Prof. Dr. Alicia Fornes gives talk

"Recognition of Historical Handwritten Documents in Low Resource Scenarios" (English)

Despite decades of research, the recognition of historical handwritten
documents is still an open problem. Current techniques are based on
deep-learning architectures, which are powerful, but they need a huge amount of
data to train. Consequently, deep learning methods are not satisfactory in case
of low resource scenarios, defined as the scarcity of labeled data for
training. This is the case of historical manuscripts, handwritten music scores,
ciphered manuscripts, or documents with rare scripts. This talk will focus on
methods designed for coping with the limitations of few labelled data, such as
few-shot learning, synthetic data generation and self-supervised learning.

Back