Abstract

Many existing recognition algorithms combine differentmodalities based on training accuracy but do not considerthe possibility of noise at test time. We describe an algo-rithm that perturbs test features so that all modalities pre-dict the same class. We enforce this perturbation to be assmall as possible via a quadratic program (QP) for con-tinuous features, and a mixed integer program (MIP) forbinary features. To efficiently solve the MIP, we provide a greedy algorithm and empirically show that its solution is very close to that of a state-of-the-art MIP solver. We evaluate our algorithm on several datasets and show that themethod outperforms existing approaches.