Trustworthy Adaptive Computing

The unit combines the expertise of the “Trustworthy and Efficient Collaborative AI” (TEC-AI) team, which focuses on trustworthy and collaborative edge AI, with that of the “Hardware Accelerators and Design” (HAND) team, which specializes in hardware-software co-design and domain-specific accelerators.

The TEC-AI team’s work focuses on developing trustworthy AI for edge and distributed environments and encompasses federated learning, federated distillation, and multi-agent reinforcement learning. The team develops explainable and formally verified AI systems and applies adaptive shielding in conjunction with machine learning to meet the requirements of safety-critical applications. Targeted efforts toward efficient AI deployment ensure that these solutions remain viable even on resource-constrained edge platforms.

The HAND team’s research expertise encompasses domain-specific AI architectures as well as compile-time middleware, profiling, and runtime support, including support for post-quantum-secure cryptography. Its hardware/software co-design approach, based on RISC-V and custom accelerators, paves the way for secure edge learning and end-to-end AI integration.