We are very excited to share that Hazy has two workshop papers accepted at ICML 2023:
The preparation of the two papers required a wide range of expertise, and we extend our gratitude to our exceptional co-workers, academic partners, and legal advisors for their valuable advice, feedback, support, and encouragement.
When Synthetic Data Met Regulation
The paper was accepted to the 1st ICML Workshop on Generative AI and Law. It focuses on the legality of synthetic data created by generative machine learning models trained on tabular personal data.
This paper delves into the legal definitions of personal data, effective and sufficient anonymization, and their translation to technical tests and requirements. Additionally, we analyse the privacy aspects of generative models and differential privacy. We argue that the combinations of these two techniques can yield synthetic data that meets the criteria for sufficient anonymization, making it anonymous and compliant with regulations.
On the Challenges of Deploying Privacy-Preserving Synthetic Data in the Enterprise
The paper was accepted to the 1st ICML Workshop on Challenges in Deploying Generative AI. It identifies and organises the challenges linked to the implementation of synthetic data in enterprise settings. Furthermore, it proposes a systematic approach to efficiently tackle these challenges.
This paper draws upon our extensive real-life experience in providing synthetic data solutions to diverse customers, enabling us to systematise the primary hurdles encountered by large enterprises. These obstacles are categorised into five groups: i) generation, ii) infrastructure & architecture, iii) governance, iv) compliance & regulation, and v) adoption. Inspired by our maturity model, we then present a strategic and systematic approach that enterprises can adopt to effectively tackle these challenges and build trust in their implemented solutions, ultimately achieving their goals.
For more details, please refer to the full versions of the papers: