Daytime Project

Multi Agent Reinforcement Tool

AI systems abound in the world around us, but very often we do not have much insight in how these systems come to their decisions. Explaining AI decisions can help gain (expert) users’ trust in the AI and entice them to provide feedback. Automatically processing that feedback can then make the AI system more robust. Datenna combines these two elements in its feedback tool.


For demonstration purposes, the web-based Datenna feedback tool handles text data, more specifically, a set of research project titles. The tool classifies the data using a basic ML model as belonging to one of eight sectors (Mathematical Science, Chemical Science, Life Sciences, Earth Science, Engineering and Materials science, Information Science, Management Science, Medical Sciences). The user can then give feedback on the classification of some random instances of the data. Aiding the user’s assessment of the model, the words that contributed to the model’s decision are highlighted. Instead of going through thousands of datapoints, the user can then simulate further expert user feedback which will trigger the model to retrain itself using reinforcement learning. All the while, the model’s accuracy is plotted on a graph which you should see increase over time. This is akin to multiple experts (e.g. service mechanics) correcting a live AI system and instantly feeding this back into a digital-twin model over an extended period of time. 

Contact Information: