An optimized hybrid deep intrusion detection model (HD-IDM) for enhancing network security
- Ahmad, Iftikhar, Imran, Muhammad, Qayyum, Abdul, Ramzan, Muhammad, Alassafi, Madini
- Authors: Ahmad, Iftikhar , Imran, Muhammad , Qayyum, Abdul , Ramzan, Muhammad , Alassafi, Madini
- Date: 2023
- Type: Text , Journal article
- Relation: Mathematics Vol. 11, no. 21 (2023), p.
- Full Text:
- Reviewed:
- Description: Detecting cyber intrusions in network traffic is a tough task for cybersecurity. Current methods struggle with the complexity of understanding patterns in network data. To solve this, we present the Hybrid Deep Learning Intrusion Detection Model (HD-IDM), a new way that combines GRU and LSTM classifiers. GRU is good at catching quick patterns, while LSTM handles long-term ones. HD-IDM blends these models using weighted averaging, boosting accuracy, especially with complex patterns. We tested HD-IDM on four datasets: CSE-CIC-IDS2017, CSE-CIC-IDS2018, NSL KDD, and CIC-DDoS2019. The HD-IDM classifier achieved remarkable performance metrics on all datasets. It attains an outstanding accuracy of 99.91%, showcasing its consistent precision across the dataset. With an impressive precision of 99.62%, it excels in accurately categorizing positive cases, crucial for minimizing false positives. Additionally, maintaining a high recall of 99.43%, it effectively identifies the majority of actual positive cases while minimizing false negatives. The F1-score of 99.52% emphasizes its robustness, making it the top choice for classification tasks requiring precision and reliability. It is particularly good at ROC and precision/recall curves, discriminating normal and harmful network activities. While HD-IDM is promising, it has limits. It needs labeled data and may struggle with new intrusion methods. Future work should find ways to handle unlabeled data and adapt to emerging threats. Also, making HD-IDM work faster for real-time use and dealing with scalability challenges is key for its broader use in changing network environments. © 2023 by the authors.
- Authors: Ahmad, Iftikhar , Imran, Muhammad , Qayyum, Abdul , Ramzan, Muhammad , Alassafi, Madini
- Date: 2023
- Type: Text , Journal article
- Relation: Mathematics Vol. 11, no. 21 (2023), p.
- Full Text:
- Reviewed:
- Description: Detecting cyber intrusions in network traffic is a tough task for cybersecurity. Current methods struggle with the complexity of understanding patterns in network data. To solve this, we present the Hybrid Deep Learning Intrusion Detection Model (HD-IDM), a new way that combines GRU and LSTM classifiers. GRU is good at catching quick patterns, while LSTM handles long-term ones. HD-IDM blends these models using weighted averaging, boosting accuracy, especially with complex patterns. We tested HD-IDM on four datasets: CSE-CIC-IDS2017, CSE-CIC-IDS2018, NSL KDD, and CIC-DDoS2019. The HD-IDM classifier achieved remarkable performance metrics on all datasets. It attains an outstanding accuracy of 99.91%, showcasing its consistent precision across the dataset. With an impressive precision of 99.62%, it excels in accurately categorizing positive cases, crucial for minimizing false positives. Additionally, maintaining a high recall of 99.43%, it effectively identifies the majority of actual positive cases while minimizing false negatives. The F1-score of 99.52% emphasizes its robustness, making it the top choice for classification tasks requiring precision and reliability. It is particularly good at ROC and precision/recall curves, discriminating normal and harmful network activities. While HD-IDM is promising, it has limits. It needs labeled data and may struggle with new intrusion methods. Future work should find ways to handle unlabeled data and adapt to emerging threats. Also, making HD-IDM work faster for real-time use and dealing with scalability challenges is key for its broader use in changing network environments. © 2023 by the authors.
- Nejati, Maryam, Amjady, Nima
- Authors: Nejati, Maryam , Amjady, Nima
- Date: 2022
- Type: Text , Journal article
- Relation: IEEE transactions on sustainable energy Vol. 13, no. 2 (2022), p. 1188-1198
- Full Text: false
- Reviewed:
- Description: Solar generation systems are globally extending in terms of scale and number, which highlights the increasing importance of solar power forecast. In this paper, a day-ahead solar power prediction method is proposed including 1) a novel feature selecting/clustering approach based on relevancy and redundancy criteria and 2) an innovative hybrid-classification-regression forecasting engine. The proposed feature selecting/clustering approach filters out irrelevant features and partitions relevant features to two separate subsets to decrease the redundancy of features. Each of these two subsets is separately trained by one forecasting engine and the final solar power prediction of the proposed method is obtained by a relevancy-based combination of these two forecasts. The proposed forecasting engine classifies the historical data based on the learnability of its constituent regression models and assigns each class of training samples to one regression model. Each regression model predicts the outputs of the test samples that belong to its class. The effectiveness of the proposed solar power prediction method is illustrated by testing on two real-world solar farms.
Mineral texture identification using local binary patterns equipped with a Classification and Recognition Updating System (CARUS)
- Aligholi, Saeed, Khajavi, Reza, Khandelwal, Manoj, Armaghani, Danial
- Authors: Aligholi, Saeed , Khajavi, Reza , Khandelwal, Manoj , Armaghani, Danial
- Date: 2022
- Type: Text , Journal article
- Relation: Sustainability (Switzerland) Vol. 14, no. 18 (2022), p.
- Full Text:
- Reviewed:
- Description: In this paper, a rotation-invariant local binary pattern operator equipped with a local contrast measure (riLBPc) is employed to characterize the type of mineral twinning by inspecting the texture properties of crystals. The proposed method uses photomicrographs of minerals and produces LBP histograms, which might be compared with those included in a predefined database using the Kullback–Leibler divergence-based metric. The paper proposes a new LBP-based scheme for concurrent classification and recognition tasks, followed by a novel online updating routine to enhance the locally developed mineral LBP database. The discriminatory power of the proposed Classification and Recognition Updating System (CARUS) for texture identification scheme is verified for plagioclase, orthoclase, microcline, and quartz minerals with sensitivity (TPR) near 99.9%, 87%, 99.9%, and 96%, and accuracy (ACC) equal to about 99%, 97%, 99%, and 99%, respectively. According to the results, the introduced CARUS system is a promising approach that can be applied in a variety of different fields dealing with classification and feature recognition tasks. © 2022 by the authors.
- Authors: Aligholi, Saeed , Khajavi, Reza , Khandelwal, Manoj , Armaghani, Danial
- Date: 2022
- Type: Text , Journal article
- Relation: Sustainability (Switzerland) Vol. 14, no. 18 (2022), p.
