Hierarchical VAE Training Metrics for Posterior Collapse Analysis

A research dataset accompanies a paper on mitigating posterior collapse in hierarchical variational autoencoders. The data likely contains training metrics, model configurations, and experimental results from the University of Amsterdam. The paper proposes a spectral smoothing technique and analyzes phase transitions in the VAE loss.

Use Cases

• Analyzing the relationship between smoothing parameters and KL divergence based on the described phase transition phenomenon.
• Studying model performance across different datasets based on the validation mentioned in the description.
• Investigating the variance of parameterizing functions in hierarchical VAEs as described in the paper.
• Benchmarking new methods against the spectral smoothing technique for solving posterior collapse.

Strengths

• The dataset is associated with peer-reviewed research published on paperswithcode.
• The description provides a clear theoretical motivation and experimental claim regarding phase transitions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

University of Amsterdam

Collection Method

Generated from machine learning experiments described in the associated research paper.