Title:
Image Space Embeddings and Generalized Convolutional Neural Networks
Abstract:
Cutting edge results in many image processing tasks now rely on Convolutional Neural Network architecture.
By developing principled embeddings into spaces of smooth images, we extend this technique to arbitrary datasets.
This talk discusses algorithms and theory for embedding arbitrary datasets into image spaces,
and we illustrate how generalizing convolutional neural networks outperform unstructured neural networks.
|