Search

The extensive growth of data in the health domain has increased the utility of Deep Learning in health. Deep learning is a highly advanced successor of artificial neural networks, having powerful computing ability. Due to the availability of fast data storage and hardware parallelism its popularity grows in the last five years. This in article presents a comprehensive literature review of research deploying deep learning medical imaging and medical NLP including tasks, pipelines, and challenges. In this work, we have presented an extensive survey of deep learning architecture deployed in the fields of medical imaging and medical natural language processing. This paper helps in identifying suitable combination of Deep learning, Natural language processing and medical imaging to enhance diagnosis. We have highlighted the major challenges in deploying deep learning in medical imaging and medical natural language processing. All the results are presented in pictorial form. This survey is very helpful for novices working in the area of health informatics.

Simple Summary: The classification is performed later by an interactively learning Swin Transformer block, the core unit for feature representation and long-range semantic information. In particular, the proposed strategy improved significantly and was very resilient while dealing with small liver pieces, discontinuous liver regions, and fuzzy liver boundaries. The experimental results confirm that the proposed APESTNet is more effective in classifying liver tumours than the current state-of-the-art models. Without compromising accuracy, the proposed method conserved resources. However, the proposed method is prone to slight over-segmentation or under-segmentation errors when dealing with lesions or tumours at the liver boundary. Therefore, our future work will concentrate on completely utilizing the z-axis information in 3D to reduce errors. Diagnosis and treatment of hepatocellular carcinoma or metastases rely heavily on accurate segmentation and classification of liver tumours. However, due to the liver tumor's hazy borders and wide range of possible shapes, sizes, and positions, accurate and automatic tumour segmentation and classification remains a difficult challenge. With the advancement of computing, new models in artificial intelligence have evolved. Following its success in Natural language processing (NLP), the transformer paradigm has been adopted by the computer vision (CV) community of the NLP. While there are already accepted approaches to classifying the liver, especially in clinical settings, there is room for advancement in terms of their precision. This paper makes an effort to apply a novel model for segmenting and classifying liver tumours built on deep learning. In order to accomplish this, the created model follows a three-stage procedure consisting of (a) pre-processing, (b) liver segmentation, and (c) classification. In the first phase, the collected Computed Tomography (CT) images undergo three stages of pre-processing, including contrast improvement via histogram equalization and noise reduction via the median filter. Next, an enhanced mask region-based convolutional neural networks (Mask R-CNN) model is used to separate the liver from the CT abdominal image. To prevent overfitting, the segmented picture is fed onto an Enhanced Swin Transformer Network with Adversarial Propagation (APESTNet). The experimental results prove the superior performance of the proposed perfect on a wide variety of CT images, as well as its efficiency and low sensitivity to noise. [ABSTRACT FROM AUTHOR]

Deep Neural Networks (DNN) has attained impressive results in various natural language processing tasks. It attracts the researchers to apply DNN in the Recommender Systems (RS). Typically, majority of the recommendation algorithms apply Collaborative Filtering (CF) to recommend the items of user interest. Recently, so many researchers have applied CF with deep learning for RS. But most of the recommendations exploit only on the implicit data like user clicks, page visit, item description and employs matrix factorization with an inner product to obtain the correlations. To improve the performance of the recommendation system, we propose a novel architecture named Outer Product Based Residual CNN. The proposed model utilizes an explicit user-item sparse rating matrix and outer product function to learns high-order correlations that exist between the users and items latent features. The experimental result shows that the proposed methods outperform the state of art methods. [ABSTRACT FROM AUTHOR]

In the field of natural language processing, machine translation is a colossally developing research area that helps humans communicate more effectively by bridging the linguistic gap. In machine translation, normalization and morphological analyses are the first and perhaps the most important modules for information retrieval (IR). To build a morphological analyzer, or to complete the normalization process, it is important to extract the correct root out of different words. Stemming and lemmatization are techniques commonly used to find the correct root words in a language. However, a few studies on IR systems for the Urdu language have shown that lemmatization is more effective than stemming due to infixes found in Urdu words. This paper presents a lemmatization algorithm based on recurrent neural network models for the Urdu language. However, lemmatization techniques for resource-scarce languages such as Urdu are not very common. The proposed model is trained and tested on two datasets, namely, the Urdu Monolingual Corpus (UMC) and the Universal Dependencies Corpus of Urdu (UDU). The datasets are lemmatized with the help of recurrent neural network models. The Word2Vec model and edit trees are used to generate semantic and syntactic embedding. Bidirectional long short-term memory (BiLSTM), bidirectional gated recurrent unit (BiGRU), bidirectional gated recurrent neural network (BiGRNN), and attention-free encoder–decoder (AFED) models are trained under defined hyperparameters. Experimental results show that the attention-free encoder-decoder model achieves an accuracy, precision, recall, and F-score of 0.96, 0.95, 0.95, and 0.95, respectively, and outperforms existing models. [ABSTRACT FROM AUTHOR]

Modern well-performing approaches to neural decoding are based on machine learning models such as decision tree ensembles and deep neural networks. The wide range of algorithms that can be utilized to learn from neural spike trains, which are essentially time-series data, results in the need for diverse and challenging benchmarks for neural decoding, similar to the ones in the fields of computer vision and natural language processing. In this work, we propose a spike train classification benchmark, based on open-access neural activity datasets and consisting of several learning tasks such as stimulus type classification, animal's behavioral state prediction, and neuron type identification. We demonstrate that an approach based on hand-crafted time-series feature engineering establishes a strong baseline performing on par with state-of-the-art deep learning-based models for neural decoding. We release the code allowing to reproduce the reported results. Author summary: Machine learning-based neural decoding has been shown to outperform traditional approaches like Wiener and Kalman filters on certain key tasks. To further the advancement of neural decoding models, such as improvements in deep neural network architectures and better feature engineering for classical ML models, there need to exist common evaluation benchmarks similar to the ones in the fields of computer vision or natural language processing. In this work, we propose a benchmark consisting of several individual neuron spike train classification tasks based on open-access data from a range of animals and brain regions. We demonstrate that it is possible to achieve meaningful results in such a challenging benchmark using the massive time-series feature extraction approach, which is found to perform similarly to state-of-the-art deep learning approaches. [ABSTRACT FROM AUTHOR]

The paper presents a machine-learning based approach to text-to-ontology mapping. We explore a possibility of matching texts to the relevant ontologies using a combination of artificial neural networks and classifiers. Ontologies are formal specifications of the shared conceptualizations of application domains. While describing the same domain, different ontologies might be created by different domain experts. To enhance the reasoning and data handling of concepts in scientific papers, finding the best fitting ontology regarding description of the concepts contained in a text corpus. The approach presented in this work attempts to solve this by selection of a representative text paragraph from a set of scientific papers, which are used as data set. Then, using a pre-trained and fine-tuned Transformer, the paragraph is embedded into a vector space. Finally, the embedded vector becomes classified with respect to its relevance regarding a selected target ontology. To construct representative embeddings, we experiment with different training pipelines for natural language processing models. Those embeddings in turn are later used in the task of matching text to ontology. Finally, the result is assessed by compressing and visualizing the latent space and exploring the mappings between text fragments from a database and the set of chosen ontologies. To confirm the differences in behavior of the proposed ontology mapper models, we test five statistical hypotheses about their relative performance on ontology classification. To categorize the output from the Transformer, different classifiers are considered. These classifiers are, in detail, the Support Vector Machine (SVM), k-Nearest Neighbor, Gaussian Process, Random Forest, and Multilayer Perceptron. Application of these classifiers in a domain of scientific texts concerning catalysis research and respective ontologies, the suitability of the classifiers is evaluated, where the best result was achieved by the SVM classifier. [ABSTRACT FROM AUTHOR]

This manuscript introduces a new concept of statistical depth function: the compositional D-depth. It is the first data depth developed exclusively for text data, in particular, for those data vectorized according to a frequency-based criterion, such as the tf-idf (term frequency–inverse document frequency) statistic, which results in most vector entries taking a value of zero. The proposed data depth consists of considering the inverse discrete Fourier transform of the vectorized text fragments and then applying a statistical depth for functional data, D. This depth is intended to address the problem of sparsity of numerical features resulting from the transformation of qualitative text data into quantitative data, which is a common procedure in most natural language processing frameworks. Indeed, this sparsity hinders the use of traditional statistical depths and machine learning techniques for classification purposes. In order to demonstrate the potential value of this new proposal, it is applied to a real-world case study which involves mapping Consolidated Framework for Implementation and Research (CFIR) constructs to qualitative healthcare data. It is shown that the D D G -classifier yields competitive results and outperforms all studied traditional machine learning techniques (logistic regression with LASSO regularization, artificial neural networks, decision trees, and support vector machines) when used in combination with the newly defined compositional D-depth. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (DCNNs) have achieved great empirical success in many fields such as natural language processing, computer vision, and pattern recognition. But there still lacks theoretical understanding of the flexibility and adaptivity of DCNNs in various learning tasks, and the power of DCNNs at feature extraction. We propose a generic DCNN structure consisting of two groups of convolutional layers associated with two downsampling operators, and a fully connected layer, which is determined only by three structural parameters. Our generic DCNNs are capable of extracting various features including not only polynomial features but also general smooth features. We also show that the curse of dimensionality can be circumvented by our DCNNs for target functions of the compositional form with (symmetric) polynomial features, spatially sparse smooth features, and interaction features. These demonstrate the expressive power of our DCNN structure, while the model selection can be relaxed comparing with other deep neural networks since there are only three hyperparameters controlling the architecture to tune. [ABSTRACT FROM AUTHOR]

Interpreting deep neural networks is of great importance to understand and verify deep models for natural language processing (NLP) tasks. However, most existing approaches only focus on improving the performance of models but ignore their interpretability. In this work, we propose a Randomly Wired Graph Neural Network (RWGNN) by using graph to model the structure of Neural Network, which could solve two major problems (word-boundary ambiguity and polysemy) of ChineseNER. Besides, we develop a pipeline to explain the RWGNNby using Saliency Map and Adversarial Attacks. Experimental results demonstrate that our approach can identify meaningful and reasonable interpretations for hidden states of RWGNN. [ABSTRACT FROM AUTHOR]

Deep Learning is one of the most popular computer science techniques, with applications in natural language processing, image processing, pattern identification, and various other fields. Despite the success of these deep learning algorithms in multiple scenarios, such as spam detection, malware detection, object detection and tracking, face recognition, and automatic driving, these algorithms and their associated training data are rather vulnerable to numerous security threats. These threats ultimately result in significant performance degradation. Moreover, the supervised based learning models are affected by manipulated data known as adversarial examples, which are images with a particular level of noise that is invisible to humans. Adversarial inputs are introduced to purposefully confuse a neural network, restricting its use in sensitive application areas such as biometrics applications. In this paper, an optimized defending approach is proposed to recognize the adversarial iris examples efficiently. The Curvelet Transform Denoising method is used in this defense strategy, which examines every subband of the adversarial images and reproduces the image that has been changed by the attacker. The salient iris features are retrieved from the reconstructed iris image by using a pre-trained Convolutional Neural Network model (VGG 16) followed by Multiclass classification. The classification is performed by using Support Vector Machine (SVM) which uses Particle Swarm Optimization method (PSO-SVM). The proposed system is tested when classifying the adversarial iris images affected by various adversarial attacks such as FGSM, iGSM, and Deepfool methods. An experimental result on benchmark iris dataset, namely IITD, produces excellent outcomes with the highest accuracy of 95.8% on average. [ABSTRACT FROM AUTHOR]

