Stochastic Approximation in the presence of heavy-tailed and long-range-dependent noise
Venkat Anantharam and Vivek Borkar1
The theory of stochastic approximation is widely used to develop algorithms convergent to global optima under noisy measurements. Typically the noise is modeled as a Brownian motion or other short range dependent process. In networks with long-range-dependent traffic, such noise models are inadequate. We are developing a theory of stochastic approximation that is broader in scope and can include heavy-tailed and long-range-dependent noise in the measurements.