Chronic Wasting Disease (CWD) Surveys

Survey Research

Use this area to describe your project.

• Date: April 2017
• Category: Missouri Department of Conservation(MDC)

Bowl Hunting

App Design

Short Description: We developed a mobile application for Android and iOS devices to collect survey data for hunters being interviewed by MDC.

Description: On-site clerks would take surveys of hunters in the Missouri area, write down all the responses on paper, and deliver it to the MDC data entry personnel, who would then have to type the collected survey data onto a database. Thus, MDC were looking for a time-saving solution to their problems. Since MDC had both Android and iOS devices, the application was aimed to be delivered quickly and for both the platforms. The project ran for 6 months, which was enough time for us to provide a working prototype of the application.

• Date: April 2017
• Category: Missouri Department of Conservation(MDC)
• People: Nishant Chettri, William Morrison, Nickolas Wergeles

Generic Survey

App Design

Use this area to describe your project.

• Date: April 2017
• Category: Missouri Department of Conservation(MDC)

Creel

App Design

Short Description: We developed a mobile application for iPads to collect survey data for fishing activities in and around Missouri waters.

Description: MDC conducts regular surveys on Missouri lakes and rivers. With clerks going out and taking surveys on paper and bringing it back to MDC to be entered into the database at a later date, it takes almost 6 months for the data to be fully uploaded on the site. Further verification from MDC adds even more time. To save time in this process MDC asked us to create a mobile application for their iPads. This application was based loosely on the Shooting Range Application (Android) developed previously by the lab for MDC. The application takes the entire survey and uploads the data to a remote server, which is hosted on AWS. This application has an added website component, which allows MDC to view, edit and verify the data being uploaded in real-time. The combination of the application and website has substantially reduced the amount of time and money it takes MDC to carry out its surveys.

• Date: April 2017
• Category: Missouri Department of Conservation(MDC)
• People: Nishant Chettri, William Morrison, Nickolas Wergeles, Zeshan Peng

Animal trajectories modeling and analysis

AI research

Use this area to describe your project.

• Date: April 2017
• Category: Missouri Department of Conservation(MDC)

Tigeraware

App Design

Short Description: A fully customizable iOS data collection application in real-time without modifying the code.

Description: This project was built using iOS Swift Development. The project was built for the Psychology Department. We first build an Android application for Alcohol Craving Studies to collect user survey and physiological data. They needed another data collection application for Mood Dysregulation and we had to build an entirely separate application. They needed a third application for St. Louis University for HIV studies and we had to build an entirely independent application. We then became wise and decided we should develop a generic data collection application where the researcher can develop the data that is collected from scratch, e.g. a survey. If they run the survey builder, then decide which users will take the survey, and launch the survey, then the survey will automatically be pushed to all the users. Then if they modify the survey and relaunch, the survey will but updated on all the devices without modifying and re-compiling the code. If they create a new survey, they could develop it from our custom online portal and launch it, without changing any of the code on the phone. Therefore, data collection for this application is customizable in real-time. We also have a dashboard to view the data with very nice data visualization, e.g. graphs. On the dashboard, administers can create and verify users, build surveys, view the results in real-time, edit surveys, and much more.

• Date: fall 2017
• Category: Psychology Department
• People: Will, Jayanth, Luke, Gaofei, Patrik, Nishant, Nick, Connor, Siyang, Weiliang

Modeling Voice Data with Machine Learning Algorithms

Machine Learning Research

Short Description: Predicting whether a sales conference call was successful selling a product based off voice and words exchanged between the seller and buyer.

Description: The business school had 3000+ phone calls where a business is trying to be successful in obtaining a meeting after the phone call to sell a product. Each phone call is a record. We take these records and segment them into pieces by speaking turns. From each segment, we extract acoustic features, like speech rate, number of pauses, decibels, etc. We use these features as input to traditional machine learning models and try to generate a prediction model based on them. Then we compare this with results from deep learning, which is the state-of-the-art predictive technology in recent years. With deep learning, we build a deep neural network with same features and let the computer determine the most important features and their inner connections. We also try to transform each segment of audio files into a spectrogram and train an image classification model based on the spectrograms. In the future, we would like to analyze the audio in real-time and give the sales associate knowledge how successful they are doing while the conversation is in progress.

