Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Is the Last Layer Sufficient for Uncertainty Quantification?
Joseph Wilson, Chris van der Heide, Liam Hodgkinson, Fred Roosta
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 |
|---|---|---|---|
| linearize only the last layer (LL-GLM) instead of the full network (DNN-GLM) | Bayesian GLM UQ · UCI regression (9 datasets) | RMSE, NLL, ECE | No clear effect |
| use last-layer GLM instead of full GLM | GPT-2 fine-tuned on IMDB · IMDB | VARROC-ID: 0.842 -> 0.823 | No clear effect |
| use last-layer GLM instead of full GLM | GPT-2 fine-tuned on IMDB · IMDB | wall-time and memory: 2.55 hr -> 0.26 hr; 71.23 GB -> 20.97 GB | Improved |
| use last-layer GLM instead of full GLM | image classifiers (LeNet, ResNet9, ResNet50) · image classification, 8 model/dataset pairs | VARROC and LPPD: average relative difference V-ID -1.14%, V-OOD 1.20% | No clear effect |
| use last-layer GLM instead of full GLM | image classifiers · image classification, 8 model/dataset pairs | time and memory: average reduction 58.44% time, 57.32% memory | Improved |
| use full NTK (DNN-GLM) instead of CK (LL-GLM) at random initialization when training points greatly exceed width | GP kernel at initialization · theory, over-sampled regime | Bayes free energy | Improved |
| train the network before forming the CK and NTK | CK vs NTK · teacher-network and Gaussian data | Bayes free energy | Improved |
| restrict weak descent to NTK versus CK parameter region | CK vs NTK · theory, weak descent | descent region | Mixed |
| compare LL-GLM and DNN-GLM against deep ensembles | GLMs · image classification | VARROC | Hurt |
Closest papers by meaning
- Epistemic Uncertainty Quantification For Pre-Trained Neural Networks (CVPR 2024 · similarity 0.93)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.89)
- Accurate and Scalable Estimation of Epistemic Uncertainty for Graph Neural Networks (ICLR 2024 · similarity 0.87)
- Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage? (NeurIPS 2024 · similarity 0.87)
- Uncertainty Estimation by Flexible Evidential Deep Learning (NeurIPS 2025 · similarity 0.87)
- Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods? (ICML 2024 · similarity 0.87)
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.
