Guided Trajectory Generation with Diffusion Models for Offline Model-based Optimization

Authors: Taeyoung Yun, Sujin Yun, Jaewoo Lee, Jinkyoo Park

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiment results demonstrate that our method outperforms competitive baselines on Design-Bench and its practical variants.
Researcher Affiliation Collaboration Taeyoung Yun1 Sujin Yun1 Jaewoo Lee1 Jinkyoo Park1,2 1Korea Advanced Institute of Science and Technology (KAIST) 2Omelet.ai
Pseudocode Yes Algorithm 1 Trajectory construction procedure of GTG
Open Source Code Yes The code is publicly available in https://github.com/dbsxodud-11/GTG.
Open Datasets Yes We empirically demonstrate that our method achieves superior performance on Design-Bench, a well-known benchmark for MBO with a variety of real-world tasks. ... Design-Bench [5] is the most widely used benchmark for evaluating MBO algorithms.
Dataset Splits No The paper mentions training models on the offline dataset and using sparse/noisy variants for evaluation but does not explicitly provide training, validation, or test dataset splits with percentages or sample counts.
Hardware Specification Yes All training is done with a single NVIDIA RTX 3090 GPU and takes approximately 30 minutes for discrete tasks and 2 hours for continuous tasks.
Software Dependencies No The paper mentions using "Adam optimizer [40]" and "temporal U-Net architecture from Diffuser [18]" but does not specify version numbers for these or other software libraries/frameworks like Python or PyTorch.
Experiment Setup Yes The hyperparameters we used for modeling and training are listed in Table 10. (Hyperparameters for Training Diffusion Models) and Table 11. (Hyperparameters for Training Proxy).