Abstract： To solve the problems that the instancebased partial label learning (IPAL) spent too much time on calculating the neighbors of each sample and the weight of the neighbors was not suitable for largescale data, a new partial label learning method was proposed based on the weighed neighbor distance to improve the calculating way of neighbor weight. To enhance the operating efficiency of the new method, the parallel computation was carried out on the reading of training set and testing set, the construction of similarity graph, the propagation of iterative markers and the prediction of test samples. The parallel model of the new method was designed and implemented in MPI cluster environment. The running efficiency and the classification accuracy of the improved serial algorithm of weighted instancebased partial label learning (WIPAL) were compared with those of IPAL, and the running time and the acceleration of the parallel algorithm of parallel weighted instancebased partial label learning (PWIPAL) were compared for different process numbers. The experimental results show that the new method can shorten the running time with good accuracy of classification. The classification accuracy of PWIPAL is the same with that of WIPAL, and the speedup ratio of runtime is gradually close to the number of processes set with the increasing of data scale. The proposed method can be used to deal with largescale data.
2018, Vol. 39 
