Abstract
Sparse generalized linear model is useful in many fields. In the research, the researchers will learn sparse generalized linear model using different algorithms. The paper determines the better algorithm for learning this model by comparing the convergence rate of mirror descent and projected gradient descent. By implementing the two algorithms and comparing the results, the researchers conclude that the mirror descent converges much faster than the projected gradient descent for learning the sparse generalized linear model. This means the mirror descent algorithm is better for learning this model.