Long term ahead prediction using neural networks

The area of long term prediction is becoming a major topic in terms of video traffic prediction. To perform long term predictions is much more difficult than the concept of single-step ahead prediction. Most of the researchers perform the long term prediction as means to predict finite number of samples to the future. In our work we propose an algorithm that can predict in theory infinite number of samples to the future (in this paper thousands of samples). We compare our algorithm with standard neural networks that utilize feed-back loop to perform long term prediction, which is common approach to long term prediction.