Architecture based on Rozell's '08 paper on a locally competitive algorithm (LCA) for Sparse Approximation: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.64.7897&rep=rep1&type=pdf
Further inspiration is taken from current neuroscience literature
We consider the relationship between representations of natural images in a temporally smooth sequence (i.e.consecutive frames in a video). Traditionally, sparse coding methods learn representations of images in isolation. Here, we learn an image’s sparse representation with the previous image’s representation as a starting point.
- Different sized dictionary elements
- Support for RGB images
- Support for Convolutional Sparse Coding
- Implement same technique in other ML architectures
Collaborators: Jeff Winchell, Dr. Edward Kim (Drexel University)