Evaluating and Designing Sparse Autoencoders by Approximating Quasi-Orthogonality

Open in new window