Summary of Verifiable Evaluations Of Machine Learning Models Using Zksnarks, by Tobin South et al.
Verifiable evaluations of machine learning models using zkSNARKs
by Tobin South, Alexander Camuto, Shrey Jain, Shayla Nguyen, Robert Mahari, Christian Paquin, Jason Morton, Alex ‘Sandy’ Pentland
First submitted to arxiv on: 5 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel method for verifying the performance of closed-source machine learning models without re-performing the benchmarking process. The approach, based on zero-knowledge proofs (zkSNARKs), allows developers to generate attestations that demonstrate a model’s accuracy, fairness, or other metrics over public inputs. The authors present a flexible proving system that can be applied to various neural network architectures and compute requirements. They showcase their method by evaluating real-world models, highlighting the challenges and design solutions involved. This work paves the way for a new paradigm in transparently evaluating private machine learning models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you buy a self-driving car that can’t be trusted to follow the rules of the road because its programming is secret. That’s a problem! In the same way, when we use AI models we can’t always trust what they’re doing because their “brain” is hidden from us. This research makes it possible to verify that an AI model works as promised without needing to see how it was programmed. They do this using special math tricks called zero-knowledge proofs. The authors show how this can be done for real-world models and discuss the challenges they faced along the way. |
Keywords
* Artificial intelligence * Machine learning * Neural network
