a PENTA Call 3 project
Machine learning on the edge nodes, for IoT security analytics and anomaly detection
Cloud platform applying machine learning techniques for sharing relevant security data
Homomorphic encryption as privacy enhancing technology for Industry 4.0
Manufacturing technologies for uniquely secure low-footprint ASICs
SunRISE targets a crucial point in future IoT systems: a comprehensive chain of security evidence gathering and dissemination. Leveraging on recent advances in semiconductor manufacturing, machine learning, and privacy-preserving technologies, SunRISE targets:
Implementation of novel Privacy-Preserving Techniques (PPT) for Machine Learning (ML)
Development of hardware accelerators for cloud and edge computing
High volume production of immutable, hard-coded, unique identities to secure IoT devices in CMOS 200nm technology
Design of system and communication architectures enabling security by design
The Importance of Data Privacy
The benefits of Machine Learning (ML) are nowadays exploited in various fields such as smart home, healthcare & cloud infrastructure, but also in various other fields such as autonomous driving, facial recognition, etc. Already today ML has become the state-of-the art solution for many safety- and security-critical applications. In the context of cybersecurity, ML is part of most state-of-the-art solutions, e.g. for anomaly, intrusion detection, allowing to identify malicious e-mails and protect users from fraudulent actions.
But all the promising benefits come at a high price: The huge amounts of data collected about each individual can easily be misused to their detriment. In the following, 3 project-related examples are given of how the misuse of data can cause considerable damage:
Energy communities: From fine grained electricity consumption data, it is possible to forecast future electricity consumption and detected anomalies in the consumption, e.g. if someone would intentionally tamper with the data. But it is also possible to observe customer behavior and infer information about when someone will most likely be at home and when not.
Cloud Infrastructure: Similar to energy communities, lots of information of can be inferred over the behavior from the warm water consumption.
Healthcare: Modern approaches to detect cancer in MRI images are often based on ML algorithms, but the health data of a patient is very privacy-sensitive.
The SunRISE project aims to develop novel Privacy-Preserving Techniques (PPT) for Machine Learning. The following algorithms are investigated in the project:
Homomorphic Encryption (HE)
Federated Machine Learning (FedML)
Secure Multi-Party Computation (MPC)
Differential Privacy (DP)
Holistic Approach to ethical and trustworthy AISource: The Moral of Algorithms
Ethics of AI systems
In 2021 NXP GE has taken its stance when it comes to data privacy and published the whitepaper The Morals of Algorithms, in which the importance of Data Privacy and NXP's vision for the coming years is explained.
NXP GE has taken action and invested in the development of hardware accelerators, that could be the foundation of a new state-of-the-art solution for data privacy. With NXP's expertise of hardware development in the security domain, NXP develops hardware accelerators that increase the performance of the computational expensive operations of homomorphic encryption schemes by parallelization of the operations. Furthermore, NXP's i.MX-family allows to efficiently perform the inference of deep neural networks on the edge, by making use of the 2.3 TOPS NPU processor. This way all privacy-sensitive data can remain on the edge nodes and doesn't have to be shared with any centralized instance. With the involvement in research projects like SunRISE, NXP is part of shaping the future standards of data privacy and collaborates with universities and industrial partners that share the same vision.