Aging, Inflammaging and the challenge of explainable Artificial Intelligence
Presentation by Claudio Franceschi, Bologna
Time: 14:00–14:45
Machine learning (ML), and deep learning (DL) in particular, is currently one of the most common approaches used in many tasks in different research areas. Deep models operate with a large amount of input data, training many layers, but in most cases, their work process is not transparent. That is why they are also called black boxes. Decision-making process in such deep architectures is difficult to explain, therefore questions about the trustworthiness of such models and the security of their deployment arise. The problem of explainability of artificial intelligence (AI) models is being actively studied, and eXplainable Artificial Intelligence (XAI) has become an important area of AI. Major goals of XAI are to develop approaches capable of describing the reasons for model decision-making, and, more profoundly, to develop interpretable and logically explainable models. XAI explanations must be understandable, reliable, and yet the models to which they are applied must retain predictive accuracy. At present, a major area of medical studies is the search for biomarkers of aging, which has become particularly active with the growth of ML and DL research. A separate task here is the prediction of biological age, which characterizes human health in various aspects and can differ from chronological age. Age is a universal attribute of all living organisms and a biologically meaningful characteristic associated with risk of mortality, disease, and general well-being. Even though each individual feature may not be explicitly related to age, a combination of features may have predictive power (Zhavoronkov and Mamoshina, 2019). Therefore, various types of data may be used to predict age, such as laboratory tests, magnetic resonance (MRI) and X-ray images, electrocardiogram (ECG) and electroencephalogram (EEG) signals, and many other inputs. Development and analysis of age predictors can help, in particular, in the study of age-related diseases (Zhavoronkov and Cantor, 2011), immunological aging, response to medications and vaccines (Zhavoronkov and Mamoshina, 2019) and in many other healthcare applications.