Speaker
Description
Climate change, loss of bio-diversity, food/water/energy security for the growing population of the world are some of the greatest environmental challenges that are facing the humanity. These challenges have been traditionally studied by science and engineering communities via process-guided models that are grounded in scientific theories. Motivated by phenomenal success of Machine Learning (ML) in advancing areas such as computer vision and language modeling, there is a growing excitement in the scientific communities to harness the power of machine learning to address these societal chal-lenges. In particular, massive amount of data about Earth and its environment is now continuously be-ing generated by a large number of Earth observing satellites, in-situ sensors as well as physics-based models. These information-rich datasets in conjunction with recent ML advances offer huge potential for understanding how the Earth's climate and ecosystem have been changing, how they are being impacted by humans actions, and for devising policies to manage them in a sustainable fashion. However, capturing this potential is contingent on a paradigm shift in data-intensive scientific discovery since the “black box” ML models often fail to generalize to scenarios not seen in the data used for training and produce results that are not consistent with scientific understanding of the phenomena.
This talk presents an overview of a new generation of machine learning algorithms, where scientific knowledge is deeply integrated in the design and training of machine learning models to accelerate scientific discovery. These knowledge-guided machine learning (KGML) techniques are fundamental-ly more powerful than standard machine learning approaches, and are particularly relevant for scien-tific and engineering problems that are traditionally addressed via process-guided (also called mecha-nistic or first principle-based) models, but whose solutions are hampered by incomplete or inaccurate knowledge of physics or underlying processes. While this talk will illustrate the potential of the KGML paradigm in the context of environmental problems (e.g., Ecology, Hydrology, Agronomy, climate sci-ence), the paradigm has the potential to greatly advance the pace of discovery in any discipline where mechanistic models are used.
