Reduce, Reuse, Recycle: Compositional Generation with Energy-Based Diffusion Models and MCMC

Authors: Yilun Du, Conor Durkan, Robin Strudel, Joshua B. Tenenbaum, Sander Dieleman, Rob Fergus, Jascha Sohl-Dickstein, Arnaud Doucet, Will Sussman Grathwohl

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our approach in settings from 2D data to high-resolution text-to-image generation. and 5 Experiments.
Researcher Affiliation Collaboration 1MIT 2Google Deepmind.
Pseudocode Yes Algorithm 1 Annealed MCMC
Open Source Code No Project webpage: https://energy-based-model.github.io/reduce-reuse-recycle/. This is a project overview page, not an explicit code repository or statement of code release for the methodology.
Open Datasets Yes We train our models on a dataset of images containing between 1 and 5 examples of various shapes taken from CLEVR (Johnson et al., 2017). and Next, we train unconditional diffusion models and a noise-conditioned classifier on Image Net.
Dataset Splits No The paper mentions using CLEVR and Image Net datasets but does not specify exact train/validation/test split percentages, sample counts, or detailed splitting methodology.
Hardware Specification Yes 10 minutes on a 8 TPUv2 cores, 8 hours on 8 TPUv2 cores, 3 days on 16 TPUv2 cores, one week on an internal text/image dataset consisting of 400 million images using 32 TPUv3 cores.
Software Dependencies No The paper mentions software like the 'Adam optimizer' but does not provide specific version numbers for any key software components or libraries.
Experiment Setup Yes For synthetic datasets, we train both score and energy based diffusion models using a small residual MLP model with 4 residual blocks, with a internal hidden dimension of 128 dimensions. We train models for 15000 iterations (10 minutes on a 8 TPUv2 cores) using the Adam optimizer with learning rate of 1e-3, and train diffusion models on 100 discrete timesteps with linear schedule of β values.