Summary of GRASP Lab Days 2022 - Faculty Research Talks presents Dr. Pratik Chaudhari

00:00:00 - 00:30:00

In the video, Dr. Pratik Chaudhari discusses his research into information geometry and how it can be used to train artificial neural networks. He also discusses how human decision-making may be similar to how neural networks learn. Additionally, he discusses how machine learning can be adapted to different languages, specifically how it can be adapted to translate between English and French and Italian.

• 00:00:00 The research conducted in Pratik Chaudhari's group focuses on understanding why learning is effective in the modern age, and how to make certain machine learning algorithms easier to train. Their work has led to the development of methods to prevent overfitting in deep learning models.
• 00:05:00 The GRASP Lab Days 2022 presentation by Dr. Pratik Chaudhari discusses research into information geometry, which is a new field combining geometry and information theory. Chaudhari explains how probability distributions can be visualized as functions, and how these functions can be used to train artificial neural networks. Finally, he discusses how human decision-making may be similar to how neural networks learn.
• 00:10:00 The speaker discusses how machine learning can be adapted to different languages, specifically how it can be adapted to translate between English and French and Italian. He also discusses how the same structure can be used across different tasks in order to make the adaptation easier.
• 00:15:00 The video discusses a research project of Dr. Pratik Chaudhari's at the Georgia Institute of Technology, in which they are mapping forests in order to understand how they are growing and capturing carbon. They also have a project investigating how two teams of multiple agents fight each other.
• 00:20:00 Pratik Chaudhari discusses research into neural radiance fields, which are very upcoming and exciting. The video also provides a brief description of a deep learning course that he teaches.
• 00:25:00 The researchers at GRASP Lab focus on using methods from information theory and learning theory to study how the brain processes information. They also focus on developing principles for understanding how small parts of the physiology of the retina work.
• 00:30:00 In this video, faculty from the GRASP Lab discuss the efficiency of biological systems. They note that while these systems are inefficient in the sense that each of the neurons do slightly different things, this is an artifact of growth and allows for more efficient performance. Additionally, they discuss the theory of how systems evolve, and point out that while it is achievable, most problems will not be solved this way.