The 2nd International Workshop "Visualization for AI Explainability" (visxai.io) was held in Vancouver, Canada on October 21, 2019 as part of the IEEE VIS conference (ieeevis.org/year/2019/welcome). This workshop, attended by nearly 200 participants, focuses on the intersection of AI and Visualization techniques. It is jointly organized by manufacturers (Google Brain and IBM) and academics (MIT, CMU, GT, Konstanz).

The article presented by Théo Jaunet "What if we Reduce the Memory of an Artificial Doom Player?" co-supervised by Romain Vuillemot (SICAL) and Christian Wolf (IMAGINE) won the best-paper award. An online demonstration of the tool presented is available via the following link: theo-jaunet.github.io/MemoryReduction.

The contribution of the article aims to show how a human being can interactively reduce the size of the memory of a Deep Reinforcement Learning model and observe its impact on the behaviour of AI. The applications of such a tool will make it possible to optimize the storage and the number of calculations required by these memories, which can be very large, for robots or autonomous cars.

Demo: theo-jaunet.github.io/MemoryReduction
Open source code: github.com/Theo-Jaunet/MemoryReduction
Théo Jaunet's website: theo-jaunet.github.io