Constrained Synthesis with Projected Diffusion Models
Authors: Jacob K Christopher, Stephen Baek, Nando Fioretto
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The proposed method recast the traditional sampling process of generative diffusion models as a constrained optimization problem, steering the generated data distribution to remain within a specified region to ensure adherence to the given constraints. These capabilities are validated on applications featuring both convex and challenging, non-convex, constraints as well as ordinary differential equations, in domains spanning from synthesizing new materials with precise morphometric properties, generating physics-informed motion, optimizing paths in planning scenarios, and human motion synthesis. |
| Researcher Affiliation | Academia | Jacob K. Christopher University of Virginia csk4sr@virginia.edu Stephen Baek University of Virginia baek@virginia.edu Ferdinando Fioretto University of Virginia fioretto@virginia.edu |
| Pseudocode | Yes | Algorithm 1: PDM 1 x0 T N(0, σT I) 2 for t = T to 1 do 3 γt σ2 t/2σ2 T 4 for i = 1 to M do 5 ϵ N(0, I); g sθ (xi 1 t , t) 6 xi t = PC(xi 1 t + γtg + 2γtϵ) 7 x0 t 1 x M t 8 return x0 0 |
| Open Source Code | Yes | The code and datasets for all experiments are provided in the supplemental materials. |
| Open Datasets | Yes | The dataset, obtained from the authors of [5], consists of 64 64 image patches... using the challenging Human ML3D dataset [12]... The performance is evaluated on two sets of maps adapted from Carvalho et al.... For this setting, we used a state-of-the-art masked conditional video diffusion model, following Voleti et al. [27]. |
| Dataset Splits | Yes | The diffusion models are trained on 1000 points with a 90/10 train/test split. |
| Hardware Specification | Yes | All sampling is run on two NVIDIA A100 GPUs. |
| Software Dependencies | No | The paper mentions various software components and models (e.g., Motion Diffuse, Motion Planning Diffusion, interior point method) but does not provide specific version numbers for these or other underlying software dependencies (like Python, PyTorch, etc.). |
| Experiment Setup | Yes | Additionally, we set M = 100 and T = 10 for each experiment. |