KASER: Knowledge-Aligned Student Error Simulator for Open-Ended Coding Tasks

In three linesKASER is a reinforcement learning method to simulate student coding errors. It uses a hybrid reward combining code similarity, error matching, and solution diversity to prevent mode collapse and capture the variety of student responses.

Summary generated by Claude — human-verified