Accelerated Material Discovery
Federico Zipoli, PhD
IBM Research Division, Cognitive Computing & Industry Solutions Dept., Zurich Research Laboratory
Monday, August 19th, 10:30 – 11:30
Görgesbau (Helmholtzstr. 9), room 229
Abstract: Experiments and theory have been widely used to drive the discovery novel material. In the last decades, also simulations contributed to speed up innovation. In silico experiments became the third pillar of research. Today, the scientific community produces continuously a large amount of data which are made available via publications. It is becoming difficult for a researcher to follow such stream of information. Many researchers believe that data are becoming the fourth pillar of research. If we could efficiently consume data, we could face a disruptive opportunity of accelerating innovation. The work I will present is focused on extracting information from different unstructured sources with the goal of building a platform which can allow a researcher, expert in the field, to efficiently retrieve and link facts. Our current work is focused on building a platform that can connect data from different sources to allow inference and analytics to drive new discovery.
In my presentation, I will illustrate our team effort at IBM Research – Zurich to develop the building blocks which can enable such platform. In particular, I will focus on data ingestion from PDF documents and extracting information from text, tables, images with the goal of building a knowledge graph suitable for Q&A.
Dr. Federico Zipoli is a Research Staff Member at IBM Research – Zurich since 2011. He graduated in Materials Science at the University of Milano-Bicocca, Milan, Italy, in 2003, and obtained his Ph.D in Nanostructures and Nanotechnology in 2006, also in Milan. At IBM, he work atomistic simulations on different type of materials, for example on phase-change materials for storage applications, solid-state electrolytes for batteries. In the last 4 years, he is focused on accelerating the discovery of new materials via a combined use of data analytics and atomistic simulations based on neural networks. He is working on natural language processing tools to extract knowledge from text and deep learning based-models for data-mining and name entity recognition. He is exploring neural-network forcefield to model the atomic interactions.