ABOUT KEML

KEML - Knowledge Enhanced Machine Learning: We are a research group at the Center for Artificial Intelligence (C4AI) at the Universidade de São Paulo, with researchers working in machine learning, logical reasoning, and knowledge representation. The goal of this research group has been to merge data-driven learning with knowledge-based reasoning. Although the field of Artificial Intelligence is now highly successful due to innovations in machine learning, more attention should be given to reasoning with formal guarantees. The group's research has led to the development of a conversational agent that combines multiple sources of information about the ocean, used in both web-based and robot-based interfaces. Currently, significant effort is dedicated to evaluating language models and their synergy with prompt engineering to ensure factuality guarantees. The team also investigates neurosymbolic combinations of probabilistic programming and artificial neural networks for automated reasoning.

PROJECTS

Conversational agents

Conversational agents are perhaps the main application for a language model ...

Project summary

Evaluation of large language models

The rapid evolution of large language models and their evident competence ...

Project summary

Argument mining

One of the traits of rational behavior is the ability to provide justifications ...

Project summary

Neuro-symbolic probabilistic logical programs

In recent decades, we have been repeatedly surprised by the performance ...

Project summary

RESOURCES

Corpora

The corpora created by the KEML team are composed primarily of texts in Portuguese.

Visit and explore them!

Datasets

Machine learning cannot be done without datasets.

We offer some ...

Models

The models we created in our work are described in this area.

Enter to get to know them ...

Knowledge graphs and ontologies

Knowledge and machine learning together is our goal.

Access structured knowledge here.

Frameworks

Over the years, we have directed some of our efforts into frameworks.

Use them ...

Codes

For the sake of reproducibility.

Description of our codes here ...