Generative Multiple-Instance Learning Models For Quantitative Electromyography