The emergence of Artificial General Intelligence (AGI) demands reasoning systems that are both expressive and computationally efficient. Neuro-Symbolic (NeSy) AI, which combines the symbolic and neural approaches, has emerged as a promising framework to bridge this gap by uniting the interpretability of symbolic AI with the adaptability of neural networks. However, NeSy AI systems face a fundamental challenge: the heterogeneous kernels (neural network and probabilistic AI computation) are unevenly explored at the current stage. While the hardware community has primarily focused on accelerating neural networks, comparatively less attention has been given to expressive probabilistic models (PMs). This thesis tries to reduce the gap by focusing the hardware acceleration of probabilistic inference techniques for such models by proposing novel architectures, algorithms, and co-design methodologies.
10/10/2025 10:00 - 12:00
ESAT Aula L
