A training apparatus (101) includes an acquisition unit (102), a synergy calculation unit (103), a feature quantity calculation unit (104), and an output unit (105). The acquisition unit (102), while a subject is performing a behavior, acquires time-series myoelectric potentials of each of muscles 1, 2, ..., m of the subject. The synergy calculation unit (103) calculates a muscle synergy matrix W, a control matrix C and an error matrix E from a myoelectric potential matrix M so that the relationship M = WC + E is satisfied. Row vectors each expressing the acquired time-series myoelectric potentials of a corresponding one of the muscles 1, 2, ..., m are arranged in rows in the myoelectric potential matrix M. N unit column vectors W (1) , W (2) ,..., W (n) are arranged in columns in the muscle synergy matrix W. N row vectors C (1) , C (2) , ..., C (n) are arranged in rows in the control matrix C. The feature quantity calculation unit (104) calculates a feature quantity indicating consistency among the unit column vectors W (1) , W (2) , ..., W (n) included in the calculated muscle synergy matrix W. The output unit (105) outputs the calculated feature quantity.
