Industrial AI for Maintenance and Repair (2014-Present) Developing innovative algorithms for performance degradation detection, maintenance effectiveness estimation, failure prediction, remaining useful life estimation, and maintenance recommendation (more at Hitachi R&D) Outcome: Two granted patents, ten pending patents and fourteen papers Impact: Two patent awards, four achievement awards Fast and Accurate Algorithms for Organizing Big Data (2013-2014)
Developed MapReduce algorithms for big data summarization and mentoring a Master’s
student in the development of MapReduce-efficient algorithms for big data clustering
Outcome: One journal paper in KAIS, Two conference papers in ICDM'13 and SDM'14, and two workshop papers in NIPS'13
Impact: Best Paper Award Runner-up in ICDM'13 Mentored a Ph.D. student in the development of innovative algorithms for domain adaptation and minimally supervised learning Impact: Best Abstract Award in NIPS'13 Workshop on New Directions in Transfer and Multi-Task Learning
Minimally-supervised Query Understanding for Intelligent Personal Assistants (2013) Managed a team of two researchers to build an innovative solution for Vestec, Inc. to understand the semantics of transcribed queries Impact: More than 75% reduction in the required training data Representative Selection for Unsupervised Data Analysis (2009–2012)
Developed novel algorithms to select a few representative instances and features from data and enhance the efficiency and effectiveness of data clustering algorithms
Outcome: One journal paper in KAIS, two papers in ICDM'11 and AISTATS'11, one workshop paper in NIPS'10
Impact: Nomination for the Best Paper Award in ICDM'11
Fast and Accurate Recognition of Banknote Serial Numbers (2011-2012) Managed a team of two researchers to develop a novel solution for NCR Canada, Ltd. to recognize banknote serial numbers at the speed of accepting/dispensing banknotes Impact: Less than 2% error and half a millisecond recognition time per character using a few training data Statistical Semantics for Enhancing Document Clustering (2008–2010)
Developed novel
measures of semantic similarity between documents to enhance different
clustering algorithms
Outcome: One journal paper
in KAIS and two papers in specialized workshops in ICDM'09 and SDM'10
Impact: Best
Paper Award in ICDM'09 Workshop on Semantic Aspects in Data Mining
High–Speed Number Recognition for Inline Print Inspection Systems (2008-2009) Developed a novel solution for Erhardt+Leimer, Ltd. to inspect prints with changing numbers at the speed of the printing process Impact: Less than 0.01% error and half a millisecond recognition time per digit using a few training data Detecting Anomalous Behaviour for Business Process Management (2005–2007)
Co-invented two IBM patents to detect and predict anomalous behaviour in business processes Outcome: Two patents were filed based on this work Impact: Selected as one of the Innovation that Matters for IBM |