Ant Group Files Patent for Privacy Computing Algorithm, Utilizing Large Language Models to Evaluate Performance

Ant Group Files Patent for Privacy Computing Algorithm, Utilizing Large Language Models to Evaluate PerformanceFinancial World, August 29, 2024 - Tianyancha intellectual property information shows that Ant Blockchain Technology (Shanghai) Co., Ltd

Ant Group Files Patent for Privacy Computing Algorithm, Utilizing Large Language Models to Evaluate Performance

Financial World, August 29, 2024 - Tianyancha intellectual property information shows that Ant Blockchain Technology (Shanghai) Co., Ltd. has filed a patent for a "Method, Apparatus, Device and Storage Medium for Evaluating the Operation of Privacy Computing Algorithms," with a publication number of CN202410383065.2, filed in March 2024.

This patent aims to provide an efficient method for evaluating the performance of privacy computing algorithms, leveraging large language model technology to enhance the accuracy and efficiency of the evaluation. Specifically, the patent describes an evaluation method consisting of the following steps:

1. Information Acquisition: Initially, the runtime information of the privacy computing algorithm being evaluated and the configuration information of the device used to run the algorithm need to be obtained. This information includes the algorithm's specific parameters, runtime environment, data types, etc.

2. Model Input: The collected algorithm runtime information and device configuration information are input into the service model of the large language model.

3. Performance Prediction: The large language model's service model predicts the performance of the privacy computing algorithm on a specific device based on the input information. This includes evaluating aspects like runtime, resource consumption, and security risks.

4. Model Training: The large language model's service model is not generated out of thin air. It is trained using a massive amount of data. This training data comprises algorithm runtime information, device configuration information, and the corresponding actual performance. Through supervised fine-tuning training on these training samples labeled with their performance, the large language model's base model gradually learns the relationship between algorithm performance and device configuration, enabling it to accurately predict based on new input information.

5. Output Results: Finally, the large language model service model outputs the predicted performance of the privacy computing algorithm on the device, serving as a reference for relevant technical personnel.

The method described in this patent leverages the powerful information processing capabilities of large language models, automating the evaluation process that previously required significant manual operations and testing. It also enhances the accuracy and efficiency of the evaluation. This technology can effectively contribute to the development of the privacy computing field, improving the security, reliability, and scalability of related technology applications, providing assurance for more secure and convenient digital services.

It's noteworthy that Ant Group has been actively investing in the privacy computing field, achieving a series of breakthroughs. This patent application highlights Ant Group's technological leadership and continuous innovation in this area. As large language model technology continues to advance and application scenarios expand, more privacy computing evaluation methods based on large language models are anticipated, providing stronger technical support for safer and more efficient digital services.