relationship in terms of absolute variation between the input and the latent variables, thereby allowing us to extract a ranking of the input variables based on their importance to the prediction task enhancing the model interpretability. We conducted numerical experiments for regression, classification, and feature selection tasks, comparing our model against popular machine learning models and the state-of-the-art deep learning-based embedding feature selection techniques. Our results demonstrate that the proposed model does not only yield an attractive prediction performance with respect to the competitors but also provides meaningful and interpretable results that potentially could assist the decision-making process in real-world applications. A PyTorch implementation of the model is available on GitHub at the following link. https://github.com/dannyzx/GRBF-NNs
... We normalized all satellite features of the int and gf time series from zero to one by parcel and year to scale the input features to the same range of values, as this improves model training (Bisphop, 1995). Weather data were normalized using all weather data from 2017-2019 to preserve the temporal and spatial variability between years and study sites. ...
The European Union’s Common Agricultural Policy (CAP) and the Habitats Directive aim to improve biodiversity in agricultural landscapes. Both policies require enormous monitoring, which can be facilitated by remote sensing. Use intensity, measured by mowing frequency is an important indicator of biodiversity in permanent grasslands. The frequency and timing of mowing can be determined using satellite remote sensing because photosynthetically active biomass changes rapidly in response to mowing. However, the rapid regrowth of grasses requires very dense satellite time series for reliable detection. Radar time series can complement optical time series and fill in cloud-related gaps to overcome this problem. Additional weather data can support the detection of grassland mowing events, as mowing events are associated with specific meteorological conditions. However, previous studies have not fully exploited both potentials or different machine learning approaches for mowing event detection.
This study presents a new transferable two-step approach to detect grassland mowing events using combined optical and SAR data and additional weather data. First, we filled cloud-related gaps in optical time series using a supervised machine learning regression with optical and SAR data. We then classified time series sequences of optical, SAR and weather data into mown and unmown using four different machine learning algorithms. We used time series of NDVI and EVI (combined Sentinel-2 and Landsat 8), SAR backscatter, six-day interferometric coherence, backscatter radar vegetation index, backscatter cross-ratio (Sentinel-1), and temperature and precipitation sums. Our test sites are distributed across Germany and cover the entire gradient of grassland use intensities.
Mowing events could be detected with F1 values of up to 89%, first cut with up to 94%. Our results show no structural advantage of infilling time series with machine learning over linearly interpolated time series. The combined Sentinel-2 and Landsat-8 time series provided dense time series with mostly median gaps less than 20 days, which proved sufficient to reliably detect mowing events. SAR data were not essential for mowing event detection in our study, but weather data improved classification results for models trained on all areas and years. However, when the model was transferred to unknown years or areas that were not used for training, SAR data improved detection accuracy, whereas weather data degrade it. Models trained on all years but not all study sites detected mowing events with an accuracy of up to F1 = 76%. Models trained with all regions but not all years detected mowing events in untrained years with F1 up to 80%.
... This algorithm uses a limited amount of computer memory, and was used here for training the Autoencoders to improve training speed. The code, besides having an implemented square-error cost function, was extended to operate on cross entropy error [14] and to use momentum. The regularization term of W 2 was added to the cost error function, for the purpose of decreasing the magnitude of the weights and help to prevent overfitting. ...
This thesis describes several unconventional methods of signal analysis for
the purpose of modeling and recognizing speech and music. This process is
commonly referred to as feature extraction and is an important step in any
machine learning task. Most of the current research on this topic involves
Fourier transform derived features. These are usually formulated as a set of
spectral features arranged according to a perceptual scale, like the mel scale,
and possibly transformed into the cepstrum domain.
The basis for this thesis lies in the use of alternative signal representation techniques derived from two signal processing methodologies. One involves sparse coding mechanisms and the matching pursuit (MP) algorithm.
The other is a novel wavelet derived feature set known as the scattering
wavelet transform (SWT).
These methods have already been applied to various signal processing tasks, involving both audio and image processing. On the other hand,
they have not been utilized in many practical settings, like the modern largevocabulary continuous speech recognition (LVCSR) systems. The sparse coding mechanisms are often used in computer vision research but rarely are they
applied to analyzing audio and even less so to speech recognition. The SWT
is a fairly novel technique and while it has been used for solving some speech
related problems it was never utilized in an actual LVCSR system.
Within the thesis, sparse coding mechanisms are studied in detail in order to verify their capacity for modeling speech signals. Several coding
mechanisms and dictionary adaptation methods are discussed and the technique that yields the highest quality of reconstruction is chosen. Similarly,
the SWT is chosen in a configuration best fitting its intended use.
Next, both of these feature sets are tested on the problem of framewise phoneme classification, representative of the issues behind the acoustic
modeling used in most speech recognition systems. The SWT is additionally
tested on two more problems: musical genre recognition and LVCSR. All
these methods are compared to the most commonly used signal processing
methods.
Various topics related to the above experiments were also discussed,
like the construction of LVCSR and various usability concerns related to
exploiting such systems in real-life situations, with an example of a dialog
system operating in a telephony environment.
This dissertation postulates four main theses. It is shown that
sparse coding can be effectively used to encode speech signals and that this
form of representation can be used to improve the performance of speech
recognition. The second thesis shows that SWT also enhances speech recognition accuracy, which is proven using the same problem that was utilized in
the first thesis. In addition to that, the third thesis demonstrates that SWT
derived feature set also improves the performance of LVCSR. The final thesis
shows that IA has a substantial significance in voice user interface (VUI)
design.
The author’s contribution to this field of science is primarily in the
novel application of the methods described above, in order to make them
usable in practical speech recognition tasks. The author’s contribution also
includes a novel approach to the conversion of sparse coding into a form which
can be applied to speech recognition and an innovative concept of exploiting
IA in the domain of VUIs.
... The most basic structural and functional unit of a neural network is a neuron, which is illustrated in figure 6. These neurons are generally stacked together in the layered structure (Anderson and McNeill, 1992) used form of FNN is the multi-layer perception (MLP) (Rumelhart et al., 1985)(Bishop et al., 1995. ...
Traffic Congestion wastes time and energy, which are the two most valuable commodities of the current century. It happens when too many vehicles try to use a transportation infrastructure without having enough capacity. However, researches indicate that adding extra lane without studying the future consequences does not improve the situation. Our goal is to add another layer of information to the traffic data, find which type of vehicles are contributing to road traffic congestion, and predict future road traffic congestion and demands based on the historical data.
We collected more than 400,000 images from traffic cameras installed in Autoroute
40, in the city of Montreal. The images were collected for five consecutive weeks from different locations from April 14, 2019, up until May 18, 2019. To process these images and extract useful information out of them, we created an object detection and classification model using the Faster RCNN algorithm. Our goal was to be able to detect different types of vehicles and see if we have traffic congestion in an image. In order to improve the accuracy and reduce the error rate, we provided multiple examples with different conditions to the model. By introducing blurry, rainy, and low light images to the model, we managed to build a robust model that could do the detection and classification task with excellent accuracy.
Furthermore, by extracting the information from the collected images, we created a
dataset of the number of vehicles in each location. After analyzing and visualizing the data, we find out the most congested areas, the behavior of the traffic flow during the day, peak hours, the contribution of each type of vehicle to the traffic, seasonality of the data, and where we can see each type of vehicle the most.
Finally, we managed to predict the total number of congestion incidents for seven days based on historical data. Besides, we were able to predict the total number of different types of vehicles on the road as well. In order to do this task, we developed multiple Regression, Deep Learning, and Time Series Forecasting models and trained them with our vehicle count dataset. Based on the experimental results, we were able to get the best predictions with the Deep Learning models and succeeded in predicting future road traffic congestion with excellent accuracy
... Experiment 1: Encoder quality. Using Fisher's class separability metric [Bishop et al. 1995] over the representation learned by the encoder, we measure the separability between the motion classes within the latent space, where a motion class is defined as submotions within a single motion file. As shown in Table 1, CALM learns to encode motions into representations with much better separation. ...
In this work, we present Conditional Adversarial Latent Models (CALM), an approach for generating diverse and directable behaviors for user-controlled interactive virtual characters. Using imitation learning, CALM learns a representation of movement that captures the complexity and diversity of human motion, and enables direct control over character movements. The approach jointly learns a control policy and a motion encoder that reconstructs key characteristics of a given motion without merely replicating it. The results show that CALM learns a semantic motion representation, enabling control over the generated motions and style-conditioning for higher-level task training. Once trained, the character can be controlled using intuitive interfaces, akin to those found in video games.
... ANN has an advantage over traditional machine learning techniques of being able to observe relationships in data by replicates the brain's capacity to recognize and sort patterns through the interconnected networks of several neurons ( Kumar et al. 2021;Xu et al. 2022;Bagheri et al. 2018;Chettry and Surawar 2021). MLP is composed of three distinct layers, which are alluded to as input layers, output layers, and possibly hidden or exclusionary layers that can be used to identify a non-linear relationship in real life (Bishop Christopher et al. 1995;Dey et al. 2021). ...
Numerous cities throughout the world are experiencing tremendous population growth in their peripheral areas, resulting in a progressive modification of landscapes and raising serious concerns about natural environments, notably forests and agricultural area. Monitoring LULC changes can assist in understanding historical trends, while simulation-based modelling shed light on possible potential future developments. Both of these tactics are indispensable and complimentary for implementing effective land use policies to mitigate the adverse ramifications of urbanization. Present area of investigation, Siliguri town one of prime trading hub of whole north-east India surrounded by ecologically sensitives zones Himalayas. To monitor land use dynamics of peri-urban spaces in Siliguri town Landsat images of 2000, 2010 and 2020 were derived from USGS and classified using Support vector machine learning algorithms. Following the quantification of the previous trend of landuse change, an integrated Artificial Neural Network (ANN) and CA-Markov chain Model was utilized to forecast LULC for the years 2030 and 2050. Eleven pertinent geographical factors, comprising topographical, socioeconomic, and connectivity information, were generated and validated using the crammer v test. The results from LULC modeling predicts as compared to 2020, the urban area is expected to increase by 48.23%, while forest areas, other vegetation cover, and agricultural areas are predicted to shrink by 9.42%, 29.83%, and 26.60% respectively, by the year 2050. The results could provide useful information about historical and potential landuse change and as well as assist local governments in formulating management strategies for the protection of ecological resource.
... The most basic structural and functional unit of a neural network is a neuron, which is illustrated in figure 6. These neurons are generally stacked together in the layered structure (Anderson and McNeill, 1992) used form of FNN is the multi-layer perception (MLP) (Rumelhart et al., 1985)(Bishop et al., 1995. ...
Traffic Congestion wastes time and energy, which are the two most valuable commodities of the current century. It happens when too many vehicles try to use a transportation infrastructure without having enough capacity. However, researches indicate that adding extra lane without studying the future consequences does not improve the situation. Our goal is to add another layer of information to the traffic data, find which type of vehicles are contributing to road traffic congestion, and predict future road traffic congestion and demands based on the historical data.
We collected more than 400,000 images from traffic cameras installed in Autoroute40, in the city of Montreal. The images were collected for five consecutive weeks from different locations from April 14, 2019, up until May 18, 2019. To process these images and extract useful information out of them, we created an object detection and classification model using the Faster RCNN algorithm. Our goal was to be able to detect different types of vehicles and see if we have traffic congestion in an image. In order to improve the accuracy and reduce the error rate, we provided multiple examples with different conditions to the model. By introducing blurry, rainy, and low light images to the model, we managed to build a robust model that could do the detection and classification task with excellent accuracy.
Furthermore, by extracting the information from the collected images, we created a dataset of the number of vehicles in each location. After analyzing and visualizing the data, we find out the most congested areas, the behavior of the traffic flow during the day, peak hours, the contribution of each type of vehicle to the traffic, seasonality of the data, and where we can see each type of vehicle the most.
Finally, we managed to predict the total number of congestion incidents for seven days based on historical data. Besides, we were able to predict the total number of different types of vehicles on the road as well. In order to do this task, we developed multiple Regression, Deep Learning, and Time Series Forecasting models and trained them with our vehicle count dataset. Based on the experimental results, we were able to get the best predictions with the Deep Learning models and succeeded in predicting future road traffic congestion with excellent accuracy.
... Two surrogate models for prediction of residual stress in the WCL of pipe girth welds are created by using an Artificial Neural Network (ANN) ensemble approach. To tackle the problem of the poor profile curvature prediction, weight decay term is not added in the cost function, as it can lead to underfittng if its hyperparameters are not tuned [32]. Instead, to avoid overfitting, the number of epochs is tuned with cross-validation. ...
... Sun et al. (2015b) utilized Granger causality for selecting machine learning features in two-dimensional space. This approach outperformed the traditional feature selection techniques like Principal Component Analysis (PCA) (Abdi and Williams (2010)), Functional Connectome (FC) (Bishop et al. (1995)), and Recursive Feature Elimination (RFE) (Guyon et al. (2002)) due to the ability of Granger causality to identify the causal connection between the input variable and the chosen time series. Nogueira et al. (2021) published a survey paper that mainly focused on the applications of causal discovery in machine learning. ...
Causal Discovery (CD) is the process of identifying the cause-effect relationships among the variables from data. Over the years, several methods have been developed primarily based on the statistical properties of data to uncover the underlying causal mechanism. In this study we introduce the common terminologies in causal discovery, and provide a comprehensive discussion of the approaches designed to identify the causal edges in different settings. We further discuss some of the benchmark datasets available for evaluating the performance of the causal discovery algorithms, available tools to perform causal discovery readily, and the common metrics used to evaluate these methods. Finally, we conclude by presenting the common challenges involved in CD and also, discuss the applications of CD in multiple areas of interest.
... By performing this data augmentation, we enlarged our data set ten times resulting in the training data size of 3840 and 560 for the Sonix-Touch and Wireless US scanners respectively. Training for both US systems was performed using the random sampling method where 50 images were randomly selected from AP scan planes dataset for testing and the rest AP scan plane images and all other US images from the rest four scan planes were used for training after augmentation (Bishop, 1995). To further solidify the identification accuracy, we repeated the process of random selection of AP scan plane image five times and averaged the identification accuracy over these five times testing (5-fold cross validation). ...
In order to reduce the total amount of radiation exposure and provide real-time guidance ultrasound has been incorporated as a potential intra-operative imaging modality into various orthopedic procedures. However, high levels of noise, various imaging artifacts, and bone boundaries appearing several millimeters in thickness hinder the success of ultrasound as an alternative imaging modality in assisting orthopedic surgery procedures. Additional difficulties are also encountered during manual operation of the ultrasound transducer during image acquisition. In this work, we proposed a combination of novel scan plane identification method, based on convolutional neural networks, and bone surface localization method. The bone surface localization approach utilizes both local phase information, a combination of three different local image phase information and signal transmission map obtained from an L1 norm based contextual regularization method. The proposed network was utilized on two different US systems and to identify five different scan planes. Validation was performed on scans obtained from 16 volunteers. The correct scan plane identification rate of over 93% has been obtained. Validation against expert segmentation achieved a mean vertebra surface localization error of 0.42 mm.
... Receiver Operating Curve (ROC), Confusion Matrix and 'Accuracy vs Standard deviation plot' are used to evaluate the performance. ROC plots the True Positive Rate (Sensitivity) and False Positive Rate (1specificity) [9], [8]. Overfitting can be detected if the accuracy of training data is much greater than the accuracy of testing data. ...
... The authors considered an imbalanced data set containing 25,332 putative off-target DNA sequences, with 152 verified positive off-targets, and the other being negative. They compared the five following machine learning algorithms -AdaBoost [97], random forest, a multi-layer perceptron [98], SVM, and decision trees [99]. In their experiments, Zhang et al. demonstrated that the ensemble-based AdaBoost algorithm was able to outperform the other predictive algorithms in terms of the area under the precision recall curve (AUPRC) and the area under the receiver operating characteristic curve (AUROC) metrics. ...
CRISPR/Cas9 (Clustered Regularly Interspaced Short Palindromic Repeats and CRISPR-associated protein 9) is a popular and effective two-component technology used for targeted genetic manipulation. It is currently the most versatile and accurate method of gene and genome editing, which benefits from a large variety of practical applications. For example, in biomedicine it has been used in research related to cancer, virus infections, pathogen detection and genetic diseases. Recent CRISPR/Cas9 research is based on data-driven models for on-and off-target prediction as a cleavage may occur at non-target sequence locations. Currently, conventional machine learning and deep learning methods are applied on a regular basis to accurately predict the sgRNA (single-guide RNA) on-target knockout efficacy and off-target profile. In this paper, we present an overview and a comparative analysis of traditional machine learning and deep learning models used in CRISPR/Cas9. We highlight the key research challenges and directions associated with target activity prediction. We discuss some recent advances in the sgRNA-DNA sequence encoding used in state-of-the-art on-and off-target prediction models. Furthermore, we present the most popular deep learning neural network architectures used in CRISPR/Cas9 prediction models. Finally, we summarize existing challenges and discuss possible future investigations in the field of on-and off-target prediction. Our paper provides valuable support for academic and industrial researchers interested in the application of machine learning methods in the field of CRISPR/Cas9 genome editing.
Remote physiological measurement (RPM) is an essential tool for healthcare monitoring as it enables the measurement of physiological signs, e.g., heart rate, in a remote setting via physical wearables. Recently, with facial videos, we have seen rapid advancements in video-based RPMs. However, adopting facial videos for RPM in the clinical setting largely depends on the accuracy and robustness (work across patient populations). Fortunately, the capability of the state-of-the-art transformer architecture in general (natural) video understanding has resulted in marked improvements and has been translated to facial understanding, including RPM. However, existing RPM methods usually need RPM-specific modules, e.g., temporal difference convolution and handcrafted feature maps. Although these customized modules can increase accuracy, they are not demonstrated for their robustness across datasets. Further, due to their customization of the transformer architecture, they cannot use the advancements made in general video transformers (GVT). In this study, we interrogate the GVT architecture and empirically analyze how the training designs, i.e., data pre-processing and network configurations, affect the model performance applied to RPM. Based on the structure of video transformers, we propose to configure its spatiotemporal hierarchy to align with the dense temporal information needed in RPM for signal feature extraction. We define several practical guidelines and gradually adapt GVTs for RPM without introducing RPM-specific modules. Our experiments demonstrate favorable results to existing RPM-specific module counterparts. We conducted extensive experiments with five datasets using intra-dataset and cross-dataset settings. We highlight that the proposed guidelines GVT2RPM can be generalized to any video transformers and is robust to various datasets.
In the analysis of high-dimensional data, applying dimensionality reduction techniques is often necessary. However, setting an suitable criterion for determining the extent of dimensionality reduction often poses a challenge. The use of automatic relevance determination (ARD) in a linear latent variable model, such as Bayesian PCA, offers a way to automatically identifying the effective dimensionality of the latent space. However, conventional ARD methods often fail to extract sparse representation of latent variables in dealing with noisy, non-Gaussian, or nonlinearly observed data, such as calcium imaging data. To encourage the sparsity of the latent space, we proposed a dual ARD method in a linear latent variable model that applies ARD priors to both loading weights and latent variables. We first detailed our dual ARD method and mathematically analyzed how the dual ARD priors promote more sparsity in the latent space. We then evaluated the performance of the dual ARD methods against existing dimensionality reduction techniques using both simulated datasets and actual calcium imaging data. While conventional methods could retrieve essential signals in linear Gaussian settings, the dual ARD method outperformed the previous models in extracting low-dimensional signals from simulated calcium imaging data that contain higher levels of nonlinear noise. In applying the dual ARD method to actual two-photon calcium imaging data, we were able to identify low-dimensional latent variables that were sufficient for performing a sound localization decoding task successfully. Additionally, decoding performance across different cortical depths reflects the varied roles that specific cortical layers play in sound localization. In conclusion, the dual ARD method is well-suited for automatically reducing dimensionality of calcium imaging data while preserving essential information for further analysis.
Streaming content advances and the appearance of online media raised the ability for massive content sharing that reaches thousands of people worldwide in a real-time fashion. Fake news spreading is nowadays the main concern of several authorities worldwide due to the negative impact and potential to induce social and political instability in our society. Therefore, fake news detection and suppression gained increased attention as an important topic in natural language processing and machine learning academic research. Regardless of the state-of-the-art methods available for fake news detection , a good corpus revealing novel language-specific counterfeit aspects is also important to exploit and distinguish between real and fake news in the context of social and political impacts for specific regions. This paper extends a previous Brazilian Portuguese corpora dataset and proposes using and comparing several deep learning and classical machine learning models to detect counterfeit content in the Portuguese language. Moreover, we propose using text summarization to achieve concise news summaries and prevent losing relevant information. This work presents an updated and balanced version of the FakeRecogna dataset for detecting fake news articles using a temporal learning approach based on efficient and well-known deep learning models.
Industrial data mining usually deals with data from different sources. These heterogeneous datasets describe the same object in different views. However, samples from some of the datasets may be lost. Then the remaining samples do not correspond one-to-one correctly. Mismatched datasets caused by missing samples make the industrial data unavailable for further machine learning. In order to align the mismatched samples, this article presents a cooperative iteration matching method (CIMM) based on the modified dynamic time warping (DTW). The proposed method regards the sequentially accumulated industrial data as the time series. Mismatched samples are aligned by the DTW. In addition, dynamic constraints are applied to the warping distance of the DTW process to make the alignment more efficient. Then a series of models are trained with the cumulated samples iteratively. Several groups of numerical experiments on different missing patterns and missing locations are designed and analyzed to prove the effectiveness and the applicability of the proposed method.
Supervised dimensionality reduction has become an important theme in the last two decades. Despite the plethora of models and formulations, there is a lack of a simple model that aims to project the set of patterns into a space defined by the classes (or categories). We set up a model where each class is represented as a 1D subspace of the vector space formed by the features. Assuming the set of classes does not exceed the cardinality of the features, the model results in multi-class supervised learning in which the features of each class are projected into the class subspace. Class discrimination is guaranteed via the imposition of the orthogonality of the 1D class sub-spaces. The resulting optimization problem—formulated as the minimization of a sum of quadratic functions on a Stiefel manifold—while being non-convex (due to the constraints), has a structure for which we can identify when we have reached a global minimum. After formulating a version with standard inner products, we extend the formulation to a reproducing kernel Hilbert space and similarly to the kernel version. Comparisons with the multi-class Fisher discriminants and principal component analysis showcase the relative merits toward dimensionality reduction.
The low quality of current jobs in Mexico and their scarcity have led to the need to undertake. Consequently, people have ceased to be employees to become entrepreneurs. However, the specialized literature ensures that there are factors that may well characterize this venture. In the chapter, a first approach is made to the risk aversion that a person faces when failing to decide to be an entrepreneur in Mexico. The information integrated by the reports of the Global Entrepreneurship Monitor (GEM) served as input so that through a Multiple Linear Regression Analysis (MLRA). During the period 2011-2021, it was verified if the factors of education, experience, knowledge, skills, age, among others, directly influence a person to make the decision to start a business in Mexico.
In this study, we introduce a method to directly invert for porosity, Vclay and hydrocarbon saturation (Shc) simultaneously from pre-stack seismic data using deep learning approach. We implemented L1 norm in the loss function for Shc estimation, added noise into synthetic seismic dataset for training, and estimated uncertainties in the inversion results by training multiple network models. UNet architecture (ResNet-18 as encoder) is used due to its ability to preserve spatial resolution. The inputs for the network are the angle stacks whereas the outputs are the petrophysical properties. We implemented mean-squared error and L1 norm as the loss functions during the training process. The L1 norm is the mean absolute values of the predicted hydrocarbon saturation, which can help promotes sparsity. The network learns on synthetic dataset. We use facies-based geostatistical simulation to generate 1D synthetic petrophysical logs. Then linking the petrophysical properties to elastic properties through rock physics model (RPM), followed by computation of reflectivities using full Zoeppritz equations at five different groups of incidence angles (0°-55°). The traces in each group are convolved with the source wavelet prior to stacking the synthetic seismograms. To increase the variability of possible scenarios, we vary the spherical variogram ranges (8,10, and 12ms), use four different types of suitable RPM, apply oil and gas cases for the hydrocarbon fluid types, and convolve with nine different sets of angle dependent source wavelets. Two synthetic datasets are prepared: Dataset1 (ideal noiseless case) and Dataset 2 (noise added to the angle stacks), and a field data. The first (MLT1) and second (MLT2) machine learning are trained on a sub-dataset in Dataset 1 and 2 respectively. Based on the field dataset, the results from MLT2 show a better prediction performance than MLT1, with an average correlation coefficient of 0.68 (porosity), 0.74 (Vclay) and 0.67 (Shc) achieved. The better results from MLT2 can be related to the nature of measured seismic which contain noise that being learnt by MLT2. For uncertainty estimation, the network (ML3) is trained for 20 times on randomly selected sub-dataset in Dataset 2 using Monte Carlo dropout technique. The uncertainty is estimated by calculating the standard deviation of the solutions provided by ML3 when applying to the field data. Uncertainty estimation allows quantification on the stability of the solutions when varying training dataset.
We present in this paper a versatile machine learning (ML) based vehicle-to-vehicle (V2V) connectivity model constructed from empirical measurements to enable the mutual exchange of data among vehicles. We portray the results of a campaign for measurements and relevant models for the V2V channel within the 5-GHz band in the form of attained RSSI and bit rate values in multiple V2V environments. To that end, we perform a parallel analysis to compare performance of an assortment of ML regression algorithms to assess their capabilities in predicting RSSI values and bit rates, including K-nearest-neighbors, AdaBoost, Regression Trees, Random Forest, SGD, SVM, and Artificial Neural Networks to predict connectivity patterns. Results in the form of numerical analysis illustrate that in our connectivity model RSSI and bitrate could be effectually predicted utilizing a subset of the group of considered ML ...
In marine vessel operations, fuel costs are major operating costs which affect the overall profitability of the maritime transport industry. The effective enhancement of using ship fuel will increase ship operation efficiency. Since ship fuel consumption depends on different factors, such as weather, cruising condition, cargo load, and engine condition, it is difficult to assess the fuel consumption pattern for various types of ships. Most traditional statistical methods do not consider these factors when predicting marine vessel fuel consumption. With technological development, different statistical models have been developed for estimating fuel consumption patterns based on ship data. Artificial Neural Networks (ANN) are some of the most effective artificial methods for modelling and validating marine vessel fuel consumption. The application of ANN in maritime transport improves the accuracy of the regression models developed for analysing interactive relationships between various factors. The present review sheds light on consolidating the works carried out in predicting ship fuel consumption using ANN, with an emphasis on topics such as ANN structure, application and prediction algorithms. Future research directions are also proposed and the present review can be a benchmark for mathematical modelling of ship fuel consumption using ANN.
