Rylan Schaeffer

Logo
Resume
Research
Learning
Blog
Teaching
Jokes
Kernel Papers


Comparison of Backprop, Weight Perturbation & Node Perturbation

Werfel, Xie and Seung 2004 offer a comparison of the convergence of Backpropagation (BP), Weight Pertuubation (WP) and Node Perturbation (NP) in a simple linear model as a function of the input dimension $N$ and output dimension $M$. They consider the following student-teacher setup, where $W^*$ is the teacher’s weight matrix and where $W$ is the student’s weight matrix, both in \(\mathbb{R}^{M \times N}\):

\[x \in \mathbb{R}^N \sim \mathcal{N}(0, I), \qquad y^* \in \mathbb{R}^M = W^* x, \qquad y = W x , \qquad L(W) = ||y - y^*||_2^2\]