Training-free Multi-objective Diffusion Model for 3D Molecule Generation

Authors: Xu Han, Caihua Shan, Yifei Shen, Can Xu, Han Yang, Xiang Li, Dongsheng Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted experiments on both single-objective and multi-objective 3D molecule generation, focusing on quantum properties, and compared our approach with the trained or fine-tuned diffusion models.
Researcher Affiliation Collaboration 1Tufts University, 2Microsoft Research Asia 3East China Normal University, 4Microsoft Research AI4Science
Pseudocode Yes The pseudo-code and hyperparameters are provided in Appendix A.1 and A.3, and the code will be published later.
Open Source Code No The pseudo-code and hyperparameters are provided in Appendix A.1 and A.3, and the code will be published later.
Open Datasets Yes We perform conditional molecule generation on QM9 (Ramakrishnan et al., 2014), a dataset of over 130K molecules and 6 corresponding quantum properties.
Dataset Splits Yes Following previous research, we split the dataset into training, valid, and test sets, each including 100K, 18K, and 13K samples respectively.
Hardware Specification No The paper mentions hardware used for training baselines ('A100 GPU') but does not specify the hardware used for running their own training-free method's experiments (e.g., inference, evaluation).
Software Dependencies No The paper does not provide specific version numbers for any software dependencies.
Experiment Setup Yes The pseudo-code and hyperparameters are provided in Appendix A.1 and A.3. In Appendix A.3, Table 3 lists 'Hyperparameters for two conditions sampling' including 'Guide from', 'w1', 'w2', 'r', and 'MC sample size'.