1: \begin{abstract}

2: We studied the effects of time-correlation of subsequent patterns 

3: on the convergence of on-line learning by a feedforward neural network with 

4: backpropagation algorithm.

5: By using chaotic time series as sequences of 

6: correlated patterns,

7: we found that the unexpected scaling of converging time with learning 

8: parameter emerges 

9: when time-correlated patterns accelerate learning process.

10: \end{abstract}

11: