EECS Graduate Student Conference Papers

Permanent URI for this collection


Recent Submissions

Now showing 1 - 5 of 106
  • Item
    Understanding spatial constraints for autonomous robotic assembly with neural logic learning
    (Wichita State University, 2021-04-02) Yan, Fujian; He, Hongsheng
    Spatial constraints of objects are one of the key elements that are required in industrial assembly. Robots deployed in conventional assembly lines are based on schema by referring to computer-aided design (CAD) software. Spatial constraints are modeled by computer-aided design (CAD) software. Compared with conventional assembly lines, autonomous robotic assembly requires robots to learn spatial constraints intelligently. Therefore, understanding spatial constraints are critical for autonomous robotic assembly. This work proposed a method to address the critical need of enabling robots to comprehend spatial constraints with a single RGB-D scan. The proposed method contains two parts: the first one generates 3D models to fulfill the missing point-cloud of a single RGB-D scan of objects with an extended generative adversary network (GAN). The second part enables robots to comprehended spatial constraints with a neural-logic network. The spatial constraints include left, right, above, below, front, behind, parallel, perpendicular, concentric, and coincident. The 3D composition model achieved 57.23% intersection over union (IoU), and the neural logic model that can learn spatial constraints achieved over 99% in comprehending all spatial constraints.
  • Item
    Shapley value-based satellite communication bandwidth allocation strategy
    (Wichita State University, 2021-04-02) Yarlagadda, Maha Lakshmi; Watkins, John Michael; Sawan, M. Edwin; Lakshmikanth, Geethalakshmi S.
    Satellite technology is essential in the field of communications and is popularly called SATCOM, short for Satellite Communication. Bandwidth and connectivity needs are the two main expensive resources in SATCOM that are constantly changing and therefore need to be efficiently managed. An expensive resource that we manage every day is money. Needs and priorities where this money can be spent are constantly changing in most of our lives. Budgeting and money management are very essential to achieve financial peace and freedom. Likewise, in SATCOM we use resource managers (like financial advisors) to adaptively adjust the bandwidth based on needs and priorities. The available research is very low on a combination of demand-driven resource allocation and satellite communication based on control theory concepts. Our goal is to support SATCOM network operation centers by proposing a very systematic framework that integrates the available controller design architectures (Linear Quadratic Regulator LQR) with a fair and adaptive resource allocation algorithm (Shapley Value-based Algorithm). This is achieved by following the rules of co-operative game theory where the competitors work in the direction of winning by forming coalitions or groups. Figure 1 shows a satellite network with Remote Terminals (RT), who are the players. When the sum of requested data rates of some of the RTs passes the threshold of the majority level, the formation of a winning coalition is enabled and this will help to evaluate the resource to be allocated to each player. The Shapley Value-based resource allocation establishes socially fair and high standard bandwidth allocation for each remote terminal and the average fairness of the complete system is high. The Satellite System Controller (SSC) is like the Money manager who decides what proportion of the available budget gets assigned based on need and priority while making sure that the quality of communication is not compromised, in any way. Mathematical modeling and simulation of the entire SATCOM network along with Shapley Value-based resource allocation are carried out using MATLAB.
  • Item
    Federated learning with differential privacy: An online mirror descent approach
    (Wichita State University, 2021-04-02) Odeyomi, Olusola T.; Záruba, Gergely V.
    Federated learning is a machine learning paradigm that provides privacy to the local data of multiple clients communicating with a central server. Federated learning has successfully been applied in natural language processing, vehicle-to-vehicle communication, social networks, healthcare predictions through medical collaborations, wireless sensor networks etc. Although federated learning has gained much research attention in recent times, there are still some open problems. Existing research work assumes that the training data of the clients are time invariant. This assumption does not hold in real-time traffic monitoring, where events occur randomly. Thus, an online learning approach must be introduced into federated learning algorithms to capture the randomness of the clients' training data. Another open problem is improving communication and computations by removing the central server in the federated learning design. More so, the model updates at the central server may be prone to adversarial attacks. Such adversarial attacks may put the clients' local data in a potential privacy risk. Lastly, the stochastic gradient algorithms commonly used for federated learning do not fully exploit the convexity of the loss functions of the clients. This work proposes an online mirror descent learning algorithm that can handle time-varying data in a decentralized federated learning setting. To provide additional privacy, local differential privacy is introduced to the setting. The convergence of the proposed algorithm is compared to some state-of-the-art federated learning algorithms. The proposed algorithm shows to converge faster to the global model for all the clients.
