Hyphylearn: A Domain Adaptation-Inspired Approach to Classification Using Limited Number of Training Samples

Abstract

The fundamental task of classification given a limited number of training data samples is considered for physical systems with known parametric statistical models. As a solution, a hybrid classification method-termed HYPHYLEARN-is proposed that exploits both the physics-based statistical models and the learning-based classifiers. The proposed solution is based on the conjecture that HYPHYLEARN would alleviate the challenges associated with the individual approaches of learning-based and statistical model-based classifiers by fusing their respective strengths. The proposed hybrid approach first estimates the unobservable model parameters using the available (suboptimal) statistical estimation procedures, and subsequently uses the physics-based statistical models to generate synthetic data. Next, the training data samples are incorporated with the synthetic data in a learning-based classifier that is based on domain-adversarial training of neural networks. Numerical results on multiuser detection, a concrete communication problem, demonstrate that HYPHYLEARN leads to major classification improvements compared to the existing stand-alone and hybrid classification methods.