In context: Researchers are turning the inventive world the other way up, exploiting synthetic intelligence and machine studying algorithms to show many duties into semi-autonomous processes. Nothing is secure from generative AI anymore, not even your native physician’s illegible writing.
Years earlier than OpenAI and different organizations began toying with AI to simply generate textual content, speech, artworks, malware, and movies, machine studying researcher Sean Vasquez was learning a 2013 paper by Google DeepMind’s Alex Graves to create “handwriting synthesis” experiments.
Vasquez archived his code on GitHub alongside along with his web-based demo. The experiment is accessible at Calligrapher.ai, which Hacker Information just lately rediscovered. The handwriting synthesis behind Calligrapher.ai employs a generative technique constructed upon a recurrent neural community (RNN).
An RNN is a category of synthetic neural networks the place connections between nodes can create a cycle permitting output from some nodes to have an effect on subsequent enter to the identical nodes. Recurrent neural networks can exhibit temporal dynamic conduct, which makes them notably helpful in duties comparable to handwriting or speech recognition. Like every other neural community, Vasquez skilled Calligrapher.ai on a reasonably giant dataset of calligraphy samples, primarily the IAM On-Line Handwriting Database.
The IAM-On database comprises “types of handwritten English textual content acquired on a whiteboard,” with samples from 221 completely different “writers” and greater than 1,700 acquired varieties. The database consists of 13,049 remoted and labeled textual content strains in “on-line” and “off-line” format, for a complete of 86,272 samples from an 11,059-word dictionary.
Calligrapher.ai can generate variable handwriting in 9 completely different kinds, whereas customers can change velocity, legibility, and stroke width sliders for additional customization. Not like conventional font varieties designed to imitate handwriting, each pattern generated by Calligrapher.ai needs to be distinctive even when the writing type is similar. Customers can obtain the ultimate consequence as an SVG vector file.
In keeping with Vasquez, the legibility slider employs a technique referred to as “adjusting the temperature of the sampling distribution” to change variation in handwriting. Outputs come from a “chance distribution,” and rising the legibility “successfully concentrates chance density round extra possible outcomes.”
Being only a demo, Calligrapher.ai is proscribed in scope regardless of its potential to create plausible handwriting patterns. Moreover, Vasquez solely skilled the underlying RNN on English language samples, so the web site is not notably good at reproducing accents generally utilized in different languages.