Master's thesis presentation. Yichen is advised by Dr. Nadiia Derevianko and Prof. Dr. Felix Dietrich.
Previous talks at the SCCS Colloquium
Yichen Tang: Sampling Based Neural Networks with Adaptive Activation Functions
SCCS Colloquium |
Simulating a complicated dynamic, or solving a partial differential equation plays an important role in scientific computing. Data oriented approaches such as neural network approximation are also widely used in this field. Sampling based networks (SWIM) offer faster training by avoiding the time-costly back propagation approaches, whereas adaptive activation functions improve the approximation capability of neural networks. In this work, we tried to combine these two advantages and provide a computation framework for sampling all parameters of adaptive neural networks, making it much easier to introduce arbitrary parametrized activation functions in SWIM.