Deep Learning

Justin Sirignano, University of Illinois at Urbana-Champaign

Usage Details

Blue Waters Trainee 001, Blue Waters Trainee 002, Blue Waters Trainee 003, Blue Waters Trainee 004, Blue Waters Trainee 005, Blue Waters Trainee 006, Blue Waters Trainee 007, Blue Waters Trainee 008, Blue Waters Trainee 009, Blue Waters Trainee 010, Blue Waters Trainee 011, Blue Waters Trainee 012, Blue Waters Trainee 013, Blue Waters Trainee 014, Blue Waters Trainee 015, Blue Waters Trainee 016, Blue Waters Trainee 017, Blue Waters Trainee 018, Blue Waters Trainee 019, Blue Waters Trainee 020, Blue Waters Trainee 021, Blue Waters Trainee 022, Blue Waters Trainee 023, Blue Waters Trainee 024, Blue Waters Trainee 025, Blue Waters Trainee 026, Blue Waters Trainee 027, Blue Waters Trainee 028, Blue Waters Trainee 029, Blue Waters Trainee 030, Blue Waters Trainee 031, Blue Waters Trainee 032, Blue Waters Trainee 033, Blue Waters Trainee 034, Blue Waters Trainee 035, Blue Waters Trainee 036, Blue Waters Trainee 037, Blue Waters Trainee 038, Blue Waters Trainee 039, Blue Waters Trainee 040, Blue Waters Trainee 041, Blue Waters Trainee 042, Blue Waters Trainee 043, Blue Waters Trainee 044, Blue Waters Trainee 045, Blue Waters Trainee 046, Blue Waters Trainee 047, Blue Waters Trainee 048, Blue Waters Trainee 049, Blue Waters Trainee 050, Blue Waters Trainee 051, Blue Waters Trainee 052, Blue Waters Trainee 053, Blue Waters Trainee 054, Blue Waters Trainee 055, Blue Waters Trainee 056, Blue Waters Trainee 057, Blue Waters Trainee 058, Blue Waters Trainee 059, Blue Waters Trainee 060, Blue Waters Trainee 061, Blue Waters Trainee 062, Blue Waters Trainee 063, Blue Waters Trainee 064, Blue Waters Trainee 065, Blue Waters Instructor 001, Blue Waters Instructor 002, Blue Waters Instructor 003, Justin Sirignano

The course “Deep Learning” will provide an introduction to neural networks and deep learning. Fundamental topics such as convolution neural networks, recurrent neural net- works, and deep reinforcement learning will be covered. The course will use TensorFlow to train deep learning models on GPUs. Recent exciting advances in deep learning will be discussed and students will implement a variety of deep learning models via homeworks and a substantial project. Deep learning has produced state-of-the-art results in a number of areas such as image classification, natural language processing, finance, and reinforcement learning.

Projects and homework assignments will include re-producing some of these results. For example, students will be asked to classify images of different types of objects/items from the raw pixels. In typical applications, neural networks may have millions of parameters and the datasets can be large. Training is consequently very computationally expensive. Fortunately, neural network training can be highly parallelized via GPUs. Without a GPU, training can be impractical for large neural networks. The allocation on Blue Waters will allow students to implement neural networks for important applications and achieve near state-of-the-art results on these applications.

The students taking the course will be Masters and PhD level. Students are expected to come from a diverse set of departments at UIUC, including industrial and systems engineering, mechanical engineering, electrical engineering, computer science, and statistics. They will have programming experience, although it is expected that it will be necessary to introduce them to the Blue Waters system and how to use it.