Most of our work has resulted in scholarly publications. On this page you can review our publications to get an idea about our work.
Long-Context Training and Positional Extrapolation: A Technical Note
July, 2026 • Technical note
Santos, Dheiver
A technical note on long-context training and positional extrapolation for large language models, covering techniques such as rotary position embedding scaling and attention modifications used to exte…
A technical note on long-context training and positional extrapolation for large language models, covering techniques such as rotary position embedding scaling and attention modifications used to extend context windows beyond their original training length.
Machine Learning Framework for Road Accident Prediction
July, 2026 • Conference proceeding • Indiana Journal of Multidisciplinary Research
Spoorthi, M
Abstract: Road accidents are a major public safety issue because they kill 17 people every hour in India. We need to find good ways to stop accidents right away because each death is linked to many in…
Abstract: Road accidents are a major public safety issue because they kill 17 people every hour in India. We need to find good ways to stop accidents right away because each death is linked to many injuries of varying severity. Accurate forecasting is essential for transportation safety management due to the concerning increase in daily incidents attributed to the rapid rise in vehicle numbers. This study develops an accident prediction model utilising data mining techniques, particularly the K-Nearest Neighbours (KNN) Classifier and Support Vector Machines (SVM), and ensembles both results to increase the accuracy. The study takes into account the environment, the condition of the roads, and the amount of traffic. By spotting trends and predicting how often accidents will happen, this method may help government agencies, transportation departments, and NGOs focus their safety measures. Contractors, public works agencies, and automakers may also use the results to make safety improvements and better infrastructure design. The proposed forecasting technique illustrates how machine learning methodologies can improve road safety by reducing accident frequencies.
Accident prediction, Machine learning, Data mining, SVM, KNN, Road accidents, Pattern of data.
A technical note on Reinforcement Learning from AI Feedback (RLAIF) and Constitutional AI, methods that use AI-generated feedback and explicit principles to align model behavior.
AS TRANSFORMAÇÕES NA GESTÃO DE PESSOAS NO PÓS-PANDEMIA E A INFLUÊNCIA SOBRE A QUALIDADE DE VIDA DOS SERVIDORES: A REALIDADE DAS UNIVERSIDADES FEDERAIS MINEIRAS
November, 2023 • Thesis
LUCAS, JORGE LUCAS SANTOS DA LUZ
A pandemia da COVID-19 representou um desafio global sem precedentes, impactando de forma abrangente e profunda todos os aspectos da vida, incluindo a saúde pública, a economia…
A pandemia da COVID-19 representou um desafio global sem precedentes, impactando de forma abrangente e profunda todos os aspectos da vida, incluindo a saúde pública, a economia e as relações sociais. Nesse contexto, as universidades foram diretamente afetadas, tendo que realizar adaptações significativas em suas atividades e rotinas diárias. A suspensão das aulas presenciais e a transição para o ensino remoto, juntamente com a interrupção de pesquisas e a necessidade de comunicação virtual, trouxeram novos desafios para estudantes, professores e demais servidores universitários. Este estudo tem como objetivo compreender as estratégias de Qualidade de Vida no Trabalho (QVT) adotadas durante a pandemia nas universidades federais mineiras e identificar como essas estratégias afetaram os seus servidores com base no modelo de QVT de Walton (1973). Primeiramente, foi realizada coleta documental nos sites das 11 universidades federais mineiras, a partir de portarias, notas técnicas, decretos e notícias relacionadas às ações relacionadas à QVT. Em seguida, foram realizadas entrevistas com 15 servidores de 5 destas instituições. Utilizando a análise de conteúdo proposta por Bardin (2016), os documentos e as entrevistas foram submetidos a uma pré-análise, análise e tratamento dos dados. Os resultados destacam a importância de medidas flexíveis e adaptativas, levando em consideração as necessidades e demandas dos servidores. As universidades demonstraram esforços em promover um ambiente de trabalho saudável e equilibrado, priorizando a saúde física e mental dos servidores. As análises vislumbraram também a relevância de políticas de suporte psicológico, como programas de orientação e apoio emocional, evidenciando a preocupação com o bem-estar global dos servidores. Além disso, a valorização do desenvolvimento profissional foi observada por meio de ações como capacitações online e incentivos para aprimoramento docente. Essas estratégias contribuíram para manter a motivação e o engajamento dos servidores durante um período adverso. Por fim, foi possível observar como as diferenças sociais e de gênero influenciam na propensão dos servidores ao teletrabalho.
Knowledge Distillation and Self-Distillation for Language Models
July, 2026 • Technical note
Santos, Dheiver
A technical note covering knowledge distillation and self-distillation techniques used to compress and improve large language models while retaining task performance.
A technical note on training sparse Mixture-of-Experts (MoE) models, covering expert routing strategies and the sparse upcycling approach for converting dense checkpoints into MoE architectures.
Parameter-Efficient Fine-Tuning: LoRA, QLoRA, and DoRA
July, 2026 • Technical note
Santos, Dheiver
A technical note comparing parameter-efficient fine-tuning techniques for large language models, including Low-Rank Adaptation (LoRA), its quantized variant QLoRA, and Weight-Decomposed Low-Rank Adapt…
A technical note comparing parameter-efficient fine-tuning techniques for large language models, including Low-Rank Adaptation (LoRA), its quantized variant QLoRA, and Weight-Decomposed Low-Rank Adaptation (DoRA).
Reinforcement Learning with Verifiable Rewards (RLVR) and GRPO for Reasoning
July, 2026 • Technical note
Santos, Dheiver
A technical note on Reinforcement Learning with Verifiable Rewards (RLVR) and Group Relative Policy Optimization (GRPO), and how these methods are used to improve multi-step reasoning capabilities in …
A technical note on Reinforcement Learning with Verifiable Rewards (RLVR) and Group Relative Policy Optimization (GRPO), and how these methods are used to improve multi-step reasoning capabilities in language models.
Direct Preference Optimization (DPO): A Technical Note
July, 2026 • Technical note
Santos, Dheiver
A technical note on Direct Preference Optimization (DPO), a method for aligning language models with human preferences without a separate reward model. It covers the training objective, how it relates…
A technical note on Direct Preference Optimization (DPO), a method for aligning language models with human preferences without a separate reward model. It covers the training objective, how it relates to RLHF, and practical trade-offs for implementation.
large language modelsdirect preference optimization
There is an increasing interest in upgrading the EModel, a parametric tool for speech quality estimation, to the wideband and super-wideband contexts. The
Contemporary models of Unmanned Aerial Vehicles (UAVs) are largely developed using simulators. In a typical scheme, a flight simulator is dovetailed with a
Undertaking engineering research can be compounding for beginning graduate students and thwarting even for seasoned researchers. With a wealth of academic
To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Functional
Always active
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
Preferences
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
Statistics
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
Marketing
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.