Options
Intellectual Property in the Age of Self-Learning Systems (Neural-IP)
Type
fundamental research project
Start Date
2019
End Date
2020
Status
ongoing
Keywords
artificial intelligence
intellectual property
deep neural networks
Description
Deep Neural Networks (DNN) are currently the key technology for autonomous vehicles, algorithmic trading, or decision-making systems in human resource, legal, and other robotic process automation scenarios. A key characteristics of these systems is their evolving nature as they are self-learning unlike traditional software, which is programmed. This, however, raises new challenges to known concepts such as intellectual property (IP) management. This project aims to re-think the current definition of IP in the context of deep neural networks. It further investigates how original work can be understood in the realm of continuously trained deep neural networks. Additionally, it aims to develop technical solutions as proof of concept to allow for similarity measurements between different DNN models. This is particularly novel, since DNNs are represented by millions of weight values. Compared to traditional software code, which consists of human-readable and human-interpretable text, the interpretation of DNN weight values present a task beyond human abilities. In summary, this project addresses an important issue with DNNs, the driving technology behind modern artificial intelligence. It highlights the unique opportunity for interdisciplinary work at University of St. Gallen (HSG) at the intersection of technology management and technological implementation.
Leader contributor(s)
Funder(s)
Range
HSG Internal
Range (De)
HSG Intern
Division(s)
Eprints ID
247843