There are errors in our paper “Constructing Multiclass Learners from Binary Learners: A Simple Black-Box Analysis of the Generalization Errors,” which appeared in ALT’05 [3]. The errors are related to our uses of union bounds. We briefly describe the problem and discuss which of our results can be shown to hold. We also provide a counter example for our previous claim.

The organizational economics literature provides two primary arguments for vertically integrating a factor of production: incentives for investment in relationship-specific assets and improved coordination due to less bargaining activity (and other factors). Although there is now a well-established body of empirical work investigating investments in specific assets, there is almost no empirical...

Article history: Received 29 April 2014 Available online 17 September 2014 Communicated by Charles Leedham-Green MSC: primary 20D06, 20G40 secondary 20B40, 20G41, 20P05, 68Q25