- Full Text:
- Reviewed:
- Description: In this paper, a rotation-invariant local binary pattern operator equipped with a local contrast measure (riLBPc) is employed to characterize the type of mineral twinning by inspecting the texture properties of crystals. The proposed method uses photomicrographs of minerals and produces LBP histograms, which might be compared with those included in a predefined database using the Kullback–Leibler divergence-based metric. The paper proposes a new LBP-based scheme for concurrent classification and recognition tasks, followed by a novel online updating routine to enhance the locally developed mineral LBP database. The discriminatory power of the proposed Classification and Recognition Updating System (CARUS) for texture identification scheme is verified for plagioclase, orthoclase, microcline, and quartz minerals with sensitivity (TPR) near 99.9%, 87%, 99.9%, and 96%, and accuracy (ACC) equal to about 99%, 97%, 99%, and 99%, respectively. According to the results, the introduced CARUS system is a promising approach that can be applied in a variety of different fields dealing with classification and feature recognition tasks. © 2022 by the authors.
Subgraph adaptive structure-aware graph contrastive learning
- Chen, Zhikui, Peng, Yin, Yu, Shuo, Cao, Chen, Xia, Feng
- Authors: Chen, Zhikui , Peng, Yin , Yu, Shuo , Cao, Chen , Xia, Feng
- Date: 2022
- Type: Text , Journal article
- Relation: Mathematics (Basel) Vol. 10, no. 17 (2022), p. 3047
- Full Text:
- Reviewed:
- Description: Graph contrastive learning (GCL) has been subject to more attention and been widely applied to numerous graph learning tasks such as node classification and link prediction. Although it has achieved great success and even performed better than supervised methods in some tasks, most of them depend on node-level comparison, while ignoring the rich semantic information contained in graph topology, especially for social networks. However, a higher-level comparison requires subgraph construction and encoding, which remain unsolved. To address this problem, we propose a subgraph adaptive structure-aware graph contrastive learning method (PASCAL) in this work, which is a subgraph-level GCL method. In PASCAL, we construct subgraphs by merging all motifs that contain the target node. Then we encode them on the basis of motif number distribution to capture the rich information hidden in subgraphs. By incorporating motif information, PASCAL can capture richer semantic information hidden in local structures compared with other GCL methods. Extensive experiments on six benchmark datasets show that PASCAL outperforms state-of-art graph contrastive learning and supervised methods in most cases.
- Authors: Chen, Zhikui , Peng, Yin , Yu, Shuo , Cao, Chen , Xia, Feng
- Date: 2022
- Type: Text , Journal article
- Relation: Mathematics (Basel) Vol. 10, no. 17 (2022), p. 3047
- Full Text:
- Reviewed:
- Description: Graph contrastive learning (GCL) has been subject to more attention and been widely applied to numerous graph learning tasks such as node classification and link prediction. Although it has achieved great success and even performed better than supervised methods in some tasks, most of them depend on node-level comparison, while ignoring the rich semantic information contained in graph topology, especially for social networks. However, a higher-level comparison requires subgraph construction and encoding, which remain unsolved. To address this problem, we propose a subgraph adaptive structure-aware graph contrastive learning method (PASCAL) in this work, which is a subgraph-level GCL method. In PASCAL, we construct subgraphs by merging all motifs that contain the target node. Then we encode them on the basis of motif number distribution to capture the rich information hidden in subgraphs. By incorporating motif information, PASCAL can capture richer semantic information hidden in local structures compared with other GCL methods. Extensive experiments on six benchmark datasets show that PASCAL outperforms state-of-art graph contrastive learning and supervised methods in most cases.
Towards robust convolutional neural networks in challenging environments
- Authors: Hossain, Md Tahmid
- Date: 2021
- Type: Text , Thesis , PhD
- Full Text:
- Description: Image classification is one of the fundamental tasks in the field of computer vision. Although Artificial Neural Network (ANN) showed a lot of promise in this field, the lack of efficient computer hardware subdued its potential to a great extent. In the early 2000s, advances in hardware coupled with better network design saw the dramatic rise of Convolutional Neural Network (CNN). Deep CNNs pushed the State-of-The-Art (SOTA) in a number of vision tasks, including image classification, object detection, and segmentation. Presently, CNNs dominate these tasks. Although CNNs exhibit impressive classification performance on clean images, they are vulnerable to distortions, such as noise and blur. Fine-tuning a pre-trained CNN on mutually exclusive or a union set of distortions is a brute-force solution. This iterative fine-tuning process with all known types of distortion is, however, exhaustive and the network struggles to handle unseen distortions. CNNs are also vulnerable to image translation or shift, partly due to common Down-Sampling (DS) layers, e.g., max-pooling and strided convolution. These operations violate the Nyquist sampling rate and cause aliasing. The textbook solution is low-pass filtering (blurring) before down-sampling, which can benefit deep networks as well. Even so, non-linearity units, such as ReLU, often re-introduce the problem, suggesting that blurring alone may not suffice. Another important but under-explored issue for CNNs is unknown or Open Set Recognition (OSR). CNNs are commonly designed for closed set arrangements, where test instances only belong to some ‘Known Known’ (KK) classes used in training. As such, they predict a class label for a test sample based on the distribution of the KK classes. However, when used under the OSR setup (where an input may belong to an ‘Unknown Unknown’ or UU class), such a network will always classify a test instance as one of the KK classes even if it is from a UU class. Historically, CNNs have struggled with detecting objects in images with large difference in scale, especially small objects. This is because the DS layers inside a CNN often progressively wipe out the signal from small objects. As a result, the final layers are left with no signature from these objects leading to degraded performance. In this work, we propose solutions to the above four problems. First, we improve CNN robustness against distortion by proposing DCT based augmentation, adaptive regularisation, and noise suppressing Activation Functions (AF). Second, to ensure further performance gain and robustness to image transformations, we introduce anti-aliasing properties inside the AF and propose a novel DS method called blurpool. Third, to address the OSR problem, we propose a novel training paradigm that ensures detection of UU classes and accurate classification of the KK classes. Finally, we introduce a novel CNN that enables a deep detector to identify small objects with high precision and recall. We evaluate our methods on a number of benchmark datasets and demonstrate that they outperform contemporary methods in the respective problem set-ups.
