From exploration to deployment - combining PyTorch and TensorFlow for Deep Learning

PyTorch vs. TensorFlow? Why not PyTorch + TensorFlow? Lets's combine deep learning frameworks and use their individual strengths for explorative and production-oriented tasks along a data science project.

Tags: Artificial Intelligence, Deep Learning & Artificial Intelligence, Data Science, Machine Learning

Scheduled on friday 14:00 in room cubus


Marcel Kurovski

Marcel Kurovski is a Data Scientist at inovex, a German IT project house focusing on digital transformation. He earned a master's degree in Industrial Engineering and Management from the Karlsruhe Institute of Technology (KIT) where he focused on computer science, machine learning and operations research.

Marcel works on novel methods to exploit deep learning for recommender systems in order to better personalize content and improve user experience. He works for clients in e-commerce and retail where he bridges the gap between proof-of-concept and scalable AI systems.

Marcel develops on the Python data science stack and also contributes to TensorFlow. His research spans recommender systems, deep learning as well as methods for approximate nearest neighbor search.


Despite the many deep learning frameworks out in the wild few have achieved widespread adoption. Two of them are TensorFlow and PyTorch. Where PyTorch relies on a dynamic computation graph TensorFlow goes for a static graph. Where TensorFlow shows greater adoption and additional useful extensions with TensorFlow Serving and TensorBoard, Pytorch proves useful trough its easy and more pythonic API.

Data scientists are confronted with explorative challenges, but also need to be aware of model deployment and production. Do we need to single out frameworks until we end up with the only one or is there a case for joint usage of two deep learning frameworks? Can we leverage the strengths of the frameworks for different tasks along the path from exploration to production?

In my talk, I want to present a case combining the benefits of PyTorch and TensorFlow using the first for explorative and latter for deployment tasks. Therefore, I will choose a common deep learning challenge and discuss the strengths and weaknesses of both frameworks along a demo that brings a model from development into production.