The Evolution of Deep Learning in Practice: The Move Toward Data-Centric Deep Learning

Deep Learning has been developed at universities, where the circumstances differ dramatically from the challenges in the field. While researchers usually start from an existing dataset and try to develop algorithms able to shave off an additional fraction of a percentage point of the result, in a practical application, like industrial quality inspection, no such dataset exists. The first challenge is to build a proper data pipeline, which has far more impact on the success of the project than utilizing the latest innovations. What innovations and best practices can support this often overlooked but critical aspect of Deep Learning?

About the presenter
Christian Eckstein is the product manager for the Deep Learning Tool at MVTec Software GmbH in Munich. In his role, he is responsible for the development and implementation of a long-term product strategy, along with continuous further development of the application possibilities of deep learning in machine vision. Eckstein's initial experience working at MVTec was as a student employee from 2009 to 2015 while studying computer science and economics. Afterward, he spent four years as a team leader and project manager, leading the implementation, integration, and international rollout of IT systems for several international companies. Today, Eckstein has acquired extensive knowledge of the possibilities and current challenges of deep learning in machine vision applications. As one of the deep learning experts within MVTec, he is constantly exploring more ways to integrate deep learning into machine vision technologies to add additional value for customers.