Background
FaceReader is a validated software package that uses computer vision technology for facial expression recognition which has become increasingly popular in academic research to expedite, scale, and decrease the cost of facial emotion analysis. In this study, we compare FaceReader analysis to human evaluator interpretation in order to define standard values for the software output.Methods
Randomly generated facial images produced by generative adversarial networks were analyzed using FaceReader and by survey participants (n=496). The age, facial emotion, and intensity of emotion as determined by the software and survey participants were recorded. Results were analyzed and compared.Results80 randomly generated images (20 children, 20 young adult, 20 middle aged, and 20 elderly; 38 male and 42 female) were included.Analysis of correlation between most common expression identified by FaceReader and the primary emotion detected by surveyors showed strong correlation (κ = 0.77, 95% CI = 0.64–0.91).On analyzing this correlation by age group, there was fair correlation in children (κ = 0.40, 95% CI = 0.078–0.72), perfect correlation in young adults (κ = 1.0, 95% CI = 1.0–1.0), strong correlation in middle aged adults (κ = 0.79, 95% CI = 0.53–1) and near perfect in elderly adults(κ = 0.9 , 95% CI = 0.7–1.0).Conclusions
We provided the first study defining the expected average values generated by FaceReader in generally smiling images. This can be used as a standard in future studies.Level of Evidence IV This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266.
This study reports continuous measurements of subsurface soil radon as well as environmental parameters for a period of three years. The survey was carried out along the active fault area in the Indo-Myanmar subduction zone in the north-eastern part which lies in the highest seismic zone of India. The wavelet-based decomposition of the environmental parameters was done using discrete wavelet transformation technique. The denoised environmental parameters by discrete wavelet transformation technique was fed as the inputs to the MLR (multiple linear regression) and MLP (multilayer perceptron) models. Residual radon was calculated and correlated with nearby seismic events. Many events of magnitude greater than or equal to 5 have occurred in the investigation area. It was possible to successfully correlate one event with the anomalous variation in soil radon. The correlated event was the only one with the shallow epicentral depth indicating that the investigated area has undergone a shallow rock fracturing due to the stress generated before the occurrence of the seismic event.
La recherche sur l’intelligence artificielle (IA) appliquée à la médecine d’urgence et son utilisation au quotidien dans les structures d’urgences (SU) ont augmenté significativement ces dernières années. L’IA doit être considérée comme un outil d’aide à la prise en charge diagnostique et thérapeutique des patients et d’amélioration de l’organisation des SU, notamment par la prise en compte de contraintes « métiers », contextuelles, relatives aux patients et plus généralement structurelles. L’IA comporte des avantages (reproductibilité, rapidité) mais aussi des risques (erreur, perte d’esprit critique). À l’image du Règlement général sur la protection des données et notamment de santé, la Commission européenne a publié un projet de règlement nommé « AI Act » pour la conception, le développement et l’utilisation des algorithmes d’IA. Elle souhaite imposer, entre autres, une garantie humaine, autrement dit une supervision humaine pour assurer la sécurité des patients, des soignants et des institutions. La mise en place d’un collège de garantie humaine pluriprofessionnel visant à garantir la supervision des outils d’IA de la conception au développement, au déploiement et à l’utilisation quotidienne permettra ainsi d’assurer durablement la sécurité des patients.
Knowledge graphs (KGs) store relational information in a flexible triplet schema and have become ubiquitous for information storage in domains such as web search, e-commerce, social networks, and biology. Retrieval of information from KGs is generally achieved through logical reasoning, but this process can be computationally expensive and has limited performance due to the large size and complexity of relationships within the KGs. Furthermore, to extend the usage of KGs to non-expert users, retrieval over them cannot solely rely on logical reasoning but also needs to consider text-based search. This creates a need for multi-modal representations that capture both the semantic and structural features from the KGs.
The primary objective of the proposed work is to extend the accessibility of KGs to non-expert users/institutions by enabling them to utilize non-technical textual queries to search over the vast amount of information stored in KGs. To achieve this objective, the research aims to solve four limitations: (i) develop a framework for logical reasoning over KGs that can learn representations to capture hierarchical dependencies between entities, (ii) design an architecture that can effectively learn the logic flow of queries from natural language text, (iii) create a multi-modal architecture that can capture inherent semantic and structural features from the entities and KGs, respectively, and (iv) introduce a novel hyperbolic learning framework to enable the scalability of hyperbolic neural networks over large graphs using meta-learning.
