Spectrum-Guided Adversarial Disparity Learning

In this work, we propose a novel end-toend to improve models' robustness on new subjects. Given that generative knowledge directed adversarial learning framework, which models usually perform better on sparse data, However, most portrays the class-conditioned intraclass disparity using two competitive subject-independent studies [1, 20, 25, 30] are still limited in considering encoding distributions and learns the purified latent codes the intraclass disparity as meaningless noise and neglect by denoising learned disparity. Furthermore, the domain knowledge the point that intraclass disparity is related to the subject and the is incorporated in an unsupervised manner to guide the optimization class type. They are still inaccurate in exhibiting the relationship and further boosts the performance. The experiments on four between the subject variation and the class, e.g., subject variation HAR benchmark datasets demonstrate the robustness and generalization within a class should be conditionally constrained. of our proposed methods over a set of state-of-the-art. We Besides, signal data may be segmented imprecisely, and the segments further prove the effectiveness of automatic domain knowledge may include gaps and noises. Further, the segments carry incorporation in performance enhancement.