The new method tackles the challenge of missing features under the multi-task learning framework. The proposed method is effective for prediction and model estimation when missing data is present.