Edoardo Pirovano

3Machine learning papers tracked by Lacuna
Is this you?

Research Overview

Edoardo Pirovano focuses on the formal verification of multi-agent systems, probabilistic swarm robotics, and neural agent-environment interactions. His work addresses the challenge of ensuring safety and robustness in systems with unbounded numbers of agents or complex stateful behaviors. The most recent evidence of his research investigates the verification of recurrent neural networks in dynamic environments.

Research Trajectory

The available evidence for Pirovano's research trajectory is highly concentrated, consisting of foundational papers published between 2018 and 2019 in premier artificial intelligence venues such as IJCAI and AAAI. His work during this period primarily addresses the mathematical guarantees required to deploy complex, multi-agent, and neural-driven systems safely.

In 2018, Pirovano's research centered on the formal verification of multi-agent systems, particularly in scenarios where the number of agents is unbounded or the environment is probabilistic. Serving as the senior author on Symbolic Synthesis of Fault-Tolerance Ratios in Parameterised Multi-Agent Systems, he contributed to the broader area of Parameterized Verification for Unbounded Multi Agent Systems. This research introduced a symbolic procedure to automatically synthesize the maximum allowable ratio of faulty agents in a system while still satisfying specifications expressed in temporal-epistemic logic. By validating the methodology on swarm robotics protocols, the work also touched upon themes such as Temporal Epistemic Reasoning for Dynamic Multi-Agent Systems.

In the same year, Pirovano co-authored Verifying Emergence of Bounded Time Properties in Probabilistic Swarm Systems. This work expanded his focus into the research direction of Probabilistic Temporal Logic for Stochastic Systems. The paper developed parameterized semantics to formally identify emergent properties in probabilistic swarms using a fragment of Probabilistic Computation Tree Logic (PCTL), demonstrating the approach's effectiveness on an ant coverage algorithm.

By 2019, Pirovano's focus evolved to address the growing intersection of formal methods and machine learning. As the senior author on Verification of RNN-Based Neural Agent-Environment Systems., he tackled the verification of stateful agents executing ReLU recurrent neural networks (RNNs). By establishing equivalences between recurrent and feed-forward networks over bounded execution traces, the research provided a sound and complete verification procedure against properties specified in a simplified Linear Temporal Logic (LTL).

Current Directions

The most recent paper in the provided evidence is from 2019, marking a distinct shift toward the formal verification of neural-symbolic and stateful machine learning systems. Based on this most recent work, Pirovano's latest documented research investigates how to provide rigorous safety guarantees for agents controlled by recurrent neural networks. This direction bridges traditional formal methods—such as bounded execution traces and temporal logic—with the continuous, stateful dynamics of modern neural architectures, aiming to ensure predictable behavior in complex agent-environment interactions.

Profile updated

See something wrong? Report a mistake