Predicting the past with Ithaca - Technology Org

The understanding of ancient inscriptions is challenging as they have been damaged over the centuries or moved from their original location. A recent paper by DeepMind proposes Ithaca, a deep neural network that can restore the missing text of damaged inscriptions, identify their original location, and help establish the date they were created. In order to work with the damaged and missing chunks of text, the model is trained using both words and the individual characters as inputs. Ithaca generates several prediction hypotheses for the text restoration task for historians to choose. It also shows its uncertainty by giving a probability distribution over all possible predictions of geographical and chronological distribution. Saliency maps identify which input sequences contribute most to a prediction.