Project Grant 2240916

This $196,176 Project Grant from the National Science Foundation's Computer and Information Science and Engineering program will support collaborative research to apply deep learning techniques to the design of new encoding and decoding methods for physical layer communication. The researchers at Princeton University aim to use deep learning tools to generate a new family of codes naturally built for finite block lengths, addressing a longstanding challenge in information theory. In parallel, studying neural network architectures as encoding and decoding procedures may provide new insights into their mathematical properties. All resulting algorithms and source code will be maintained in an online repository. Outcomes will also be used to develop new curricula. This award seeks to advance the development of reliable communication through independent scientific investigation into finite block length information theory and the mathematics underlying deep learning models.