Timezone: »

Iterative Teaching by Label Synthesis
Weiyang Liu · Zhen Liu · Hanchen Wang · Liam Paull · Bernhard Schölkopf · Adrian Weller

@ None

In this paper, we consider the problem of iterative machine teaching, where a teacher provides examples sequentially based on the current iterative learner. In contrast to previous methods that have to scan over the entire pool and select teaching examples from it in each iteration, we propose a label synthesis teaching framework where the teacher randomly selects input teaching examples (e.g., images) and then synthesizes suitable outputs (e.g., labels) for them. We show that this framework can avoid costly example selection while still provably achieving exponential teachability. We propose multiple novel teaching algorithms in this framework. Finally, we empirically demonstrate the value of our framework.

Author Information

Weiyang Liu (University of Cambridge)
Zhen Liu (University of Montreal, MILA)
Hanchen Wang (University of Cambridge)

Hey, I am a 2nd Year PhD Student in Info Engineering (i.e. Computer Science) at Cambridge, where I work with Joan Lasenby and Adrian Weller on Geometric Deep Learning (3D Vision, Graph) and Statistical Learning. Previously I worked with Xinran Wang, Ali Javey and Hao Dong on various topics such as Electronic/Photovoltaic Devices and Computational Biology. My PhD is supported via donations from Kathy Xu and Cambridge Trust.

Liam Paull (Université de Montréal)
Bernhard Schölkopf (MPI for Biological Cybernetics)
Adrian Weller (University of Cambridge )

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors