Artificial Intelligence and Data Analytics

Artificial Intelligence and Data Analytics

Through the use of artificial intelligence algorithms, ARI researchers are creating systems that learn to detect and classify objects, identify embodied signals for health and wellness, and control processes in manufacturing. ARI researchers are also working on methods for analyzing data to make analysis more powerful and efficient. Read about some of our past and current projects below.

The AIFARMS Open World Semi-Supervised Object Detection project tackles the problem of the unlabeled data containing previously unseen classes of objects in images. The goal is to develop a method that can 1) detect instances of a new class of objects in unlabeled images and 2) add them back to the model for training. We explore the potential of outlier detection algorithms to recognize novel classes of objects. Being able to group/cluster similar objects after they have been identified as outliers constitutes an essential step in this process.

The Foxconn Interconnected Technology (FIT) and the University of Illinois funded Center for Networked Intelligent Components (C-NICE) focuses on Smart Manufacturing, the development of technology for smart devices, including those used in manufacturing, medical environments, as well as homes and self-driving vehicles.

Our team is part of the Human-Robot COllaBOraTion - Interactive Manipulation for Industrial Robotic project, together with Picking and Assembly, Safety and Digital Twin, and Planning groups. Our role focuses on software design recommendations to the Illinois team with the aim to facilitate technology adoption at FIT including providing training material for FIT engineers. We also assist with code integration mainly with Robot Operating System (ROS) and the development of simulation capabilities.

Scientific journal articles contain a wealth of data that, due to its size, is difficult to summarize using traditional approaches. One of the goals of the Polymer Informatics project, a collaboration between Materials and Manufacturing and Algorithm & Software Applied Research Institute groups, is to create focused summaries of scientific articles that report polymer experiment results that can assist with the creation of new hypotheses. We employ natural language processing methods to create such summaries. Alongside these efforts, the Polymer Informatics project is developing an active learning framework that can enable researchers to select new samples and efficiently explore a polymer design space. The project uses the Python framework and Community Resource for Innovation in Polymer Technology (CRIPT) scalable data model.

As machine learning and computer vision are rapidly improving, the algorithms used to detect and classify objects in imagery have improved as well. Because there are a number of approaches to the task of object detection, it is not always obvious which algorithm will be most suitable for specific tasks. This project - Detecting and Classifying Airplanes for Rapid Automation, Validation, and Verification (RAVVE) - in collaboration with Sandia National Labs (SNL) seeks to solve this problem with a software framework called SeaScape, which can test and train machine learning algorithms in a consistent manner. To support this effort, the Applied Research Institute built and tested a machine learning algorithm to identify and classify airplanes in optical overhead images, packaged our model and documentation, and sent it to SNL to implement within the SeaScape framework.