Generative Modeling with Phase Stochastic Bridge

Authors: Tianrong Chen, Jiatao Gu, Laurent Dinh, Evangelos Theodorou, Joshua M. Susskind, Shuangfei Zhai

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our model yields comparable results in image generation and notably outperforms baseline methods, particularly when faced with a limited Number of Function Evaluations. ... We achieve competitive results compared to DM approaches equipped with specifically designed fast sampling techniques on image datasets, particularly in small NFE settings.
Researcher Affiliation Collaboration 1Georgia Tech, 2Apple
Pseudocode Yes Algorithm 1 Training; Algorithm 2 Sampling
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the described methodology.
Open Datasets Yes CIFAR-10 (Krizhevsky et al., 2009), AFHQv2 (Choi et al., 2020) and Image Net (Deng et al., 2009)
Dataset Splits No The paper references well-known datasets but does not explicitly provide the training/test/validation dataset splits (e.g., percentages, sample counts, or citations to specific predefined splits) required for reproduction.
Hardware Specification Yes We use 8 Nvidia A100 GPU for all experiments.
Software Dependencies No The paper mentions the Adam W optimizer but does not provide specific version numbers for software dependencies such as Python, PyTorch, or CUDA used for running the experiments.
Experiment Setup Yes We use Adam W(Loshchilov & Hutter, 2017) as our optimizer and Exponential Moving Averaging with the exponential decay rate of 0.9999. We use 8 Nvidia A100 GPU for all experiments. For further, training setup, please refer to Table.6. Table 6: Additional experimental details [includes] Training Iter Learning rate Batch Size network architecture