Over the past year, zk-SNARKs have progressed beyond expectations. While the general consensus is that these innovations are years away, applications, such as ZK-EVM, are emerging. Enhancements to zk-SNARKs have made it possible to explore new use cases for blockchains, and in particular, we are paying close attention to research using zk-SNARKs to solve many pressing problems posed by the increased use of machine learning and artificial intelligence.
As machine learning gains popularity, it is being used in a wide variety of applications. However, the credibility of its forecasts and reliance on opaque data sources has become a major concern. The ability to replicate a model claiming high accuracy is difficult, and there is no guarantee of consistency and correctness of predictions in actual production.
This article aims to briefly describe why there is a lot of interest in zk-SNARK-based machine learning (ZK-ML) systems, and discuss some potential applications of the technology.
Why is ZK-ML needed?
With supervised machine learning, the input is given to a model that has been trained with specific parameters. The model then produces output that can be used by other systems. Thanks to lightweight machine learning frameworks and formats like ONNX, it is now possible to run these inferences on edge devices, such as mobile phones or IoT devices, instead of sending input data to centralized servers. This improves scalability and privacy for users.
However, it is important to note that both the inputs and parameters of machine learning models are often kept private and hidden from public view. This is because input data may contain sensitive information, such as personal financial or biometric data, and model parameters may also contain sensitive information, such as biometric authentication parameters.
On the other hand, downstream systems that consume the output of the ML model, such as on-chain smart contracts, need to be able to verify that the input was processed correctly to produce the claimed output.
The combination of machine learning and the zkSNARK protocol provides a new solution to these seemingly contradictory requirements.
ZK-ML use cases
There are many papers discussing how we can use zk-SNARKs to improve our future machine learning. The ZK-ML community has provided a very useful decision tree that lets us consider various use cases for this technique.
This decision tree is based on the intersection of two criteria: the need for privacy and computational integrity, and a heuristic optimization problem solved using machine learning. In other words, decision trees are used to determine suitability for use cases involving ZKML, where privacy and computational integrity are important, and where machine learning techniques are used to solve heuristic optimization problems,
Here are some ways how zk can be used for ML model innovation:
Privacy Preserving Machine Learning
zk-SNARKs can be used to train machine learning models without exposing private data to the model's creator or user. This allows the development of models that can be used in sensitive or regulated industries such as healthcare or finance without compromising the privacy of individuals using personal data.
Verifiable Machine Learning
zk-SNARKs can be used to prove that a machine learning model was trained on a specific dataset, or that a specific model was used to make predictions, without revealing details of the training data or the model. This can increase trust in the results of machine learning models, which is important in applications such as credit scoring or medical diagnosis.
Security Machine Learning
zk-SNARKs can be used to protect the integrity of machine learning models by ensuring that the model has not been tampered with or replaced with a different model. This is useful in applications where models are deployed in untrusted environments such as edge devices or public clouds.
Possible applications of ZKonduit (EZKL)
Projects like ZKonduit are seeing ZK-ML as the key to giving blockchain eyes, smart contracts exercising judgment, one-man oracles, and generally getting data on-chain in a scalable way. Using ZK-ML oracles provides an easier, faster, and more efficient way to transfer data off-chain to the blockchain, greatly increasing the potential for bringing data on-chain. ZK-ML can enable "smart judges" to interpret ambiguous events. This could lead to unimaginable new use cases for Web3, but the following are just a few that have been discussed recently:
ZK KYC
Be able to prove that a person's identity matches the corresponding ID card and that the ID number is not on a sanction list. While the technology is available, regulators may not accept it, as they currently require banks to "know" their customers, rather than just verify that they are not on sanctions lists. This is new territory for regulators, who must take steps to prevent unwanted players from using decentralized projects.
Fraud Check
Smart contracts or abstract accounts add a ZK-ML fraud spam check for detecting anomalous behavior. This means that zero-knowledge machine learning techniques can be used to detect and prevent fraudulent or spam activity by analyzing patterns of activity and comparing them to known patterns of fraudulent or spam activity. This can help ensure the security and integrity of the system by detecting and preventing malicious activity.
Make DAO Autonomous
in conclusion
in conclusion
Integrating zero-knowledge proofs into AI systems can provide new levels of security and privacy for users and the companies that use these systems. By enabling AI to prove the validity of its decisions without revealing the underlying data or algorithms, zero-knowledge proofs can help mitigate the risk of data breaches and malicious attacks. Additionally, they can help build trust in AI systems by providing a transparent and verifiable way to prove their fairness and accuracy.
As the field of artificial intelligence continues to grow and expand, the application of zero-knowledge proofs will become increasingly important to ensure the safe and responsible deployment of these powerful technologies.
