The scale, heterogeneity and velocity of data far transcends human comprehension. We rely on machines to learn from raw data, using mathematical models to mine the experiential patterns needed for applications involving prediction, classification, recommendation, event detection, decision making, and so forth.
Ultimately we need to translate the resulting store of numeric data compiled by the machine back into a format that humans can comprehend and act upon.
However, traditional methods for learning, mining and visualization face unprecedented challenges — and unprecedented opportunities — as data volume and computational capacity continue to grow. This research line will focus on emerging themes relating to learning, mining and visualization:
-Addressing robustness and data quality issues, first to improve the quality, structure and scope of training data provided as input to the learning process, to learn to detect conflicting, anomalous or otherwise low-credibility information.
-Enhancing contemporary learning models, e.g. in the context of Deep Learning, devising new strategies for sequential, reinforcement and imitation learning in order to make the most of available data; and in the context of social networks, developing mechanisms able to cope with missing data or bias sampling, to name some current challenges.
-Proposing new visualizations and interactive media that transform “machine experience” into a form that humans can, explore, conceptualize, share and act upon.