Dissertation advances offline Apprenticeship Learning for human-centric AI

This dissertation introduces novel frameworks for offline Apprenticeship Learning (AL) designed for human-centric environments like Intelligent Tutoring Systems (ITSs) and clinical healthcare. The proposed methods, including EM-EDM and THEMES, focus on improving reward modeling by clustering demonstrations based on latent reward functions and partitioning trajectories into coherent sub-trajectories. Further advancements with HALIDE and CCHAL address demonstration quality by integrating continuous quality signals and attributing quality to individual decisions, outperforming existing baselines on various datasets. A real-world classroom evaluation of StuEDM demonstrated significant learning gains and preserved student agency. AI

IMPACT Establishes offline AL as a practical, data-efficient, and ethically viable paradigm for AI in education and healthcare.

RANK_REASON The item is an arXiv preprint detailing academic research on offline apprenticeship learning frameworks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• electronic health records
• Em Edmundo Vieira
• Halide
• Intelligent Tutoring Systems
• Md Mirajul Islam
• StuEDM
• Themes
• Tsqaltubo

• paper
• other

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