Increase In F1 Score Research Articles

Evaluation of methods to detect circular RNAs from single-end RNA-sequencing data

BackgroundCircular RNA (circRNA), a class of RNA molecule with a loop structure, has recently attracted researchers due to its diverse biological functions and potential biomarkers of human diseases. Most of the current circRNA detection methods from RNA-sequencing (RNA-Seq) data utilize the mapping information of paired-end (PE) reads to eliminate false positives. However, much of the practical RNA-Seq data such as cross-linking immunoprecipitation sequencing (CLIP-Seq) data usually contain single-end (SE) reads. It is not clear how well these tools perform on SE RNA-Seq data.ResultsIn this study, we present a systematic evaluation of six advanced RNA-based methods and two CLIP-Seq based methods for detecting circRNAs from SE RNA-Seq data. The performances of the methods are rigorously assessed based on precision, sensitivity, F1 score, and true discovery rate. We investigate the impacts of read length, false positive ratio, sequencing depth and PE mapping information on the performances of the methods using simulated SE RNA-Seq simulated datasets. The real datasets used in this study consist of four experimental RNA-Seq datasets with ≥100bp read length and 124 CLIP-Seq samples from 45 studies that contain mostly short-read (≤50bp) RNA-Seq data. The simulation study shows that the sensitivities of most of the methods can be improved by increasing either read length or sequencing depth, and that the levels of false positive rates significantly affect the precision of all methods. Furthermore, the PE mapping information can improve the method’s precision but can not always guarantee the increase of F1 score. Overall, no method is dominant for all SE RNA-Seq data. The RNA-based methods perform better for the long-read datasets but are worse for the short-read datasets. In contrast, the CLIP-Seq based methods outperform the RNA-Seq based methods for all the short-read samples. Combining the results of these methods can significantly improve precision in the CLIP-Seq data.ConclusionsThe results provide a systematic evaluation of circRNA detection methods on SE RNA-Seq data that would facilitate researchers’ strategies in circRNA analysis.

Incorporating uncertainty in learning to defer algorithms for safe computer-aided diagnosis

Deep neural networks are increasingly being used for computer-aided diagnosis, but erroneous diagnoses can be extremely costly for patients. We propose a learning to defer with uncertainty (LDU) algorithm which identifies patients for whom diagnostic uncertainty is high and defers them for evaluation by human experts. LDU was evaluated on the diagnosis of myocardial infarction (using discharge summaries), the diagnosis of any comorbidities (using structured data), and the diagnosis of pleural effusion and pneumothorax (using chest x-rays), and compared with ‘learning to defer without uncertainty information’ (LD) and ‘direct triage by uncertainty’ (DT) methods. LDU achieved the same F1 score as LD but deferred considerably fewer patients (e.g. 36% vs. 69% deferral rate for diagnosing pleural effusion with an F1 score of 0.96). Furthermore, even when many patients were assigned the wrong diagnosis with high confidence (e.g. for the diagnosis of any comorbidities) LDU achieved a 17% increase in F1 score, whereas DT was not applicable. Importantly, the weight of the defer loss in LDU can be easily adjusted to obtain the desired trade-off between diagnostic accuracy and deferral rate. In conclusion, LDU can readily augment any existing diagnostic network to reduce the risk of erroneous diagnoses in clinical practice.

Tree-KGQA: An Unsupervised Approach for Question Answering Over Knowledge Graphs

Most Knowledge Graph-based Question Answering (KGQA) systems rely on training data to reach their optimal performance. However, acquiring training data for supervised systems is both time-consuming and resource-intensive. To address this, in this paper, we propose <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Tree-KGQA</b> , an unsupervised KGQA system leveraging pre-trained language models and tree-based algorithms. Entity and relation linking are essential components of any KGQA system. We employ several pre-trained language models in the entity linking task to recognize the entities mentioned in the question and obtain the contextual representation for indexing. Furthermore, for relation linking we incorporate a pre-trained language model previously trained for language inference task. Finally, we introduce a novel algorithm for extracting the answer entities from a KG, where we construct a forest of interpretations and introduce tree-walking and tree disambiguation techniques. Our algorithm uses the linked relation and predicts the tree branches that eventually lead to the potential answer entities. The proposed method achieves 4.5% and 7.1% gains in F1 score in entity linking tasks on LC-QuAD 2.0 and LC-QuAD 2.0 (KBpearl) datasets, respectively, and a 5.4% increase in the relation linking task on LC-QuAD 2.0 (KBpearl). The comprehensive evaluations demonstrate that our unsupervised KGQA approach outperforms other supervised state-of-the-art methods on the WebQSP-WD test set (1.4% increase in F1 score) - without training on the target dataset.

Improve Representation for Cross-Language Clone Detection by Pretrain Using Tree Autoencoder

With the rise of deep learning in recent years, many code clone detection (CCD) methods use deep learning techniques and achieve promising results, so is cross-language CCD. However, deep learning techniques require a dataset to train the models. The dataset is typically small and has a gap between real-world clones due to the difficulty of collecting datasets for cross-language CCD. This creates a data bottleneck problem: data scale and quality issues will cause that model with a better design can still not reach its full potential. To mitigate this, we propose a tree autoencoder (TAE) architecture. It uses unsupervised learning to pretrain with abstract syntax trees (ASTs) of a large-scale dataset, then fine-tunes the trained encoder in the downstream CCD task. Our proposed TAE contains a tree Long Short-Term Memory (LSTM) encoder and a tree LSTM decoder. We design a novel embedding method for AST nodes, including type embedding and value embedding. In the training of TAE, we present an “encode and decode by layers” strategy and a node-level batch size design. For the CCD dataset, we propose a negative sampling method based on probability distribution. The experimental results on two datasets verify the effeteness of our embedding method, as well as that TAE and its pretrain enhance the performance of the CCD model. The node context information is well captured, and the reconstruction accuracy of the node-value reaches 95.45%. TAE pretrain improves the performance of CCD with a 4% increase in F1 score, which alleviates the data bottleneck problem.

GCS-YOLOV4-Tiny: A lightweight group convolution network for multi-stage fruit detection.

Fruits require different planting techniques at different growth stages. Traditionally, the maturity stage of fruit is judged visually, which is time-consuming and labor-intensive. Fruits differ in size and color, and sometimes leaves or branches occult some of fruits, limiting automatic detection of growth stages in a real environment. Based on YOLOV4-Tiny, this study proposes a GCS-YOLOV4-Tiny model by (1) adding squeeze and excitation (SE) and the spatial pyramid pooling (SPP) modules to improve the accuracy of the model and (2) using the group convolution to reduce the size of the model and finally achieve faster detection speed. The proposed GCS-YOLOV4-Tiny model was executed on three public fruit datasets. Results have shown that GCS-YOLOV4-Tiny has favorable performance on mAP, Recall, F1-Score and Average IoU on Mango YOLO and Rpi-Tomato datasets. In addition, with the smallest model size of 20.70 MB, the mAP, Recall, F1-score, Precision and Average IoU of GCS-YOLOV4-Tiny achieve 93.42 ± 0.44, 91.00 ± 1.87, 90.80 ± 2.59, 90.80 ± 2.77 and 76.94 ± 1.35%, respectively, on F. margarita dataset. The detection results outperform the state-of-the-art YOLOV4-Tiny model with a 17.45% increase in mAP and a 13.80% increase in F1-score. The proposed model provides an effective and efficient performance to detect different growth stages of fruits and can be extended for different fruits and crops for object or disease detections.

IEPet: A Lightweight Multiscale Infrared Environmental Perception Network

In recent years, the development of unmanned driving technology requires continuous progress in environment perception technology. Aiming at the key research direction of infrared environment perception in unmanned driving technology, a lightweight real-time detection network model for infrared environment perception, IEPet, is proposed. The model backbone adds the BottleneckCSP module and the proposed DCAP attention module, which can significantly improve the detection ability and spatial position perception ability while maintaining light weight. At the same time, the model improves the detection accuracy by using a 3-scales detection head. Comparative experiments on the unmanned driving data set show that compared with the lightweight model YOLOv4-tiny, the model proposed in this paper has an increase in F1 Score by 1.48% and an average detection accuracy by 6.37% to reach 84.31%. And the model is lighter. It shows that the proposed IEPet model can better meet the excellent performance required for infrared environment perception.