The proposed work is distinct from current research because it models the logical flow of textual queries in hyperbolic space and uses it to perform complex reasoning over large KGs. The models developed in this work are evaluated on both the standard research setting of logical reasoning, as well as, real-world scenarios of query matching and search, specifically, in the e-commerce domain.
In summary, the proposed work aims to extend the accessibility of KGs to non-expert users by enabling them to use non-technical textual queries to search vast amounts of information stored in KGs. To achieve this objective, the work proposes the use of multi-modal representations that capture both semantic and structural features from the KGs, and a novel hyperbolic learning framework to enable scalability of hyperbolic neural networks over large graphs. The work also models the logical flow of textual queries in hyperbolic space to perform complex reasoning over large KGs. The models developed in this work are evaluated on both the standard research setting of logical reasoning and real-world scenarios in the e-commerce domain.
This paper provides a review of models commonly used over the years in the study of microscopic models of material corrosion mechanisms, data mining methods and the corrosion-resistant performance control of structural steels. The virtual process of material corrosion is combined with experimental data to reflect the microscopic mechanism of material corrosion from a nano-scale to macro-scale, respectively. Data mining methods focus on predicting and modeling the corrosion rate and corrosion life of materials. Data-driven control of the corrosion resistance of structural steels is achieved through micro-alloying and organization structure control technology. Corrosion modeling has been used to assess the effects of alloying elements, grain size and organization purity on corrosion resistance, and to determine the contents of alloying elements.
Patriotism, an important component of the Chinese national spirit, has inspired generations of Chinese to strive for national prosperity. Promoting patriotism and implementing patriotic education is an eternal topic. If the youth is robust, the country will be strong. Because college students are the vital force of the country and the hope of the nation, it is especially important to cultivate their patriotism. China is facing new challenges, with profound changes in domestic and foreign situations, rapid technological development and increasingly frequent Internet exchanges. The patriotic education environment has also become more complex under the impact of undesirable Western culture. With the external and internal influences, further patriotic education for college students still faces many challenges. We should face up to the problems in contemporary patriotism education in higher education institutions, explore the solutions and cultivate patriotism among college students in the new era. Therefore, the study of patriotism education in higher education institutions has important theoretical and practical significance. This paper mainly collates literature through literature and historical research method, and the combination of theory and practice, analyzes the problems and causes of patriotism education in higher education institutions with contemporary society, college and family education as well as the characteristics of college students themselves, and puts forward targeted countermeasures for solutions. In the main body, this paper is divided into four parts. First of all, there is an introduction, which mainly includes the background and significance of research, the current status of domestic and international research, research methods, innovations and deficiencies. The framework of the paper was determined to be based on relevant domestic and international studies and the theory was well prepared for the article. The second part mainly elaborates the theories and necessity of patriotism education for college students, mainly including the connotation and characteristics of patriotism education. The third part presents a comprehensive analysis of the problem of patriotic education of college students and its causes from the social environment, patriotic education in higher education institutions, family education and college students themselves. The fourth part is the core of this paper, the practical effect of patriotic education in higher education institutions is ensured, by summarizing the relevant theories and proposing effective technologies and methods against the corresponding problems.
We used deep neural networks to process the mass spectrometry imaging (MSI) data of mouse muscle (young vs aged) and human cancer (tumor vs normal adjacent) tissues, with the aim of using explainable artificial intelligence (XAI) methods to rapidly identify biomarkers that can distinguish different classes of tissues, from several thousands of metabolite features. We also modified classic neural network architectures to construct a deep convolutional neural network that is more suitable for processing high-dimensional MSI data directly, instead of using dimension reduction techniques, and compared it to seven other machine learning analysis methods' performance in classification accuracy. After ascertaining the superiority of Channel-ResNet10, we used a novel channel selection-based XAI method to identify the key metabolite features that were responsible for its learning accuracy. These key metabolite biomarkers were then processed using MetaboAnalyst for pathway enrichment mapping. We found that Channel-ResNet10 was superior to seven other machine learning methods for MSI analysis, reaching > 98% accuracy in muscle aging and colorectal cancer datasets. We also used a novel channel selection-based XAI method to find that in young and aged muscle tissues, the differentially distributed metabolite biomarkers were especially enriched in the propanoate metabolism pathway, suggesting it as a novel target pathway for anti-aging therapy.
ResearchGate has not been able to resolve any references for this publication.