Efficient machine learning: models and accelerations