Research map / Probabilistic and scientific modeling / Variational and generative sampling
Understanding Instance-based Interpretability of Variational Auto-Encoders
Zhifeng Kong, Kamalika Chaudhuri
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use VAE-TracIn self-influence ranking to flag extra samples | β-VAE · MNIST with EMNIST extra samples | detection AUC: 0.887 | Improved |
| use VAE-TracIn self-influence ranking to flag extra samples | β-VAE · CIFAR-10 with CelebA extra samples | detection AUC: 0.760 | Improved |
| increase latent dimension from 64 to 128 (ablation) | VAE-TracIn · MNIST | self-influence top-1 frequency: 0.992 -> 1.000 | Improved |
| increase latent dimension from 64 to 128 (ablation) | VAE-TracIn · CIFAR-10 | self-influence top-1 frequency: 0.609 -> 0.602 | Hurt |
| increase latent dimension from 64 to 128 (ablation) | VAE-TracIn · MNIST synthesized major/minor components | major-component top-1 score: 0.422 -> 0.467 | Improved |
| replace perceptual autoencoder nearest-neighbor ranking with VAE-TracIn | MNIST synthesized major/minor components · MNIST synthesized major/minor components | major-component top-1 score: 0.633 -> 0.422 | Hurt |
| replace perceptual autoencoder nearest-neighbor ranking with VAE-TracIn | MNIST synthesized major/minor components · MNIST synthesized major/minor components | minor-component rank 25% quantile: 6943 -> 1097 | Improved |
| estimate influence with VAE-TracIn | β-VAE · CIFAR-10 subclasses | Pearson correlation with loss change: 0.882, 0.910, 0.895, 0.934, 0.828, 0.865, 0.908, 0.852, 0.861, 0.613 | Improved |
| use VAE-TracIn to identify the strongest self-proponent | β-VAE · CIFAR-10 subclasses (average) | self-influence top-1 frequency: 0.998 | Improved |
| use VAE-TracIn to identify the strongest self-proponent | β-VAE · CIFAR-10 | self-influence top-1 frequency: 0.609; 0.602 | Mixed |
Closest papers by meaning
- The Autoencoding Variational Autoencoder (NeurIPS 2020 · similarity 0.88)
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.88)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.86)
- Variational Autoencoders with Jointly Optimized Latent Dependency Structure (ICLR 2019 · similarity 0.86)
- Automatically Interpreting Millions of Features in Large Language Models (ICML 2025 · similarity 0.86)
- A Statistical Analysis of Wasserstein Autoencoders for Intrinsically Low-dimensional Data (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
