Variational Wasserstein Barycenters with C-cyclical Monotonicity Regularization

Authors: Jinjin Chi, Zhiyao Yang, Ximing Li, Jihong Ouyang, Renchu Guan

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show theoretical convergence analysis and demonstrate the superior performance of VWB-CMR on synthetic data and real applications of subset posterior aggregation.
Researcher Affiliation Academia Jinjin Chi1,2, Zhiyao Yang1,2, Ximing Li1,2*, Jihong Ouyang1,2, Renchu Guan1,2 1 College of Computer Science and Technology, Jilin University, China 2 Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, China
Pseudocode Yes Algorithm 1: Optimization of VWB-CMR
Open Source Code No The paper provides links to the codebases of comparison methods (CRWB, CWB, SCWB) in footnotes but does not provide a link to the open-source code for their proposed VWB-CMR method.
Open Datasets Yes Following (Li et al. 2020), we use Poisson regression for the task of predicting the hourly number of bike rentals using features such as the day of the week and weather conditions 4. We consider the posterior distribution on the 8-dimensional regression coefficients for the Poisson model. We firstly randomly split the full data... (Footnote 4: http://archive.ics.uci.edu/ml/datasets/Bike+Sharing+Dataset)
Dataset Splits Yes We firstly randomly split the full data into 5 subsets with equal-size and get 105 samples from each subset posterior using the Stan library (Carpenter et al. 2017).
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments. It mentions neural network architectures but no specific GPU/CPU models or compute resources.
Software Dependencies No The paper mentions using 'Adam method' for adjusting the learning rate and the 'Stan library' (Carpenter et al. 2017) for sampling. However, it does not specify version numbers for these software components.
Experiment Setup Yes In our method, we use Adam method to adjust the learning rate, where parameters β1=0.9, β2=0.999 and α=0.001. For the training, the number of iterations and batch size are 20000 and 1024, respectively. The dual potentials {ϕn, ψn}N n=1 are parameterized as neural networks with two fully-connected hidden layers (D 128 256 D) using Re LU activations. We consider quadratic regularization and report the average results of 5 independent runs for all experiments.