Omar U. Florez

Email: [email protected] Doing state-of-the-art research on enabling natural conversations between humans and devices. Implementing meta-learning and memory augmented recurrent neural networks to learn with small data, as we humans do, and to deal with catastrophic forgetting. Formerly, I was a Senior Research Manager at Capital One and Machine Learning Research Scientist at Intel Labs. Focused on teaching computers to understand user context in images (visual questioning answering), audio (prediction of acoustic events), and text (image captioning, dialog systems). I use Deep Learning for accurate predictions and Bayesian models for discovering interpretable hypotheses. IBM Research granted me the Innovation Award on Scalable Data Analytics in 2010, which funded my PhD thesis "Real-time learning of rules and prediction of dangerous scenes on traffic surveillance videos". On the news: http://goo.gl/9ptn4U I contributed to the Yahoo's Pig Latin open source project to include metadata (aspect oriented features like privacy, reliability, and time) into the FILTER, JOIN, GROUP, and UNION transformations of the execution engine. Pig Latin is a cloud computing language for massive data analysis based on Hadoop. Specialties: – Deep Learning – Statistical Machine Learning – Recommendation Algorithms – Multimedia Information Retrieval – Image Processing/Computer Vision – Timeseries Analysis