Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Mixture of HyperGenerators for Learning Distributions over Functions
Batuhan Koyuncu, Pablo Sanchez Martin, Ignacio Peis, Pablo M. Olmos, Isabel Valera
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 |
|---|---|---|---|
| Replace per-sample functa optimization with a single forward-pass encoder (VaMoH) for reconstruction | reconstruction inference · PolyMNIST, ShapeS3D, CelebA-HQ, Era5, Chairs | inference time (s): 2.29x to 3.64x faster than Functa (3 steps); 7.11x to 11.28x faster than Functa (10 steps) | Improved |
| Use VaMoH instead of Functa for reconstruction | image-like datasets · PSNR test set, Era5 and ShapeS3D | PSNR | Improved |
| Use VaMoH instead of GASP for generation | unconditional generation · CelebA-HQ 64x64 | FID: 66.27 vs 14.01 (GASP tr) | Hurt |
| Use VaMoH instead of baselines for generation | unconditional generation · ShapeS3D 64x64 | FID: 56.25 vs 57.81 (Functa) and 118.66 (GASP) | Improved |
| Use VaMoH instead of baselines for generation | unconditional generation · ShapeS3D 64x64 | recall: 0.64 vs 0.13 (Functa) and 0.16 (GASP) | Improved |
| Use VaMoH instead of baselines for generation | unconditional generation · CelebA-HQ 64x64 | recall: 0.0 vs 0.43 (GASP) | Hurt |
| Flow-based (planar flow) prior replacing standard Gaussian prior on z (ablation) | VaMoH latent prior · CelebA (qualitative) | sample realism / reconstruction-generation alignment | Improved |
| Mixture with K=1 instead of K>1 hypergenerators (ablation) | VaMoH · CelebA-HQ reconstruction | reconstruction quality (qualitative) | Hurt |
| Increase mixture components to K=10 (ablation) | VaMoH · CelebA-HQ reconstruction | reconstruction quality (qualitative) | No clear effect |
| Point dropout during training (ablation) | VaMoH encoder · CelebA-HQ patch imputation | imputation quality (qualitative) | Improved |
| Use Functa with normalizing-flow generator on PolyMNIST and CelebA-HQ instead of VaMoH | Functa baseline · PolyMNIST | generation quality | Hurt |
Closest papers by meaning
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.88)
- VAEM: a Deep Generative Model for Heterogeneous Mixed Type Data (NeurIPS 2020 · similarity 0.87)
- Learning Hierarchical Variational Autoencoders With Mutual Information Maximization for Autoregressive Sequence Modeling (TPAMI 2022 · similarity 0.87)
- On the Limitations of Multimodal VAEs (ICLR 2022 · similarity 0.87)
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.86)
- Trading Information between Latents in Hierarchical Variational Autoencoders (ICLR 2023 · similarity 0.86)
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.
