While machines that analyze movement work well, models that anticipate are the future of machine learning. Using static relations to predict future movement results in...
While machines that analyze movement work well, models that anticipate are the future of machine learning. Using static relations to predict future movement results in less accurate forecasting because relations between static entities change over time. This technology uses dynamic Neural Relational Inference to predict an outcome for each static relation over time, resulting in more accurate forecasts.
This suite of tools can be used to improve the performance of a football program. The tools focus on training and preparation and incorporate mobile, virtual reality, and...
This suite of tools can be used to improve the performance of a football program. The tools focus on training and preparation and incorporate mobile, virtual reality, and data analytics to get the most out of the football team. Players can see custom formations in a mobile or VR setting and can take the perspective of any position on the field plus several aerial views. This allows for players to easily prepare for opponents and better train by seeing formations from their respective positions on the field. Additionally, this suite includes a team management system. This tool allows coaches to visualize the depth chart and incorporates a play/game-based grading system.
Researchers at the University of Illinois have developed a new software for size and shape characterization of riprap materials using image analysis. Riprap is a...
Researchers at the University of Illinois have developed a new software for size and shape characterization of riprap materials using image analysis. Riprap is a collection of large broken stones or concrete used along roadways or shoreline as a means of erosion and sediment control. Riprap analysis is currently done manually and is labor intensive and time consuming. This software was developed based on deep learning techniques and provides user-independent image segmentation analysis. This software provides more advances modules for the size and shape characterization than other available methods.
Dr. Schwing has developed a model with a temporal backbone that allows video object segmentation. This allows for the model to compare images frame by frame and create a...
Dr. Schwing has developed a model with a temporal backbone that allows video object segmentation. This allows for the model to compare images frame by frame and create a propagated image in order to identify occluded objects more accuractely. The previous method of a single image analysis sometimes caused issues in the object getting lost or misidentified by the algorithms. This technology is particularly useful in autonomous driving features for lane changes, and other maneuvers. It also has potential uses in tracking software for security purposes.
Researchers have created a Generative Adversarial Network (GAN) for anomaly detection in manufacturing. By deploying a method that clusters data and individually assesses...
Researchers have created a Generative Adversarial Network (GAN) for anomaly detection in manufacturing. By deploying a method that clusters data and individually assesses manifolds for anomalies, this method significantly outperforms other GANs in testing. Within particular datasets, D-AnoGAN can be trained to detect industry-specific anomalies with a very low error rate.
