Timezone: »

Human-in-the-Loop Interpretability Prior
Isaac Lage · Andrew Ross · Samuel J Gershman · Been Kim · Finale Doshi-Velez

Thu Dec 06 07:45 AM -- 09:45 AM (PST) @ Room 517 AB #119

We often desire our models to be interpretable as well as accurate. Prior work on optimizing models for interpretability has relied on easy-to-quantify proxies for interpretability, such as sparsity or the number of operations required. In this work, we optimize for interpretability by directly including humans in the optimization loop. We develop an algorithm that minimizes the number of user studies to find models that are both predictive and interpretable and demonstrate our approach on several data sets. Our human subjects results show trends towards different proxy notions of interpretability on different datasets, which suggests that different proxies are preferred on different tasks.

Author Information

Isaac Lage (Harvard)
Andrew Ross (Harvard University)
Samuel J Gershman (Harvard University)
Been Kim (Google)
Finale Doshi-Velez (Harvard)

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

More from the Same Authors