Reachability analysis for neural agent-environment systems
File(s)
Published version
Author(s)
Lomuscio, AR
akitunde, M
maganti, L
Pirovano, E
Type
Conference Paper
Abstract
We develop a novel model for studying agent-environment
systems, where the agents are implemented via feed-forward
ReLU neural networks. We provide a semantics and develop
a method to verify automatically that no unwanted states are
reached by the system during its evolution. We study several
reachability problems for the system, ranging from one-step
reachability, to fixed multi-step and arbitrary-step to study
the system evolution. We also study the decision problem of
whether an agent, realised via feed-forward ReLU networks
will perform an action in a system run. Whenever possible,
we give tight complexity bounds to decision problems intro-
duced. We automate the various reachability problems stud-
ied by recasting them as mixed-integer linear programming
problems. We present an implementation and discuss the ex-
perimental results obtained on a range of test cases.
systems, where the agents are implemented via feed-forward
ReLU neural networks. We provide a semantics and develop
a method to verify automatically that no unwanted states are
reached by the system during its evolution. We study several
reachability problems for the system, ranging from one-step
reachability, to fixed multi-step and arbitrary-step to study
the system evolution. We also study the decision problem of
whether an agent, realised via feed-forward ReLU networks
will perform an action in a system run. Whenever possible,
we give tight complexity bounds to decision problems intro-
duced. We automate the various reachability problems stud-
ied by recasting them as mixed-integer linear programming
problems. We present an implementation and discuss the ex-
perimental results obtained on a range of test cases.
Date Issued
2018-11-28
Date Acceptance
2018-07-11
Citation
2018
Publisher
Association for the Advancement of Artificial Intelligence
Copyright Statement
© 2018, Association for the Advancement of Artificial
Intelligence (www.aaai.org). All rights reserved.
Intelligence (www.aaai.org). All rights reserved.
Sponsor
Royal Academy Of Engineering
Defence Advanced Research Projects Agency (UK)
Identifier
https://aaai.org/ocs/index.php/KR/KR18/paper/view/17991
Grant Number
CIET 1718/26
Ref: FA8750-18-C-0095
Source
16th International Conference on Principles of Knowledge Representation and Reasoning
Publication Status
Published
Start Date
2018-10-27
Finish Date
2018-11-02
Coverage Spatial
Tempe, Arizona, USA
Date Publish Online
2018-11-28