Our paper on Set Prediction was accepted at International Conference on Learning Representations (ICLR 2021). This research is about predicting sets such as reconstructing point clouds or identifying the relevant subset from a large set of samples.

One example is to identify which of the input elements constitute the subset that shares a common feature. E.g., the subset of faces that wear glasses, or have gray hair, etc., - without telling the model which commonality to look for.

The scientific contributions are (a) to analyze sets which can have different input and output sizes, and, (b) to deal with ambiguity (e.g., both the subsets of glasses and gray hair are plausible). More details and paper will be shared soon.

This research is a collaboration with the University of Amsterdam (professor Cees Snoek). PhD student David Zhang is the first author and is supervised by UvA and TNO. Part of the NWO Efficient Deep Learning research program.