Research map / Probabilistic and scientific modeling / Variational and generative sampling
Auxiliary Deep Generative Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add second stochastic layer (L=1 to L=2) to AVAE (ablation) | AVAE · MNIST unsupervised | log-likelihood bound: -84.59 -> -82.97 | Helped |
| increase importance-weighted samples 1 to 50 in training | IWAE L=1 · MNIST unsupervised | log-likelihood bound: -86.76 -> -84.78 | Helped |
| AVAE (L=1) vs IWAE (L=1, IW=1) and VAE+NF | IWAE L=1 · MNIST unsupervised | log-likelihood bound: -86.76 -> -84.59 | Helped |
| remove warm-up of KL term (ablation) | AVAE · MNIST unsupervised | activation of auxiliary variables | Hurt |
| add auxiliary variable to M2 (ADGM vs M2 with same hyperparameters) (ablation) | M2 · MNIST 100 labels | classification error | Helped |
| make auxiliary variable deterministic (ablation) | ADGM · MNIST 100 labels | classification error | Hurt |
| use ADGM instead of SDGM on MNIST | SDGM · MNIST 100 labels | test error (%): 1.32 -> 0.96 | Helped |
| SDGM (skip connections, two stochastic layers) vs ADGM on SVHN | ADGM · SVHN 1000 labels | test error (%): 22.86 -> 16.61 | Helped |
| SDGM vs ADGM on NORB | ADGM · NORB 1000 labels | test error (%): 10.06 -> 9.40 | Helped |
| SDGM vs VAT on SVHN | VAT · SVHN 1000 labels | test error (%): 24.63 -> 16.61 | Helped |
| omit SVHN extra set from training (ablation) | SDGM · SVHN 1000 labels | test error (%): 16.61 -> 29.82 | Hurt |
| ADGM vs VAT on NORB | VAT · NORB 1000 labels | test error (%): 9.88 -> 10.06 | Hurt |
Closest papers by meaning
- An Architecture for Deep, Hierarchical Generative Models (NeurIPS 2016 · similarity 0.89)
- Deep Generative Learning via Variational Gradient Flow (ICML 2019 · similarity 0.88)
- Bridging Explicit and Implicit Deep Generative Models via Neural Stein Estimators (NeurIPS 2021 · similarity 0.88)
- Deep Generative Modeling on Limited Data with Regularization by Nontransferable Pre-trained Models (ICLR 2023 · similarity 0.88)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.87)
- On Implicit Regularization in $β$-VAEs (ICML 2020 · similarity 0.87)
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.