BinaryCorrelatorLearningRule.hpp

Go to the documentation of this file.

#ifndef BINARYCORRELATORLEARNINGRULE_HPP_
#define BINARYCORRELATORLEARNINGRULE_HPP_

#include <optimization/OptimizationFunction.hpp>

namespace mic {
namespace neural_nets {
namespace learning {

template <typename eT=float>
class BinaryCorrelatorLearningRule : public mic::neural_nets::optimization::OptimizationFunction<eT> {
public:
    BinaryCorrelatorLearningRule(size_t rows_, size_t cols_) {
        delta = MAKE_MATRIX_PTR(eT, rows_, cols_);
        delta->zeros();
    }

    // Virtual destructor - empty.
    virtual ~BinaryCorrelatorLearningRule() { }


    virtual mic::types::MatrixPtr<eT> calculateUpdate(mic::types::MatrixPtr<eT> x_, mic::types::MatrixPtr<eT> y_, eT ni_aa = 0.1) {
        // Calculate N_on - sum active bits.
        size_t N_on = y_->sum();
        // Calculate ni_ia = ni_ai.
        eT ni_ia = ni_aa * N_on * (x_->rows() - N_on);

        delta->setZero();

        // Calculate deltas.
        for (size_t b=0; b< (size_t)x_->cols(); b++) { // for batch size.
            for (size_t i=0; i< (size_t)x_->rows(); i++) {
                for (size_t j=0; j< (size_t)y_->rows(); j++) {
                    if ((*y_)(j,b) && (*x_)(i,b))
                        (*delta)(j,i) += ni_aa;
                    else if ((*y_)(j,b) || (*x_)(i,b))
                        (*delta)(j,i) -= ni_ia;
                }//: for
            }//: for
        }//: for

        return delta;
    }


protected:
    mic::types::MatrixPtr<eT> delta;

};

} //: namespace learning
} /* namespace neural_nets */
} /* namespace mic */

#endif /* BINARYCORRELATORLEARNINGRULE_HPP_ */