Ruth Stern

Transforming the loss-function to a v-shape in order to accelerate gradient descent.

Autonomous recognition of regions to select the optimal algorithm for gradient descent optimizers.

Approximating with Chebyshev Decomposition in AGD

Currently, AGD has no third order approximation of drag (Mu) and uses Taylor Expansion to approximate the second order (H). We propose a novel approach to use Chebyshev expansion to simplify and increase the speed of AGD.