Informative Example Selection for In-Context Learning

Explore the critical role of example selection in In-context Learning (ICL) for large language models (LLMs) in this 59-minute talk presented by Shivanshu Gupta from UCI at USC Information Sciences Institute. Delve into the importance of ICL as a training-free interface for LLMs, especially those that are intractable to train or hidden behind APIs. Examine the limitations of current approaches to selecting in-context examples and discover new, efficient methods for choosing informative examples that demonstrate salient information necessary for solving test inputs. Learn about the speaker's research findings across various tasks and LLMs, showcasing how selecting informative examples can significantly improve ICL performance. Gain insights from Shivanshu Gupta, a Computer Science Ph.D. Candidate at UC Irvine, with experience at LinkedIn and Microsoft Research India, as he shares his expertise in systematic generalization, in-context learning, and multi-step reasoning capabilities of large language models.

Informative Example Selection for In-Context Learning