Pablo Piantanida

Biography

Pablo Piantanida studies when more information, computation or communication leads to a better decision. Drawing on information theory and statistical decision theory, he develops methods to compare learned representations before their eventual use is known. His research on reliability examines when a model should defer a prediction and how to audit generative AI with a limited testing budget. He also studies how agents with different observations can reason together and decide which messages are worth exchanging.

Research topics

• Learning, representation and reasoning
• Uncertainty and model evaluation
• Multi-agent reasoning and communication