145.  Computational Intelligence Development Towards a Distributed Continuous Learning Architecture  

Abstract:

The current direction of artificial intelligence development is focused on scaling data and computing processes. As important as this direction is, it does not provide the explainability that is necessary for sharing the knowledge among peers. This would be very beneficial within the medical world. The patients’ data is distributed among many hospitals, clinics, and research centres and lacks a unified form. Moreover, due to GDPR restrictions, we cannot share such sensitive data freely, and it must be processed locally, which can cause a problem when scientists want to process the data and see a bigger picture. AI models often lack ways to explain itself and to share the knowledge gained from a dataset. Due to limited resources on hospitals premises, we ask ourselves if we can distribute the work on “fast” and “slow” processes, such as classification, and rules discovery and inference, respectively. The standard federated learning methods might be hard to implement into the real-world hospitals and research centres due to serious variations in data, its form and software solutions implemented locally. We hypothesise an architecture that allows to perform computations locally, on hospitals infrastructure, without sharing the patients’ data, but shares the gained knowledge and compare treatment paths and results within the network of research centres and hospitals.

About the author:

Filip completed his Master’s degree in Computer Science at AGH University in cooperation with Sano. For his thesis, he examined a design of a federated text search engine for Biomedical literature analysis. At Sano he works on Computational Intelligence and Distributed Systems design for medical centres. As part of his career, he is interested in topics such as Cloud Computing, Large Scale Computing and Natural Language Processing.