Yad Vashem to use deep learning to organize its digital media and touch more people worldwide.
Yad Vashem, the world’s preeminent Holocaust memorial center, is dedicated to keeping alive for future generations the memory of the 6 million Jews who perished at the hands of the German Nazis and their collaborators.
But its World Holocaust Remembrance Center — a source for documentation used by scholars worldwide — is overwhelmed with difficult-to-find digital media documenting the lives of victims and survivors.
The Jerusalem-based organization is turning to AI to help identify, organize and link photos and other historical documents amid its ocean of data, for easier discovery. That’s because the documentation, gathered over decades of submissions and discoveries, and now almost fully digitized, is a source for Holocaust scholars globally.
A destination for a million visitors each year — six U.S. presidents have visited the site — Yad Vashem has archives that include unique, searing video testimonies, short films, photos, personal written accounts, Nazi documentation, and audio files. In addition to remembering Hitler’s victims, it pays tribute to the non-Jews who put their lives at risk trying to save them.
People worldwide last week recognized Holocaust Remembrance Day.
Twice the Data of Library of Congress
Its 800 million digital assets — which comprise over 4 petabytes of data (more than twice that held by the U.S. Library of Congress) — make it a daunting challenge for the institution to keep up with indexing this history for researchers, let alone reach a younger generation.
Using deep neural networks, Yad Vashem’s team can let image-recognition algorithms help index and categorize its digital history. This could lead to finding new connections and stories on Holocaust victims, according to Michael Lieber, chief information officer at Yad Vashem.
Lieber is optimistic that AI will help better identify resources to tell stories of Holocaust victims and survivors on its social media accounts. That could help keep it in touch with younger audiences, he said.
He’s also hopeful that researchers may use deep learning in ways to surface new historical information that couldn’t otherwise be discovered.
“We are among the first institutions in the world dealing with cultural heritage that decided to have a digital copy of everything because that is the way to get to a much wider audience globally,” said Lieber.
Improving Search for Family History
Many individuals visit Yad Vashem to research what happened to grandparents and great grandparents and piece together their family history. The problem is that the collection of digitized data, which could double in years to come, is difficult to search.
Yad Vashem’s technology team aims to change that by tapping into deep learning driven by high performance computing.
It plans to harness the supercomputing power of the NVIDIA DGX-1 AI system to help organize and augment its history using deep learning. DGX-1 offers the power of hundreds of CPU-based servers in a single system capable of over a petaFLOP of AI computing power.
The DGX-1 puts Yad Vashem alongside the world’s most innovative organizations deploying AI to address their challenges, said Yuval Mazor, senior solutions architect at NVIDIA.
“They get tangible benefits from the application of AI,” he said. “For example, Yad Vashem can use video analytics for understanding and predicting museum traffic and the impact of individual exhibits, as well as for extracting deep insights from the wealth of historical data,” he said. “These can help Yad Vashem in its primary mission, which is to reach and educate as many people as possible.”
Unsupervised learning holds the promise for trained neural networks to create meta-tags for digital artifacts, allowing deep learning to connect the dots on all kinds of information, Lieber said.
“If you manage to locate a prison card in the Mauthausen camp, the system will know that it is an inmate card,” he said. “It will direct you to the relevant data fields and documents, and you will be able to locate and identify types of documents and provide additional information without human intervention.”
The alternative would be to have legions of people label hundreds of millions of digital media assets and continue to keep track and make updates on databases.
NVIDIA research and development staff in Israel is partnering with Yad Vashem on the effort.