Research map / Architectures and efficiency / RNNs and normalization
Scaling Up Parameter Generation: A Recurrent Diffusion Approach
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes RPG, a recurrent diffusion framework that generates full neural network parameters, up to hundreds of millions, on a single GPU.
Problem
Parameter generation has struggled to scale to the size of today's large vision and language models.
Approach
Partitions a network's parameters into non-overlapping tokens and uses a recurrent diffusion process to generate full parameter sets at large scale, rather than generating small networks only.
Key results
- Generates full neural network parameters up to hundreds of millions on a single GPU.
Closest papers by meaning
- Scaling Diffusion Transformers Efficiently via $\mu$P (NeurIPS 2025 · similarity 0.87)
- LaVin-DiT: Large Vision Diffusion Transformer (CVPR 2025 · similarity 0.87)
- Sketch Then Paint: Hierarchical Reinforcement Learning for Diffusion Multi-Modal Large Language Models (NeurIPS 2026 · similarity 0.87)
- Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.86)
- Scaling Exponents Across Parameterizations and Optimizers (ICML 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.