- Description: Doctor of Philosophy
- Authors: Hossain, Md Tahmid
- Date: 2021
- Type: Text , Thesis , PhD
- Full Text:
- Description: Image classification is one of the fundamental tasks in the field of computer vision. Although Artificial Neural Network (ANN) showed a lot of promise in this field, the lack of efficient computer hardware subdued its potential to a great extent. In the early 2000s, advances in hardware coupled with better network design saw the dramatic rise of Convolutional Neural Network (CNN). Deep CNNs pushed the State-of-The-Art (SOTA) in a number of vision tasks, including image classification, object detection, and segmentation. Presently, CNNs dominate these tasks. Although CNNs exhibit impressive classification performance on clean images, they are vulnerable to distortions, such as noise and blur. Fine-tuning a pre-trained CNN on mutually exclusive or a union set of distortions is a brute-force solution. This iterative fine-tuning process with all known types of distortion is, however, exhaustive and the network struggles to handle unseen distortions. CNNs are also vulnerable to image translation or shift, partly due to common Down-Sampling (DS) layers, e.g., max-pooling and strided convolution. These operations violate the Nyquist sampling rate and cause aliasing. The textbook solution is low-pass filtering (blurring) before down-sampling, which can benefit deep networks as well. Even so, non-linearity units, such as ReLU, often re-introduce the problem, suggesting that blurring alone may not suffice. Another important but under-explored issue for CNNs is unknown or Open Set Recognition (OSR). CNNs are commonly designed for closed set arrangements, where test instances only belong to some ‘Known Known’ (KK) classes used in training. As such, they predict a class label for a test sample based on the distribution of the KK classes. However, when used under the OSR setup (where an input may belong to an ‘Unknown Unknown’ or UU class), such a network will always classify a test instance as one of the KK classes even if it is from a UU class. Historically, CNNs have struggled with detecting objects in images with large difference in scale, especially small objects. This is because the DS layers inside a CNN often progressively wipe out the signal from small objects. As a result, the final layers are left with no signature from these objects leading to degraded performance. In this work, we propose solutions to the above four problems. First, we improve CNN robustness against distortion by proposing DCT based augmentation, adaptive regularisation, and noise suppressing Activation Functions (AF). Second, to ensure further performance gain and robustness to image transformations, we introduce anti-aliasing properties inside the AF and propose a novel DS method called blurpool. Third, to address the OSR problem, we propose a novel training paradigm that ensures detection of UU classes and accurate classification of the KK classes. Finally, we introduce a novel CNN that enables a deep detector to identify small objects with high precision and recall. We evaluate our methods on a number of benchmark datasets and demonstrate that they outperform contemporary methods in the respective problem set-ups.
- Description: Doctor of Philosophy
A machine learning approach for prediction of pregnancy outcome following IVF treatment
- Hassan, Md Rafiul, Al-Insaif, Sadiq, Hossain, Muhammad, Kamruzzaman, Joarder
- Authors: Hassan, Md Rafiul , Al-Insaif, Sadiq , Hossain, Muhammad , Kamruzzaman, Joarder
- Date: 2020
- Type: Text , Journal article
- Relation: Neural Computing and Applications Vol. 32, no. 7 (2020), p. 2283-2297
- Full Text: false
- Reviewed:
- Description: Infertility affects one out of seven couples around the world. Therefore, the best possible management of the in vitro fertilization (IVF) treatment and patient advice is crucial for both patients and medical practitioners. The ultimate concern of the patients is the success of an IVF procedure, which depends on a number of influencing attributes. Without any automated tool, it is hard for the practitioners to assess any influencing trend of the attributes and factors that might lead to a successful IVF pregnancy. This paper proposes a hill climbing feature (attribute) selection algorithm coupled with automated classification using machine learning techniques with the aim to analyze and predict IVF pregnancy in greater accuracy. Using 25 attributes, we assessed the prediction ability of IVF pregnancy success for five different machine learning models, namely multilayer perceptron (MLP), support vector machines (SVM), C4.5, classification and regression trees (CART) and random forest (RF). The prediction ability was measured in terms of widely used performance metrics, namely accuracy rate, F-measure and AUC. Feature selection algorithm reduced the number of most influential attributes to nineteen for MLP, sixteen for RF, seventeen for SVM, twelve for C4.5 and eight for CART. Overall, the most influential attributes identified are: ‘age’, ‘indication’ of fertility factor, ‘Antral Follicle Counts (AFC)’, ‘NbreM2’, ‘method of sperm collection’, ‘Chamotte’, ‘Fertilization rate in vitro’, ‘Follicles on day 14’ and ‘Embryo transfer day.’ The machine learning models trained with the selected set of features significantly improved the prediction accuracy of IVF pregnancy success to a level considerably higher than those reported in the current literature. © 2018, The Natural Computing Applications Forum.
Action research to implement an Indigenous health curriculum framework
- Wilson, Cath, Heinrich, Liesl, Heidari, Parvaneh, Adams, Karen
- Authors: Wilson, Cath , Heinrich, Liesl , Heidari, Parvaneh , Adams, Karen
- Date: 2020
- Type: Text , Journal article
- Relation: Nurse education today Vol. 91, no. (2020), p. 104464-104464
- Full Text: false
- Reviewed:
- Description: In recent decades Indigenous health curriculum frameworks have been developed, however, few studies about their implementation exist. This study aimed to employ critical theory and action research to understand how an Indigenous health curriculum framework could be applied and associated learning and teaching iteratively improved. Three action research cycles where conducted from 2017 to 2019. Student reaction (satisfaction and engagement) was collected via survey 2017–2019. Student learning was collated 2018–2019 via self-perception survey (knowledge, attitude, confidence, commitment) multi-choice questions (knowledge) and content analysis of apply and analyse activities (skill). The teaching team met annually to reflect on findings and plan enhancements to learning and teaching. Over 2017–2019 there was a pattern of improved student reaction and learning. Connecting this research to Faculty level committees led to widening success and improved sustainability of the practice. The online unit and workshop delivery were scalable, overcame a barrier of educator skill and confidence to teach this area, allowed for quality content control and provided data for analysis. Interestingly, learning gained from this unit matched that described as occurring from student placements in health settings with high numbers of Indigenous people. Student learning occurred across the Framework three levels (novice, intermediate and entry to practice) suggesting that the taxonomy of the Framework does not necessarily align with the reality of learning and teaching. Vertical implementation of the five learning domains would benefit from alignment with training evaluation models and validated assessment to understand learning that has occurred rather than the teaching that has been taught. In this study health profession accreditation bodies had driven the imperative for an Indigenous health program and curriculum. Research on Indigenous health learning and teaching relating to behaviour and results in workplaces is needed.
Enhancing linear time complexity time series classification with hybrid bag-of-patterns
- Liang, Shen, Zhang, Yanchun, Ma, Jiangang
- Authors: Liang, Shen , Zhang, Yanchun , Ma, Jiangang
- Date: 2020
- Type: Text , Conference paper
- Relation: 25th International Conference on Database Systems for Advanced Applications, DASFAA 2020 Vol. 12112 LNCS, p. 717-735
- Full Text: false
- Reviewed:
- Description: In time series classification, one of the most popular models is Bag-Of-Patterns (BOP). Most BOP methods run in super-linear time. A recent work proposed a linear time BOP model, yet it has limited accuracy. In this work, we present Hybrid Bag-Of-Patterns (HBOP), which can greatly enhance accuracy while maintaining linear complexity. Concretely, we first propose a novel time series discretization method called SLA, which can retain more information than the classic SAX. We use a hybrid of SLA and SAX to expressively and compactly represent subsequences, which is our most important design feature. Moreover, we develop an efficient time series transformation method that is key to achieving linear complexity. We also propose a novel X-means clustering subroutine to handle subclasses. Extensive experiments on over 100 datasets demonstrate the effectiveness and efficiency of our method. © 2020, Springer Nature Switzerland AG.
Fraud detection for online banking for scalable and distributed data
- Authors: Haq, Ikram
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: Online fraud causes billions of dollars in losses for banks. Therefore, online banking fraud detection is an important field of study. However, there are many challenges in conducting research in fraud detection. One of the constraints is due to unavailability of bank datasets for research or the required characteristics of the attributes of the data are not available. Numeric data usually provides better performance for machine learning algorithms. Most transaction data however have categorical, or nominal features as well. Moreover, some platforms such as Apache Spark only recognizes numeric data. So, there is a need to use techniques e.g. One-hot encoding (OHE) to transform categorical features to numerical features, however OHE has challenges including the sparseness of transformed data and that the distinct values of an attribute are not always known in advance. Efficient feature engineering can improve the algorithm’s performance but usually requires detailed domain knowledge to identify correct features. Techniques like Ripple Down Rules (RDR) are suitable for fraud detection because of their low maintenance and incremental learning features. However, high classification accuracy on mixed datasets, especially for scalable data is challenging. Evaluation of RDR on distributed platforms is also challenging as it is not available on these platforms. The thesis proposes the following solutions to these challenges: • We developed a technique Highly Correlated Rule Based Uniformly Distribution (HCRUD) to generate highly correlated rule-based uniformly-distributed synthetic data. • We developed a technique One-hot Encoded Extended Compact (OHE-EC) to transform categorical features to numeric features by compacting sparse-data even if all distinct values are unknown. • We developed a technique Feature Engineering and Compact Unified Expressions (FECUE) to improve model efficiency through feature engineering where the domain of the data is not known in advance. • A Unified Expression RDR fraud deduction technique (UE-RDR) for Big data has been proposed and evaluated on the Spark platform. Empirical tests were executed on multi-node Hadoop cluster using well-known classifiers on bank data, synthetic bank datasets and publicly available datasets from UCI repository. These evaluations demonstrated substantial improvements in terms of classification accuracy, ruleset compactness and execution speed.
- Description: Doctor of Philosophy
- Authors: Haq, Ikram
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: Online fraud causes billions of dollars in losses for banks. Therefore, online banking fraud detection is an important field of study. However, there are many challenges in conducting research in fraud detection. One of the constraints is due to unavailability of bank datasets for research or the required characteristics of the attributes of the data are not available. Numeric data usually provides better performance for machine learning algorithms. Most transaction data however have categorical, or nominal features as well. Moreover, some platforms such as Apache Spark only recognizes numeric data. So, there is a need to use techniques e.g. One-hot encoding (OHE) to transform categorical features to numerical features, however OHE has challenges including the sparseness of transformed data and that the distinct values of an attribute are not always known in advance. Efficient feature engineering can improve the algorithm’s performance but usually requires detailed domain knowledge to identify correct features. Techniques like Ripple Down Rules (RDR) are suitable for fraud detection because of their low maintenance and incremental learning features. However, high classification accuracy on mixed datasets, especially for scalable data is challenging. Evaluation of RDR on distributed platforms is also challenging as it is not available on these platforms. The thesis proposes the following solutions to these challenges: • We developed a technique Highly Correlated Rule Based Uniformly Distribution (HCRUD) to generate highly correlated rule-based uniformly-distributed synthetic data. • We developed a technique One-hot Encoded Extended Compact (OHE-EC) to transform categorical features to numeric features by compacting sparse-data even if all distinct values are unknown. • We developed a technique Feature Engineering and Compact Unified Expressions (FECUE) to improve model efficiency through feature engineering where the domain of the data is not known in advance. • A Unified Expression RDR fraud deduction technique (UE-RDR) for Big data has been proposed and evaluated on the Spark platform. Empirical tests were executed on multi-node Hadoop cluster using well-known classifiers on bank data, synthetic bank datasets and publicly available datasets from UCI repository. These evaluations demonstrated substantial improvements in terms of classification accuracy, ruleset compactness and execution speed.
- Description: Doctor of Philosophy
Imbalanced data classification and its application in cyber security
- Authors: Moniruzzaman, Md
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: Cyber security, also known as information technology security or simply as information security, aims to protect government organizations, companies and individuals by defending their computers, servers, electronic systems, networks, and data from malicious attacks. With the advancement of client-side on the fly web content generation techniques, it becomes easier for attackers to modify the content of a website dynamically and gain access to valuable information. The impact of cybercrime to the global economy is now more than ever, and it is growing day by day. Among various types of cybercrimes, financial attacks are widely spread and the financial sector is among most targeted. Both corporations and individuals are losing a huge amount of money each year. The majority portion of financial attacks is carried out by banking malware and web-based attacks. The end users are not always skilled enough to differentiate between injected content and actual contents of a webpage. Designing a real-time security system for ensuring a safe browsing experience is a challenging task. Some of the existing solutions are designed for client side and all the users have to install it in their system, which is very difficult to implement. In addition, various platforms and tools are used by organizations and individuals, therefore, different solutions are needed to be designed. The existing server-side solution often focuses on sanitizing and filtering the inputs. It will fail to detect obfuscated and hidden scripts. This is a realtime security system and any significant delay will hamper user experience. Therefore, finding the most optimized and efficient solution is very important. To ensure an easy installation and integration capabilities of any solution with the existing system is also a critical factor to consider. If the solution is efficient but difficult to integrate, then it may not be a feasible solution for practical use. Unsupervised and supervised data classification techniques have been widely applied to design algorithms for solving cyber security problems. The performance of these algorithms varies depending on types of cyber security problems and size of datasets. To date, existing algorithms do not achieve high accuracy in detecting malware activities. Datasets in cyber security and, especially those from financial sectors, are predominantly imbalanced datasets as the number of malware activities is significantly less than the number of normal activities. This means that classifiers for imbalanced datasets can be used to develop supervised data classification algorithms to detect malware activities. Development of classifiers for imbalanced data sets has been subject of research over the last decade. Most of these classifiers are based on oversampling and undersampling techniques and are not efficient in many situations as such techniques are applied globally. In this thesis, we develop two new algorithms for solving supervised data classification problems in imbalanced datasets and then apply them to solve malware detection problems. The first algorithm is designed using the piecewise linear classifiers by formulating this problem as an optimization problem and by applying the penalty function method. More specifically, we add more penalty to the objective function for misclassified points from minority classes. The second method is based on the combination of the supervised and unsupervised (clustering) algorithms. Such an approach allows one to identify areas in the input space where minority classes are located and to apply local oversampling or undersampling. This approach leads to the design of more efficient and accurate classifiers. The proposed algorithms are tested using real-world datasets. Results clearly demonstrate superiority of newly introduced algorithms. Then we apply these algorithms to design classifiers to detect malwares.
