Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes

Heteroscedastic Variational Last Layers

NeurIPS 2026 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Variational Bayesian deep learning · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes heteroscedastic variational last layers (t-VBLL) that perform variational inference over the aleatoric noise covariance of a neural network's last layer for cheap uncertainty quantification.

Problem

Heteroscedastic (input-dependent aleatoric) uncertainty quantification in neural networks, which is typically costly or unstable to obtain with standard methods.

Approach

Builds on Variational Bayesian Last Layers (VBLL) with deterministic training objectives, extending them with t-VBLL layers that run variational inference for the aleatoric noise covariance, so heteroscedastic uncertainty comes at near-deterministic cost.

Experimental setup

Builds on: Variational Bayesian Last Layers

Closest papers by meaning

• Variational Bayesian Last Layers (ICLR 2024 · similarity 0.94)
• Bayesian Optimization via Continual Variational Last Layer Training (ICLR 2025 · similarity 0.88)
• Neural Conditional Probability for Uncertainty Quantification (NeurIPS 2024 · similarity 0.87)
• Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.87)
• Enhancing Uncertainty Estimation and Interpretability with Bayesian Non-negative Decision Layer (ICLR 2025 · similarity 0.86)
• Maxitive Donsker-Varadhan Formulation for Possibilistic Variational Inference (NeurIPS 2026 · similarity 0.86)