Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
On amortizing convex conjugates for optimal transport
Authors: Brandon Amos
ICLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | I show that combining amortized approximations to the conjugate with a solver for fine-tuning significantly improves the quality of transport maps learned for the Wasserstein-2 benchmark by Korotin et al. (2021a) and is able to model many 2-dimensional couplings and flows considered in the literature. Sect. 5 shows that amortizing and fine-tuning the conjugate results in state-of-the-art performance in all of the tasks proposed in the Wasserstein-2 benchmark by Korotin et al. (2021a). |
| Researcher Affiliation | Industry | Brandon Amos Meta AI |
| Pseudocode | Yes | Algorithm 1 Learning Wasserstein-2 dual potentials with amortized and ο¬ne-tuned conjugation. Algorithm 2 CONJUGATE(f, y, xinit) |
| Open Source Code | Yes | All of the baselines, methods, and solvers in this paper are available at http://github.com/facebookresearch/w2ot. |
| Open Datasets | Yes | Wasserstein-2 benchmark by Korotin et al. (2021a) and samples from generative models trained on Celeb A (Liu et al., 2015). |
| Dataset Splits | No | The paper refers to using a benchmark and sampling, but does not explicitly provide percentages or counts for training, validation, and test dataset splits. |
| Hardware Specification | Yes | wall-clock time for the entire training run measured on an NVIDIA Tesla V100 GPU |
| Software Dependencies | No | The core set of tools in Python (Van Rossum and Drake Jr, 1995; Oliphant, 2007) enabled this work, including Hydra (Yadan, 2019), JAX (Bradbury et al., 2018), Flax (Heek et al., 2020), Matplotlib (Hunter, 2007), numpy (Oliphant, 2006; Van Der Walt et al., 2011), and pandas (Mc Kinney, 2012). |
| Experiment Setup | Yes | Tables 4 and 5 detail the main hyper-parameters for the Wasserstein-2 benchmark experiments. I tried to keep these consistent with the choices from Korotin et al. (2021a), e.g. using the same batch sizes, number of training iterations, and hidden layer sizes for the potential. |