- Description: Doctor of Philosophy
- Authors: Moniruzzaman, Md
- Date: 2020
- Type: Text , Thesis , PhD
- Full Text:
- Description: Cyber security, also known as information technology security or simply as information security, aims to protect government organizations, companies and individuals by defending their computers, servers, electronic systems, networks, and data from malicious attacks. With the advancement of client-side on the fly web content generation techniques, it becomes easier for attackers to modify the content of a website dynamically and gain access to valuable information. The impact of cybercrime to the global economy is now more than ever, and it is growing day by day. Among various types of cybercrimes, financial attacks are widely spread and the financial sector is among most targeted. Both corporations and individuals are losing a huge amount of money each year. The majority portion of financial attacks is carried out by banking malware and web-based attacks. The end users are not always skilled enough to differentiate between injected content and actual contents of a webpage. Designing a real-time security system for ensuring a safe browsing experience is a challenging task. Some of the existing solutions are designed for client side and all the users have to install it in their system, which is very difficult to implement. In addition, various platforms and tools are used by organizations and individuals, therefore, different solutions are needed to be designed. The existing server-side solution often focuses on sanitizing and filtering the inputs. It will fail to detect obfuscated and hidden scripts. This is a realtime security system and any significant delay will hamper user experience. Therefore, finding the most optimized and efficient solution is very important. To ensure an easy installation and integration capabilities of any solution with the existing system is also a critical factor to consider. If the solution is efficient but difficult to integrate, then it may not be a feasible solution for practical use. Unsupervised and supervised data classification techniques have been widely applied to design algorithms for solving cyber security problems. The performance of these algorithms varies depending on types of cyber security problems and size of datasets. To date, existing algorithms do not achieve high accuracy in detecting malware activities. Datasets in cyber security and, especially those from financial sectors, are predominantly imbalanced datasets as the number of malware activities is significantly less than the number of normal activities. This means that classifiers for imbalanced datasets can be used to develop supervised data classification algorithms to detect malware activities. Development of classifiers for imbalanced data sets has been subject of research over the last decade. Most of these classifiers are based on oversampling and undersampling techniques and are not efficient in many situations as such techniques are applied globally. In this thesis, we develop two new algorithms for solving supervised data classification problems in imbalanced datasets and then apply them to solve malware detection problems. The first algorithm is designed using the piecewise linear classifiers by formulating this problem as an optimization problem and by applying the penalty function method. More specifically, we add more penalty to the objective function for misclassified points from minority classes. The second method is based on the combination of the supervised and unsupervised (clustering) algorithms. Such an approach allows one to identify areas in the input space where minority classes are located and to apply local oversampling or undersampling. This approach leads to the design of more efficient and accurate classifiers. The proposed algorithms are tested using real-world datasets. Results clearly demonstrate superiority of newly introduced algorithms. Then we apply these algorithms to design classifiers to detect malwares.
- Description: Doctor of Philosophy
New gene selection algorithm using hypeboxes to improve performance of classifiers
- Bagirov, Adil, Mardaneh, Karim
- Authors: Bagirov, Adil , Mardaneh, Karim
- Date: 2020
- Type: Text , Journal article
- Relation: International Journal of Bioinformatics Research and Applications Vol. 16, no. 3 (2020), p. 269-289
- Full Text: false
- Reviewed:
- Description: The use of DNA microarray technology allows to measure the expression levels of thousands of genes in one single experiment which makes possible to apply classification techniques to classify tumours. However, the large number of genes and relatively small number of tumours in gene expression datasets may (and in some cases significantly) diminish the accuracy of many classifiers. Therefore, efficient gene selection algorithms are required to identify most informative genes or groups of genes to improve the performance of classifiers. In this paper, a new gene selection algorithm is developed using marginal hyberboxes of genes or groups of genes for each tumour type. Informative genes are defined using overlaps between hyberboxes. The results on six gene expression datasets demonstrate that the proposed algorithm is able to considerably reduce the number of genes and significantly improve the performance of classifiers. © 2020 Inderscience Enterprises Ltd.
- Authors: Greenslade, Penelope
- Date: 2019
- Type: Text , Journal article
- Relation: South Australian naturalist Vol. 93, no. 2 (2019), p. 103-110
- Full Text: false
- Reviewed:
- Description: 'Temeritas regalis' (Womersley, 1939) was formerly common in the Mt Lofty Ranges extending from as far north as Alligator Gorge/Mt Remarkable/Wirrabara in the Southern Flinders to Myponga, 60 km south of Adelaide. The species is a South Australian endemic and belongs to a pantropical wide genus. It was collected several times in the 1930's and fourteen times in the 1970's but it has not been seen since 1978. In spite of collecting efforts from its typical habitat in the Mt Lofty Ranges and Southern Flinders Ranges and at Mt Bold on 21 occasions in the last 20 years, no further specimens have been found. It appears that populations have shrunk in size or may have disappeared altogether since the 1980s due to a number of factors mainly: climate change, drought, habitat alteration and/or expansion of invasive plants, invertebrates and 'Phytophthora'. Consequently, it is suggested that the apparent population decrease and possible local extinctions are real and, as a result, this species complies with the IUCN criteria A1 and B, and should be listed as endangered under the South Australian legislation.
Multi-source cyber-attacks detection using machine learning
- Taheri, Sona, Gondal, Iqbal, Bagirov, Adil, Harkness, Greg, Brown, Simon, Chi, Chihung
- Authors: Taheri, Sona , Gondal, Iqbal , Bagirov, Adil , Harkness, Greg , Brown, Simon , Chi, Chihung
- Date: 2019
- Type: Text , Conference proceedings , Conference paper
- Relation: 2019 IEEE International Conference on Industrial Technology, ICIT 2019; Melbourne, Australia; 13th-15th February 2019 Vol. 2019-February, p. 1167-1172
- Full Text:
- Reviewed:
- Description: The Internet of Things (IoT) has significantly increased the number of devices connected to the Internet ranging from sensors to multi-source data information. As the IoT continues to evolve with new technologies number of threats and attacks against IoT devices are on the increase. Analyzing and detecting these attacks originating from different sources needs machine learning models. These models provide proactive solutions for detecting attacks and their sources. In this paper, we propose to apply a supervised machine learning classification technique to identify cyber-attacks from each source. More precisely, we apply the incremental piecewise linear classifier that constructs boundary between sources/classes incrementally starting with one hyperplane and adding more hyperplanes at each iteration. The algorithm terminates when no further significant improvement of the separation of sources/classes is possible. The construction and usage of piecewise linear boundaries allows us to avoid any possible overfitting. We apply the incremental piecewise linear classifier on the multi-source real world cyber security data set to identify cyber-attacks and their sources.
