I loved Hugo Casselles-Dupré’s presentation in #TalkingRobotics:
https://www.youtube.com/watch?v=i4ovDv8DdzE

He makes a #robot learn better from another robot’s demonstration by leveraging #pragmatics! #machinelearning #ReinforcementLearning

Using #Bayesian inference, the teacher would deduce the goal behind its candidate demonstration and check that it’s not ambiguous for the learner. The learner also learns faster by guessing the intended goal of the teacher.

#TalkingRobotics 21 - Shiwali Mohan : https://www.youtube.com/watch?v=QrhfwbBSO4Q

@shiwali showcases AILEEN, a cognitive architecture for interactive task learning. It is built on top on the SOAR architecture (from the 90’s!), and leverages symbolic knowledge, #NLP and analogical pattern-matching to learn concepts from observation and written user inputs, with very few examples. And they made it work on #robots.

This is why classical #AI is interesting for learning despite the trends in #DeepLearning.