The proposed topic seeks to explore how an algorithm for big data visualization might be developed that combines the desirable properties of the existing algorithms for mining of big data. As depicted in Figure 1, the aim is to create an interactive visualization tool for big data that incorporates user feedback into the algorithm in order to give some degree of freedom for the user to manipulate the visualization process according to their preference. Similar to how SOM, MDS, and PCA preserve neighborhoods, distances, and variances, respectively, this visualization tool allows user to inject domain knowledge and user preference to the visualization process.

A possible approach to handle big data is to only keep points that would be relevant for the visualization and discard the rest, just like the approach of BFR. The problem would be on how to decide which points are relevant and which points are not. The decision can be based on the inputs of the user.

A similar work is reported in \cite{sumi1997computer} on computer-aided thinking where MDS was used to visualize thought spaces which can be manipulated by the user to allow idea formulation. As shown in Figure 2, users can arrange points in the thought spaces, fix points to certain locations, and the rest of the points will position themselves with respect to the fixed points.

The concept may be related to the game of TextTwist where you can rearrange letters, which in their case are words, to discover and formulate solutions that might not be obvious in some arrangements.