Research map / Image generation / Autoregressive and latent generation
Latent Constraints: Learning to Generate Conditionally from Unconditional Generative Models
Jesse Engel, Matthew Hoffman, Adam Roberts
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 |
|---|---|---|---|
| Add a latent-distance penalty to the conditional actor (ablation) | CelebA latent constraints CGAN · CelebA conditional generation | Accuracy: 0.942 -> 0.928 | Hurt |
| Add a latent-distance penalty to the conditional actor (ablation) | CelebA latent constraints CGAN · CelebA conditional generation | z_MSE: 80.7 -> 17.0 | Improved |
| Use a smaller actor-critic with 3 layers of 256 units (ablation) | CelebA latent constraints CGAN · CelebA conditional generation | Accuracy: 0.942 -> 0.926 | Hurt |
| Train an actor for the C Major pitch constraint | Melody VAE · Melody generation, C Major | Pitch reward: 0.579 (0.43%) -> 0.991 (70.8%) | Improved |
| Add the note-density constraint to the C Major actor | Melody VAE · Melody generation, density ≥192 | Density reward: 0.417 (0.04%) -> 0.985 (84.9%) | Improved |
| Set the CelebA VAE likelihood variance to σ_x² = 1e-2 rather than 1e-1 (ablation) | CelebA VAE · CelebA VAE | ELBO: -11475 -> 15080 | Improved |
| Set the CelebA VAE likelihood variance to σ_x² = 1e-3 rather than 1e-2 (ablation) | CelebA VAE · CelebA VAE | ELBO: 15080 -> 14140 | Hurt |
Closest papers by meaning
- Conditional Generative Modeling via Learning the Latent Space (ICLR 2021 · similarity 0.89)
- Ctrl-U: Robust Conditional Image Generation via Uncertainty-aware Reward Modeling (ICLR 2025 · similarity 0.85)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.85)
- A Theory for Conditional Generative Modeling on Multiple Data Sources (ICML 2025 · similarity 0.85)
- Return of Unconditional Generation: A Self-supervised Representation Generation Method (NeurIPS 2024 · similarity 0.85)
- Condition-Aware Neural Network for Controlled Image Generation (CVPR 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.
