Uploaded May 2025 | Updated September 2026, 1 day ago
Abstract: Discrete diffusion has achieved state-of-the-art performance, outperforming or approaching autoregressive models on standard benchmarks. In this work, we introduce Discrete Diffusion with Planned Denoising (DDPD), a novel framework that separates the generation process into two models: a planner and a denoiser. At inference time, the planner selects which positions to denoise next by identifying the most corrupted positions in need of denoising, including both initially corrupted and those requiring additional refinement. This plan-and-denoise approach enables more efficient reconstruction during generation by iteratively identifying and denoising corruptions in the optimal order. DDPD outperforms traditional denoiser-only mask diffusion methods, achieving superior results on language modeling benchmarks such as text8, OpenWebText, and token-based generation on ImageNet 256×256. Notably, in language modeling, DDPD significantly reduces the performance gap between diffusion-based and autoregressive methods in terms of generative perplexity.
Bio: I am a postdoctoral researcher at MIT in Rafael Gómez-Bombarelli ’s group, and I also collaborate with Tommi Jaakkola ’s group. I received my PhD from Princeton University in 2023, advised by Ryan P. Adams and Peter J. Ramadge . In summer 2021, I interned at Meta Research with Ben Letham and Eytan Bakshy in the Adaptive Experimentation team. I did my undergrad at National University of Singapore and worked with Sinno Jialin Pan at Nanyang Technological University, Singapore.
Abstract: Discrete diffusion has achieved state-of-the-art performance, outperforming or approaching autoregressive models on standard benchmarks. In this work, we introduce Discrete Diffusion with Planned Denoising (DDPD), a novel framework that separates the generation process into two models: a planner and a denoiser. At inference time, the planner selects which positions to denoise next by identifying the most corrupted positions in need of denoising, including both initially corrupted and those requiring additional refinement. This plan-and-denoise approach enables more efficient reconstruction during generation by iteratively identifying and denoising corruptions in the optimal order. DDPD outperforms traditional denoiser-only mask diffusion methods, achieving superior results on language modeling benchmarks such as text8, OpenWebText, and token-based generation on ImageNet 256×256. Notably, in language modeling, DDPD significantly reduces the performance gap between diffusion-based and autoregressive methods in terms of generative perplexity.
Bio: I am a postdoctoral researcher at MIT in Rafael Gómez-Bombarelli ’s group, and I also collaborate with Tommi Jaakkola ’s group. I received my PhD from Princeton University in 2023, advised by Ryan P. Adams and Peter J. Ramadge . In summer 2021, I interned at Meta Research with Ben Letham and Eytan Bakshy in the Adaptive Experimentation team. I did my undergrad at National University of Singapore and worked with Sinno Jialin Pan at Nanyang Technological University, Singapore.
