Pith. sign in

REVIEW 3 cited by

No Soundness in the Real World: On the Challenges of the Verification of Deployed Neural Networks

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2506.01054 v1 pith:52MBQUTA submitted 2025-06-01 cs.LG

No Soundness in the Real World: On the Challenges of the Verification of Deployed Neural Networks

classification cs.LG
keywords soundnessfloatingnetworkspointverificationverifiersboundingdeployed
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The ultimate goal of verification is to guarantee the safety of deployed neural networks. Here, we claim that all the state-of-the-art verifiers we are aware of fail to reach this goal. Our key insight is that theoretical soundness (bounding the full-precision output while computing with floating point) does not imply practical soundness (bounding the floating point output in a potentially stochastic environment). We prove this observation for the approaches that are currently used to achieve provable theoretical soundness, such as interval analysis and its variants. We also argue that achieving practical soundness is significantly harder computationally. We support our claims empirically as well by evaluating several well-known verification methods. To mislead the verifiers, we create adversarial networks that detect and exploit features of the deployment environment, such as the order and precision of floating point operations. We demonstrate that all the tested verifiers are vulnerable to our new deployment-specific attacks, which proves that they are not practically sound.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Lipschitz-Based Robustness Certification Under Floating-Point Execution

    cs.LG 2026-03 conditional novelty 7.0 partial

    Lipschitz-based robustness certificates that assume real arithmetic can be unsound under floating-point execution; a formal FP-aware theory and certifier close that gap for dense ReLU networks.

  2. Lipschitz-Based Robustness Certification Under Floating-Point Execution

    cs.LG 2026-03 conditional novelty 7.0

    Classical Lipschitz robustness certificates are unsound under floating-point execution; this paper derives sound corrected conditions, a Rocq formalisation, and a practical certifier for dense ReLU networks.

  3. TAO: Tolerance-Aware Optimistic Verification for Floating-Point Neural Networks

    cs.CR 2025-10 conditional novelty 7.0

    A tolerance-aware optimistic verification protocol makes floating-point neural-network inference verifiable on heterogeneous GPUs by accepting outputs within per-operator error bounds and resolving disputes via a Merk...