Beyond Accuracy: Behavioral Testing of NLP Models with CheckList

Explore the innovative CheckList methodology for testing NLP models in this 35-minute talk by Marco Túlio Ribeiro, Senior Researcher at Microsoft Research. Discover a task-agnostic approach inspired by software engineering principles that goes beyond traditional accuracy metrics. Learn about intriguing bugs uncovered in both commercial and research models, including those from tech giants like Microsoft, Amazon, and Google, as well as popular models like BERT and RoBERTA. Gain insights into the effectiveness of CheckList through case studies and user feedback from researchers and engineers. Understand how this powerful tool can enhance the testing and debugging process for NLP models, benefiting both practitioners and researchers in the field.

Beyond Accuracy: Behavioral Testing of NLP Models with CheckList