Yanzhen Qu | Colorado Technical University (original) (raw)

Papers by Yanzhen Qu

2019 International Conference on Computational Science and Computational Intelligence (CSCI)

In this paper, we aim to demonstrate a practical mechanism to determine the boundary which can gu... more In this paper, we aim to demonstrate a practical mechanism to determine the boundary which can guide the design and implementation of privacy protection through perturbation. We have leveraged the strategy of detecting adversarial examples through a set of detection methods to find the "blind corners" of detection, and use them as the guidance of design and implementation of perturbation for privacy protection. To ensure the confidence of this approach, we have created a detection mechanism which has a very high accuracy and precision. First we use a classical statistical method to detect most of adversarial examples. Then, in considering that small examples sets will impact the confidence of statistical tests, we applied second detection method to discover the adversarial examples escaped from the first round detection by comparing similarity of the loss curves of training on the original data and tested data. Experiment results have shown that our enhanced detection mechanism not only extends the capability of detecting more adversarial examples in a much broad scope of perturbation, but also provides us the set of boundaries, represented by the values of a set of thresholds of exploring the "blind corners" of the detection methods, which can be practically used to effectively direct the design and implementation of perturbation for privacy protection. This is also the main contribution of this paper.

2021 International Conference on Computational Science and Computational Intelligence (CSCI)

European Journal of Electrical Engineering and Computer Science

The Time Series Classification Residual Network (TSCResNet) deep learning model is introduced in ... more The Time Series Classification Residual Network (TSCResNet) deep learning model is introduced in this study to improve the classification performance in the human activity recognition (HAR) problem. Specifically in the context of closed-set classification where all labels or classes are present during the model training phase. This contrasts with open-set classification where new, unseen activities are introduced to the HAR system after the training phase. The proposed TSCResNet model is evaluated with the benchmark PAMAP2 Physical Activity Monitoring Dataset. By using the same quantitative methods and data preprocessing protocols as previous research in the field of closed-set HAR, results show that the TSCResNet model architecture is able to achieve improved classification results across accuracy, the weighted F1-score, and mean F1-score.

Data warehousing is an important part of the enterprise information system. Business intelligence... more Data warehousing is an important part of the enterprise information system. Business intelligence (BI) relies on data warehouses to improve business performance. Data quality plays a key role in BI. Source data is extracted, transformed, and loaded (ETL) into the data warehouses periodically. The ETL operations have the most crucial impact on the data quality of the data warehouse. ETL-related data warehouse architectures including structure-oriented layer architectures and enterpriseview data mart architecture were studied in the literature. Existing architectures have the layer and data mart components but do not make use of design patterns; thus, those approaches are inefficient and pose potential problems. This paper relays how to use design patterns to improve data warehouse architectures.

2020 International Conference on Computational Science and Computational Intelligence (CSCI), 2020

With the continued expansion and adoption of multi-factor authentication technologies within orga... more With the continued expansion and adoption of multi-factor authentication technologies within organizations across industries, there exists a need to determine the relationship between composite vulnerability exposures and the multiplicity of authentication factors. To meet this need, we propose a power curve formula for the relationship between these variables, demonstrating the wellness of fit to generalized data sets. We provide an extension to the Common Vulnerability Scoring System (CVSS) v3 calculator, allowing the characterization of combined Common Vulnerability Exposure (CVE) in an objective and repeatable means, demonstrating the capabilities of that extent. This paper explores the potential for future work derived from the proposed mathematical formula representation of the relationship between authentication factor multiplicity and composite vulnerability scores.

2017 3rd IEEE International Conference on Computer and Communications (ICCC), 2017

The development of affective computing system for soaring flight simulation training requires a m... more The development of affective computing system for soaring flight simulation training requires a mechanism for determining pilot affective patterns. These patterns may be comprised of multiple elements, including emotion, pilot performance level, and physiological responses. This article proposes an approach to quantify these elements during the performance of flight training maneuvers. To validate this, a sample population of pilots performed flight tasks in a soaring flight simulator while wearing physiological sensors. Pilots reported subjective emotional stress using a 10-point numeric scale. A 5-point numeric scale was developed to transform flight task performance criteria into a unified scale. The use of common quantitative units for emotion and performance provides a simple mechanism for computational analysis and comparison of values that would otherwise be either qualitative (i.e. emotion) or measured in different units (i.e. flight tasks). The research provides a foundatio...

2020 International Conference on Computational Science and Computational Intelligence (CSCI), 2020

Subsequent to the data deluge of the internet era and the recent advancement in big data technolo... more Subsequent to the data deluge of the internet era and the recent advancement in big data technologies, it is easy to affirm the continuous application of such technological innovation to tackling a wide array of students' educational needs. The field of artificial intelligence and machine learning have improved education learning outcomes. However, the problem of generalized traditional supportive collaboration scripts for all students irrespective of the student's learning traits and position on the learning spectrum leads to less than optimum result in their educational pursuits. This paper presents a novel approach that uses data mining algorithm to optimize the selection of educational resources for students based on their learning traits and the six factors that cofound instructional content and delivery with a focus on students with learning disabilities for STEM subjects.

