Two contributions from the joint team of LIS with BMW, PoliTorino, and KIT have been accepted to the DATE 2022 conference. Congratulation to all the members of the team for their joint effort.
The first paper entitled "Mind the Scaling Factors: Resilience Analysis of Quantized Adversarially Robust CNNs" was authored by Nael Fasfous, Lukas Frickenstein, Michael Neumeier, Manoj Rohit Vemparala, Alexander Frickenstein, Emanuele Valpreda, Maurizio Martina and Walter Stechele. The authors investigate the negative effect of adversarial training on the resilience of convolutional neural networks (CNNs). While the weaknesses of the adversarially trained CNN are identified, a protection method is proposed to improve the fault resilience against bit flips.
The second paper entitled "AnaCoNGA: Analytical HW-CNN Co-design using Nested Genetic Algorithms" has been authored by Nael Fasfous, Manoj Rohit Vemparala, Alexander Frickenstein, Emanuele Valpreda, Driton Salihu, Julian Höfer, Anmol Singh, Naveen-Shankar Nagaraja, Hans-Joerg Voegel, Nguyen Anh Vu Doan, Maurizio Martina, Juergen Becker and Walter Stechele. The paper proposes a genetic algorithm approach where hardware architecture search for bit-serial FPGA accelerators and quantization search are nested into a combined design space exploration. By evaluating the hardware design Pareto-front of each considered quantization strategy, the search space is reduced considerably. This is enabled by a suitable hardware execution model.
The DATE conference is the main European event bringing together designers and design automation users, researchers and vendors, as well as specialists in the hardware and software design, test and manufacturing of electronic circuits and systems. DATE puts a strong emphasis on both technology and systems, covering ICs/SoCs, reconfigurable hardware and embedded systems, and embedded software.