Congratulations to our winners of the 2020 Dean’s High Impact Award for Undergraduate Research from the College of Liberal Art and Sciences!
John Wahlig, advised by Prof. Jia (Kevin) Liu, received the 2020 Dean’s High Impact Award for Undergraduate Research from the College of Liberal Art and Sciences. This research is motivated by the recent rapid growth of large-scale distributed deep learning (DDL) (a form of machine learning that lets computer programs learn patterns and adapt their performance) frameworks (e.g., Google's TensorFlow, MXNet, etc.), which exploit the massive parallelism of computing clusters to expedite the training and inference phases of deep learning systems. In a networked computing cluster that supports a large number of deep learning jobs, Wahlig and Liu strive to answer a key question: How to design efficient scheduling algorithms to allocate resources across different machines to minimize the overall job processing time, i.e., allowing computers to process as many tasks as efficiently as possible. In this research, Wahlig and Liu proposed to develop a suite of online scheduling algorithms that jointly optimize resource allocation and locality decisions for distributed deep learning training and inference phases. The goal of this research is to develop theoretically provable (near) delay-optimal scheduling and resource allocation optimization algorithms for RNN-based (recursive neural network) distributed deep learning based on cell-based batching in the inference phase.
McKenna Goffinet and Karthik Subbarao received the 2020 Dean's High Impact Award for Undergraduate Research from the College of Liberal Arts and Sciences. The students will be advised by Tavanapong and her Ph.D. students, Lei Qi and Mohammed Khaleel. During the project, they will learn and integrate a new interpretable quantification learning (IQ) framework developed by Tavanapong's research team with data acquisition, data labeling, data analysis, and visualization in Google Cloud Platform. Given a set of documents with corresponding class labels as training samples, the IQ framework produces a quantifier that estimates the proportion of documents in each class given a batch of documents without class labels. The IQ framework produced the smallest error in class distribution estimation compared to existing state-of-the-art methods in our study. Quantification learning has applications across subject domains such as tracking the prevalence of diseases, trending topics in customer demands, and policy areas of interest to state legislators.
Nicholas Stout, advised by Dr. Adisak Sukul, received the 2020 Dean's High Impact Award for Undergraduate Research from the College of Liberal Arts and Sciences. They are proposing to develop the “Online Data Science Interactive Practice” software system to support the learning and practicing of Data Science courses. The software will allow students to be able to practice and submit Data Science assignments or short projects online. The student will be able to interact with the software and get the feedback in real-time. The software will also perform auto grader action. It would help speed up the grading process and allow Data Science courses’ instructors to focus more on improving the content. It will also allow the instructor to see the portion of the assignments/projects that have been challenging to the student.