A neural network where all parameters are active for every input, in contrast to sparse architectures like mixture-of-experts that selectively activate different parts.