We developed a new framework which captures the teaching process via preference functions. We found new connections between teaching complexity of a family defined in this frame work, and VC dimension.
In contrast to classical teaching algorithms, we have worked on teaching scenarios where the teacher isn't fully aware of the true target hypothesis, and tried to develop robust algorithms for teaching.
We investigated several Reinforcement Learning settings with presence of adversary, who was perturbing the states which the agent was viewing.
We have worked on finding a curriculum of environments in Reinforcement Learning setting, in order too teach a human learner faster.