A5063254356- Advanced Graph Neural Networks
- Ferroelectric and Negative Capacitance Devices
- Graph Theory and Algorithms
- Advanced Memory and Neural Computing
- Caching and Content Delivery
Yonsei University
2021-2023
Graph convolutional networks (GCNs) are becoming increasingly popular as they overcome the limited applicability of prior neural networks. One recent trend in GCNs is use deep network architectures. As opposed to traditional GCNs, which only span around two five layers deep, modern now incorporate tens hundreds with help residual connections. From such we find an important characteristic that exhibit very high intermediate feature sparsity. This reveals a new opportunity for accelerators...
GCNs (Graph Convolutional Networks) are becoming increasingly popular in the field of neural networks due to their ability analyze many kinds irregular data. Along with rapid growth, there various accelerators being proposed mitigate huge computational requirements. Often, key bottleneck executing is at random accesses posed on wide feature array. Vertex tiling a technique address issue, but has drawback putting too much repetition data and hard tune parameters. In such regard, we propose...
Graph convolutional networks (GCNs) are becoming increasingly popular as they can process a wide variety of data formats that prior deep neural cannot easily support. One key challenge in designing hardware accelerators for GCNs is the vast size and randomness their access patterns which greatly reduces effectiveness limited on-chip cache. Aimed at improving cache by mitigating irregular accesses, studies often employ vertex tiling techniques used traditional graph processing applications....
Graph convolutional networks (GCNs) are becoming increasingly popular as they can process a wide variety of data formats that prior deep neural cannot easily support. One key challenge in designing hardware accelerators for GCNs is the vast size and randomness their access patterns which greatly reduces effectiveness limited on-chip cache. Aimed at improving cache by mitigating irregular accesses, studies often employ vertex tiling techniques used traditional graph processing applications....
Graph convolutional networks (GCNs) are becoming increasingly popular as they overcome the limited applicability of prior neural networks. A GCN takes input an arbitrarily structured graph and executes a series layers which exploit graph's structure to calculate their output features. One recent trend in GCNs is use deep network architectures. As opposed traditional only span around two five deep, modern now incorporate tens hundreds with help residual connections. From such GCNs, we find...