Jason Mcguirk

Title: Jason McGuirk and His Work in Model Understandability.

Introduction

Jason McGuirk is an inventor based in Dublin, Ireland, associated with Optum, Inc. His patent work focuses on making machine-learning models easier to understand through visualizations and representations of model behavior.

Latest Patents

McGuirk is named on one patent, titled “Model Understandability.” The invention describes using interpolation to smooth animation frames by generating synthetic Shapley values, which are then used to create animations. It also describes generating an embedding space to help users investigate model training through corresponding Shapley values. Activations from a specially configured long short-term memory (LSTM) network are mapped into this space using an embedded representation generated by a specially configured autoencoder. The embedding space and animations can be displayed in an interface and customized to focus on particular data.

Career Highlights

The available information identifies McGuirk’s work through this patent and his association with Optum, Inc. The invention addresses model interpretability by combining Shapley-value-based explanations, animation, and an interactive embedding space.

Collaborations

McGuirk’s listed coworkers include Joao Gabriel Teixeira Nogueira and Johnathon E Schultz. The patent’s described approach brings together techniques for model analysis and visualization.

Conclusion

Jason McGuirk’s patent work centers on helping users investigate and understand machine-learning models through interactive visual explanations and representations.