A spatially continuous diffusion model of visual working memory