AI Paper Atlas

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

Meta-Learning Theory-Informed Inductive Biases using Deep Kernel Gaussian Processes

Bahti Zakirov, Gasper Tkacik

ICLR 2026 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Structured Gaussian process models

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 changedWhereMeasuredEffect
use meta-learned theory-informed kernel instead of plain RBF GPdeep kernel GP · Mouse RGC responses (86 neurons)test Pearson correlation: Wilcoxon p=6.14e-03 at N=64, p=9.37e-05 at N=1452Improved
replace meta-learned feature extractor with random weights (ablation)Theory-Informed Kernel · Mouse RGC responses (86 neurons)test Pearson correlation: median +0.049 at N=512Improved
meta-train on theory-derived efficient-coding tasks instead of generic natural-image PCA tasks (ablation)Theory-Informed Kernel · Mouse RGC responses (86 neurons)test Pearson correlation: Wilcoxon p=5.04e-03 at N=128, p=2.47e-01 at N=1452Improved
add task-adaptive linear heads (ablation)Theory-Informed Kernel · Mouse RGC responses (86 neurons)test Pearson correlation: Wilcoxon p=2.77e-10 at N=32Improved
add meta-learned deep feature extractor on top of identity (linear heads on raw pixels) (ablation)Theory-Informed Kernel · Mouse RGC responses (86 neurons)test Pearson correlation: Wilcoxon p=6.67e-09 at N=1024Improved
use theory-informed kernel instead of LN model at very low dataTheory-Informed Kernel · Mouse RGC responses (N<=8 and N<16)test Pearson correlationHurt
use pretrained ResNet-18 (ImageNet or SimCLR) frozen features with fitted heads (transfer learning)RGC response prediction · Mouse RGC responses (86 neurons)test Pearson correlationHurt
use theory-informed kernel instead of RBF GP for uncertaintydeep kernel GP · Mouse RGC responses (86 neurons)NLPDImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy