PAPER 4
Probabilistic Numeric Convolutional Neural Networks
Marc Finzi (NYU),
Roberto Bondesan (Qualcomm AI Research*),
Max Welling (formerly Qualcomm AI Research)
ICLR 2021
Summary
Continuous input signals (e.g., images and time series) that are irregularly sampled or have missing values are challenging for existing deep learning methods. Probabilistic Numeric Convolutional Neural Networks proposes another advanced type of CNN. This CNN borrows concepts from probabilistic numeric theory and represents features as Gaussian processes, providing a probabilistic description of discretization error.
Citation
@article{finzi2020probabilistic, title={Probabilistic numeric convolutional neural networks}, author={Finzi, Marc and Bondesan, Roberto and Welling, Max}, journal={arXiv preprint arXiv:2010.10876}, year={2020}}
Results
The solution shows a 3x reduction error on computer vision tasks from the previous state-of-the-art results on a benchmark dataset.
Looking for more papers with code?
* Qualcomm AI Research is an initiative of Qualcomm Technologies, Inc.
