James Bergstra

Career

Bergstra completed a master's degree in 2006 under Douglas Eck, on algorithms for classifying recorded music by genre, and doctoral studies at the University of Montreal in July 2011 under Yoshua Bengio, with a dissertation on incorporating complex cells into deep learning models. During graduate work, Bergstra co-developed Theano, an open-source optimizing compiler that can use GPUs for high-performance computation.²¹

Bergstra then spent a year as a Research Scholar in David Cox's computer and biological vision group at Harvard's Rowland Institute, and moved on to a postdoctoral position under Chris Eliasmith in the Centre for Theoretical Neuroscience at the University of Waterloo.¹²

Research

Much of Bergstra's research concerns hyperparameter optimization. "Random Search for Hyper-Parameter Optimization" (2012), with Yoshua Bengio, showed empirically and theoretically that randomly chosen trials are more efficient than a grid. "Algorithms for Hyper-Parameter Optimization" (NIPS 2011), with Rémi Bardenet, Bengio and Balázs Kégl, introduced sequential methods based on the expected improvement criterion for configuring neural networks and deep belief networks.⁵⁶

Bergstra is first author of the 2015 journal paper describing the Hyperopt Python library, whose license names Bergstra as copyright holder, and co-authored Hyperopt-Sklearn for automatic configuration of scikit-learn models. At Waterloo, Bergstra also co-authored Nengo, a Python tool for building large-scale functional brain models.²⁴

Work and contributions

• Hyperopt Open-source code

  First author of "Hyperopt: a Python library for model selection and hyperparameter optimization" (2015), as listed on the University of Waterloo profile; the project's BSD-style license names James Bergstra as copyright holder.²⁴

• Theano Open-source code

  Co-developed Theano, according to the University of Waterloo profile, and co-author of "Theano: new features and speed improvements" (2012).²³⁷

Timeline

1. 2006Completed a master's degree under Douglas Eck on classifying recorded music by genre.²
2. 2011Co-authored "Algorithms for Hyper-Parameter Optimization" at NIPS 2011.⁶
3. Jul 2011Completed doctoral studies at the University of Montreal under Yoshua Bengio.²
4. 2012Published "Random Search for Hyper-Parameter Optimization" with Yoshua Bengio in JMLR.⁵
5. Jul 2012Gave a UBC talk on hyperparameter optimization as a Research Scholar at Harvard's Rowland Institute.¹
6. Nov 2012Co-authored "Theano: new features and speed improvements".³
7. 2014Co-authored Nengo, a Python tool for building large-scale functional brain models.²
8. 2015Published the Hyperopt library paper as first author in Computational Science & Discovery.²

Documented roles

The sources do not confirm a current role; these are the roles they document.

• Research Scholar, David Cox's biological and computer vision group at the Rowland Institute, Harvard University¹²
• Postdoc, Centre for Theoretical Neuroscience, University of Waterloo²

Companies in their history

• Harvard University 8 people in the wiki

Related people

• David BauConnected to Harvard University
• Hendrik StrobeltConnected to Harvard University
• Sasha RushConnected to Harvard University
• Stephen CasperConnected to Harvard University
• Azeem AzharConnected to Harvard University
• Piero MolinoML frameworks and libraries · Developer tools
• Matthew WatsonML frameworks and libraries · Developer tools
• Nikita ShulgaML frameworks and libraries · Developer tools
• William FalconML frameworks and libraries · Developer tools
• Thomas J. FanML frameworks and libraries · Developer tools