`

Timezone: »

 
Poster
Learning with Labeling Induced Abstentions
Kareem Amin · Giulia DeSalvo · Afshin Rostamizadeh

Fri Dec 10 08:30 AM -- 10:00 AM (PST) @ None #None

Consider a setting where we wish to automate an expensive task with a machine learning algorithm using a limited labeling resource. In such settings, examples routed for labeling are often out of scope for the machine learning algorithm. For example, in a spam detection setting, human reviewers not only provide labeled data but are such high-quality detectors of spam that examples routed to them no longer require machine evaluation. As a consequence, the distribution of examples routed to the machine is intimately tied to the process generating labels. We introduce a formalization of this setting, and give an algorithm that simultaneously learns a model and decides when to request a label by leveraging ideas from both the abstention and active learning literatures. We prove an upper bound on the algorithm's label complexity and a matching lower bound for any algorithm in this setting. We conduct a thorough set of experiments including an ablation study to test different components of our algorithm. We demonstrate the effectiveness of an efficient version of our algorithm over margin sampling on a variety of datasets.

Author Information

Kareem Amin (Google Research)
Giulia DeSalvo (Google Research)
Afshin Rostamizadeh (Google Research)

More from the Same Authors