Voyager: A Training Free Approach for Generating Diverse Datasets using LLMs
Avinash Amballa, Yashas Malur Saidutta, Chi-Heng Lin, Vivek Kulkarni, Srinivas Chappidi
Samsung Research America
Voyager: A Training Free Approach for Generating Diverse Datasets using LLMs
Avinash Amballa, Yashas Malur Saidutta, Chi-Heng Lin, Vivek Kulkarni, Srinivas Chappidi
Samsung Research America
Large language models (LLMs) are increasingly being used to generate synthetic datasets for the evaluation and training of downstream models. However, prior work has noted that such generated data lacks diversity. In this paper, we propose Voyager, a novel principled approach to generate diverse datasets. Our approach is iterative and directly optimizes a mathematical quantity that optimizes the diversity of the dataset using the machinery of determinantal point processes. Furthermore, our approach is training-free, applicable to closed-source models, and scalable. In addition to providing theoretical justification for the working of our method, we also demonstrate through comprehensive experiments that Voyager significantly outperforms popular baseline approaches by providing a 1.5-3x improvement in diversity.
— Think of it as sending out explorers to map a territory
— You want to map a large area. You send explorers out on successive voyages.
— When an explorer returns, you keep their region only if it's genuinely new compared to what you've already mapped.
— To stay efficient, you don't compare against everything — you keep a small anchor set of the most representative regions seen so far.
— Reject a region? You tell future explorers to go elsewhere — steering the search toward unexplored ground.
If you find our project useful, please consider citing:
@inproceedings{amballa-etal-2026-voyager,
title = "{VOYAGER}: A Training Free Approach for Generating Diverse Datasets using {LLM}s",
author = "Amballa, Avinash and
Saidutta, Yashas Malur and
Lin, Chi-Heng and
Kulkarni, Vivek and
Chappidi, Srinivas",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.acl-long.784/",
doi = "10.18653/v1/2026.acl-long.784",
pages = "17224--17245",
ISBN = "979-8-89176-390-6",
abstract = "Large language models (LLMs) are increasingly being used to generate synthetic datasets for the evaluation and training of downstream models. However, prior work has noted that such generated data lacks diversity. In this paper, we propose VOYAGER, a novel principled approach to generate diverse datasets. Our approach is iterative and directly optimizes a mathematical quantity that optimizes the diversity of the dataset using the machinery of determinantal point processes. Furthermore, our approach is training-free, applicable to closed-source models, and scalable. In addition to providing theoretical justification for the working of our method, we also demonstrate through comprehensive experiments that VOYAGER improves diversity by $\mathbf{1.5}$-$\mathbf{3}$ times compared to popular baseline approaches."
}