MBZUAI: Neural network for imitation handwriting

2024: Product Announcement

At the end of December 2023, specialists from the Mohamed bin Zayed University of Artificial Intelligence in the UAE (MBZUAI) announced the creation of a neural network capable of imitating human handwriting. The developers registered their technology with the United States Patent and Trademark Office (USPTO).

The project participants note that previous systems for imitating human handwriting were based on the use of the so-called generative-adversarial network (GAN). This is a machine learning algorithm built on a combination of two neural networks, one of which generates samples and the other tries to distinguish right from wrong. The handwriting formed by the GAN reflects the general style of writing - for example, the slope with which a person writes letters, or the width of strokes. However, difficulties arise with the reconstruction of ligatures.

A neural network has been created that imitates human handwriting

Instead of GAN, the researchers used a vision converter -- a type of neural network designed for computer vision tasks. The proposed method allows you to identify and reproduce all the features of the handwriting of a particular person. The model does not require a lot of data to learn - it is enough to analyze only a few paragraphs of text with original handwriting. As of mid-January 2024, the neural network is capable of recognizing and generating handwritten text in English and French, but in the future, the developers intend to add support for Arabic.

The researchers say the technology could be used to efficiently generate large amounts of data in order to train other AI models working with handwritten texts. In addition, the system can help people with injuries or any medical conditions that make them unable to pick up a pen. At the same time, it is necessary to create protective mechanisms that prevent the use of the tool for unauthorized and fraudulent purposes.^[1]

Notes