Planning to Explore: Curiosity-Driven Planning for LLM Test Generation
How curiosity-driven planning, grounded in a 35-year thread of information-theoretic exploration, helps LLMs break through the coverage plateau in automated test …
(More)
About a Research Visit to KAUST
Reflections on a research visit to KAUST in Saudi Arabia, working with Jurgen Schmidhuber’s lab on curiosity-driven exploration and its application to …
(More)
Grounding LLM Reasoning with Knowledge Graphs
We discuss the integration of LLMs with Knowledge Graphs (KGs) to ground their reasoning capabilities. It highlights the importance of KGs in providing structured …
(More)
Towards a Reliable and Trustworthy MultiAgent Debate
LLMs tend to be overconfident, always providing an answer to the user’s query, even when the model is unsure about the answer. What happens when we ask the …
(More)
Exploring the Frontier of Uncertainty with Large Language Models: The Role of Known-Unknown Questions
LLMs tend to be overconfident, always providing an answer to the user’s query, even when the model is unsure about the answer. What happens when we ask the …
(More)
Neural Methods for Logical Reasoning over Knowledge Graphs
In this work, we explore and try to improve how we can answer complex logical queries over knowledge graphs. We call them complex logical queries because the …
(More)