Approximating sparse binary matrices in the cut-norm