Babak Barazandeh (R), Winner of Best Paper at the IEEE Data Science Workshop, with DSW organizer Antonio García Marques.
PhD student in the Daniel J. Epstein Department of Industrial and Systems Engineering, Babak Barazandeh has been awarded Best Paper at the recent Institute of Electrical and Electronics (IEEE) Data Science Workshop (DSW 2019).
Held in Minneapolis, Minnesota from June 2-5, DSW 2019 featured leading researchers from academia and industry working on data science problems.
Barazandeh works within Assistant Professor Meisam Razaviyayn’s Optimization for Data Driven Science (ODDS) research group, with a research interest in developing algorithms for non-convex learning problems such as mixture models or Generative Adversarial Networks (GANs). He obtained his MS from Virginia Tech in 2017 and BS from Sharif University of Technology in 2014 in Statistics and Electrical Engineering, respectively.
Barazandeh’s winning paper, titled Training generative networks using random discriminators, was written with Razaviyayn and Maziar Sanjabi, a research scientist at Electronic Arts.
Barazandeh’s paper deals with Generative Adversarial Networks (GANs), which have significantly advanced the performance of many machine learning and artificial intelligence tasks in recent years. These networks have applications in improving the quality of medical or astronomical images, creating artificial paintings or music, developing virtual commentators/news anchors on sport events, or developing defense mechanisms against adversarial and cyber-attacks to neural networks.
However training these models is notoriously difficult and requires many computational resources and tuning a wide range of hyper-parameters. Barazandeh’s winning paper proposes a new model that relies on the use of random neural networks as discriminators.
The paper is available here.