• Journal of the Korea Society of Computer and Information
• /
• v.29 no.11
• /
• pp.97-105
• /
• 2024

Lots of tasks including physical systems modeling, chemical reaction prediction, and relation extraction require dealing with higher-order relations. Graph neural networks (GNNs) are favorite models for relational data but they have inherent limits due to their focus on pairwise relationships. Topological data analysis (TDA) provides insight into the "shape" of data (or underlying data topology). TDA aims to infer information about data manifold such as connectivity and offers higher-dimensional analog of graphs. Topological deep learning (TDL) combines various deep learning techniques with TDA. TDL enables us to formulate simplicial complex and cell complex through techniques such as low-dimensional embedding and attention. In this paper, we summarize recent achievements especially on simplicial complex and cell complex. We also provide succinct descriptions of related concepts.