TLDR: A new research paper introduces a privacy-preserving platform and an AI-powered crystal counting tool that enable small and medium-sized manufacturers to securely share proprietary data with researchers. This collaboration allows for the development and deployment of innovative AI solutions, such as the automated crystal analysis tool for food manufacturing, which significantly improves efficiency and accuracy in quality control while ensuring the confidentiality of sensitive business data.
Manufacturers, especially small and medium-sized enterprises (SMEs), often face a dilemma: they need advanced data tools, particularly those powered by Artificial Intelligence (AI), to stay competitive. However, deep-seated concerns about data privacy and the confidentiality of their proprietary information prevent them from sharing data with researchers who could develop these crucial tools.
A new research paper introduces a groundbreaking solution to this challenge: a privacy-preserving platform designed to bridge the gap between manufacturers and AI researchers. This platform allows manufacturers to securely share their data, enabling researchers to develop innovative AI tools. Crucially, these tools can then be deployed back onto the platform, offering solutions with robust privacy and confidentiality guarantees, ensuring proprietary data remains secure.
The paper illustrates this innovative approach through a practical case study in the food manufacturing industry. Specifically, it addresses the labor-intensive process of counting food crystals in microscope images for quality control. Traditionally, this task was performed manually, requiring significant human effort and time. The researchers developed and deployed an AI-powered crystal analysis tool that automates this process, making it significantly faster and more accurate.
This new tool automatically characterizes crystal size distribution and counts from microscope images, even removing natural imperfections from sample preparation. A machine learning model was specifically developed to count high-resolution translucent crystals and agglomerations. The algorithm was then packaged into a user-friendly, web-based application, secured by the underlying privacy-preserving platform, MyDataCan. This allows manufacturers to utilize the tool on their factory floor while ensuring their sensitive data remains confidential.
The MyDataCan Platform: Enabling Secure Collaboration
At the heart of this solution is MyDataCan, a privacy-preserving data-sharing platform developed by the Public Interest Tech Lab at Harvard University. Originally designed to give individuals control over their personal data, MyDataCan’s capabilities have been adapted for the manufacturing sector. It enables an “AI Data Community” model where manufacturers can upload their data to their private “can.” Researchers can then apply machine learning algorithms to this data, and the results are returned securely to the manufacturer’s MyDataCan.
The platform ensures data privacy and confidentiality throughout the entire process—during upload, transit to/from AI tools, analysis, and storage—by encrypting all data. Neither the platform nor other users can decrypt or access the raw data. MyDataCan employs advanced privacy-enhancing technologies such as homomorphic encryption, federated learning analysis, secure multi-party computation, and zero-knowledge proofs. Virtual machines created for analysis are destroyed once the process is complete, further enhancing security.
Also Read:
- VeFIA: A New Framework for Auditing AI Inference in Collaborative Systems
- The Rise of AI Agents and Agentic AI: A New Era for Manufacturing
Impact and Future Directions
The crystal counting tool has been successfully adopted by the manufacturer for daily operations, leading to substantial benefits. Operators can now analyze an entire crystal image in approximately 30 seconds, a tenfold decrease in time compared to the previous manual method. This has resulted in significant time savings, improved accuracy and consistency in seed counts, and increased confidence in product quality.
The success of this project highlights the potential for privacy-preserving AI tools to transform manufacturing. The manufacturer is interested in expanding the tool’s use to other factories and applying it to later stages of the manufacturing process where crystal agglomeration occurs. The image processing algorithm’s versatility also suggests its applicability to diverse crystallization processes in pharmaceuticals, food processing, and particle processing. This research demonstrates a viable path for fostering collaboration between researchers and manufacturers, laying a foundation for future AI advancements across various industries while upholding critical data confidentiality.
For more detailed information, you can read the full research paper here.
