A New Way of Making Data Processing Verifiably Better

Discover how a framework from IMDEA Software and partners may just revolutionise verifiable computing, enhancing AI and image processing efficiency, privacy, and scalability for research solutions.

Researchers from the Institute IMDEA Software, Universidad Carlos III de Madrid, and NEC Laboratories Europe have unveiled a new potential framework that significantly boosts the efficiency and practicality of verifiable computing. Detailed in their latest paper, “Modular Sumcheck Proofs with Applications to Machine Learning and Image Processing,” and presented at the ACM conference on computer and communications security, this framework tackles the longstanding issues of scalability and modularity in cryptographic proof systems.

Verifiable computation is a set of cryptographic techniques ensuring the correctness of data processing by third parties, such as cloud servers or companies, without compromising data privacy. These techniques are vital for validating data integrity in various applications, including proving the authenticity of image edits, verifying AI predictions, and confirming the exclusive use of customer-provided data in financial assessments or research collaborations.

Traditionally, verifiable computation has faced challenges. General proof systems, while broad in application, often struggle with scalability when managing large datasets. Alternatively, bespoke solutions offer enhanced efficiency but suffer from compatibility issues that hinder their integration into broader data processing chains.

The newly introduced framework by the research team proposes a novel solution to these challenges by employing a modular approach to verifiable computation of sequential operations. Central to this innovation is a cryptographic primitive known as the Verifiable Evaluation Scheme (VE), designed to bridge the performance gap between general-purpose and custom-tailored solutions.

One of the standout applications of this new framework is in artificial intelligence. The researchers have developed a VE specifically tailored for convolution operations, a fundamental component of most AI models, such as convolutional neural networks (CNNs). “Our protocol can seamlessly integrate into any data processing chain, ensuring complete verification of outputs, such as AI predictions,” explained David Balbás, a PhD student at IMDEA Software and a researcher involved in the study.

Moreover, the framework extends to image processing, enabling efficient verification of image edits, including complex operations like cropping, blurring, and rescaling. Damien Robissout, a research programmer at IMDEA Software and co-author of the study, noted, “Our benchmarks indicate that our proofs are five times faster to generate and ten times faster to verify than existing solutions, marking a significant advancement in the field.”

This technological leap not only enhances the operational efficiency of cryptographic proofs but also broadens their application, ensuring data integrity, fairness, and privacy across various sectors. “In today’s world, ensuring technological trust is paramount, and our framework addresses this need effectively,” stated Maribel González Vasco, a Professor of Excellence at UC3M.

The open-source nature of the application and its modular design offer vast potential for adaptation and integration into diverse tools within data processing chains. This adaptability makes the framework a robust candidate for deployment in areas ranging from financial ethics and personal data protection to AI regulation.

You can read their full paper Modular Sumcheck Proofs with Applications to Machine Learning and Image Processing in 2023 ACM SIGSAC Conference on Computer and Communications.

Staff Writer

Our in-house science writing team has prepared this content specifically for Lab Horizons

Leave a Reply

Your email address will not be published. Required fields are marked *