Biography

Michael Shtelma is a software engineer passionate about all data related topics, especially data engineering and data science. He loves coding in Scala and Python. Currently, Michael is working at Databricks in Frankfurt, Germany.

Talk

Deep Learning at Scale: Distributed Training and Hyperparameter Search for Image Recognition Problems

Training complex image recognition model on a large dataset using one machine can be long and cumbersome. This talk focuses on methods and libraries, which allow us to train models on a dataset that does not fit into memory, or maybe even on the disk using multiple GPUs or even nodes. The ways of using multiple GPUs and nodes will be discussed and tradeoffs between different approaches will be compared.
This talk includes a live demonstration of distributed training of image recognition model on large data set using Horovod and Petastorm on Databricks platform.

Session Keywords