The past few years have seen breakthroughs in computer vision: human performance can now be attained on certain tasks thanks to convolutional neural networks, at the cost of greater volumes of labeled training data. One can argue that the bottleneck in computer vision is no longer in engineering visual recognition algorithms but in creating sufficiently large, labeled training sets for each task. When labels are expensive to obtain, this can render the task infeasible to solve.
Alexandre Dalyac and Robert Hogan address this issue through a combination of dimensionality reduction, information retrieval, and domain adaptation techniques packaged in a software product that acts as a human-algorithm interface to facilitate transfer of expertise from human to machine. Alexandre and Robert present the results of their solution, which currently show that the labeling burden can be reduced by up to two orders of magnitude, and take a deeper dive into domain adaptation.
Alex Dalyac is cofounder and CEO of Tractable. Previously, he worked as a quantitative researcher at London hedge fund Toscafund. Alex holds an MS in computer science from Imperial College London and was a recipient of the Philips Prize in computer science.
Robert Hogan is a data scientist at Tractable. Robert holds a PhD in theoretical particle physics and cosmology from King’s College London and an MSc in quantum fields and fundamental forces from Imperial College London and has been the recipient of numerous awards, including the Institute of Physics Earnshaw Medal 2011.
©2016, O’Reilly UK Ltd • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org
Apache Hadoop, Hadoop, Apache Spark, Spark, and Apache are either registered trademarks or trademarks of the Apache Software Foundation in the United States and/or other countries, and are used with permission. The Apache Software Foundation has no affiliation with and does not endorse, or review the materials provided at this event, which is managed by O'Reilly Media and/or Cloudera.