MetaQuotes :
An error occurred during ORL optimization.Neural Networks Made Easy (Part 84): Reversible Normalisation (RevIN) has been published:
Author by: Dmitriy Gizlyk
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Neural Networks Made Easy (Part 84): Reversible Normalization (RevIN).
We already know that pre-processing of the input data plays a major role in the stability of model training. To process "raw" input data online, we often use a batch normalization layer. But sometimes we need a reverse procedure. In this article, we discuss one of the possible approaches to solving this problem.
We have already encountered a similar problem when training various autoencoder models. In those cases, we found a solution in using the original data after normalization as targets. However, in this case we need data describing subsequent states of the environment that are different from the input data. One of the methods for solving this problem was proposed in the paper "Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift".
The authors of the paper propose a simple yet effective normalization and denormalization method: Reversible Instantaneous Normalization (RevIN). The algorithm first normalizes the input sequences and then denormalizes the output sequences of the model to solve timeseries forecasting problems associated with distribution shift. RevIN is symmetrically structured to return the original distribution information to the model output by scaling and shifting the output in the denormalization layer in an amount equivalent to the shifting and scaling of the input data in the normalization layer.
RevIN — is a flexible, trainable layer that can be applied to any arbitrarily chosen layers, effectively suppressing non-stationary information (mean and variance of an instance) in one layer and restoring it in another layer of nearly symmetric position, such as input and output layers.
Author: Dmitriy Gizlyk