Mapping coastal wetlands in the Yellow River Delta, China during 2008–2019: impacts of valid observations, harmonic regression, and critical months

ABSTRACT Spartina alterniflora has become one of the top exotic invasive plants in coastal China. Accurate and timely mapping of coastal wetlands is critical to understand the spatiotemporal dynamics of S. alterniflora invasion and native species degradation. Due to the humid weather and frequent cloud cover in coastal wetlands, pixel-based classifications based on time-series satellite observations have been increasingly utilized to mitigate this problem. However, the spatial and temporal variations in valid observations might affect pixel-based classification accuracies, while this issue has been paid little attention in previous research. In this paper, we conducted annual mapping of the Yellow River Delta (YRD) wetland from 2008 to 2019 based on time-series Landsat 5/7/8 images using harmonic regression analysis and random forest classification on Google Earth Engine. We further analysed how valid observations and temporal distribution of observations affected the classification accuracy. Our results showed that the overall classification accuracies ranged from 87.25% in 2012 to 92.36% in 2018. The mapping results showed that S. alterniflora expanded from 39.91 ha to 4672.38 ha from 2008 to 2019, encroaching seagrass beds with 902.32 ha and encroaching S.salsa and bare flat with 3730.15 ha. This is the first time that maps of coastal wetlands of YRD are produced at an annual time step. Our analysis showed that the number of valid observations affected classification accuracies both temporally and spatially. Years with higher density of observations witnessed higher overall accuracies compared to the years with fewer observations. Pixels with more observations had a higher chance to be correctly classified compared to the pixels with fewer observations. Harmonic regression features helped improve classification accuracies (increase in overall accuracies from 0.15% to 3.32%), especially for vegetation types (increase in F-score from 0.19% to 3.86%). Greater number of valid observations enhances the importance of harmonic regression features in classification. The critical months identified were March, July, and October, and the combinations of these months achieved better accuracies (91.24%) than using all other observations (89.68%) in 2018, suggesting that temporal distributions of observations could be more important than the number of observations for harmonic-based classification of coastal wetlands such as YRD.

A Deep Learning Enabled Multi-Class Plant Disease Detection Model Based on Computer Vision

In this paper, a deep learning enabled object detection model for multi-class plant disease has been proposed based on a state-of-the-art computer vision algorithm. While most existing models are limited to disease detection on a large scale, the current model addresses the accurate detection of fine-grained, multi-scale early disease detection. The proposed model has been improved to optimize for both detection speed and accuracy and applied to multi-class apple plant disease detection in the real environment. The mean average precision (mAP) and F1-score of the detection model reached up to 91.2% and 95.9%, respectively, at a detection rate of 56.9 FPS. The overall detection result demonstrates that the current algorithm significantly outperforms the state-of-the-art detection model with a 9.05% increase in precision and 7.6% increase in F1-score. The proposed model can be employed as an effective and efficient method to detect different apple plant diseases under complex orchard scenarios.

Multimodal Classification of Parkinson's Disease in Home Environments with Resiliency to Missing Modalities.

Parkinson’s disease (PD) is a chronic neurodegenerative condition that affects a patient’s everyday life. Authors have proposed that a machine learning and sensor-based approach that continuously monitors patients in naturalistic settings can provide constant evaluation of PD and objectively analyse its progression. In this paper, we make progress toward such PD evaluation by presenting a multimodal deep learning approach for discriminating between people with PD and without PD. Specifically, our proposed architecture, named MCPD-Net, uses two data modalities, acquired from vision and accelerometer sensors in a home environment to train variational autoencoder (VAE) models. These are modality-specific VAEs that predict effective representations of human movements to be fused and given to a classification module. During our end-to-end training, we minimise the difference between the latent spaces corresponding to the two data modalities. This makes our method capable of dealing with missing modalities during inference. We show that our proposed multimodal method outperforms unimodal and other multimodal approaches by an average increase in -score of 0.25 and 0.09, respectively, on a data set with real patients. We also show that our method still outperforms other approaches by an average increase in -score of 0.17 when a modality is missing during inference, demonstrating the benefit of training on multiple modalities.

Systematic Evaluation of Design Choices for Deep Facial Action Coding Across Pose

The performance of automated facial expression coding is improving steadily. Advances in deep learning techniques have been key to this success. While the advantage of modern deep learning techniques is clear, the contribution of critical design choices remains largely unknown, especially for facial action unit occurrence and intensity across pose. Using the The Facial Expression Recognition and Analysis 2017 (FERA 2017) database, which provides a common protocol to evaluate robustness to pose variation, we systematically evaluated design choices in pre-training, feature alignment, model size selection, and optimizer details. Informed by the findings, we developed an architecture that exceeds state-of-the-art on FERA 2017. The architecture achieved a 3.5% increase in F1 score for occurrence detection and a 5.8% increase in Intraclass Correlation (ICC) for intensity estimation. To evaluate the generalizability of the architecture to unseen poses and new dataset domains, we performed experiments across pose in FERA 2017 and across domains in Denver Intensity of Spontaneous Facial Action (DISFA) and the UNBC Pain Archive.

Feature-based evidential reasoning for probabilistic risk analysis and prediction

Risk analysis plays an important role in quality control in engineering projects for the consideration of time, cost, safety, and the environment. This study proposes a feature-based evidential reasoning approach for probabilistic risk analysis and prediction, incorporating the learning process of belief degrees and estimation of the judgment quality. Firstly, classifiers are trained to estimate the probabilistic risk from sub-groups of factors. Secondly, the judgment from each classifier is evaluated according to the classifier’s performance which is characterized by the importance weight and reliability. Finally, the judgments from classifiers are fused via evidential reasoning to give the overall probabilistic risk classification result. The proposed approach displays superior performance on the dataset from Wuhan Metro with a 16% increase in precision, a 6% increase in recall, and an 8% increase in F1-score, compared to the direct model without information fusion. The fused model achieves a classification accuracy of 0.86 on the testing samples, which is better than the direct model. Besides, the model shows good error tolerance for wrongly classified results from classifiers without information fusion. The model has an acceptable performance even when the dataset is challenging to conduct classification tasks due to high overlapping areas in the attribute space.

SelfHAR

Machine learning and deep learning have shown great promise in mobile sensing applications, including Human Activity Recognition. However, the performance of such models in real-world settings largely depends on the availability of large datasets that captures diverse behaviors. Recently, studies in computer vision and natural language processing have shown that leveraging massive amounts of unlabeled data enables performance on par with state-of-the-art supervised models. In this work, we present SelfHAR, a semi-supervised model that effectively learns to leverage unlabeled mobile sensing datasets to complement small labeled datasets. Our approach combines teacher-student self-training, which distills the knowledge of unlabeled and labeled datasets while allowing for data augmentation, and multi-task self-supervision, which learns robust signal-level representations by predicting distorted versions of the input. We evaluated SelfHAR on various HAR datasets and showed state-of-the-art performance over supervised and previous semi-supervised approaches, with up to 12% increase in F1 score using the same number of model parameters at inference. Furthermore, SelfHAR is data-efficient, reaching similar performance using up to 10 times less labeled data compared to supervised approaches. Our work not only achieves state-of-the-art performance in a diverse set of HAR datasets, but also sheds light on how pre-training tasks may affect downstream performance.

Improved Influenza Diagnostics through Thermal Contrast Amplification.

Influenza poses a serious health threat and creates an economic burden for people around the world. The accurate diagnosis of influenza is critical to the timely clinical treatment of patients and the control of outbreaks to protect public health. Commercially available rapid influenza diagnostic tests (RIDTs) that are operated by visual readout are widely used in clinics to screen influenza infections, but RIDTs suffer from imperfect analytical sensitivity, especially when the virus concentration in the sample is low. Fortunately, the sensitivity can be simply improved through an add-on signal amplification step, i.e., thermal contrast amplification (TCA). To demonstrate the advantage of TCA for influenza diagnosis, we conducted a prospective cohort study on 345 clinical specimens collected for influenza A and B testing during the 2017–2018 influenza season. All samples were tested using the Quidel QuickVue Influenza A + B test, followed by a TCA readout, and then confirmatory polymerase chain reaction testing. Through the TCA detecting sub-visual weak positives, TCA reading improved the overall influenza sensitivity by 53% for influenza A and 33% for influenza B over the visual RIDTs readings. Even though the specificity was compromised slightly by the TCA protocol (relative decrease of 0.09% for influenza A and 0.01% for influenza B), the overall performance was still better than that achieved by visual readout based on comparison of their plots in receiver operating characteristic space and F1 scores (relative increase of 14.5% for influenza A and 12.5% for influenza B). Performing a TCA readout on wet RIDTs also improved the overall TCA performance (relative increase in F1 score of 48%). Overall, the TCA method is a simple and promising way to improve the diagnostic performance of commercial RIDTs for infectious diseases, especially in the case of specimens with low target analytes.