• Date: fall 2017
• Category: Business School
• People: Zeshan, Peng, Nick, Zhihao Zhang, Zhengyu Li, Moesaeah King, Yingda Liu

Bird Recognition and Counting

Deep Learning Research

Short Description: Using deep learning to detect and then count the number of birds in aerial images. We developed our own custom algorithm to label the images automatically to train the algorithm.

Description: Object detection is a critical problem in computer vision domain, lots of researchers have been working on this problem for a long time. In recent years, deep learning on object detection has achieved a big breakthrough. In the yearly ImageNet competition, teams around the world in computer vision are using different types of Deep Neural Network to figure out how to detect the object from the image. In this project, the data collected from Missouri Department of Conservation are taken from aerial. The goal of this project is to count how many birds in each aerial image. The difficulty of the project is that the object size is too small compared with ImageNet tasks so that most of the state-of-art network cannot be applied. In addition, the shape types of the bird also vary in each image. In this project, we will develop our own deep learning and image process algorithms to handle all these problems and all the system can be automatically run in the real time.

• Date: fall 2017
• Category: Missouri Department of Conservation(MDC)
• People: Peng, Max, Yang, Nick

Protein Secondary And Tertiary Structure Trediction

Deep Learning Research

Short Description: We are using deep learning for several computational biology problems including protein secondary and tertiary structure prediction, loop modeling, contact prediction, beta-turn and psi-phi angle prediction, and many others.

Description: Deep learning has been playing an important role in many fields in recent years. We are applying deep learning techniques to computational biology problems such as secondary and tertiary structure prediction, loop modeling, contact prediction, beta-turn and psi-phi angle prediction, etc., and have achieved the state-of-the-art performance on some of these problems. We have been working on various large, deep and advanced neural networks to do the prediction, and by improving the network structures and integrating Generative Adversarial networks we can further improve the prediction results. We are also applying novel feature engineering on larger data sets to improve the prediction results. Several conference and journal papers have been published or are under review.

• Date: fall 2017
• Category: EECS
• People: Chao Fang, Zhaoyu Li, Junlin Wang, Wenbo Wang

Fish Type Recognition And Length Estimation

AI Research

Short Description: A computer vision-based solution for automatic fish detection, classification and measurement without interference of human.

Description: We are designing an AI system consisting of unified deep neural networks to deal with fish recognition and measurement tasks. The system has the ability to localize fish in images and categorize them automatically. What's more, with the help of reference object, the system can even measure the length of the fish. We are aiming to make the system as robust as possible to harsh environmental conditions such as bad weathers and illumination. A prototype of the system was ranked in Top 1% in Kaggle "The Nature Conservancy Fisheries Monitoring" Competition.

• Date: fall 2017
• Category: Missouri Department of Conservation(MDC)
• People: Guang, Peng, Nick, Fuming

Template-based Methods For Protein Model Quality

3D Structure Research

Short Description: For evaluating the quality of the predicted protein models, we developed two innovative methods (MUfoldQA_S & MUfoldQA_C) and they ranked No.1 in the Critical Assessment of Techniques for Protein Structure Prediction (CASP) competition, which is the word-wide biannual protein structure prediction contest, in their respective categories.

Description: Protein’s functionality is highly relevant to its 3D structure. The current experimental methods to find 3D structure for specific proteins are both expensive and time consuming, resulting that we are only able to acquire less than 0.15% of the known protein sequences’ structures. Computational protein structure prediction is the only practical way to bridge this gap between the known sequences and known structures. After decades of improving the protein structure prediction algorithms, we are able to generate an enormous amount of high quality candidate models for each target protein sequence, which raised another problem: how to select the best models?

To tackle this problem, we developed two innovative protein quality assessment methods: MUfoldQA_S and MUfoldQA_C. The former is able to directly use structural information from known protein fragments as the basis for scoring the query model without building any reference model. And the latter combines the information from the known protein structure with information from other predicted protein 3D models to generate a multi-model based score. To verify their real-world performance, we enrolled these two algorithms to the Critical Assessment of Techniques for Protein Structure Prediction (CASP) experiment, the “Olympics” of the protein structure prediction community, which is a renowned biannual international competition held every two years since 1994 designed to establish a golden standard to evaluate state of art methods for protein structure prediction. In the final result, MUfoldQA_S overall No. 1 in one QA category, and MUfoldQA_C ranked No. 1 in one category and No. 2 in another category.

• Date: fall 2017
• Category: EECS
• People: Chao Fang, Zhaoyu Li, Junlin Wang, Wenbo Wang