https://repositorio.ufu.br/handle/123456789/5142 Fri, 18 Sep 2026 22:06:12 GMT 2026-09-18T22:06:12Z https://repositorio.ufu.br/handle/123456789/50340 Title: Domain-specific small language models for creative text generation: a case study on the backrooms Abstract: Large Language Models (LLMs) occupy a prominent place among Machine Learning models that have been revolutionizing modern life as an indispensable support tool in the most diverse areas of human activity. Nevertheless, the popularization of large LLMs faces a significant limiting challenge: the high cost associated with their use, whether through expensive hardware, cloud processing or paid APIs. In this light, the main objective of this work is to propose an approach to generate smaller language models, which can be trained in more modest hardware and that are effective in generating texts related to specific domains. The case study chosen for that are The Backrooms, an online legend which describes an alternative reality with different levels that are described with very unique, surreal and rich descriptions. For this, a Backrooms dataset, was created and used to produce two small language models (SLMs) to generate Backrooms level descriptions: Backrooms-Llama, conceived from the fine-tuning of the open source pretrained model Llama 1B; and Tiny Backrooms, a decoder-only architecture pretrained from scratch and then fine-tuned. The performance of these two models was evaluated against the famous and much larger DeepSeek-V3. The generated level descriptions of the three models were evaluated by using chatGPT 4o-mini as judge and human judges. Additionally, a brand new evaluation method, which consists on generating pictures with the output of each competitor model and, then, evaluating these pictures alongside their descriptions with chatGPT 4o, was proposed. The results found are very promising, showing that, even though DeepSeek performed better in general, Backrooms-Llama and Tiny Backrooms, operating in a much more modest architecture, also got mainly relevant evaluations and were able to learn how to generate Backrooms level descriptions. Such results indicate that, in creative generative tasks, a lower cost domain-specific SLM, when properly optimized, can achieve performance comparable to that of consecrated LLMs. Thu, 18 Jun 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/50340 2026-06-18T00:00:00Z https://repositorio.ufu.br/handle/123456789/50302 Title: Desenvolvimento de plataforma para auxiliar nas aulas de anatomia da madeira Thu, 07 May 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/50302 2026-05-07T00:00:00Z https://repositorio.ufu.br/handle/123456789/50299 Title: Análise da maturidade em gestão de projetos e previsão de desempenho operacional em uma startup de tecnologia Fri, 31 Jul 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/50299 2026-07-31T00:00:00Z https://repositorio.ufu.br/handle/123456789/50297 Title: Investigação de diferentes abordagens do Algoritmo Evolutivo aplicado ao problema de predição de proteínas no modelo HP-3D Abstract: Evolutionary Algorithms (EAs) are metaheuristics inspired by biological evolution, widely used to solve complex optimization problems, especially those involving multiple conflicting objectives. This work aims to investigate and compare the performance of different EA variants, considering both single-objective and multiobjective approaches, applied to the Protein Structure Prediction (PSP) problem using the three-dimensional hydrophobic-polar (HP-3D) model. To this end, different approaches were computationally implemented and evaluated, highlighting multiobjective variants based on multiple tables, such as the AEMMT. The algorithms were evaluated using metrics such as solution fitness, number of evaluations, and success rate. The results show that multiobjective approaches outperform single-objective ones, especially in more complex instances. As a main contribution, this study provides a systematic comparative analysis that identifies the most robust and efficient algorithmic configurations for the problem in the HP-3D model, corroborating the literature and reinforcing the potential of multiobjective EAs for this application. Thu, 19 Mar 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/50297 2026-03-19T00:00:00Z