PAC Learning on Thresholds over {0, 1, ..., d}