The Evolution of Language Models Applied to Emotion Analysis of Arabic Tweets

The field of natural language processing (NLP) has witnessed a boom in language representation models with the introduction of pretrained language models that are trained on massive textual data then used to fine-tune downstream NLP tasks. In this paper, we aim to study the evolution of language representation models by analyzing their effect on an under-researched NLP task: emotion analysis; for a low-resource language: Arabic. Most of the studies in the field of affect analysis focused on sentiment analysis, i.e., classifying text into valence (positive, negative, neutral) while few studies go further to analyze the finer grained emotional states (happiness, sadness, anger, etc.). Emotion analysis is a text classification problem that is tackled using machine learning techniques. Different language representation models have been used as features for these machine learning models to learn from. In this paper, we perform an empirical study on the evolution of language models, from the traditional term frequency–inverse document frequency (TF–IDF) to the more sophisticated word embedding word2vec, and finally the recent state-of-the-art pretrained language model, bidirectional encoder representations from transformers (BERT). We observe and analyze how the performance increases as we change the language model. We also investigate different BERT models for Arabic. We find that the best performance is achieved with the ArabicBERT large model, which is a BERT model trained on a large dataset of Arabic text. The increase in F1-score was significant +7–21%.

Re-Ranking System with BERT for Biomedical Concept Normalization

In recent years, various neural network architectures have been successfully applied to natural language processing (NLP) tasks such as named entity normalization. Named entity normalization is a fundamental task for extracting information in free text, which aims to map entity mentions in a text to gold standard entities in a given domain-specific ontology; however, the normalization task in the biomedical domain is still challenging because of multiple synonyms, various acronyms, and numerous lexical variations. In this study, we regard the task of biomedical entity normalization as a ranking problem and propose an approach to rank normalized concepts. We additionally employ two factors that can notably affect the performance of normalization, such as task-specific pre-training (Task-PT) and calibration approach. Among five different biomedical benchmark corpora, our experimental results show that our proposed model achieved significant improvements over the previous methods and advanced the state-of-the-art performance for biomedical entity normalization, with up to 0.5% increase in accuracy and 1.2% increase in F-score.

LCBPA: An Enhanced Deep Neural Network-Oriented Bug Prioritization and Assignment Technique Using Content-Based Filtering

Software maintenance is an important phase of a development life cycle that needs to be essentially performed in order to avoid the software failure. To systematically handle the bugs (defects), the software development organization develops a bug report that demonstrates the vulnerabilities from the software under test. However, manually handling the bug reports is a laborious, tedious, and time-consuming task. Moreover, the bug repository receives large numbers of bug reports on daily basis, which demands to timely fix the found and received bugs. Motivated by this, current work proposes an automated bug prioritization and assignment technique, called LCBPA ( <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L</b> ong short-term memory, <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C</b> ontent-based filtering for <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</b> ug <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</b> rioritization and <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A</b> ssignment). To perform the bug prioritization, we employed Long Short-Term Memory (LSTM) to predict the priority of the bug report. In contrast, for bug assignment, we used content-based filtering, where the prioritized bug reports are automatically assigned to the developers based on their previous knowledge. The performance of the proposed bug prioritization model is determined by comparing with the state-of-the-art bug prioritization techniques, and measured using the evaluation metrics including Precision, Recall and F1-score. Similarly, the effectiveness of the bug assignment model is evaluated by defining various case scenarios. The results show that the proposed LCBPA technique outperforms the current state-of-the-art bug prioritization techniques (with a 22% increase in F1-score), and also efficiently handles the bug assignment problem compared to the existing bug assignment techniques.

Abstract 13797: Enabling Advanced Real-world Evidence in Heart Failure: A Pilot Study Defining Preferred Approaches to Electronic Health Record Data Use

