Research map / Probabilistic and scientific modeling / Causal discovery
Mitigating the Effects of Non-Identifiability on Inference for Bayesian Neural Networks with Latent Variables
Yaniv Yacoby, Weiwei Pan, Finale Doshi-Velez
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 |
|---|---|---|---|
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Heavy Tail | test log-likelihood: -1.867 ± 0.078 -> -1.426 ± 0.042 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Goldberg | test log-likelihood: -1.026 ± 0.056 -> -0.963 ± 0.041 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Williams | test log-likelihood: -1.033 ± 0.156 -> -0.414 ± 0.184 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Yuan | test log-likelihood: -1.278 ± 0.164 -> -1.211 ± 0.083 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Depeweg | test log-likelihood: -2.342 ± 0.048 -> -1.973 ± 0.049 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Abalone | test log-likelihood: -0.843 ± 0.071 -> -0.831 ± 0.086 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Airfoil | test log-likelihood: -0.512 ± 0.083 -> -0.462 ± 0.056 | Improved |
| Use NCAI instead of mean-field VI | BNN+LV · Energy Efficiency | test log-likelihood: 0.573 ± 0.288 -> 0.898 ± 0.452 | Hurt |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Lidar | test log-likelihood: 0.129 ± 0.131 -> 0.263 ± 0.11 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Wine Quality Red | test log-likelihood: -1.709 ± 0.22 -> -0.849 ± 0.038 | Improved |
| Constrain latent noise to be less dependent on inputs and closer to its prior | BNN+LV · Yacht | test log-likelihood: 0.638 ± 0.121 -> 0.832 ± 0.077 | Improved |
| Select nonzero constraint strengths rather than fixing them at zero (ablation) | BNN+LV NCAI · Heavy Tail | test log-likelihood: -1.481 ± 0.018 -> -1.426 ± 0.042 | Improved |
Closest papers by meaning
- Do Bayesian Neural Networks Actually Behave Like Bayesian Models? (ICML 2025 · similarity 0.83)
- Concept Modulation Models: A Unified Framework for Identifiability and Extrapolation (NeurIPS 2026 · similarity 0.83)
- Enhancing Uncertainty Estimation and Interpretability with Bayesian Non-negative Decision Layer (ICLR 2025 · similarity 0.82)
- Identifiable Latent Polynomial Causal Models through the Lens of Change (ICLR 2024 · similarity 0.82)
- Adjustment for Confounding using Pre-Trained Representations (ICML 2025 · similarity 0.82)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.82)
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.
