Automate reading of human handwriting
From handwritten documents to digital text
In three steps, our system is able to identify lines of text in a document, read the characters in each line, and choose the words that will maximize the overall coherence of the text.
Beyond neural networks: sensible language model with better performance
We add an optimization layer that models the language directly, choosing which words should be read from the overall context. This leads to more sensible and more accurate readings than traditional approaches to decipher words from neural networks outputs.
Fueled by different data sources
Fueled by this variety of data sources, our system reads 90% of the words correctly on documents written in a writing style unseen in training.
Motivated by pressing real-world applications
Why is the Interpretable AI solution unique?
Our solution builds on several state-of-the-art parts of the data science toolkit: neural networks, classical machine learning, and optimization
The choices of our optimization algorithm are auditable for further manual inspection