¹Department of Mathematical Sciences, Shaoxing College of Arts and Sciences, Shaoxing, China
²Mathematics & Science College, Shanghai Normal University, Shanghai, China
³School of Mathematics and LPMC, Nankai University, Tianjin, China

---

Abstract—It is known that kernel regularized online learning has the advantages of low complexity and simple calculations, and thus is accompanied with slow convergence and low accuracy. Often the algorithm are designed with the help of gradient of the loss function, the complexity of the loss may influence the convergence. In this paper, we show, at some extent, the strong convexity can increase the learning rates.

Index Terms—Online learning, Strong convex loss, Learning rates

[PDF]

Cite: Baohuai Sheng, Liqin Duan, Peixin Ye, "Strong Convex Loss Can Increase the Learning Rates of Online Learning," Journal of Computers vol. 9, no. 7, pp. 1606-1611, 2014.