Gábor Bartók

Title: Gábor Bartók: Inventor in Multi-Task Machine Learning

Introduction

Gábor Bartók is an inventor associated with Zürich, Switzerland, and Google Inc. His patent work addresses flexible parameter sharing in multi-task machine learning.

Latest Patents

Bartók is credited with 1 patent: “Flexible Parameter Sharing for Multi-Task Learning.” The patent describes systems and methods for training a multi-task machine-learned model. The method selects a task, uses test input and connection probability matrices in a forward pass to generate a sample distribution of test outputs, and trains model components based partly on that distribution. It also trains a connection probability matrix through a backward pass using a straight-through Gumbel-softmax approximation.

Career Highlights

Bartók’s listed patent reflects work on methods for sharing model parameters flexibly across tasks. The approach uses connection probability matrices as part of the model-training process.

Collaborations

Krzysztof Stanislaw Maziarz is identified as Bartók’s coworker. Both are associated with Google Inc.

Conclusion

Gábor Bartók’s patent contribution focuses on flexible parameter sharing for multi-task learning, combining task-specific training with a method for learning connection probabilities.