The paper "Revisiting Non-Autoregressive Transformers for Efficient Image Synthesis" by Zanlin Ni et al. addresses the limitations of non-autoregressive Transformers (NATs) in image synthesis, particularly their inferior performance compared to diffusion models. The authors propose AutoNAT, an automatic framework that optimizes training and inference strategies for NATs, aiming to enhance their efficiency and performance. By formulating the design of these strategies as a unified optimization problem, AutoNAT avoids the need for heuristic-driven rules, which are often suboptimal and labor-intensive. The method is validated on four benchmark datasets (ImageNet-256 & 512, MS-COCO, and CC3M), demonstrating significant improvements in both generation quality and computational efficiency. AutoNAT achieves results comparable to the latest diffusion models while reducing inference costs by up to 5 times. The paper also includes detailed experimental results and ablation studies to support the effectiveness of AutoNAT.The paper "Revisiting Non-Autoregressive Transformers for Efficient Image Synthesis" by Zanlin Ni et al. addresses the limitations of non-autoregressive Transformers (NATs) in image synthesis, particularly their inferior performance compared to diffusion models. The authors propose AutoNAT, an automatic framework that optimizes training and inference strategies for NATs, aiming to enhance their efficiency and performance. By formulating the design of these strategies as a unified optimization problem, AutoNAT avoids the need for heuristic-driven rules, which are often suboptimal and labor-intensive. The method is validated on four benchmark datasets (ImageNet-256 & 512, MS-COCO, and CC3M), demonstrating significant improvements in both generation quality and computational efficiency. AutoNAT achieves results comparable to the latest diffusion models while reducing inference costs by up to 5 times. The paper also includes detailed experimental results and ablation studies to support the effectiveness of AutoNAT.