About his Research at CISC he says
Active machine learning is a new paradigm of machine learning where we seek to use a human in the loop in order to improve the efficiency and performance of our machine learning model. We do this by iteratively querying a human to provide feedback and information to the AI system, and using this information to become better at not only machine learning but also querying the user in the next iteration. Currently I am developing such an Active learning technology for safety critical scenarios, namely anomaly and fault detection in the industry. This problem has two parts which need to be handled respectively, the anomaly detection part requires detection methods which can be improved with active learning, and secondly the safety critical part which can be achieved by using explainable and reliable ML methods. Bayesian learning is one form of learning which naturally incorporates active learning and provides us with explainable and reliable algorithms. Currently, I am investigating the use of Bayesian Deep learning for fault detection in the semi-conductor manufacturing datasets. BDL gives us predictions with certainty(error) estimates and thus can be used as a reliable algorithm. BDL can easily be improved by adding a querying algorithm and thus turning it into Bayesian deep active learning, therefore our default choice to proceed with the investigations.
My progress can be found at GitHub.com/deveshjawla/PhD_Project_2021. A comparison of BDL, and BDAL with the current state of the art(Random forests) on the SECOM dataset is sought in the first year’s publication.”
Co-author of the paper: “Safety-critical systems in the automotive sector: pros and cons in the current state-ofthe-art of human performance assessment“