reconstruction-loss

This paper provides a mathematical analysis of superposition in neural networks, deriving upper and lower bounds on L2 reconstruction loss for simple autoencoders with power activation functions, corroborating empirical findings by Elhage et al.