The research group is contributing to establish lifelong machine learning capabilities as a key and unsolved challenge is that much of the previous work in both supervised and unsupervised learning assumes that training data have already been collected and used as training data to make predictions. At test time, the ML algorithm makes predictions without taking into account the fact that the circumstances under which the data are acquired may have changed and humans should be involved to continuously adapt to new data, new processes and new sensors and technologies, which is prohibitive for reliable flexibility.
Lifelong learning and lifelong computing approaches will enable capabilities to operate, adapt and evolve learning systems autonomously throughout its lifetime in harsh and continuously changing environments. This novelty should replace the actual manual ML system design with a dynamic software architecture that implements well-composed learning processes, which are dynamically optimized and evolving in a holistic lifetime manner.
Implementation of lifelong and self-explainable ML in the industrial contexts will improve productivity without compromising on product quality.
Another works in the research groups involve novel and recent progress in ML, such as automated ML, which applies algorithm configuration for constructing ML pipelines and optimizing hyperparameters of ML algorithms to tune configurations for optimizing performance on a given set of problem instances. An automated data-driven ML algorithm optimization will enable cognitive capabilities for predictive and autonomous operations.
