Sho Sonoda, Research Scientist of the Deep Learning Theory Team (PI: Taiji Suzuki) received the FY2021 Young Presentation Award of the Japan Society for Industrial and Applied Mathematics.
Title: A general solution of an integral equation with a continuous neural network
To understand the generalization performance of deep learning, it is a key step to characterize typical solutions obtained through deep learning. In this talk, we consider an integral equation corresponding to the learning problem of a depth-2 neural network, and explain the method to derive its general solution by using the Fourier expression.
For more information, please see the following site.
The Japan Society for Industrial and Applied Mathematics
*Only available in Japanese