Human-in-the-Loop Generation of Adversarial Texts: A Case Study on Tibetan Script
HITL-GAT is an interactive system for human-in-the-loop generation of adversarial texts to build and update robust benchmarks for lower-resourced languages. It involves iteratively constructing victim models, generating adversarial examples, creating high-quality benchmarks through human annotation, and evaluating robustness. A case study on Tibetan script showed its utility.