Achille FokouePrincipal Research Staff Member and Master Inventor IBM
Achille Fokoue is a Principal Research Staff Member and Master Inventor at IBM Research AI in Yorktown Heights, where he leads the Foundations of AI Reasoning group. He has over 18 years of research experience in knowledge representation and reasoning focused on developing theories, algorithms, standards, and systems for scaling reasoning over large and expressive knowledge bases that tolerate inconsistencies and uncertainties inherent in KBs populated from unstructured sources. He has lead various research efforts on applying machine learning and knowledge representation and reasoning in many domains including the Tiresias project that aggregates information from multiple data sources to predict adverse drug reactions. He is a co-editor of the OWL 2 Web Ontology Language Profiles specification, and has authored or co-authored over 90+ scientific reports and manuscripts that have been cited, in aggregate, more than 2400 times.
Contact Information: firstname.lastname@example.org
Unifying Learning and Reasoning
Over the past decade Deep Learning has been extremely successful at leveraging large amount of available training data to reach and, in some cases, surpass human level performance at pattern detection oriented tasks in a variety of domains including vision, speech, and machine translation. However, this learning based approach to Artificial Intelligence has proven insufficient to successfully tackle relatively simple tasks requiring a deeper level of understanding and reasoning (both common sense and formal reasoning): e.g., answering elementary school science questions. In this talk, I will compare and contrast deep learning against traditional knowledge & reasoning approaches, which used to be the core component of early AI systems. I will then present some approaches, explored at IBM Research AI, to address some key limitations of today’s AI systems by bridging the gap between learning and reasoning with systems that learn to reason and use knowledge and symbolic reasoning to learn more efficiently (i.e., with less data while providing explanations).