Customer support is one of the main aspects of the user experience for online services. However, the rise of natural language processing techniques, the industry is looking at automated chatbot solutions to provide quality services to an ever-growing user base. In Sri Lanka, Ceylon Electricity Board website is one of the largest websites that customers use always to get information about electricity services. Hence, a chatbot system is very essential in CEB website. This paper presents a study about implementing and evaluating of a chatbot model for CEB website. This study implements virtual conversation agent based on deep learning algorithm which is multilayer perceptron neural network and a special text dataset for conversations about CEB services. The conversation agent model is made by utilizing the natural language processing techniques to facilitate the processing of user messages. The output of this research is the response from the chatbot and identify the best testing method to get highest accuracy for chatbot model. The chatbot model achieves the highest accuracy with the number of epochs set to 2000 and the learning rate value of 0.01 on response context data training so that it gets 78.8% accuracy. [ABSTRACT FROM AUTHOR]

This paper presents a modified neural model for topic detection from a corpus and proposes a new metric to evaluate the detected topics. The new model builds upon the embedded topic model incorporating some modifications such as document clustering. Numerical experiments suggest that the new model performs favourably regardless of the document's length. The new metric, which can be computed more efficiently than widely-used metrics such as topic coherence, provides variable information regarding the understandability of the detected topics. [ABSTRACT FROM AUTHOR]

One of the major tasks of natural language processing is sentiment analysis. The web is a source of unstructured and rich informa-tion with thousands of opinions and reviews. Individuals, businesses, and governments can all benefit from recognizing sentiment. As part of this study, we propose a deep learning-based approach for sentiment analysis on drug product review data obtained from the UCI machine learning repository. As an alternative to deep learning models, this architecture integrates glove word embedding with convolutional neural networks (CNN). Word2vec and GloVe word embedding schemes have been evaluated empirically for their predictive performance in CNN architectures. Based on a comparison of the deep learning architecture with RoBERTa, itcan be seen that BERT architecture outperforms both of them in training and validation. However, CNN models using Glove word embedding provided superior results in testing. [ABSTRACT FROM AUTHOR]

Recently, hyperbolic deep neural networks (HDNNs) have been gaining momentum as the deep representations in the hyperbolic space provide high fidelity embeddings with few dimensions, especially for data possessing hierarchical structure. Such a hyperbolic neural architecture is quickly extended to different scientific fields, including natural language processing, single-cell RNA-sequence analysis, graph embedding, financial analysis, and computer vision. The promising results demonstrate its superior capability, significant compactness of the model, and a substantially better physical interpretability than its counterpart in the euclidean space. To stimulate future research, this paper presents a comprehensive review of the literature around the neural components in the construction of HDNN, as well as the generalization of the leading deep approaches to the hyperbolic space. It also presents current applications of various tasks, together with insightful observations and identifying open questions and promising future directions. [ABSTRACT FROM AUTHOR]

There is an increasing interest in developing artificial intelligence (AI) systems to process and interpret electronic health records (EHRs). Natural language processing (NLP) powered by pretrained language models is the key technology for medical AI systems utilizing clinical narratives. However, there are few clinical language models, the largest of which trained in the clinical domain is comparatively small at 110 million parameters (compared with billions of parameters in the general domain). It is not clear how large clinical language models with billions of parameters can help medical AI systems utilize unstructured EHRs. In this study, we develop from scratch a large clinical language model—GatorTron—using >90 billion words of text (including >82 billion words of de-identified clinical text) and systematically evaluate it on five clinical NLP tasks including clinical concept extraction, medical relation extraction, semantic textual similarity, natural language inference (NLI), and medical question answering (MQA). We examine how (1) scaling up the number of parameters and (2) scaling up the size of the training data could benefit these NLP tasks. GatorTron models scale up the clinical language model from 110 million to 8.9 billion parameters and improve five clinical NLP tasks (e.g., 9.6% and 9.5% improvement in accuracy for NLI and MQA), which can be applied to medical AI systems to improve healthcare delivery. The GatorTron models are publicly available at: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/models/gatortron_og. [ABSTRACT FROM AUTHOR]

We study a new technique for solving the fundamental challenge in nanophotonic design: fast and accurate characterization of nanoscale photonic devices with minimal human intervention. Much like the fusion between Artificial Intelligence and Electronic Design Automation (EDA), many efforts have been made to apply deep neural networks (DNN) such as convolutional neural networks to prototype and characterize next-gen optoelectronic devices commonly found in Photonic Integrated Circuits. However, state-of-the-art DNN models are still far from being directly applicable in the real world: e.g., DNN-produced correlation coefficients between target and predicted physical quantities are about 80%, which is much lower than what it takes to generate reliable and reproducible nanophotonic designs. Recently, attention-based transformer models have attracted extensive interests and been widely used in Computer Vision and Natural Language Processing. In this work, we for the first time propose a Transformer model (POViT) to efficiently design and simulate photonic crystal nanocavities with multiple objectives under consideration. Unlike the standard Vision Transformer, our model takes photonic crystals as input data and changes the activation layer from GELU to an absolute-value function. Extensive experiments show that POViT significantly improves results reported by previous models: correlation coefficients are increased by over 12 % (i.e., to 92.0 % ) and prediction errors are reduced by an order of magnitude, among several key metric improvements. Our work has the potential to drive the expansion of EDA to fully automated photonic design (i.e., PDA). The complete dataset and code will be released to promote research in the interdisciplinary field of materials science/physics and computer science. [ABSTRACT FROM AUTHOR]

