Improved Algorithms for Linear Stochastic Bandits