Deep Additive Kernel (DAK)

This repository implements Deep Additive Kernel (DAK) model in "From Deep Additive Kernel Learning to Last-Layer Bayesian Neural Networks via Induced Prior Approximation" (AISTATS 2025).

Model

Model architecture of Deep Additive Kernel (DAK).

Benchmark

• DNN (torchvision)
• NN+SVGP (GPyTorch)
• SV-DKL (GPyTorch)
• AV-DKL (gpinfuser folder from this repo)
• DAK (dak folder)

Usage

To reproduce the experiments, first install the required packages.

$ pip install -r requirement.txt

Toy Example

Jupyter notebook for the toy example: examples/notebooks/2_DKL_example.ipynb

UCI Regression

$ cd examples/uci
$ python run_uci.py 

Image Classification

• MNIST

$ cd examples/mnist
$ python run_mnist.py 

• CIFAR

$ cd examples/cifar
$ python run_cifar.py 

Citation

If you find our work relevant to your research, please cite:

@inproceedings{zhao2025deep,
  title={From Deep Additive Kernel Learning to Last-Layer Bayesian Neural Networks via Induced Prior Approximation},
  author={Zhao, Wenyuan and Chen, Haoyuan and Liu, Tie and Tuo, Rui and Tian, Chao},
  booktitle={The 28th International Conference on Artificial Intelligence and Statistics},
  year={2025}
}