International Journal of Computer Networks & Communications, 2019

European Journal of Electrical Engineering and Computer Science

A large amount of data is available on Twitter that can be used to manage different types of risk... more A large amount of data is available on Twitter that can be used to manage different types of risks in financial institutions. This paper shows how machine learning algorithms can be applied to analyze large unstructured data and train a model to make a future prediction on tweets to categorize them by risk type and use sentiment analysis to understand the risk type. This model reads each tweet and categorizes them by risk using a specified dictionary and adds sentiment analysis to show the risk type seen in each tweet. Logistic regression used in this research helped to formulate the prediction model. Twitter data from 2019 was used to train and test a supervised machine learning algorithm and once the model started predicting tweets efficiently, it was used to predict twitter data from 2022 in our experimental research. Our experiment confirmed that Twitter data can be used to manage risk with the right type of modeling using machine learning techniques.

2016 IEEE 2nd International Conference on Big Data Security on Cloud (BigDataSecurity), IEEE International Conference on High Performance and Smart Computing (HPSC), and IEEE International Conference on Intelligent Data and Security (IDS), 2016

With the materialization of the internet of things (IoT), big data analytic and cloud computing s... more With the materialization of the internet of things (IoT), big data analytic and cloud computing services give rise to extra breadth in the assessment of more secure computing environments, better resource management and vulnerability analysis. In order to accurately assess the vulnerability of Bluetooth low energy (BLE) wireless network enabled IoT systems, we have proposed a novel approach to extend the calculation formula for Authentication which is one of variables used in the conventional base score equations of the Common Vulnerability Scoring System (CVSS) v2 proposed by the National Infrastructure Advisory Council. Through an example BLE wireless network based shopping cart IoT system we have demonstrated the weakness of the current CVSS v2 base score equations and how to overcome the weakness through our extension.

2020 International Conference on Computational Science and Computational Intelligence (CSCI)

In recent years, machine learning algorithms have been more and more used in healthcare industry,... more In recent years, machine learning algorithms have been more and more used in healthcare industry, especially in research areas involving human participants such as clinical trials and areas where the data is too expensive to collect. This research project has conducted a comparative study on three well-known machine learning methods: Logistic Regression (LR), Support Vector Machine (SVM), and Naïve Bayes (NB) against the same dataset for predicting breast cancer prognosis in improving clinical trials. The experiment results have provided a comprehensive view of the patient’s risk levels and risk factors to clinicians that benefit in effective and efficient treatment. This research has also demonstrated that different machine learning algorithms against the same dataset for breast cancer prognosis can have a difference in both performance and accuracy. Therefore, the comparative study on different machine learning algorithms can identify the most suitable machine learning algorithm to achieve cost-effective clinical trials.

2020 International Conference on Computational Science and Computational Intelligence (CSCI), 2020

Traditional customer payment service scheduling approaches cannot cope with the modern demand for... more Traditional customer payment service scheduling approaches cannot cope with the modern demand for timely, high-quality service due to the disruption of big data within small and medium-sized payment solution providers (SaMS-PSP). While many customers have access to modern technologies to lodge their service requests easily and fast, SaMS-PSPs do not have equally automated big data-driven capabilities to handle the growing demands of these service requests. To effectively improve SaMS-PSP’s customer payment service requests processing speeds, personnel optimization, throughput, and low latency scheduling, we have developed a new customer payment service request scheduling algorithm via matching request priority with the best personnel to handle the request based on data analytics through machine learning. Our experiments and testing have confirmed the merits of this new algorithm. We are also in the process of applying this new algorithm in real-world payment operations.

2019 IEEE Frontiers in Education Conference (FIE), 2019

In recent years, to meet the growing needs of upgrading computer-related competence of working ad... more In recent years, to meet the growing needs of upgrading computer-related competence of working adults, many online Bachelor of Science of Computer Science (BSCS) or Bachelor of Science of Information Technology (BSIT) degree programs have been created. However, how to effectively support these online students is still a challenge. This is due to special needs such as customizing the learning to individual backgrounds, supporting various programming and technical hands-on labs, and promoting teamwork engagement, etc. We have conducted a project integrated three initiatives to create an Online Learning Environment (OLE) suited to BSCS and BSIT degree programs: (a) using the adaptive learning software to customize the learning content based on each student’s existing academic foundation; (b) using Amazon’s Web Services to set up the hands-on labs; and (c) using discussion board assignments to promote the engagement of student-student and student-faculty, to train students to be compete...

