1: \begin{abstract}

2: Unsupervised data representation and visualization using tools from topology is an active and growing field 

3: %at the intersection 

4: of Topological Data Analysis (TDA) 

5: %, non-convex optimization 

6: and data science.

7: Its most prominent line of work is based on the so-called {\em Mapper graph},

8: which is a combinatorial graph 

9: %built from the data set, and 

10: whose topological structures (connected components, branches, loops) are in correspondence with those of the data itself.

11: While highly generic and applicable, its use has been hampered so far by the manual tuning of its many

12: parameters---among these, a crucial

13: %, which is often required in order to get relevant outputs.

14: one is the so-called {\em filter}: it is a continuous function whose variations on the data set are the main

15: ingredient for both building the Mapper representation and assessing the presence and sizes of its topological structures.

16: However, while a few parameter tuning methods have already been investigated for the other Mapper parameters (i.e., resolution, gain, clustering), 

17: there is currently no method for tuning the filter itself. In this work, we build on a recently proposed 

18: optimization framework incorporating topology to provide the first filter optimization scheme for Mapper graphs.

19: In order to achieve this, we propose a relaxed and more general version of the Mapper graph, whose convergence

20: properties are investigated. Finally, we demonstrate the usefulness of our approach by optimizing Mapper 

21: graph representations on several datasets, and showcasing the superiority of the optimized representation over arbitrary ones.

22: \end{abstract}

23: