In our work, we sought to improve upon the standard feed-forward deep learning model by augmenting them with biologically inspired concepts of sparsity, top-down feed- back, and lateral inhibition. We define our model as a sparse coding problem using hierarchical layers. We solve the sparse coding problem with an additional top-down feedback error driving the dynamics of the neural network.View CVPR 2018 Publication on arXiv PetaVision - Bio-inspired Neural Networks
Virtual batting practice is an interdisciplinary effort led by Mark Jupina, PhD, involving faculty and students in the Colleges of Engineering and Liberal Arts and Sciences that will provide players on the Villanova baseball team the opportunity to take virtual batting practice. We have been working on developing accurate models and animation of MLB pitchers and environments for virtual reality.Learn more at Philly.com
We are researching novel ways to use computer vision, crowdsourcing, and deep learning algorithms to teach machines to interact with humans, improve affect recognition, and perform general-purpose video emotion classification.Learn more at Vinereactor.org
An illustration of a constructed graph between two images. An image is hierarchically segmented into a number of layers where the intra-image affinities, W, are defined between neighboring superpixels, weighted by the length of the shared border shown in red. The hierarchical constraints between layer segmentations are illustrated by the yellow connections. These connections are defined in my constraint matrix, C. The inter-image affinities, R, are made between images at their coarsest level of segmentation. A fully connected graph is considered and then the number of edges are trimmed, as illustrated in green. (Note: yellow and green connections are visualizations and not the actual edges).