Kezhen Chen and Ken Forbus Win Best Paper Award at KR2ML Workshop at NeurIPS 2019

PhD student Kezhen Chen and Northwestern Engineering’s Ken Forbus, Walter P. Murphy Professor of Computer Science, won a Best Paper Award at the KR2ML Workshop at NeurIPS 2019, held in December in Vancouver, BC. Kezhen Chen

Chen and Forbus’s winning paper entitled “TP-N2F: Tensor Product Representation for Natural to Formal Language Generation” is coauthored by Qiuyuan Huang (Microsoft Research), Hamid Palangi (Microsoft Research), Jianfeng Gao (Microsoft Research), and Paul Smolensky (Johns Hopkins University). 

While deep learning has seen a number of successes, such as in speech recognition and machine translation, deep learning models are harder to trust in higher stakes situations — such as self-driving cars and medical diagnosis — because their results do not come with explanations, according to the paper. 

Paul Smolensky’s Tensor Product Representation can provide some of the advantages of explainability for symbolic representations with the advantages of learnability provided by deep learning, the research shows. Ken Forbus

“There are decades of evidence across cognitive science suggesting that symbols and relationships are part of our mental contents,” Chen said. “Understanding how much symbols can be embedded within deep neural networks is one of the exciting new areas in machine learning where we can gain the advantages of both.”

McCormick News Article