notesum.ai

Published at December 10

MAPLE: A Framework for Active Preference Learning Guided by Large Language Models

cs.LG

cs.AI

cs.CL

Released Date: December 10, 2024

Authors: Saaduddin Mahmud¹, Mason Nakamura¹, Shlomo Zilberstein¹

Aff.: ¹University of Massachusetts Amherst

Arxiv: http://arxiv.org/pdf/2412.07207v1

Refer to caption

	HomeGrid
Model	Test Accuracy	Expected Cost $\Delta$
B-REX	$0.82\pm 0.04$	$1.40\pm 0.24$
Mistral-7B-Ins.	$\mathbf{0.92}\pm\mathbf{0.02}$	$0.67\pm 0.13$
GPT-4o	$0.87\pm 0.03$	$0.89\pm 0.20$
GPT-4o-mini	$0.83\pm 0.04$	$3.90\pm 0.66$
Gemini-1.5-Pro	$0.87\pm 0.04$	$1.00\pm 0.18$
Mistral-Large-2	$0.85\pm 0.04$	$\mathbf{0.55\pm 0.14}$