Session

DescriptionThis session presents state-of-the-art research in architecture design focusing on tackling irregular data access problems incurred in graph neural networks and accelerator architectures. One presentation deals with a Video Frame Interpolation accelerator whose energy efficiency surpasses a RTX 4090 GPU. The second half of the session offers a slew of Graph Neural Network (GNN) centric accelerators, dealing with the common irregular data access issues encountered in these GNNs.