Presentation 1-5 (invited talk), 16:00~16:45.
Speaker
Haeun Moon (Seoul National University)
Title
Topological Data Analysis and Machine Learning
Abstract
Topological Data Analysis (TDA) provides a way to capture the shape of data through its topological features. A key tool in TDA is persistent homology, which tracks how these features appear and disappear across multiple scales. In recent years, TDA has shown great promise in supporting and enriching machine learning applications. This talk will explore how TDA can be used to enhance machine learning, focusing on two main aspects: feature extraction and model evaluation.
Although persistent homology encodes rich structural information, it is not always straightforward to use directly in learning algorithms. I will introduce several techniques that transform topological summaries into forms suitable for neural networks and other models, enabling visualization, dimensionality reduction, and predictive analysis.
Lastly, I will show how TDA can also be applied to evaluate data and model performance. By analyzing the topology of generated or transformed data, TDA-based methods provide new and reliable ways to assess the quality of learning models, especially in generative modeling contexts.