Learning Gene Regulatory Networks via Globally Regularized Risk Mimimization