Emotion detection (ED) and sentiment analysis (SA) play a vital role in identifying an individual's level of interest in any given field. Humans use facial expressions, voice pitch, gestures, and words to convey their emotions. Emotion detection and sentiment analysis in English and Chinese have received much attention in the last decade. Still, poor-resource languages such as Urdu have been mostly disregarded, which is the primary focus of this research. Roman Urdu should also be investigated like other languages because social media platforms are frequently used for communication. Roman Urdu faces a significant challenge in the absence of corpus for emotion detection and sentiment analysis because linguistic resources are vital for natural language processing. In this study, we create a corpus of 1021 sentences for emotion detection and 20,251 sentences for sentiment analysis, both obtained from various areas, and annotate it with the aid of human annotators from six and three classes, respectively. In order to train large-scale unlabeled data, the bag-of-word, term frequency-inverse document frequency, and Skip-gram models are employed, and the learned word vector is then fed into the CNN-LSTM model. In addition to our proposed approach, we also use other fundamental algorithms, including a convolutional neural network, long short-term memory, artificial neural networks, and recurrent neural networks for comparison. The result indicates that the CNN-LSTM proposed method paired with Word2Vec is more effective than other approaches regarding emotion detection and evaluating sentiment analysis in Roman Urdu. Furthermore, we compare our based model with some previous work. Both emotion detection and sentiment analysis have seen significant improvements, jumping from an accuracy of 85% to 95% and from 89% to 93.3%, respectively. [ABSTRACT FROM AUTHOR]

Involving students in scientific modeling practice is one of the most effective approaches to achieving the next generation science education learning goals. Given the complexity and multirepresentational features of scientific models, scoring student‐developed models is time‐ and cost‐intensive, remaining one of the most challenging assessment practices for science education. More importantly, teachers who rely on timely feedback to plan and adjust instruction are reluctant to use modeling tasks because they could not provide timely feedback to learners. This study utilized machine learning (ML), the most advanced artificial intelligence (AI), to develop an approach to automatically score student‐drawn models and their written descriptions of those models. We developed six modeling assessment tasks for middle school students that integrate disciplinary core ideas and crosscutting concepts with the modeling practice. For each task, we asked students to draw a model and write a description of that model, which gave students with diverse backgrounds an opportunity to represent their understanding in multiple ways. We then collected student responses to the six tasks and had human experts score a subset of those responses. We used the human‐scored student responses to develop ML algorithmic models (AMs) and to train the computer. Validation using new data suggests that the machine‐assigned scores achieved robust agreements with human consent scores. Qualitative analysis of student‐drawn models further revealed five characteristics that might impact machine scoring accuracy: Alternative expression, confusing label, inconsistent size, inconsistent position, and redundant information. We argue that these five characteristics should be considered when developing machine‐scorable modeling tasks. [ABSTRACT FROM AUTHOR]

Mobile user profiling has drawn significant attentions from various disciplines. To deeply understand the mobile users, based on users' application (app) text data in the smartphone, we propose a semi-supervised learning method to infer mobile user profiles or user demographic attributes. App text has the characteristics of short text length and no word order, which leads to the problems of sparse semantics and lack of context, etc. To address these problems, we build two heterogeneous graphs with different scale features for the corpus, using app name (word) and app installation text list (document) as the nodes of the graph, and using three rules to build edges. Thus, text classification is transformed into multi-graph node classification. Then, we propose a co-learning graph convolutional network (C-GCN) based on selective-scale attention (SS-Attention) to realize the extraction of spatial features of graphs. SS-Attention can enhance the representation learning of global important (word-level) nodes by splitting and fusing operations. Experimental results demonstrated that, without using pre-training embedding, C-GCN outperforms state-of-the-art models across real data. [ABSTRACT FROM AUTHOR]

Deep neural networks have proven to be effective in various domains, especially in natural language processing and image processing. However, one of the challenges associated with using deep neural networks includes the long design time and expertise needed to apply these neural networks to a particular domain. The research presented in this paper investigates the automation of the design of the deep neural network pipeline to overcome this challenge. The deep learning pipeline includes identifying the preprocessing needed, the feature engineering technique, the neural network to use and the parameters for the neural network. A selection pertubative hyper-heuristic (SPHH) is used to automate the design pipeline. The study also examines the reusability of the generated pipeline. The effectiveness of transfer learning on the generated designs is also investigated. The proposed approach is evaluated for text processing—namely, sentiment analysis and spam detection—and image processing—namely, maize disease detection and oral lesion detection. The study revealed that the automated design of the deep neural network pipeline produces just as good, and in some cases better, performance compared to the manual design, with the automated design requiring less design time than the manual design. In the majority of instances, the design was not reusable; however, transfer learning achieved positive transfer of designs, with the performance being just as good or better than when transfer learning was not used. [ABSTRACT FROM AUTHOR]

Motivation for this work comes from the longest-running Serbian television quiz show called TV Slagalica and more specifically from one of its games named associations. In the associations game, two players attempt to guess a solution given several clue words. There is a large number of publicly available game scenarios that were used to evaluate applicability of trained artificial neural networks to predict possible solutions. Material used for the network training was obtained through unconventional sources as no professional text corpus exists for Serbian language. Under outlined schemes, it is observed that solution words come up within 2% or less of the training vocabulary, depending on the method of data preparation. Data preparation and neural network training specifics are further outlined to demonstrate effects of each technique used. Even though the results obtained are below human-level performance, they can nevertheless be useful for puzzle creation. [ABSTRACT FROM AUTHOR]

