Development of AI Based Recommendation Engine for Embedded Coder
Department of Electrical and Computer Engineering, Prairie View A&M University
Location: Rudder Tower - Room 410
Time: May 21, 2018 - 2:00-3:00pm
Due to modern technological advancement in computing theory and manufacture capability, microcontroller based embedded devices have been widely adopted in various applications. Many successful embedded IC designs are still being used and updated at a very fast speed, which creates tremendous challenges for embedded programmers. They need both programming syntax and hardware architecture knowledge to be able to write efficient codes. Although automatic programming engine for pure software development has been explored, not much has been done to embedded coder because of the complexity. Therefore, our investigation focuses on how to developed a smart recommendation engine that assists embedded programmers to quickly search and query related code and instructions.
In this recommendation engine, the index is generated by the programmer's input code, relevant codes on open source websites and online forums are taking into consideration. Since recent advancements in artificial intelligent (AI) and deep learning techniques have shed new light on data analytics, we are able to employ AI based clustering and classification algorithms to category similar codes and solutions. Then, through the Hamming distance and K-means, the most relevant hardware information and registers used are found. We use Euclidean distance, cosine similarity, and Pearson correlation coefficient to calculate the correlation between the programmer code and the searched code respectively. A deep learning model is established to train the weights of various recommendation methods. Through continuous training and testing of a large number of code segments, followed by switching, partitioning, layering, and adjustment, the final similarity scores of different programs are generated. The scores further guide the recommendation priority to ensure the optimal selection. Our proposed deep learning model has clear advantages over the single recommendation method and the linear regression approach. The results of our code recommendation methodology could help embedded programmers find the code they need more quickly and accurately.
Suxia Cui is an associate professor in the Department of Electrical and Computer Engineering at Prairie View A&M University (PVAMU). She received her B.S. and M.S. degrees in Electrical Engineering from Beijing University of Technology in 1996 and 1999 respectively. She joined PVAMU right after she obtained her Ph.D. degree in Computer Engineering from Mississippi State University in 2003. Her research interests include image processing, data compression, wavelets, computer vision, robotics, and computing education. She published papers, as well as served as reviewers for IEEE, ASEE journals and conferences. Dr. Cui’s work has been sponsored by NSF, DOD, DOE, and USDA. She is a senior member of IEEE, and a member of HKN.