Apple researchers propose doubly sub-linear interactive proofs for verifying large inputs
New theoretical framework enables ultra-fast proof generation and verification for approximate assertions about massive datasets, with potential implications for scalable verification in machine learning systems.
3 sources · cross-referenced
- Apple ML Research published a paper introducing doubly sub-linear interactive proofs of proximity (dsIPPs).
- Proof generation and verification require reading only a sub-linear portion of the input, enabling efficient validation of approximate assertions about large datasets.
- The framework supports properties decidable by constant-width read-once oblivious branching programs, Hamming weight approximation, and relaxed bipartiteness in bounded-degree graphs.
- Work is scheduled for presentation at ITCS in July 2026.
Apple’s Machine Learning Research team published a paper introducing doubly sub-linear interactive proofs of proximity (dsIPPs), a theoretical framework designed to enable ultra-fast proof generation and verification for approximate assertions about large input objects.
In the proposed system, proof generation requires reading only a sub-linear portion of the input, while verification is even faster, requiring access to an even smaller subset of the data. This contrasts with traditional methods that demand full or near-full inspection of inputs to validate properties.
The authors construct dsIPPs for properties decidable by constant-width read-once oblivious branching programs (ROOBPs), as well as for approximate verification of an input’s Hamming weight and a relaxation of bipartiteness in bounded-degree graph models.
The framework is positioned within the property testing literature, where the honest prover can make the verifier accept inputs that satisfy the property, but no dishonest prover can fool the verifier into accepting inputs that are far from the property.
The paper is authored by Noga Amir, Oded Goldreich, and Guy N. Rothblum, with Goldreich and Rothblum affiliated with the Weizmann Institute of Science. The work is slated for presentation at the Innovations in Theoretical Computer Science (ITCS) conference in July 2026.
The research highlights a broader theme in Apple’s ML Research: developing methods to handle large-scale data efficiently, including related work on interactive proofs for general distribution properties and formal proof systems like Hilbert.
- Jul 19, 2026 · Apple — Machine Learning Research
Apple researchers propose interactive proof systems to verify distribution property claims with sublinear overhead
Trust79 - Jul 19, 2026 · Apple — Machine Learning Research
Apple researchers show separation between testing and verification for location-invariant properties
Trust84 - Jul 18, 2026 · Apple — Machine Learning Research
Apple proposes unlearning method that cuts computational cost by up to 50% by focusing on low-influence data points
Trust79