Software developer interested in artificial intelligence, data analysis and visualization. I believe the best software is open source. Most comfortable programming in Python, Java and some C++.Most interested in: Artificial Intelligence, Machine Learning, Big Data.View the profile
About the talk
Deep learning models can perform significantly better when trained on very large datasets. Manually labeling large collections of real-world pictures is expensive and may not include outlier scenarios that can be relevant for models expected to work on complex environments. One way to tackle this problem is to use synthetic data which can be generated by simulating relevant scenarios using a game engine. And this can be applied in many industries, not just the gaming industry.
In this talk, Unity’s principal ML engineer will explore recent advances in machine learning and explain the role game engines play in the future of computer vision. He’ll summarize existing tools that can be used to generate perfectly labeled synthetic datasets for different perception tasks to reduce the cost of acquiring the labeled examples needed to train the model. It is still early days, but several research projects published in top conferences show how, in some cases, a model trained on real-world data is outperformed by a model trained on 100% synthetic data at a fraction of the cost. Cesar will provide takeaways to executives in other industries about how they can leverage these findings.
Buy this talk
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.