Introduction: Real world evidence (RWE) is increasingly used for regulatory and market access decision-making. In heart failure (HF), typical structured datasets have limitations in data accuracy and identifying relevant patient characteristics. Understanding which characteristics require enhancement from unstructured data and how to validly apply extraction methods will improve the definition of complex patient cohorts. Hypothesis: Augmenting structured with unstructured electronic health record (EHR) data may overcome challenges in accurately identifying relevant HF patient characteristics. Methods: Using EHR data from 4,288 primary care encounters, 20 clinical concepts were defined a priori by 3 HF experts. A reference standard was generated through chart abstraction, with each record reviewed by at least two annotators. Inter-rater reliability (IRR) was measured by Cohen’s kappa. EHR structured data (EHR-S) extracted with traditional query techniques and EHR unstructured (EHR-U) data extracted with artificial intelligence (AI) technologies were tested for accuracy against the reference standard. Results: In EHR-S, recall ranged from 0-95.1% and precision from 52.9-100%. In EHR-U data processed using AI, recall ranged from 80.4-99.7% and precision from 82.3-100%. Results demonstrated a 45.1% absolute difference and 98.1% relative increase in F1-score (Table). Reference standard IRR was 95.3%. Conclusions: RWE credibility and applicability relies on accurate identification of a patient cohort. This study suggests that readily available data sources may not accurately identify patient phenotypes in HF. Novel means of using AI with EHR-U may improve such efforts, particularly for conditions and symptoms. This approach offers a pathway for defining highly accurate HF cohorts that may be useful in studies with narrowly defined or complex phenotypes, such as those where inclusion and exclusion criteria are specific and outcomes require validity.

Speeded-up Convolution Neural Network for classification tasks using multiscale 2-dimensional decomposition

In this paper, we propose a strategy for network simplification and acceleration. First, we propose to generate a suitable resized image using multiscale patching in the first convolutional layer, which can then be used for the rest of the network. We use p convolutional filters that operate on patches of size m × n, and we first select all the possible non-superposed m × n patches from the available images. If the number of such patches is not sufficient, the remaining ones are collected using scales a × b or c × d such that ab = cd = mn or a = 2 m or b = 2n. Patches generated from the former condition are directly extracted from the images, while the downsampled results are used in the latter case.We also introduce a 2-dimensional decomposition for patch compression, by stacking all the available image patches along the columns and applying a 2D PCA decomposition. Finally, a layer weight decomposition technique followed by module-based finetuning is adopted, for a new fast module-based CNN model.Extensive evaluations using public data sets like MNIST, Pascal VOC, and WebCASIA, and with different state-of-the art CNN architectures like Open-Face, VGG and DarkNet verify that our proposed model is able to accelerate the training process and even to provide higher classification accuracies for small-sized datasets. We obtain 8% increase in Top-1 and Top-5 recognition rates, and 5% increase in F1 score over general interpolation-based resizing.

An ontology-based learning approach for automatically classifying security requirements

Although academia has recognized the importance of explicitly specifying security requirements in early stages of system developments for years, in reality, many projects mix security requirements with other types of requirements. Thus, there is a strong need for precisely and efficiently classifying such security requirements from other requirements in requirement specifications. Existing studies leverage lexical evidence to build probabilistic classifiers, which are domain-dependent by design and cannot effectively classify security requirements from different application domains. In this paper, we propose an ontology-driven learning approach to automatically classify security requirements. Our approach consists of a conceptual layer and a linguistic layer, which understands security requirements based on not only lexical evidence but also conceptual domain knowledge. In particular, we apply a systematic approach to identify linguistic features of security requirements based on an extended security requirements ontology and linguistic knowledge, connecting the conceptual layer with the linguistic layer. Such linguistic features are then used to train domain-independent security requirements classifiers by using machine learning techniques. We have carried out a series of experiments to evaluate the performance and generalization ability of our proposal against existing approaches. The results of the experiments show that the proposed approach outperforms existing approaches with a significant increase of F1 score (0.63 VS. 0.44) when the training dataset and the testing dataset come from different application domains, i.e., the classifiers trained by our approach can be generalized to classify security requirements from different domains.

TDP: Two-dimensional perceptron for image recognition

Convolutional neural network (CNN) is widely applied to different areas due to good recognition performance. However, convolution operation is a complex computation and consumes the bulk of processing time for CNN. It is still a hot problem how to develop a novel model with good recognition performance for deep learning. Here, we propose a novel model, namely, two-dimensional perceptron (TDP), to get direct input of two-dimensional data for further processing. A TDP has a new network architecture and an innovative computation process of hidden neurons. In cases with the same number of hidden neurons, compared with multilayer perceptron (MLP), TDP achieves good recognition performance with 1×-36× speedup and a decrease of parameters by exceeding 97% on MNIST and COIL-20 datasets. Meanwhile, TDP obtains 1%–32% improvement of recognition accuracy in comparison to CNN on CIFAR-10 and SVHN datasets. Furthermore, on INFUSE dataset, TDP has an increase of F1 score by up to almost 11% in comparison with MLP and CNN. The results indicate that TDP is a promising and novel model with excellent recognition performance.