Generate Quick
"Generate" research focuses on leveraging large language models (LLMs) and other generative models to create various forms of data, including text, code, images, and even simulated physical phenomena, often to augment existing datasets or address data scarcity in specific domains. Current research emphasizes improving the quality, controllability, and safety of generated content, exploring techniques like retrieval-augmented generation, fine-tuning with diverse instruction sets, and incorporating external knowledge bases to mitigate issues like hallucinations and biases. This work has significant implications for various fields, enabling more efficient data collection, improved model training, and the development of novel applications in areas such as healthcare, autonomous driving, and e-commerce.
Papers
First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI
Sourav Banerjee, Anush Mahajan, Ayushi Agarwal, Eishkaran Singh
Filter-then-Generate: Large Language Models with Structure-Text Adapter for Knowledge Graph Completion
Ben Liu, Jihai Zhang, Fangquan Lin, Cheng Yang, Min Peng
Align, Generate, Learn: A Novel Closed-Loop Framework for Cross-Lingual In-Context Learning
Mateo Alejandro Rojas, Rafael Carranza