Zhichao Zhu

We propose a second-order goodness function based on effective dimensionality for Forward-Forward learning, eliminating the need for negative samples and showing that noise enhances performance and inference robustness.