AI researchers have been created new techniques for training machine-learning algorithm while ensure patient’s privacy.
An innovative solution for privacy issues has emerged from the absorption of DeepMind’s health division in Google.
In response to the privacy concerns about the disclosure of patient data as Google moved its subsidiary DeepMind Health division into the main company, a split neutral network has been introduced as the latest advanced method to process data with confidentiality.
Its basic operation mechanism is similar to the process of data encryption. The neutral networks divide the data processing into separate stages that can be carried out by different people: one person starts training the machine learning model then another person can finish it. Based on this mechanism, patients’ raw data could be processed in hospitals and other medical institutions to form the training of the partway model in the first stage. These half-trained models would then be sent to a centralized location, either the clouds services, Google, or another company, to be completed in the final stages of training.
The superior advantage is that due to the obfuscation of medical data in the first stage, the centralized location can only be seen the output of the half-processed model plus the model itself. The raw patient data, therefore, is secure with the initial constitutions and the hospitals would benefit from a final model trained on a combination of every participating institution’s data.
This approach with split neural networks has been found to involve considerably fewer computational resources to train while producing much more accurate models.
Privacy issues stemming from AI have raised worries among users. AI is just a tool, and it is essential for people to closely supervise and moderate its operation with transparency. The importance of AI applications in key sectors including healthcare is the focus of Layer 7 – Business Applications for All of Society of the AIWS 7-Layer Model developed by the Michael Dukakis Institute.
