Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Variational Bayesian Unlearning
Quoc Phong Nguyen, Bryan Kian Hsiang Low, Patrick Jaillet
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 |
|---|---|---|---|
| minimize reverse KL instead of EUBO (ablation) | Bayesian unlearning without adjusted likelihood (lambda=0) · Synthetic linear regression | KL divergence: 266.68 -> 0.0158 | Improved |
| add adjusted likelihood (lambda 0 -> 0.1) to EUBO (ablation) | EUBO unlearning · Synthetic linear regression | KL divergence: 266.68 -> 0.0899 | Improved |
| add adjusted likelihood to reverse KL (lambda 0 -> 0.1) (ablation) | rKL unlearning · Synthetic linear regression | KL divergence: 0.0158 -> 0.0600 | Hurt |
| add adjusted likelihood to EUBO (lambda 0 -> 1e-9) (ablation) | EUBO unlearning · Moon sparse GP classification | avg predictive KL | Improved |
| add adjusted likelihood to EUBO (lambda 0 -> 1e-20) (ablation) | EUBO unlearning · Airline sparse GP | KL divergence: 2629.71 -> 1384.96 | Improved |
| add adjusted likelihood to rKL (lambda 0 -> 1e-13) (ablation) | rKL unlearning · Airline sparse GP | KL divergence: 455.11 -> 367.12 | Improved |
| reverse KL instead of EUBO (ablation) | unlearning · Airline sparse GP | KL divergence: 2629.71 -> 455.11 | Improved |
| replace Gaussian posterior with normalizing flow (ablation) | EUBO and rKL on logistic regression · Banknote authentication | avg predictive KL | Improved |
| unlearning with rKL at lambda=0 vs no unlearning (ablation) | Fashion MNIST logistic regression · Fashion MNIST | avg predictive KL | Improved |
| EUBO at lambda=0 vs adjusted lambda (ablation) | EUBO unlearning · Fashion MNIST | avg predictive KL | Hurt |
Closest papers by meaning
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.87)
- Hessian-Free Online Certified Unlearning (ICLR 2025 · similarity 0.87)
- Exact Unlearning via Quantized Sufficient Statistics (NeurIPS 2026 · similarity 0.87)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.86)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.86)
- Generalized Variational Inference via Optimal Transport (AAAI 2024 · 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.
