ARPA Computation Platform Q1 Update: Better Performance, Usability, and Next Steps

SINGAPORE–(BUSINESS WIRE)–ARPA, a blockchain-based privacy-preserving computation network, released its Quarter One Update on March 24th, 2021. Over the past years and during the first quarter of 2021, ARPA developers and researchers have made tremendous efforts to upgrade ARPA�s secure computation modules to achieve better performance and usability.

According to ARPA’s previous business development with other financial companies and data processing entities, the most frequently used applications of privacy-preserving companies are large-scale private set intersection (PSI) and linear regression (LR). With this in mind, ARPA prioritized optimizing these specific applications. After a thorough survey of existing frameworks and implementations, ARPA developed a series of experiments on these tools in its current real-world project settings. By integrating additional modules into its computation system, the performance of large-scale data analysis on the ARPA computation platform was improved by around 30%.

Private set intersection (PSI) and linear regression (LR)

PSI is a cryptographic technique that allows two parties to compute their sets’ intersection without revealing anything else. LR is a simple but efficient approach that shows the relationship between several factors, such as adolescent obesity and physical activities. Medical, financial, and other risk-sensitive entities are very much dependent on such analysis tools.

However, the required massive data collection raises concerns on privacy, trade secrets, and regulation. These are the ideal application scenarios of multi-party computation (MPC). ARPA’s secure computation platform is designed to process these analyses, and now ARPA is improving the processing performance in the fields of asymmetry database and parallelization.

Computation performance has been tested on several public datasets such as Boston real-estate pricing, cancer linear regression, and CDC behavioral risk factor data. ARPA extended the dataset with synthetic data to million entries. The experiment was conducted on three powerful AWS instances located in different regions. To test the asymmetric database PSI performance, ARPA limited the inquirer’s computation power to a single thread and allocated multiple threads to the inquired party. The inquired dataset is more than ten million entries, while the inquiry dataset is comparatively smaller. The result shows that if the smaller database is one thousand times smaller than the larger one, then a similar performance will occur with plaintext computation. As for LR, thanks to the fixed-point arithmetic and instruction vectorization ARPA employed, the throughput raised around 35% because of the drastic reduction in communication overhead.

What’s Next?

Next, ARPA will conduct gene sequence alignment experiments on the Complete Genomics public dataset. The datasets for previous performance tests are mainly synthetic, and ARPA would like to perform a real-world million-entry data analysis such as pairwise genetic alignment. This kind of application can help with disease risk evaluation without compromising patients’ privacy.

There are many good things to anticipate from ARPA in 2021 as more progress updates are coming out soon!

Contacts

Media Information:

Yemu Xu

about@arpachain.io
https://arpachain.io