Differentiable Neural Computers (DNCs) are a recent concept of deep neural networks that are enhanced by a memory module. This allows them to read and write information for future use, which is considered a major step towards a more general algorithmic learner.
Jörg Franke (Understand AI / KIT) employs and advance DNCs for his work in Question Answering. He will introduce us to these exciting models and share insights on their behavior as well as his advancements to the DNCs.
Here's a collection of material if you'd like to familiarize yourself with the topic prior to this talk:
A blog entry on DNCs by DeepMind: https://deepmind.com/blog/differentiable-neural-computers/
DeepMind's DNC Nature publication: http://www.nature.com/nature/journal/v538/n7626/full/nature20101.html
Jörg’s ADNC ACL/MRQA publication: https://arxiv.org/abs/1807.02658
ADNC Github repository: https://github.com/joergfranke/ADNC