2020 International Conference on Computational Science and Computational Intelligence (CSCI), 2020

With the development of digital image processing technology based on deep learning, the potential... more With the development of digital image processing technology based on deep learning, the potential risk of using related technologies to threaten the security of multimedia information is increasing. Because the generated human face effect largely depends on the completeness of the input sample set, most of the current deep forgery models have the problem of human side-face collapse. This paper has studied the deep forgery technology of Deepfacelab and Faceswap, and adjusts the original auto-encoder-based model architecture to a generative adversarial network. By using the harmonic mean of cross entropy and mean square error as the loss function, the improved model can reduce the probability of some frames being discarded during training. Meanwhile, by adjusting key characteristics and the weights of features in different frames, it further optimizes the cross-dataset detection performance. Experimental results have shown that the improved model can keep more facial details while sti...

2017 3rd IEEE International Conference on Computer and Communications (ICCC), 2017

Among all the popular V-words such as volume, velocity, and variety, etc. that describe the prope... more Among all the popular V-words such as volume, velocity, and variety, etc. that describe the properties or dimensions of the big data, both volume and velocity can be measured, however so far variety has not found a way to become quantifiable. Although, it has been used to capture the most important aspect of the big data — the complexity of the big data caused by various relationships embedded in the data. This paper has applied Kolmogorov's Complexity Theory as the foundation to develop a practical measurement to quantify the impact of the variety of a big data set D towards a specific processing target r, i is an integer. Using the Inverse Compression Ratio defined by comparing the complexity of and the complexity of entire Relationships set R of the D, where R is the abstraction of the variety of the D, and r, belong to R. This measurement can help computing resource planning for conducting big data analytics.

The deficiency in the ability for instructors in the Cloud based E-Learning environments to accur... more The deficiency in the ability for instructors in the Cloud based E-Learning environments to accurately determine a student’s affective status has resulted in the inability to provide effective feedback to student. Feedback is important in learning as it allows a student to learn from their mistakes and helps build their academic confidence. In this paper, we have proposed an affective based E-Learning framework that uses fuzzy logic and emoticons to determine a student’s affective status in a Cloud-based E-Learning Environment. This framework uses three emotions represented using emoticons, which are “excited”, “tired”, and “sad” to accurately detect a student’s emotion during their learning process.

2020 International Conference on Computational Science and Computational Intelligence (CSCI), 2020

Both the industry and academia agree on the immense contribution of big data analytics and machin... more Both the industry and academia agree on the immense contribution of big data analytics and machine learning to competitive businesses. The payment industry would benefit from big data analytics and machine learning capabilities to harness their customers’ opinions through sentiment analysis, thereby customizing their services and products to fit their customers’ preferences. However, the challenge is implementing this competitive edge in small and medium-sized payment solution providers. This paper proposes a deep learning-based customer sentiment analysis model and a related (SaMS-PSP) algorithm that implements sentiment analysis within SaMS-PSP. Through experiments, we have demonstrated that our model has a super performance advantage over conventional machine learning methods and is more suited to handle "big data" applications such as customer sentiment analysis. This research has demonstrated that the sentiment analysis emotional polarity score can be used in a value-...

The internet age has led to an explosion of technologies that has made digital copying of media r... more The internet age has led to an explosion of technologies that has made digital copying of media relatively easy, compared to the cost of production, often leading to widespread piracy. One mitigation to this problem is through watermarking. An ideal watermark should be invisible and robust, so it can be left behind on the host image as a poison pill, possibly laced with collateral to deter distribution. Common watermarking techniques often rely on spatial domain embedding, which often requires difficult tradeoffs between robustness and visibility. The closer the watermark to the center, the more robust, but also the more conspicuous. The closer to the margin, the easier to crop off without requiring major tradeoffs in viewing quality. The visibility also attracts the attention of many would be attackers. To ensure invisibility and robustness, this paper will focus on frequency domain embedding and robust recovery of attacked image by using deep residual learning, then explore the re...

Parallel genetic algorithms have been used to optimize many functions in science and engineering.... more Parallel genetic algorithms have been used to optimize many functions in science and engineering. However, they have not received wide acceptance due to their complex and often inefficient configuration requirements. Parallel genetic algorithms have been ported to general purpose graphical processing units on a limited basis, again due to the difficulty in configuration. Most applications of parallel genetic algorithms on graphical processing units take only computational performance into account. This is in contrast with tradition parallel genetic algorithm work that takes only convergence into consideration. Each genre has a strong following but there is no combined algorithm that takes both convergence and performance into account. This paper will present a method to determine efficient deme size and deme number that can be calculated during runtime with no human interaction. It takes convergence theory methodology as the basis to center the deme size and deme number to a well co...

While multifactor authentication technologies continue to advance and adoption rates for those te... more While multifactor authentication technologies continue to advance and adoption rates for those technologies increase, there exists a need to characterize the composite vulnerability score for complete authentication solutions. To meet this need, we propose an extension to the Common Vulnerability Scoring System (CVSS) v3 calculator to provide an aggregate score for any metric category, enabling organizations and researchers to succinctly determine the composite vulnerability impact of authentication factor multiplicity. This chapter has presented a novel mathematical approach and demonstrated the approach through a real-world application which is a comparative study on the composite vulnerability of two different multifactor authentication technologies.