Bae Ho Email & Phone Number

Why do data bias and data hallucination occur?'Synthetic data' has emerged as a trend. Original data often contains sensitive personal information, which limits its full utilization. However, synthetic data, which is artificially created to mimic real data, is relatively free from these constraints.Synthetic data is a real-like data. It poses a lower risk of personal data exposure and can generate vast amounts of data by simulating various scenarios, thus overcoming performance issues due to data shortages in AI models.Yet, there is a major irony: As AI technology has advanced, the performance of synthetic data has also improved rapidly. However, requiring synthetic data to closely mimic original data has led to the creation of synthetic data that closely resembles the original, inadvertently making it vulnerable to attacks and revealing information about the original data.This was a concern for Cubig.And to address it, we implemented differential privacy as a security mechanism from the start of synthetic data generation, producing irreversible synthetic data that cannot be traced back to the original. Differential privacy is a robust security technology used by institutions like the U.S. Census Bureau and Apple. It maintains security regardless of subsequent processing, but applying it to generative AI traditionally led to a sharp decline in synthetic data performance. However, Cubig has solved this issue with our unique solution, maintaining state-of-the-art performance.DM me if you’re Interested in learning how Cubig’s technology stays secure against any attack.