2018 Choice Prediction Competition (CPC18)
Subscribe to Decision Science News by Email (one email per week, easy unsubscribe)
WINNERS GET TO BE CO-AUTHORS ON THE PAPER
Ori Plonsky, Reut Apel, Ido Erev, Eyal Ert, and Moshe Tennenholtz announce
You are invited to participate in the 2018 choice prediction competition (CPC18) for human decision making. The main goal of this competition is to improve our understanding of the ways by which behavioral decision research can contribute to the derivation of useful predictions of human decision making, above and beyond what is possible using data-driven machine learning tools (if at all possible).
CPC18 distinguishes between two very different prediction tasks: predicting the aggregate population behavior in an unfamiliar choice problem, and predicting the individual behavior in a familiar choice problem. Specifically, CPC18 includes two parallel competition tracks, and you are invited to participate in either one, or better yet, in both. A second goal of the competition is to then understand what type of models are better suited to handle each type of task.
The rules of the competition and further details are given in https://cpc18.wordpress.com. A white paper summarizing the current stage of the competition is provided here. The deadline for final submissions is May 12, 2018 (but for one of the tracks, a partial submission must be made by May 8th; see website for details). To compete, you are required to register by April 10th.
As in some of the previous choice prediction competitions, the prize for the winners is an invitation to be a co-author of the paper that summarizes the competition.
The competition’s basic idea is as follows. We previously collected a large dataset of human choices between monetary gambles, under risk and under ambiguity, with and without feedback. This dataset includes over 500,000 individual consequential choices. Almost all of this data is publicly available at https://zenodo.org/record/845873#.WeDg9GhSw2x, and can (and probably should) be used to develop and train your predictive models.
In those experiments, each decision maker faced many problems, and the two tracks differ with respect to the exact prediction challenge:
In the track Individual behavior, familiar problems the task is to predict the individual behavior of a small portion of these decision makers in some of the problems they faced. Therefore, a small portion of the data already collected will be used as the competition data in that track and is thus not available. The goal in this track is to predict, as accurately as possible, the individual behavior reflected in that data.
In the track Aggregate behavior, unfamiliar problems the task is to predict the aggregate choice rates in a new experiment with new problems that we will run (during March-April 2018). As in some of the previous choice prediction competitions, the submissions should be computer programs that read the parameters of the choice problems as input, and derive the predicted choice rates as output.
We hope that you are up for the challenge!
Photo credit: https://flic.kr/p/4jbFC8
