Speaker: Xiaoqiang LIN & Zhaoxuan WU, National University of Singapore
Bio: Xiaoqiang LIN's research focuses on collaborative machine learning, data valuation, and zeroth-order optimization. His first-authored papers have been accepted to ICML and IJCAI.
Zhaoxuan WU's research interests include large language models, collaborative machine learning (e.g., data valuation, federated learning, incentives, fairness), and resource-efficient machine learning (e.g., Bayesian optimization).
Abstract: We adopt a neural bandit algorithm which replaces the GP in BO by an NN surrogate to optimize instructions for black-box LLMs. More importantly, the neural bandit algorithm allows us to naturally couple the NN surrogate with the hidden representation learned by a pre-trained transformer (i.e., an open-source LLM), which significantly boosts its performance. These motivate us to propose our INSTruction optimization usIng Neural bandits Coupled with Transformers (INSTINCT) algorithm. We perform instruction optimization for ChatGPT and use extensive experiments to show that our INSTINCT consistently outperforms the existing methods in different tasks, such as in various instruction induction tasks and the task of improving the zero-shot chain-of-thought instruction.
