About Us

We are interested in explainable deep learning and interactive visualization.  Latent space visualizations can help improve the explainability of AI models by providing a visual representation of the hidden or intermediate features learned by the model. These visualizations can make it easier to understand the relationships between input data points and the model's internal structure, which can be especially useful for complex models like deep learning networks.

Some of our Projects

• Cxr8Explorer utilizes the Anat-0-Mixer control, which is an interactive latent space visualization of patient geometries

• pheonixrt is an embodiment of our variational inverse planning algorithm, utilizing the physician's intent dose targets as a prior and estimates the posterior (actual) DVH using a gradient-friendly approximation.

  • The commonly used Kullback-Liebler divergence provides a robust means of performing the estimation.

  • we are currently exploring ways that the variational objective can be used as part of a compositional energy-based model, which can provide a powerful means of combining multiple knowledge sources to guide synthesis in an appropriate latent space.

• ALGT prolog predicates for the verification of geometric planning parameters.

We are also interested in architectural patterns for medical and scientific visualization that help organize software components and their interactions to efficiently manage, process, and visualize complex medical or scientific data.

Copyright (C) 2025 2nd Messenger Systems