Beyond Shannon Inspired Approaches to Limits of Learning
Abstract
Abstract: We will compute the limits of learning for various data scenarios such as high dimensions, purely high dimensions, sparse data, etc. We will provide numerical methods for the computation of limits of learning when explicit formulas are hard to compute. We will also calculate the impact of. tutor, learner s limitations, cost of learning, etc, on the limits of learning. Moreover, we will develop real-time algorithms that can approach these limits. Using our theories, we will develop neural networks that change their structures based on data in order to achieve performance close to the fundamental limits of learning (we refer to these as Organic Machines). These organic machines (in contrast to deep networks) will be able to achieve good performance with limited data with potentially changing statistics (e.g. between the training and test statistics).
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1810134
Entities
People
- Vahid Tarokh
Organizations
- Army Contracting Command
- Defense Advanced Research Projects Agency
- Duke University