Mitigating Modality Collapse in Multimodal VAEs via Impartial Optimization

Authors: Adrian Javaloy, Maryam Meghdadi, Isabel Valera

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our training framework to several multimodal VAE models, losses and datasets from the literature, and empirically show that our framework significantly improves the reconstruction performance, conditional generation, and coherence of the latent space across modalities.
Researcher Affiliation Academia 1Department of Computer Science, Saarland University, Germany 2MPI for Software Systems, Saarland, Germany.
Pseudocode Yes Algorithm 1 Backward pass within the impartiality block.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We consider 12 datasets collected from the UCI (Dua & Graff, 2017) and R (R Core Team, 2021) repositories, covering a wide range of dataset sizes and likelihoods.
Dataset Splits Yes We randomly split the data into training (70 %), validation (10 %), and testing (20 %).
Hardware Specification No The paper mentions training times in hours but does not specify any hardware details such as GPU models, CPU types, or other computing resources used for the experiments.
Software Dependencies No The paper mentions using specific optimizers like 'AMSGrad (Reddi et al., 2018)' and 'Adam' but does not provide version numbers for these or any other software dependencies such as programming languages or libraries.
Experiment Setup Yes For each experiment, we train the model for 30 epochs and a batch size of 128. We use AMSGrad (Reddi et al., 2018) with a learning rate of 0.001.