I present QoI-aware encoders-decoders from our Python library - PCAfold! They allow for improved representation learning from high-dimensional datasets.
Our paper:
www.cell.com/patterns/fulltext/S2666-3899(23)00243-X
0:00 The need for QoI-informed representation learning
2:32 How can inform projections by QoIs?
4:21 Application to reduced-order modeling
5:40 Constructing projection-dependent QoIs
7:10 Animation of projecting a synthetic dataset
7:44 Animation of varying projection-dependent QoI
8:24 Why does it work?
9:39 QoIAwareProjection class
12:14 Initialize with PCA
13:14 Print model summary & train
14:04 Access the optimized weights and biases
15:26 Plot losses
Check out the software documentation below:
pcafold.readthedocs.io/
You can download the software through GitLab:
gitlab.multiscale.utah.edu/common/PCAfold
or GitHub:
github.com/kamilazdybal/PCAfold
PCAfold is published in SoftwareX:
www.sciencedirect.com/science/article/pii/S2352711020303435
www.sciencedirect.com/science/article/pii/S2352711023001437
In the next videos, we are going to dive deeper into particular applications of PCAfold functionalities. Stay tuned!
✨ If you'd like to support my efforts, you can become a member of our community and gain access to exclusive content!
buymeacoffee.com/kamilazdybal