CNTK: fast deep learning for research and production
29th March 11:00 AM
Speaker: William Darling, applied scientist at Microsoft, Munich, Germany
Abstract: The last several years have seen an explosion in deep learning research and associated breakthrough results. A large part of this growth has been supported and enabled by the availability of DL toolkits that lowered barriers of entry for researchers and practitioners. DL toolkits allow users to concentrate on how to author neural networks, while the toolkit worries about how to train and execute them efficiently. Despite their widespread availability, however, there often still exists a meaningful gap between creating a DL model for a research task and scaling it up and ultimately serving it. In this talk I will present the Microsoft Cognitive Toolkit (CNTK) and how it goes beyond competing DL software with implicit handling of time, automatic mini-batching of variable length sequences, and fast data-parallel training. CNTK might be the tool that your workflow has been missing to bring your research to production.