The clinical application of detecting COVID-19 factors is a challenging task. The existing named entity recognition models are usually trained on a limited set of named entities. Besides clinical, the non-clinical factors, such as social determinant of health (SDoH), are also important to study the infectious disease. In this paper, we propose a generalizable machine learning approach that improves on previous efforts by recognizing a large number of clinical risk factors and SDoH. The novelty of the proposed method lies in the subtle combination of a number of deep neural networks, including the BiLSTM-CNN-CRF method and a transformer-based embedding layer. Experimental results on a cohort of COVID-19 data prepared from PubMed articles show the superiority of the proposed approach. When compared to other methods, the proposed approach achieves a performance gain of about 1–5% in terms of macro- and micro-average F1 scores. Clinical practitioners and researchers can use this approach to obtain accurate information regarding clinical risks and SDoH factors, and use this pipeline as a tool to end the pandemic or to prepare for future pandemics. [ABSTRACT FROM AUTHOR]

Mapping the pharmaceutically significant entities on natural language to standardized terms/concepts is a key task in the development of the systems for pharmacovigilance, marketing, and using drugs out of the application scope. This work estimates the accuracy of mapping adverse reaction mentions to the concepts from the Medical Dictionary of Regulatory Activity (MedDRA) in the case of adverse reactions extracted from the reviews on the use of pharmaceutical products by Russian-speaking Internet users (normalization task). The solution we propose is based on a neural network approach using two neural network models: the first one for encoding concepts, and the second one for encoding mentions. Both models are pre-trained language models, but the second one is additionally tuned for the normalization task using both the Russian Drug Reviews (RDRS) corpus and a set of open English-language corpora automatically translated into Russian. Additional tuning of the model during the proposed procedure increases the accuracy of mentions of adverse drug reactions by 3% on the RDRS corpus. The resulting accuracy for the adverse reaction mentions mapping to the preferred terms of MedDRA in RDRS is 70.9% F 1 -micro. The paper analyzes the factors that affect the accuracy of solving the task based on a comparison of the RDRS and the CSIRO Adverse Drug Event Corpus (CADEC) corpora. It is shown that the composition of the concepts of the MedDRA and the number of examples for each concept play a key role in the task solution. The proposed model shows a comparable accuracy of 87.5% F 1 -micro on a subsample of RDRS and CADEC datasets with the same set of MedDRA preferred terms. [ABSTRACT FROM AUTHOR]

The rapid development of natural language processing and deep learning techniques has boosted the performance of related algorithms in several linguistic and text mining tasks. Consequently, applications such as opinion mining, fake news detection or document classification that assign documents to predefined categories have significantly benefited from pre-trained language models, word or sentence embeddings, linguistic corpora, knowledge graphs and other resources that are in abundance for the more popular languages (e.g., English, Chinese, etc.). Less represented languages, such as the Kazakh language, balkan languages, etc., still lack the necessary linguistic resources and thus the performance of the respective methods is still low. In this work, we develop a model that classifies scientific papers written in the Kazakh language using both text and image information and demonstrate that this fusion of information can be beneficial for cases of languages that have limited resources for machine learning models' training. With this fusion, we improve the classification accuracy by 4.4499% compared to the models that use only text or only image information. The successful use of the proposed method in scientific documents' classification paves the way for more complex classification models and more application in other domains such as news classification, sentiment analysis, etc., in the Kazakh language. [ABSTRACT FROM AUTHOR]

Deep neural networks have proven to be effective in solving computer vision and natural language processing problems. To fully leverage its power, manually designed network templates, i.e., Residual Networks, are introduced to deal with various vision and natural language tasks. These hand-crafted neural networks rely on a large number of parameters, which are both data-dependent and laborious. On the other hand, architectures suitable for specific tasks have also grown exponentially with their size and topology, which prohibits brute force search. To address these challenges, this paper proposes a quantum dynamic optimization algorithm to find the optimal structure for a candidate network using Quantum Dynamic Neural Architecture Search (QDNAS). Specifically, the proposed quantum dynamics optimization algorithm is used to search for meaningful architectures for vision tasks and dedicated rules to express and explore the search space. The proposed quantum dynamics optimization algorithm treats the iterative evolution process of the optimization over time as a quantum dynamic process. The tunneling effect and potential barrier estimation in quantum mechanics can effectively promote the evolution of the optimization algorithm to the global optimum. Extensive experiments on four benchmarks demonstrate the effectiveness of QDNAS, which is consistently better than all baseline methods in image classification tasks. Furthermore, an in-depth analysis is conducted on the searchable networks that provide inspiration for the design of other image classification networks. [ABSTRACT FROM AUTHOR]

Recent research in novelty detection focuses mainly on document-level classification, employing deep neural networks (DNN). However, the black-box nature of DNNs makes it difficult to extract an exact explanation of why a document is considered novel. In addition, dealing with novelty at the word level is crucial to provide a more fine-grained analysis than what is available at the document level. In this work, we propose a Tsetlin Machine (TM)-based architecture for scoring individual words according to their contribution to novelty. Our approach encodes a description of the novel documents using the linguistic patterns captured by TM clauses. We then adapt this description to measure how much a word contributes to making documents novel. Our experimental results demonstrate how our approach breaks down novelty into interpretable phrases, successfully measuring novelty. [ABSTRACT FROM AUTHOR]

