Reproducibility of deep learning models in cognitive computational neuroscience

Recent years have seen rapid development of neuroscientific research projects using deep neural networks as a modeling framework to explain human cognitive and brain function. A growing concern — both in the research fields of cognitive neuroscience and deep learning— is whether findings can be reproduced using identical or similar models and datasets, as should be expected. However, each domain has its own research objectives and aims to reproduce patterns relating to distinct types of scientific claims. Cognitive computational neuroscience makes claims about human cognitive and neural processing, while deep learning research focuses on the performance of artificial neural networks. The issues around the reproducibility of these claims thus need to be tackled in idiosyncratic ways. 

In this talk, I will describe these issues in detail, and outline existing tools for ensuring reproducibility at the intersection of deep learning and cognitive neuroscience.