Timezone: »

 
Spotlight
Sanity Simulations for Saliency Methods
Joon Kim · Gregory Plumb · Ameet Talwalkar

Wed Jul 20 10:45 AM -- 10:50 AM (PDT) @ Room 309
in SA: Trustworthy Machine Learning »

Saliency methods are a popular class of feature attribution explanation methods that aim to capture a model's predictive reasoning by identifying "important" pixels in an input image. However, the development and adoption of these methods are hindered by the lack of access to ground-truth model reasoning, which prevents accurate evaluation. In this work, we design a synthetic benchmarking framework, SMERF, that allows us to perform ground-truth-based evaluation while controlling the complexity of the model's reasoning. Experimentally, SMERF reveals significant limitations in existing saliency methods and, as a result, represents a useful tool for the development of new saliency methods.

Keywords: SA: Accountability, Transparency and Interpretability

Author Information

Joon Kim (Carnegie Mellon University)
Gregory Plumb (Carnegie Mellon University)
Ameet Talwalkar (Carnegie Mellon University)

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

More from the Same Authors