Particle Guidance: non-I.I.D. Diverse Sampling with Diffusion Models

Authors: Gabriele Corso, Yilun Xu, Valentin De Bortoli, Regina Barzilay, Tommi S. Jaakkola

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we test the framework both in the setting of conditional image generation, where we are able to increase diversity without affecting quality, and molecular conformer generation, where we reduce the state-of-the-art median error by 13% on average.
Researcher Affiliation Academia Gabriele Corso 1, Yilun Xu1, Valentin de Bortoli2, Regina Barzilay1, Tommi Jaakkola1 1CSAIL, Massachusetts Institute of Technology, 2ENS, PSL University
Pseudocode No The paper contains mathematical equations and derivations but does not include any blocks explicitly labeled as 'Pseudocode' or 'Algorithm', nor does it present structured steps in a code-like format.
Open Source Code Yes The code is available at https://github.com/gcorso/particle-guidance.
Open Datasets Yes We use the validation set in COCO 2014 [Lin et al., 2014] for evaluation, and the CLIP [Hessel et al., 2021]/Aesthetic score [Team, 2022] (higher is better) to assess the text-image alignment/visual quality, respectively.
Dataset Splits Yes We use the validation set in COCO 2014 [Lin et al., 2014] for evaluation... we finetune the inference parameters for particle guidance and the other ablation experiments on a random subset of 200 molecules out of 30433 from the validation set.
Hardware Specification No The paper mentions 'running inference on GPU' and 'GPU memory' but does not specify any particular GPU model, CPU model, or other detailed hardware specifications used for experiments.
Software Dependencies No The paper mentions models like 'Stable Diffusion v1.5' and tools like 'RDKit' and 'OMEGA', but it does not provide specific version numbers for any key software dependencies or libraries required to reproduce the experiments.
Experiment Setup Yes We apply an Euler solver with 30 steps to solve for the ODE version of particle guidance... We set the hyper-parameter αt to 8σ(t) in particle guidance (feature) and 30σ(t)2 in particle guidance (pixel). We use an Euler solver with 30 NFE in all the experiments.