Data-Driven Verification and Learning Under Uncertainty

Introduction

Reinforcement learning (RL) agents learn to behave optimally via trial and error, without the need to encode complicated behavior explicitly. However, RL generally lacks mechanisms to constantly ensure correct behavior regarding sophisticated task and safety specifications.

Formal Verification

Formal verification (FV), and in particular model checking, provides formal guarantees on a system's correctness based on rigorous methods and precise specifications. Despite active development by researchers from all over the world, fundamental challenges obstruct the application of FV to RL so far.

Key Challenges

We identify three key challenges that frame the objectives of this proposal:

1. Curse of Dimensionality: Complex environments with large degrees of freedom induce large state and feature spaces. This curse of dimensionality poses a longstanding problem for verification.

2. Idealized State Spaces: Common approaches for the correctness of RL systems employ idealized discrete state spaces. However, realistic problems are often continuous.

3. Uncertainty in Real-World Environments: Knowledge about real-world environments is inherently uncertain. To ensure safety, correctness guarantees need to be robust against such imprecise knowledge about the environment.

Project Objectives

The main objective of the DEUCE project is to develop novel and data-driven verification methods that tightly integrate with RL.

Approach

To cope with the curse of dimensionality, we devise learning-based abstraction schemes that distill the system parts that are relevant for correctness. We employ and define models whose expressiveness captures various types of uncertainty. These models are the basis for formal and data-driven abstractions of continuous spaces.

Mechanisms

We provide model-based FV mechanisms that ensure safe and correct exploration for RL agents.

Conclusion

DEUCE will elevate the scalability and expressiveness of verification towards real-world deployment of reinforcement learning.