Multiple Matrix Learning Machine with Five Aspects of Pattern Information

This paper proposes an effective Multiple Matrix Learning Machine with Five Aspects of Pattern Information (MMFI). First aspect lies in the class label of each training or validation pattern. Second aspect is the values of components for each pattern. Third aspect is the relationship between patterns in the local regions of input space. Fourth aspect is the representation information and discriminant roles of different matrix representations for patterns. Fifth aspect is the information of patterns in each matrix representation learning. The innovations of the proposed MMFI are: (1) establishing a pattern-dependent function in the matrix learning so as to realize different roles of patterns for the first time; (2) adopting five aspects of pattern information so that a more feasible learning machine can be trained. The advantages of MMFI are: (1) proposing a new nonlinear learning machine which is different from the state-of-the-art kernelization one; (2) achieving a statistically superior classification performance than those learning machines without the introduction of five aspects of pattern information; (3) possessing a lower or comparable computational-complexity than other compared multiple matrix learning machines.