Efficiently Enforcing Diversity in Multi-Output Structured Prediction
Abner Guzman-Rivera, Pushmeet Kohli, Dhruv Batra, Rob A. Rutenbar · 2014
This paper proposes a novel method for effi-ciently generating multiple diverse predictions for structured prediction problems. Existing methods like SDPPs or DivMBest work by mak-ing a series of predictions where each predic-tion is made after considering the predictions that came before it. Such approaches are inher-ently sequential and computationally expensive. In contrast, our method, Diverse Multiple Choice Learning, learns a set of models to make multi-ple independent, yet diverse, predictions at test-time. We achieve this by including a diversity en-couraging term in the loss function used for train-ing the models. This approach encourages diver-sity in the predictions while preserving compu-tational efficiency at test-time. Experimental re-sults on a number of challenging problems show that our method learns models that not only pre-dict more diverse results than competing meth-ods, but are also able to generalize better and pro-duce results with high test accuracy. 1