- Description: Proceedings of the IEEE International Conference on Industrial Technology
- Authors: Taheri, Sona , Gondal, Iqbal , Bagirov, Adil , Harkness, Greg , Brown, Simon , Chi, Chihung
- Date: 2019
- Type: Text , Conference proceedings , Conference paper
- Relation: 2019 IEEE International Conference on Industrial Technology, ICIT 2019; Melbourne, Australia; 13th-15th February 2019 Vol. 2019-February, p. 1167-1172
- Full Text:
- Reviewed:
- Description: The Internet of Things (IoT) has significantly increased the number of devices connected to the Internet ranging from sensors to multi-source data information. As the IoT continues to evolve with new technologies number of threats and attacks against IoT devices are on the increase. Analyzing and detecting these attacks originating from different sources needs machine learning models. These models provide proactive solutions for detecting attacks and their sources. In this paper, we propose to apply a supervised machine learning classification technique to identify cyber-attacks from each source. More precisely, we apply the incremental piecewise linear classifier that constructs boundary between sources/classes incrementally starting with one hyperplane and adding more hyperplanes at each iteration. The algorithm terminates when no further significant improvement of the separation of sources/classes is possible. The construction and usage of piecewise linear boundaries allows us to avoid any possible overfitting. We apply the incremental piecewise linear classifier on the multi-source real world cyber security data set to identify cyber-attacks and their sources.
- Description: Proceedings of the IEEE International Conference on Industrial Technology
- Cai, Xin-Qiang, Zhao, Peng, Ting, Kai-Ming, Mu, Xin, Jiang, Yuan
- Authors: Cai, Xin-Qiang , Zhao, Peng , Ting, Kai-Ming , Mu, Xin , Jiang, Yuan
- Date: 2019
- Type: Text , Conference proceedings
- Relation: 2019 IEEE International Conference on Data Mining (ICDM) , Beijing, China, 8-11 Nov. 2019 p. 970-975
- Full Text: false
- Reviewed:
- Description: This paper re-examines existing systems in streaming classification with emerging new classes (SENC) problems, where new classes that have not been used to train a classifier may emerge in a data stream. We identify that existing systems have an unspecified assumption that emerging new classes are geometrically far from known classes, or instances of known classes are densely distributed, in the feature space. Using a class separation indicator alpha, we refine the SENC problem into an alpha-SENC problem, where alpha indicates a geometric distance between two classes in the feature space. We show that while most existing systems work well in high-alpha SENC problems (i.e., a new class is geometrically far from a known class or instances of known classes are densely distributed), they perform poorly in low-alpha SENC problems. To solve low-alpha SENC problems effectively, we propose an approach using nearest neighbor ensembles or SENNE. We demonstrate that SENNE is able to handle both the low-alpha and high-alpha SENC problems which can appear at different times in a single data stream.
Pervasive blood pressure monitoring using Photoplethysmogram (PPG) sensor
- Riaz, Farhan, Azad, Muhammad, Arshad, Junaid, Imran, Muhammad, Hassan, Ali, Rehman, Saad
- Authors: Riaz, Farhan , Azad, Muhammad , Arshad, Junaid , Imran, Muhammad , Hassan, Ali , Rehman, Saad
- Date: 2019
- Type: Text , Journal article
- Relation: Future Generation Computer Systems Vol. 98, no. (2019), p. 120-130
- Full Text:
- Reviewed:
- Description: Preventive healthcare requires continuous monitoring of the blood pressure (BP) of patients, which is not feasible using conventional methods. Photoplethysmogram (PPG) signals can be effectively used for this purpose as there is a physiological relation between the pulse width and BP and can be easily acquired using a wearable PPG sensor. However, developing real-time algorithms for wearable technology is a significant challenge due to various conflicting requirements such as high accuracy, computationally constrained devices, and limited power supply. In this paper, we propose a novel feature set for continuous, real-time identification of abnormal BP. This feature set is obtained by identifying the peaks and valleys in a PPG signal (using a peak detection algorithm), followed by the calculation of rising time, falling time and peak-to-peak distance. The histograms of these times are calculated to form a feature set that can be used for classification of PPG signals into one of the two classes: normal or abnormal BP. No public dataset is available for such study and therefore a prototype is developed to collect PPG signals alongside BP measurements. The proposed feature set shows very good performance with an overall accuracy of approximately 95%. Although the proposed feature set is effective, the significance of individual features varies greatly (validated using significance testing) which led us to perform weighted voting of features for classification by performing autoregressive modeling. Our experiments show that the simplest linear classifiers produce very good results indicating the strength of the proposed feature set. The weighted voting improves the results significantly, producing an overall accuracy of about 98%. Conclusively, the PPG signals can be effectively used to identify BP, and the proposed feature set is efficient and computationally feasible for implementation on standalone devices. © 2019 Elsevier B.V.
- Authors: Riaz, Farhan , Azad, Muhammad , Arshad, Junaid , Imran, Muhammad , Hassan, Ali , Rehman, Saad
- Date: 2019
- Type: Text , Journal article
- Relation: Future Generation Computer Systems Vol. 98, no. (2019), p. 120-130
- Full Text:
- Reviewed:
- Description: Preventive healthcare requires continuous monitoring of the blood pressure (BP) of patients, which is not feasible using conventional methods. Photoplethysmogram (PPG) signals can be effectively used for this purpose as there is a physiological relation between the pulse width and BP and can be easily acquired using a wearable PPG sensor. However, developing real-time algorithms for wearable technology is a significant challenge due to various conflicting requirements such as high accuracy, computationally constrained devices, and limited power supply. In this paper, we propose a novel feature set for continuous, real-time identification of abnormal BP. This feature set is obtained by identifying the peaks and valleys in a PPG signal (using a peak detection algorithm), followed by the calculation of rising time, falling time and peak-to-peak distance. The histograms of these times are calculated to form a feature set that can be used for classification of PPG signals into one of the two classes: normal or abnormal BP. No public dataset is available for such study and therefore a prototype is developed to collect PPG signals alongside BP measurements. The proposed feature set shows very good performance with an overall accuracy of approximately 95%. Although the proposed feature set is effective, the significance of individual features varies greatly (validated using significance testing) which led us to perform weighted voting of features for classification by performing autoregressive modeling. Our experiments show that the simplest linear classifiers produce very good results indicating the strength of the proposed feature set. The weighted voting improves the results significantly, producing an overall accuracy of about 98%. Conclusively, the PPG signals can be effectively used to identify BP, and the proposed feature set is efficient and computationally feasible for implementation on standalone devices. © 2019 Elsevier B.V.