Limited availability of medical imaging datasets is a vital limitation when using "data hungry" deep learning to gain performance improvements. Dealing with the issue, transfer learning has become a de facto standard, where a pre-trained convolution neural network (CNN), typically on natural images (e.g., ImageNet), is finetuned on medical images. Meanwhile, pre-trained transformers, which are self-attention-based models, have become de facto standard in natural language processing (NLP) and state of the art in image classification due to their powerful transfer learning abilities. Inspired by the success of transformers in NLP and image classification, large-scale transformers (such as vision transformer) are trained on natural images. Based on these recent developments, this research aims to explore the efficacy of pre-trained natural image transformers for medical images. Specifically, we analyze pre-trained vision transformer on CheXpert and pediatric pneumonia dataset. We use CNN standard models including VGGNet and ResNet as baseline models. By examining the acquired representations and results, we discover that transfer learning from the pre-trained vision transformer shows improved results as compared to pre-trained CNN which demonstrates a greater transfer ability of the transformers in medical imaging. [ABSTRACT FROM AUTHOR]

To improve the R&D process by reducing duplicated bug tickets, we used the idea of composing a BERT encoder as a Siamese network to create a system for finding similar existing tickets. We proposed several different methods of generating artificial ticket pairs to augment the training set. Two phases of training were conducted. The first showed that only approximately 9% of pairs were correctly identified as certainly similar. Only 48% of the test samples were found to be pairs of similar tickets. With fine-tuning, we improved that result to 81%, which is a number describing a set of common decisions between the engineer in the company and the solution presented. With this tool, engineers in the company receive a specialized instrument with the ability to evaluate tickets related to a security bug at a level close to an experienced company employee. Therefore, we propose a new engineering application in corporate practice in a very important area with Siamese network methods that are widely known and recognized for their efficiency. [ABSTRACT FROM AUTHOR]

This paper introduces a diagnostic model that effectively diagnoses in fourteen different stages, by fusing functional magnetic resonance imaging (fMRI) and structural MRI (sMRI) information. Georgios Theodoridis and Athanasios Tsadiras from the Aristotle University of Thessaloniki, Greece, have authored the seventh paper which is entitled "Applying machine learning techniques to predict and explain subscriber churn of an online drug information platform." Machine learning (ML) and more specifically deep learning (DL) algorithms are considered among the most paramount technologies of both artificial intelligence (AI) and 4 SP th sp industrial revolution. This paper provides an in-depth comparison of various machine learning (ML) techniques and advanced preprocessing methods, in an effort to successfully perform online subscriber churn prediction. [Extracted from the article]

Intrinsically disordered regions (IDRs) in proteins are the regions that lack a stable two-dimensional or three-dimensional structure. Due to their high degree of flexibility, these regions are required for a variety of cellular functions. The IDRs are determined experimentally using X-ray crystallography and NMR. Numerous computational techniques for IDR prediction have been developed, but due to the certain unknown amino acid properties and interactions, these techniques have a low predictive rate. IDR-amino acid enrichment analyses on protein chains with varying structural and physicochemical properties have shed light on a variety of features of IDRs. Additionally, repetitions of certain specific amino acids that appear to be preferentially present in the IDRs were observed. Following that, a deep neural network model inspired by natural language processing techniques was trained to learn amino acid features for the classification of IDRs and non-IDRs. We portrayed amino acids as single letters that comprise the proteome's language. Our method outperformed other IDR prediction tools currently available. This study can assist researchers in comprehending IDR constituents and utilizing NLP techniques for processing genomic or proteomic language in order to gain additional insight from them. The executable package and the dataset can be found at https://doi.org/10.24433/CO.3457808.v1. [ABSTRACT FROM AUTHOR]

Lifelong experiences and learned knowledge lead to shared expectations about how common situations tend to unfold. Such knowledge of narrative event flow enables people to weave together a story. However, comparable computational tools to evaluate the flow of events in narratives are limited. We quantify the differences between autobiographical and imagined stories by introducing sequentiality, a measure of narrative flow of events, drawing probabilistic inferences from a cutting-edge large language model (GPT-3). Sequentiality captures the flow of a narrative by comparing the probability of a sentence with and without its preceding story context. We applied our measure to study thousands of diary-like stories, collected from crowdworkers, about either a recent remembered experience or an imagined story on the same topic. The results show that imagined stories have higher sequentiality than autobiographical stories and that the sequentiality of autobiographical stories increases when the memories are retold several months later. In pursuit of deeper understandings of how sequentiality measures the flow of narratives, we explore proportions of major and minor events in story sentences, as annotated by crowdworkers. We find that lower sequentiality is associated with higher proportions of major events. The methods and results highlight opportunities to use cutting-edge computational analyses, such as sequentiality, on large corpora of matched imagined and autobiographical stories to investigate the influences of memory and reasoning on language generation processes. [ABSTRACT FROM AUTHOR]

As the world's population is aging and there is a shortage of sufficient caring manpower, the development of intelligent care robots is a feasible solution. At present, plenty of care robots have been developed, but humanized care robots that can suitably respond to the individual behaviors of elderly people, such as pose, expression, gaze, and speech are generally lacking. To achieve the interaction, the main objectives of this study are: (1) conducting a literature review and analyzing the status quo on the following four core tasks of image and speech recognition technology: human pose recognition, human facial expression recognition, eye gazing recognition, and Chinese speech recognition; (2) proposing improvement strategies for these tasks based on the results of the literature review. The results of the study on these improvement strategies will provide the basis for using human facial expression robots in elderly care. [ABSTRACT FROM AUTHOR]

Deep learning based on neural networks has been widely used in image recognition, speech recognition, natural language processing, automatic driving, and other fields and has made breakthrough progress. FPGA stands out in the field of accelerated deep learning with its advantages such as flexible architecture and logic units, high energy efficiency ratio, strong compatibility, and low delay. In order to track the latest research results of neural network optimization technology based on FPGA in time and to keep abreast of current research hotspots and application fields, the related technologies and research contents are reviewed. This paper introduces the development history and application fields of some representative neural networks and points out the importance of studying deep learning technology, as well as the reasons and advantages of using FPGA to accelerate deep learning. Several common neural network models are introduced. Moreover, this paper reviews the current mainstream FPGA-based neural network acceleration technology, method, accelerator, and acceleration framework design and the latest research status, pointing out the current FPGA-based neural network application facing difficulties and the corresponding solutions, as well as prospecting the future research directions. We hope that this work can provide insightful research ideas for the researchers engaged in the field of neural network acceleration based on FPGA. [ABSTRACT FROM AUTHOR]

