deep-ensembles

This repository provides the code used to create the results presented in "Global canopy height regression and uncertainty estimation from GEDI LIDAR waveforms with deep ensembles".

Code for the paper "Beyond Deep Ensembles: A Large-Scale Evaluation of Bayesian Deep Learning under Distribution Shift"

Single-cell Consensus Clusters of Encoded Subspaces

Ensemble deep learning of embeddings for clustering multimodal single-cell omics data

HARNet: Towards On-Device Incremental Learning using Deep Ensembles on Constrained Devices for Human Activity Recognition

Official code for the TMLR 2025 paper: "On Joint Regularization and Calibration in Deep Ensembles"

Code for the paper "Something for (almost) nothing: improving deep ensemble calibration using unlabeled data"

Label-free confidence estimation for neural networks - benchmarking MSP, MC Dropout, EDL, and Deep Ensembles with a novel unsupervised uncertainty metric.

Testbed to compare simple deep ensembles

Reproduction of Lakshminarayanan et al. (2017) deep ensembles — UCI regression, MNIST/notMNIST OOD detection, and calibration under adversarial attack

PINN framework for aerospace asset risk pricing: monotonic RUL prediction, deep ensemble uncertainty, Monte Carlo residual value, lease pricing, portfolio optimisation. Results: RUL 290±90 cycles, residual 0.19M , l e a s e r a t e 0.19M, lease rate 300k/month. Modular for real NGAFID/FAA data.

Brain Tumor Segmentation with Deep Learning and Deep Ensembles

Reliability analysis for MICCAI BraTS-GoAT 2026, comparing single-model confidence against deep-ensemble disagreement on calibration and error detection under graded synthetic acquisition shift.

Reproduction code + artifacts for “When Do Deep Ensembles Improve Robustness to Spurious Correlations?” (TU Delft, 2026).

Implementation of MC-Dropout and Deep Ensembles for uncertainty estimation in regression (UCI Energy) and classification (Fashion-MNIST) tasks. Includes OOD detection with notMNIST.

Fast stress prediction with built-in reliability: a GNN surrogate that knows when to trust itself and when to defer to FEA simulation.