Recently, the method DeepView [1] was proposed to visualize the decision function of high-dimensional classifiers in 2 dimensions. Thereby, descriminative dimensionality reduction is used to obtain a projections of the data points and the decision function. Several thesis topics are possible in this context:

• investigate popular networks with deepview
• extend deepview for the application to non
• standard data/models such as graph neural network or time series
• apply deepview in a transfer learning setting

Literature

1. Schulz, Alexander, Fabian Hinder, and Barbara Hammer. “Deepview: Visualizing classification boundaries of deep neural networks as scatter plots using discriminative dimensionality reduction.” https://www.ijcai.org/proceedings/2020/319