Qualificação de mestrado da discente Sara Câmara, dia 11/08 às 14: 00 na Sala de Seminários Decom

Qualificação de mestrado da discente Sara Câmara, dia 11/08 às 14: 00 na Sala de Seminários Decom - ICEB III

Título: MULTIOBJECTIVE OPTIMIZATION FOR PROMPTS USING EVOLUTIONARY ALGORITHMS

Resumo: This work investigates the optimization of prompts for LLMs through a multiobjective evolutionary (EMO) approach. Prompt engineering is crucial for unlocking the full potential of LLMs; however, manually designing effective prompts can be complex and time-consuming. While automated methods show promise, they typically focus on a single objective, neglecting the crucial trade-off between task performance (such as accuracy) and computational cost (measured by the number of tokens). To address this gap, we introduce the MOPrompt framework, which simultaneously optimizes prompts to maximize accuracy and minimize token count. Our methodology employs an LLM (GPT-4o Mini) to perform semantic genetic operations like crossover and mutation, enabling the generation of contextually relevant and syntactically valid prompt variations. Experiments were conducted on a sentiment classification task in portuguese, utilizing Gemma-2B and Sabiazinho-3 models as evaluators. Our results demonstrate MOPrompt's superiority over a single-objective baseline approach. Notably, for the Sabiazinho model, MOPrompt achieved the same maximum accuracy (0.97) as the best single-objective solution, but with a 31% reduction in token cost. We conclude that the multi-objective approach employed is a more effective methodology for prompt engineering, exposing crucial trade-offs between cost and performance, and offering a set of Pareto-optimal solutions for real-world LLM applications.

Data: 11/08  Horário: 14h

Local: Sala de Seminários DECOM - ICEB III

Banca Avaliadora: Prof. Dr. Gladston Moreira (Orientador), Prof. Dr. Eduardo Luz (Co-orientador), Profa. Dra. Valéria C. Santos e Prof. Dr. Ivan R. Meneghini