Title: Machine Learning on FPGAs - Opportunities and Perspectives
Speaker: Jürgen Teich, Computer Science, Friedrich-Alexander-Universität Erlangen Nürnberg (FAU), Germany
Abstract: The ever-increasing computing power of GPUs has fueled machine learning based on artificial neural networks. AI is ubiquitous and dominates many fields of application, such as computer vision and natural language processing, but it is also gaining momentum in vehicles as well as industrial automation and maintenance. Here, FPGAs are considered a viable solution that can be highly customized to the application needs for implementing ML tasks on embedded systems or edge devices, which typically have stringent performance and power requirements.
This talk first provides an overview of the landscape of compilation and design automation flows targeting FPGA-based accelerators. Here, we classify existing approaches into overlay-based and dedicated ones, and examine their supported types of neural networks, interoperability with common ML frameworks, and which non-functional properties are in focus. Next, we provide recent examples of significantly speeding up ML applications on RISC-V processors by implementing highly optimized custom functional units in programmable logic. Finally, we briefly introduce spiking neural networks (SNNs) and address whether FPGA implementations of SNNs are superior to their CNN counterparts.
Title: FPGA design for network security
Speaker: Nele Mentens, Applied Cryptography and Security, Leiden Institute of Advanced Computer Science (LIACS), Belgium
Abstract: Individuals and organizations increasingly use online services and applications. As a result, the amount of Internet traffic continuously increases. Unfortunately, this had also led to an escalation of criminal activities in computer networks. Security mechanisms need to be able to respond to this growing attack surface, even in high-bandwidth networks. Research has shown that FPGAs can help in achieving highly effective and highly efficient network security solutions. This talk will give an overview of recent developments in the area of FPGA design for high-speed network security.
Title: DISCRETION: disruptive secure communications for European Defence
Speaker: Catarina Bastos, Deimos Engenharia, Portugal
Abstract: DISCRETION intends to develop an SDN solution integrating QKD capabilities to support optical secure communications, in a way that European Defence can benefit from these technologies to be effective not only for the network situational awareness, but also for the cyber situational awareness. Cipher Machines will be the components responsible for assuring data protection in DISCRETION, enabling real-time data encryption and decryption, using key material provided by the SDN-QKD plane as well as pre-shared keys. It shall be considered the red-black architecture of the military networks. For mobility and tactical scenarios SDR solutions shall be analysed and integrated into the SDN solution to cover radio network segments and support secure communication services in mobile scenarios. DISCRETION solution is based on SDN and QKD that allow to improve security and resilience in exchange of information and in communication services.