Research map / Optimization / Implicit bias and generalization
Inverting Deep Generative models, One layer at a time
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 |
|---|---|---|---|
| Use layer-wise ℓ∞ LP inversion instead of gradient descent | ReLU generator inversion · Random networks, noiseless; varying latent dimension k | recovery success rate | Helped |
| Use layer-wise LP inversion instead of gradient descent | MNIST generator inversion · MNIST | CPU time: 72 -> 0.5 | Helped |
| Use ℓ∞ LP rather than gradient descent | ReLU generator inversion · Synthetic data with uniformly bounded noise | relative recovery error | Helped |
| Use ℓ∞ LP rather than ℓ1 LP or gradient descent | ReLU generator inversion · Synthetic data with large-variance observation noise | relative recovery error | Hurt |
| Use ℓ∞ LP, ℓ1 LP, or LP relaxation (ablation) | ReLU generator inversion · Synthetic noisy inversion | relative recovery error | No clear effect |
| Relax strict ReLU-configuration constraints with LP relaxation (ablation) | ℓ1 LP-style inversion · Synthetic noisy inversion | relative recovery error | No clear effect |
| Use layer-wise LP inversion rather than gradient descent | MNIST generator inversion · MNIST noisy reconstruction | reconstruction quality | Helped |
| Use layer-wise LP inversion rather than gradient descent | MNIST generator inversion · MNIST noisy and masked observations | recovery quality | Helped |
| Use ℓ1 LP rather than ℓ∞ LP | ReLU generator inversion · Synthetic data with large-variance observation noise | relative recovery error | Helped |
Closest papers by meaning
- PLUGIn: A simple algorithm for inverting generative models with recovery guarantees (NeurIPS 2021 · similarity 0.91)
- Invertibility of Convolutional Generative Networks from Partial Measurements (NeurIPS 2018 · similarity 0.91)
- Activation-Descent Regularization for Input Optimization of ReLU Networks (ICML 2024 · similarity 0.85)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.85)
- Generative Model Inversion Through the Lens of the Manifold Hypothesis (NeurIPS 2025 · similarity 0.85)
- Sparse Model Inversion: Efficient Inversion of Vision Transformers for Data-Free Applications (ICML 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.