Despite that pre-trained word embedding models have advanced a wide range of natural language processing applications, they ignore the contextual information and meaning within the text. In this paper, we investigate the potential of the pre-trained Arabic BERT (Bidirectional Encoder Representations from Transformers) model to learn universal contextualized sentence representations aiming to showcase its usefulness for Arabic text Multi-class categorization. We propose to exploit the pre-trained AraBERT for contextual text representation learning in two different ways, transfer learning model and feature extractor. On the one hand, we employ the Arabic BERT (AraBERT) model after fine-tuning its parameters on the OSAC datasets to transfer its knowledge for the Arabic text categorization. On the other hand, we inquire into AraBERT performance, as a feature extractor model, by combining it with several classifiers, including CNN, LSTM, Bi-LSTM, MLP, and SVM. Finally, we conduct an exhaustive set of experiments comparing two BERT models, namely AraBERT and multilingual BERT. The findings show that the fine-tuned AraBERT model accomplishes state-of-the-art performance results and attains up to 99% in terms of F1-score and accuracy. [ABSTRACT FROM AUTHOR]

Visual Question Answering (VQA) models fail catastrophically on questions related to the reading of text-carrying images. However, TextVQA aims to answer questions by understanding the scene texts in an image–question context, such as the brand name of a product or the time on a clock from an image. Most TextVQA approaches focus on objects and scene text detection, which are then integrated with the words in a question by a simple transformer encoder. The focus of these approaches is to use shared weights during the training of a multi-modal dataset, but it fails to capture the semantic relations between an image and a question. In this paper, we proposed a Scene Graph-Based Co-Attention Network (SceneGATE) for TextVQA, which reveals the semantic relations among the objects, the Optical Character Recognition (OCR) tokens and the question words. It is achieved by a TextVQA-based scene graph that discovers the underlying semantics of an image. We create a guided-attention module to capture the intra-modal interplay between the language and the vision as a guidance for inter-modal interactions. To permit explicit teaching of the relations between the two modalities, we propose and integrate two attention modules, namely a scene graph-based semantic relation-aware attention and a positional relation-aware attention. We conduct extensive experiments on two widely used benchmark datasets, Text-VQA and ST-VQA. It is shown that our SceneGATE method outperforms existing ones because of the scene graph and its attention modules.

Grammatical error correction (GEC) is one of the major tasks in natural language processing (NLP) which has recently attracted great attention from researchers. The performance of universal languages such as English and Chinese in the GEC system has improved significantly. This could be attributed to the large number of powerful applications supported by neural network models and pretrained language models. Referring to the satisfactory results of the universal language in the GEC task and the lack of research on the GEC task for low-resource languages, especially Indonesian, this paper proposes an automatic model for Indonesian grammar correction based on the Transformer architecture which can be applied to other low-resource language texts. Furthermore, we build a large corpus of the Indonesian language that can be utilized for evaluating the next Indonesian GEC task. We evaluate the models in this dataset, and the results show that the Transformer-based automatic error correction model achieved significant and satisfactory results compared with the results of previous research models. [ABSTRACT FROM AUTHOR]

Emotions play an important role in human cognition and are commonly associated with perception, logical decision making, human interaction, and intelligence. Emotion and stress detection is an emerging topic of interest and importance in the research community. With the availability of portable, cheap, and reliable sensor devices, researchers are opting to use physiological signals for emotion classification as they are more prone to human deception, as compared to audiovisual signals. In recent years, deep neural networks have gained popularity and have inspired new ideas for emotion recognition based on electroencephalogram (EEG) signals. Recently, widespread use of transformer-based architectures has been observed, providing state-of-the-art results in several domains, from natural language processing to computer vision, and object detection. In this work, we investigate the effectiveness and accuracy of a novel transformer-based architecture, called perceiver, which claims to be able to handle inputs from any modality, be it an image, audio, or video. We utilize the perceiver architecture on raw EEG signals taken from one of the most widely used publicly available EEG-based emotion recognition datasets, i.e., DEAP, and compare its results with some of the best performing models in the domain. [ABSTRACT FROM AUTHOR]

Background: Cardiovascular disease (CVD) is a serious disease that endangers human health and is one of the main causes of death. Therefore, using the patient's electronic medical record (EMR) to predict CVD automatically has important application value in intelligent assisted diagnosis and treatment, and is a hot issue in intelligent medical research. However, existing methods based on natural language processing can only predict CVD according to the whole or part of the context information of EMR. Results: Given the deficiencies of the existing research on CVD prediction based on EMRs, this paper proposes a risk factor attention-based model (RFAB) to predict CVD by utilizing CVD risk factors and general EMRs text, which adopts the attention mechanism of a deep neural network to fuse the character sequence and CVD risk factors contained in EMRs text. The experimental results show that the proposed method can significantly improve the prediction performance of CVD, and the F-score reaches 0.9586, which outperforms the existing related methods. Conclusions: RFAB focuses on the key information in EMR that leads to CVD, that is, 12 risk factors. In the stage of risk factor identification and extraction, risk factors are labeled with category information and time attribute information by BiLSTM-CRF model. In the stage of CVD prediction, the information contained in risk factors and their labels is fused with the information of character sequence in EMR to predict CVD. RFAB makes well use of the fine-grained information contained in EMR, and also provides a reliable idea for predicting CVD. [ABSTRACT FROM AUTHOR]

