Multi-Instance Multi-Label (MIML) learning is new machine learning framework where an example is described by multiple instances and associated with multiple classes of labels. A method of designing MIML algorithm is to identify its equivalence in the traditional supervised learning framework, using multi-instance learning or multi-label learning as the bridge, then using an algorithm for training and modeling. But in the degradation process, there will be the loss of information, then affecting the classification accuracy. This study improves the MIMLSVM algorithm by using a multi-label algorithm GLOCAL that considers both global and local correlations. The experimental results show that the improved algorithm has achieved sound classification results.