Analysis of Classifiers for Prediction of Type II Diabetes Mellitus
- Barhate, Rahul, Kulkarni, Pradnya
- Authors: Barhate, Rahul , Kulkarni, Pradnya
- Date: 2018
- Type: Text , Conference proceedings , Conference paper
- Relation: 4th International Conference on Computing, Communication Control and Automation, ICCUBEA 2018
- Full Text:
- Reviewed:
- Description: Diabetes mellitus is a chronic disease and a health challenge worldwide. According to the International Diabetes Federation, 451 million people across the globe have diabetes, with this number anticipated to rise up to 693 million people by 2045. It has been shown that 80% of the complications arising from type II diabetes can be prevented or delayed by early identification of the people who are at risk. Diabetes is difficult to diagnose in the early stages as its symptoms grow subtly and gradually. In a majority of the cases, the patients remain undiagnosed until they are admitted for a heart attack or begin to lose their sight. This paper analyzes the different classification algorithms based on a patient's health history to aid doctors identify the presence of as well as promote early diagnosis and treatment. The experiments were conducted on Pima Indian Diabetes data set. Various classifiers used include K Nearest Neighbors, Logistic Regression, Decision Trees, Random Forest, Gradient Boosting, Support Vector Machine and Neural Network. Results demonstrate that Random Forests performed well on the data set giving an accuracy of 79.7%. © 2018 IEEE.
- Description: E1
- Authors: Barhate, Rahul , Kulkarni, Pradnya
- Date: 2018
- Type: Text , Conference proceedings , Conference paper
- Relation: 4th International Conference on Computing, Communication Control and Automation, ICCUBEA 2018
- Full Text:
- Reviewed:
- Description: Diabetes mellitus is a chronic disease and a health challenge worldwide. According to the International Diabetes Federation, 451 million people across the globe have diabetes, with this number anticipated to rise up to 693 million people by 2045. It has been shown that 80% of the complications arising from type II diabetes can be prevented or delayed by early identification of the people who are at risk. Diabetes is difficult to diagnose in the early stages as its symptoms grow subtly and gradually. In a majority of the cases, the patients remain undiagnosed until they are admitted for a heart attack or begin to lose their sight. This paper analyzes the different classification algorithms based on a patient's health history to aid doctors identify the presence of as well as promote early diagnosis and treatment. The experiments were conducted on Pima Indian Diabetes data set. Various classifiers used include K Nearest Neighbors, Logistic Regression, Decision Trees, Random Forest, Gradient Boosting, Support Vector Machine and Neural Network. Results demonstrate that Random Forests performed well on the data set giving an accuracy of 79.7%. © 2018 IEEE.
- Description: E1
Effective and efficient kernel-based image representations for classification and retrieval
- Authors: Karmakar, Priyabrata
- Date: 2018
- Type: Text , Thesis , PhD
- Full Text:
- Description: Image representation is a challenging task. In particular, in order to obtain better performances in different image processing applications such as video surveillance, autonomous driving, crime scene detection and automatic inspection, effective and efficient image representation is a fundamental need. The performance of these applications usually depends on how accurately images are classified into their corresponding groups or how precisely relevant images are retrieved from a database based on a query. Accuracy in image classification and precision in image retrieval depend on the effectiveness of image representation. Existing image representation methods have some limitations. For example, spatial pyramid matching, which is a popular method incorporating spatial information in image-level representation, has not been fully studied to date. In addition, the strengths of pyramid match kernel and spatial pyramid matching are not combined for better image matching. Kernel descriptors based on gradient, colour and shape overcome the limitations of histogram-based descriptors, but suffer from information loss, noise effects and high computational complexity. Furthermore, the combined performance of kernel descriptors has limitations related to computational complexity, higher dimensionality and lower effectiveness. Moreover, the potential of a global texture descriptor which is based on human visual perception has not been fully explored to date. Therefore, in this research project, kernel-based effective and efficient image representation methods are proposed to address the above limitations. An enhancement is made to spatial pyramid matching in terms of improved rotation invariance. This is done by investigating different partitioning schemes suitable to achieve rotation-invariant image representation and the proposal of a weight function for appropriate level contribution in image matching. In addition, the strengths of pyramid match kernel and spatial pyramid are combined to enhance matching accuracy between images. The existing kernel descriptors are modified and improved to achieve greater effectiveness, minimum noise effects, less dimensionality and lower computational complexity. A novel fusion approach is also proposed to combine the information related to all pixel attributes, before the descriptor extraction stage. Existing kernel descriptors are based only on gradient, colour and shape information. In this research project, a texture-based kernel descriptor is proposed by modifying an existing popular global texture descriptor. Finally, all the contributions are evaluated in an integrated system. The performances of the proposed methods are qualitatively and quantitatively evaluated on two to four different publicly available image databases. The experimental results show that the proposed methods are more effective and efficient in image representation than existing benchmark methods.
- Description: Doctor of Philosophy
- Authors: Karmakar, Priyabrata
- Date: 2018
- Type: Text , Thesis , PhD
- Full Text:
- Description: Image representation is a challenging task. In particular, in order to obtain better performances in different image processing applications such as video surveillance, autonomous driving, crime scene detection and automatic inspection, effective and efficient image representation is a fundamental need. The performance of these applications usually depends on how accurately images are classified into their corresponding groups or how precisely relevant images are retrieved from a database based on a query. Accuracy in image classification and precision in image retrieval depend on the effectiveness of image representation. Existing image representation methods have some limitations. For example, spatial pyramid matching, which is a popular method incorporating spatial information in image-level representation, has not been fully studied to date. In addition, the strengths of pyramid match kernel and spatial pyramid matching are not combined for better image matching. Kernel descriptors based on gradient, colour and shape overcome the limitations of histogram-based descriptors, but suffer from information loss, noise effects and high computational complexity. Furthermore, the combined performance of kernel descriptors has limitations related to computational complexity, higher dimensionality and lower effectiveness. Moreover, the potential of a global texture descriptor which is based on human visual perception has not been fully explored to date. Therefore, in this research project, kernel-based effective and efficient image representation methods are proposed to address the above limitations. An enhancement is made to spatial pyramid matching in terms of improved rotation invariance. This is done by investigating different partitioning schemes suitable to achieve rotation-invariant image representation and the proposal of a weight function for appropriate level contribution in image matching. In addition, the strengths of pyramid match kernel and spatial pyramid are combined to enhance matching accuracy between images. The existing kernel descriptors are modified and improved to achieve greater effectiveness, minimum noise effects, less dimensionality and lower computational complexity. A novel fusion approach is also proposed to combine the information related to all pixel attributes, before the descriptor extraction stage. Existing kernel descriptors are based only on gradient, colour and shape information. In this research project, a texture-based kernel descriptor is proposed by modifying an existing popular global texture descriptor. Finally, all the contributions are evaluated in an integrated system. The performances of the proposed methods are qualitatively and quantitatively evaluated on two to four different publicly available image databases. The experimental results show that the proposed methods are more effective and efficient in image representation than existing benchmark methods.
- Description: Doctor of Philosophy
Fog computing: Survey of trends, architectures, requirements, and research directions
- Naha, Ranesh, Garg, Saurabh, Georgakopoulos, Dimitrios, Jayaraman, Prem, Gao, Longxiang, Xiang, Yong, Ranjan, Rajiv
- Authors: Naha, Ranesh , Garg, Saurabh , Georgakopoulos, Dimitrios , Jayaraman, Prem , Gao, Longxiang , Xiang, Yong , Ranjan, Rajiv
- Date: 2018
- Type: Text , Journal article
- Relation: IEEE access Vol. 6, no. (2018), p. 47980-48009
- Full Text: false
- Reviewed:
- Description: Emerging technologies such as the Internet of Things (IoT) require latency-aware computation for real-time application processing. In IoT environments, connected things generate a huge amount of data, which are generally referred to as big data. Data generated from IoT devices are generally processed in a cloud infrastructure because of the on-demand services and scalability features of the cloud computing paradigm. However, processing IoT application requests on the cloud exclusively is not an efficient solution for some IoT applications, especially time-sensitive ones. To address this issue, Fog computing, which resides in between cloud and IoT devices, was proposed. In general, in the Fog computing environment, IoT devices are connected to Fog devices. These Fog devices are located in close proximity to users and are responsible for intermediate computation and storage. One of the key challenges in running IoT applications in a Fog computing environment are resource allocation and task scheduling. Fog computing research is still in its infancy, and taxonomy-based investigation into the requirements of Fog infrastructure, platform, and applications mapped to current research is still required. This survey will help the industry and research community synthesize and identify the requirements for Fog computing. This paper starts with an overview of Fog computing in which the definition of Fog computing, research trends, and the technical differences between Fog and cloud are reviewed. Then, we investigate numerous proposed Fog computing architectures and describe the components of these architectures in detail. From this, the role of each component will be defined, which will help in the deployment of Fog computing. Next, a taxonomy of Fog computing is proposed by considering the requirements of the Fog computing paradigm. We also discuss existing research works and gaps in resource allocation and scheduling, fault tolerance, simulation tools, and Fog-based microservices. Finally, by addressing the limitations of current research works, we present some open issues, which will determine the future research direction for the Fog computing paradigm.
Improving deep forest by confidence screening
- Pang, Ming, Ting, Kaiming, Zhao, Peng, Zhou, Zhi-Hua
- Authors: Pang, Ming , Ting, Kaiming , Zhao, Peng , Zhou, Zhi-Hua
- Date: 2018
- Type: Text , Conference proceedings
- Relation: 2018 Ieee International Conference on Data Mining; Singapore, Singapore; 17th-20th November 2018 p. 1194-1199
- Full Text:
- Reviewed:
- Description: Most studies about deep learning are based on neural network models, where many layers of parameterized nonlinear differentiable modules are trained by backpropagation. Recently, it has been shown that deep learning can also be realized by non-differentiable modules without backpropagation training called deep forest. The developed representation learning process is based on a cascade of cascades of decision tree forests, where the high memory requirement and the high time cost inhibit the training of large models. In this paper, we propose a simple yet effective approach to improve the efficiency of deep forest. The key idea is to pass the instances with high confidence directly to the final stage rather than passing through all the levels. We also provide a theoretical analysis suggesting a means to vary the model complexity from low to high as the level increases in the cascade, which further reduces the memory requirement and time cost. Our experiments show that the proposed approach achieves highly competitive predictive performance with significantly reduced time cost and memory requirement by up to one order of magnitude.
- Authors: Pang, Ming , Ting, Kaiming , Zhao, Peng , Zhou, Zhi-Hua
- Date: 2018
- Type: Text , Conference proceedings
- Relation: 2018 Ieee International Conference on Data Mining; Singapore, Singapore; 17th-20th November 2018 p. 1194-1199
- Full Text:
- Reviewed:
- Description: Most studies about deep learning are based on neural network models, where many layers of parameterized nonlinear differentiable modules are trained by backpropagation. Recently, it has been shown that deep learning can also be realized by non-differentiable modules without backpropagation training called deep forest. The developed representation learning process is based on a cascade of cascades of decision tree forests, where the high memory requirement and the high time cost inhibit the training of large models. In this paper, we propose a simple yet effective approach to improve the efficiency of deep forest. The key idea is to pass the instances with high confidence directly to the final stage rather than passing through all the levels. We also provide a theoretical analysis suggesting a means to vary the model complexity from low to high as the level increases in the cascade, which further reduces the memory requirement and time cost. Our experiments show that the proposed approach achieves highly competitive predictive performance with significantly reduced time cost and memory requirement by up to one order of magnitude.
Large scale modeling of genetic networks using gene knockout data
- Youseph, Ahammed, Chetty, Madhu, Karmakar, Gour
- Authors: Youseph, Ahammed , Chetty, Madhu , Karmakar, Gour
- Date: 2018
- Type: Text , Conference proceedings
- Relation: 2018 Australasian Computer Science Week Multiconference, ACSW 2018; Brisbane, Australia; 29th January-2nd February 2018; published in ACM International Conference Proceedings Series
- Full Text: false
- Reviewed:
- Description: Gene regulatory network (GRN) represents a set of genes and their regulatory interactions. The inference of the regulatory interactions between genes is usually carried out as an optimization problem using an appropriate mathematical model and the time-series gene expression data. Among the various models proposed for GRN inference, our recently proposed Michaelis-Menten kinetics based ODE model provides a good trade-off between the computational complexity and biological relevance. This model, like other known GRN models, also uses an evolutionary algorithm for parameter estimation. Since the search space for large networks is huge, leading to a low accuracy of inference, it is important to reduce the search region for improved performance of the optimization algorithm. In this paper, we propose a classification method using gene knockout data to eliminate a large infeasible region from the optimization search area. We also propose a method for partial inference of regulations when all the regulators of a given regulated gene are unregulated genes. The proposed method is evaluated by reconstructing in silico networks of large sizes. © 2018 ACM.