2018/11/08 by Daeyoung Choi, Choi, Daeyoung, Kyung‐Eun Lee +5
Computer Science · #FOS: Computer and information sciences #Face and Expression Recognition #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and ELM #Neural Networks and Applications
paper · pdf · doi:10.48550/arxiv.1811.03666
openalex publication_date 2018/11/08 · openalex created_date 2022/09/19 · openalex updated_date 2026/07/28
In this study, the effects of eight representation regularization methods are\ninvestigated, including two newly developed rank regularizers (RR). The\ninvestigation shows that the statistical characteristics of representations\nsuch as correlation, sparsity, and rank can be manipulated as intended, during\ntraining. Furthermore, it is possible to improve the baseline performance\nsimply by trying all the representation regularizers and fine-tuning the\nstrength of their effects. In contrast to performance improvement, no\nconsistent relationship between performance and statistical characteristics was\nobservable. The results indicate that manipulation of statistical\ncharacteristics can be helpful for improving performance, but only indirectly\nthrough its influence on learning dynamics or its tuning effects.\n