Checking Stability of Neural NARX Models: An Interval Approach

Driescher, A.; Korn, Ulrich

doi:10.1016/s1474-6670(17)43495-1

Cited by 2 publications

(1 citation statement)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Techniques related to our inner-loop reachability analysis have been used for stability or reachability analysis in systems that are otherwise hard to analyze analytically. Reachability analysis for FFNNs based on abstract interpretation domains, interval arithmetic, or set inversion has been used in rule extraction and neural net stability analysis [5,18,66,74] and continues to be relevant, e.g., for verification of multi-layer perceptrons [56], estimating the reachable states of closed-loop systems with multi-layer perceptrons in the loop [78], estimating the domain of validity of neural networks [2], and analyzing security of neural networks [71]. While these works provide methods to extract descriptions that faithfully reflect behavior of the network, they do not generally ensure descriptions are comprehensible by end-users, do not explore the practice of strengthening descriptions by ignoring the effects of measure-zero sets, and do not consider varying description abstraction.…”

Section: Related Work and Discussionmentioning

confidence: 99%

Fanoos: Multi-Resolution, Multi-Strength, Interactive Explanations for Learned Systems

David¹,

Mitsch²

2020

Preprint

View full text Add to dashboard Cite

Machine learning becomes increasingly important to tune or even synthesize the behavior of safety-critical components in highly nontrivial environments, where the inability to understand learned components in general, and neural nets in particular, poses serious obstacles to their adoption. Explainability and interpretability methods for learned systems have gained considerable academic attention, but the focus of current approaches on only one aspect of explanation, at a fixed level of abstraction, and limited if any formal guarantees, prevents those explanations from being digestible by the relevant stakeholders (e.g., end users, certification authorities, engineers) with their diverse backgrounds and situation-specific needs. We introduce Fanoos, a flexible framework for combining formal verification techniques, heuristic search, and user interaction to explore explanations at the desired level of granularity and fidelity. We demonstrate the ability of Fanoos to produce and adjust the abstractness of explanations in response to user requests on a learned controller for an inverted double pendulum and on a learned CPU usage model.This material is based upon work supported by the United States Air Force and DARPA under Contract No. FA8750-18-C-0092. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the United States Air Force and DARPA. We would like to thank David Held for pointing us to the rl-baselines-zoo repository. Also, we would like to thank Nicholay Topin for supporting our spirits at some key junctures of this work.

show abstract