In the last year, there have been a number of attempts to train deep CNNs on the ImageNet dataset in the shortest time possible. (The most recent attempt managed to do it in 15 minutes.) But all of these attempts happen on custom clusters, which are out of the reach of most data scientists.
One of the key advantages of the cloud is being able to scale out compute resources as required. Mathew Salvaris, Miguel Gonzalez-Fierro, and Ilia Karmanov offer a comparison of two platforms for running distributed deep learning training in the cloud. Both utilize Docker containers, making it possible to run any deep learning framework on them. You’ll examine the performance of each as the number of nodes scales and learn some tips and tricks as well as some pitfalls to watch out for.
The first is a service called Batch AI, which uses the Azure Batch infrastructure to easily run deep learning jobs at scale across GPUs. The second is an open source toolkit that allows data scientists to spin up clusters in a turnkey fashion. It utilizes Kubernetes and Grafana for easy job scheduling and monitoring. This solution has been used in daily production for Microsoft internal groups. Mathew, Miguel, and Ilia use these training platforms to train a ResNet network on the ImageNet dataset using each of the following frameworks: CNTK, TensorFlow (Horovod), PyTorch, MxNet, and Chainer. They then compare and contrast the performance. The examples presented can also be used as templates for your own deep learning problems.
Mathew Salvaris is a data scientist at Microsoft. Previously, Mathew was a data scientist for a small startup that provided analytics for fund managers and a postdoctoral researcher at UCL’s Institute of Cognitive Neuroscience, where he worked with Patrick Haggard in the area of volition and free will, devising models to decode human decisions in real time from the motor cortex using electroencephalography (EEG). He also held a postdoctoral position in the University of Essex’s Brain Computer Interface Group, where he worked on BCIs for computer mouse control. Mathew holds a PhD in brain computer interfaces and an MSc in distributed artificial intelligence.
Miguel González-Fierro is a senior data scientist at Microsoft UK, where he helps customers leverage their processes using big data and machine learning. Previously, he was CEO and founder of Samsamia Technologies, a company that created a visual search engine for fashion items, allowing users to find products using images instead of words, and founder of the Robotics Society of Universidad Carlos III, which developed different projects related to UAVs, mobile robots, small humanoids competitions, and 3D printers. Miguel also worked as a robotics scientist at Universidad Carlos III of Madrid and King’s College London, where his research focused on learning from demonstration, reinforcement learning, computer vision, and dynamic control of humanoid robots. He holds a BSc and MSc in electrical engineering and an MSc and PhD in robotics.
Ilia Karmanov is a data scientist working on applying machine learning and deep learning solutions in industry. He is particularly interested in statistical theory behind deep learning. Ilia holds an MSc in economics from the London School of Economics.
Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?
Join the conversation here (requires login)
©2018, 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. • email@example.com