Researchers develop selective prediction for knowledge tracing models

Researchers have developed a method to improve the responsible deployment of Knowledge Tracing (KT) models by enabling them to identify uncertain predictions. By integrating a selective prediction layer using Monte Carlo Dropout, the models can defer predictions that are likely to be incorrect. This approach significantly boosts accuracy and AUC without retraining, while also ensuring fairness across different student abilities and question difficulties. The study found that model-derived uncertainty is a far more effective signal for deferral than traditional psychometric methods. AI

IMPACT Enhances responsible AI deployment by enabling models to recognize and defer uncertain predictions, improving accuracy and fairness.

RANK_REASON Academic paper introducing a new methodology for knowledge tracing models.

Read on arXiv cs.LG →

• arXiv
• BALD
• Item Response Theory
• Monte Carlo Dropout

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →