Improving Denoising Diffusion Probabilistic Models via Exploiting Shared Representations