Quantifying Uncertainties of Deep Neural Networks and Its Applications

Speaker: Dr. Jae Oh Woo

From: Samsung SDS Research America

Abstract

In this presentation titled "Quantifying Uncertainties of Deep Neural Networks and Its Applications," we delve into the critical task of measuring uncertainties in Deep Neural Networks. The talk will commence with a concise overview of various methodologies for quantifying uncertainty, underlining their significance in deep learning. Following this, we will explore applying these techniques to enhance active learning and language model instruction-tuning through the lens of curriculum learning. By integrating uncertainty quantification into these areas, we aim to demonstrate how it can significantly contribute to more efficient learning strategies and improved model performance. This presentation seeks to illuminate the pivotal role of uncertainty measurement in advancing the capabilities of neural networks, fostering a deeper understanding and encouraging further exploration in this area of research.

For more info, please follow this link.