English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Despite the tremendous advances that have been made in the last decade ondeveloping useful machine-learning applications, their wider adoption has beenhindered by the lack of strong assurance guarantees that can be made abouttheir behavior. In this paper, we consider how formal verification techniquesdeveloped for traditional software systems can be repurposed for verificationof reinforcement learning-enabled ones, a particularly important class ofmachine learning systems. Rather than enforcing safety by examining andaltering the structure of a complex neural network implementation, ourtechnique uses blackbox methods to synthesizes deterministic programs, simpler,more interpretable, approximations of the network that can nonethelessguarantee desired safety properties are preserved, even when the network isdeployed in unanticipated or previously unobserved environments. Ourmethodology frames the problem of neural network verification in terms of acounterexample and syntax-guided inductive synthesis procedure over theseprograms. The synthesis procedure searches for both a deterministic program andan inductive invariant over an infinite state transition system that representsa specification of an application's control logic. Additional specificationsdefining environment-based constraints can also be provided to further refinethe search space. Synthesized programs deployed in conjunction with a neuralnetwork implementation dynamically enforce safety conditions by monitoring andpreventing potentially unsafe actions proposed by neural policies. Experimentalresults over a wide range of cyber-physical applications demonstrate thatsoftware-inspired formal verification techniques can be used to realizetrustworthy reinforcement learning systems with low overhead.

An Inductive Synthesis Framework for Verifiable Reinforcement Learning