Multi-Task Meta-Learning Modification with Stochastic Approximation