  • Item
    Obesity classification from facial images using deep learning
    (Wichita State University, 2021-04-02) Siddiqui, Hera; Siddiqui, Hera; Rattani, Ajita; Cure Vellojin, Laila N.; Woods, Nikki Keene; Lewis, Rhonda K.; Twomey, Janet M.; Smith-Campbell, Betty; Hill, Twyla J.
    INTRODUCTION: Obesity is a serious health problem that is on the rise both in the United States and globally. Obesity is frequently defined using the clinical Body Mass Index (BMI) ratio of height and weight. Overweight individuals have a BMI between 25-30, and those over 30 are classified as obese. Obesity can lead to heart disease, type 2 diabetes, and many other serious health conditions. Self-diagnostic face-based solutions are being investigated for obesity classification and monitoring. PURPOSE: To classify obesity status based on facial images using deep learning-based convolutional neural networks (CNNs). METHODS: The four CNNs (VGG16, ResNet50, DenseNet121, and MobileNetV2) used in this study were pre-trained on three public datasets (ImageNet, VGGFace, and VGGFace2). Using the above CNNs, we extracted deep features from the FIW-BMI and VisualBMI datasets annotated with BMI information. The deep features from 8298 images in the FIW-BMI dataset along with BMI values were then used to train a Support Vector Classification (SVC) classifier. The trained SVC model was tested on 4206 different images from the VisualBMI dataset for the validation. RESULTS: CNNs trained on ImageNet dataset obtained an initial accuracy (percentage of correct obese and non-obese classifications) in the range 64% to 72%. Accuracy of 84% to 86% was obtained by using CNNs trained on VGGFace dataset. 86% accuracy was obtained by concatenating features from pre-trained (VGGFace) and fine-tuned (FIW-BMI) model. ResNet-50 trained on VGGFace2 dataset obtained an accuracy of 91% when features from the original image datasets were used and 92% accuracy when features were fused from the original image with the horizontally flipped image. The fused image modifications resulted in a model with Sensitivity, Specificity, and Precision of 0.90, 0.94, and 0.95, respectively. Mean Absolute Error (MAE) of this model in predicting BMI is 3.16 and area under the curve (AUC) is 0.97. CONCLUSION: Obesity can be predicted from facial images using deep learning models with a promising accuracy. SVC models trained on deep features extracted from models pre-trained on VGGFace2 dataset performed better than models pre-trained on ImageNet dataset. ResNet-50 (pre-trained on VGGFace2) obtained the highest accuracy of 92% by combining features from the original image and horizontally flipped image. These models when deployed on smartphones can help individuals in monitoring their obesity status, BMI, and weight changes.
  • Item
    Visualization analysis of OSPTE comments
    (Wichita State University, 2021-04-02) Reddy, Muskula Sai Deep; Abdinnour, Sue
    The teaching evaluation instrument used by Wichita State University for Face-to-Face classes is called SPTE (Student Perception Teaching Evaluation). When WSU moved to online teaching, the social science lab at WSU developed the OSPTE (O stands for Online classes) that was administered online. Students rated the teachers and typed in their comments, as opposed to writing them by hand with SPTE. The typed comments are a source of data that can be analyzed and visualized to give faculty feedback on their teaching. We developed an online app that can do just that. Once we receive the paper packet of OPSTE evaluations for a course, we scan the comments pages as an image into a PDF file. We then take the PDF file and convert it to a Word document using an online tool (we provide a tutorial on how to do that safely). We then strip the file of all headers, footers, titles, and question statements i.e., we just keep the student comments then save the file as a TEXT file. We go to the online application and drag/drop the text file to the appropriate box or use the Select Files option to browse and upload the file. A sentiment analysis of the comments is performed in the background and the faculty will get a visualization dashboard displayed on the webpage. This includes a word cloud, count of popular words in comments, and count of positive and negative words in the comments. This app is built using python which takes the text file that contains students comments as input and uses available packages in python such as nltk package to remove stop words, to generate output visualization we use word cloud, ploty and matplotlib packages and finally dash and flask packages are used to display output on dashboard. The app design and functionality will be discussed, as well as future work.