A deep neural network has multiple layers to learn more complex patterns and is built to simulate the activity of the human brain. Currently, it provides the best solutions to many problems in image recognition, speech recognition, and natural language processing. The present study deals with the topological properties of deep neural networks. The topological index is a numeric quantity associated to the connectivity of the network and is correlated to the efficiency and accuracy of the output of the network. Different degree-related topological indices such as Zagreb index, Randic index, atom-bond connectivity index, geometric-arithmetic index, forgotten index, multiple Zagreb indices, and hyper-Zagreb index of deep neural network with a finite number of hidden layers are computed in this study. [ABSTRACT FROM AUTHOR]

The purposes of this study were to determine if 1) recurrent neural networks designed for multivariate, time-series analyses outperform traditional linear and non-linear machine learning classifiers when classifying athletes based on competition level and sport played, and 2) athletes of different sports move differently during non-sport-specific movement screens. Optical-based kinematic data from 542 athletes were used as input data for nine different machine learning algorithms to classify athletes based on competition level and sport played. For the traditional machine learning classifiers, principal component analysis and feature selection were used to reduce the data dimensionality and to determine the best principal components to retain. Across tasks, recurrent neural networks and linear machine learning classifiers tended to outperform the non-linear machine learning classifiers. For all tasks, reservoir computing took the least amount of time to train. Across tasks, reservoir computing had one of the highest classification rates and took the least amount of time to train; however, interpreting the results is more difficult compared to linear classifiers. In addition, athletes were successfully classified based on sport suggesting that athletes competing in different sports move differently during non-sport specific movements. Therefore, movement assessment screens should incorporate sport-specific scoring criteria. [ABSTRACT FROM AUTHOR]

Personality is generally expressed as the combination of behavior, emotion, motivation, and thoughts that aim to describe different aspects of human behavior based on a few stable and measurable characteristics. Considering the fact that our personality has a remarkable impact in our daily life, automatic recognition of a person's personality attributes can provide many essential practical applications in various aspects of cognitive science. Although various methods have been recently proposed for the task of personality recognition, most of them have mainly focused on human-designed statistical features and they did not make use of rich semantic information existing in users' generated texts while not only these contents can specify its writer's internal thought and emotion but also can be assumed as the most direct way for people to state their feeling and opinion in an understandable form. In order to make use of contextualized information as well as overcoming the complexity and handcraft feature requirement of previous methods, a deep learning based method for the task of contextualized personality recognition is proposed in this paper. Among various deep neural networks, Convolutional Neural Networks (CNN) have demonstrated profound efficiency in natural language processing and especially personality detection. Owing to the fact that various filter sizes in CNN may influence its performance, we decided to combine CNN with AdaBoost, a classical ensemble algorithm, to consider the possibility of using the contribution of various filter lengths and gasp their potential in the final classification via combining various classifiers with respective filter size using AdaBoost. Our proposed method was validated on the Essay and YouTube datasets by conducting a series of experiments and the empirical results demonstrated the superiority of our proposed method on both datasets compared to both machine learning and deep learning methods for the task of personality recognition. [ABSTRACT FROM AUTHOR]

The probability distributions that statistical methods use to represent uncertainty fail to capture all of the uncertainty that may be relevant to decision making. A simple way to adjust probability distributions for the uncertainty not represented in their models is to average the distributions with a uniform distribution or another distribution of maximum uncertainty. A decision-theoretic framework leads to averaging the distributions by taking the means of the logit transforms of the probabilities. That method does not prevent convergence to the truth, as does taking the means of the probabilities themselves. The mean-logit approach to moderating distributions is applied to natural language processing performed by a deep neural network. [ABSTRACT FROM AUTHOR]

We make daily comments on online platforms (e.g., social networks), and such natural language texts often contain sentiment (e.g., positive and negative) for certain aspects (e.g., food and service). If we can automatically extract the aspect-based sentiment from the texts, then it will help many services or products to overcome their limitations of particular aspects. There have been studies of aspect sentiment classification (ASC) that finds sentiment towards particular aspects. Recent studies mostly adopt deep-learning models or graph neural networks as these techniques are capable of capturing linguistic patterns that contributed to performance improvement in various natural language processing tasks. In this paper, for the ASC task, we propose a new hybrid architecture of graph convolutional network (GCN) and recurrent neural network. We design a gate mechanism that jointly models word embeddings and syntactic representation of sentences. By experimental results on five datasets, we show that the proposed model outperforms other recent models and also verify that the gate mechanism contributes to the performance improvement. The overall F1 scores that we achieved is 66.64∼76.80%. [ABSTRACT FROM AUTHOR]

Speech recognition has now become ubiquitous and plays an inevitable role in almost all sectors. Numerous works have been proposed on speech recognition; however, more accurate transcriptions are not possible. Exploration of various studies related to spell correction implies that several kinds of research have been carried out in this field but still it is a very challenging problem. This led to the need for a new spell corrector framework capable of leveraging the performance of the automatic speech recognition (ASR) system. The proposed work unveils state-of-the-art Bidirectional Encoder Representations from Transformers (BERT) based spell correction module developed on top of the deep recurrent neural network (RNN) based ASR system. The impact of BERT-based spell correction on the ASR system is evaluated on three different accent datasets in the perspective of word error rate (WER), character error rate (CER), and Bilingual evaluation understudy (BLEU) score. The experimental results inferred that the enhanced spell correction module is efficacious in detecting and correcting spell errors, by achieving the WER of 5.025% on librispeech corpus, 6.35% on voxforge, and 7.05% on NPTEL corpus. [ABSTRACT FROM AUTHOR]