Norman Yarvin
Papers
1
Total Citations
31
H-Index
1
About
Norman Yarvin is a researcher whose work in the early 1990s contributed foundational insights into the architecture and capabilities of artificial neural networks. His key research area centers on computational learning theory and network design, particularly exploring how the choice of activation functions within feedforward networks affects their approximation and estimation performance. Yarvin’s most cited paper, “Networks with Learned Unit Response Functions” (1991, 31 citations), systematically tested networks using functions more complex than standard sigmoids—including polynomial functions—to evaluate their learning capacity. This work pushed beyond the conventional focus on sigmoidal units, offering early evidence that network expressivity could be enhanced by learning the unit response functions themselves. While modest in citation count, Yarvin’s study is notable for its forward-looking approach to adaptive activation functions, a concept that has since become central to modern deep learning research. His contributions remain relevant for students and researchers exploring flexible network architectures and the theoretical underpinnings of neural computation.
Research Focus
Key Achievements
Top Papers
- 1Networks with Learned Unit Response Functions31 citations · 1991