Reviews: Collaborative PAC Learning

I thank the authors for their feedback and clarification and my positive evaluation of this paper remains unchanged. The authors give theoretical guarantees for collaborative learning with shared information under the PAC regime. They show that the sample complexity for (\epsilon,\delta)-PAC learning k classifiers for a shared problem (or a size k majority-voting ensemble) only needs to grow by a factor of approximately 1 log(k) (or 1 log 2(k)) rather than a factor of k as naive theory might predict. The authors provide pseudocode for two algorithms treated by the theory. These were correct as far as I could see, though I didn't implement either.