Description
This talk will provide an overview of the key challenges and trends in the productization of machine learning systems, including concepts such as reproducibility, explainability and orchestration. The talk will also provide a high level overview of several key open source tools and frameworks available to tackle these issues, which have been identifyed putting together the Awesome Machine Learning Operations list (https://github.com/EthicalML/awesome-machine-learning-operations).
The reproducibility piece will cover key motivations as well as practical requirements for model versioning, together with tools that allow data scientists to achieve version control of model+config+data to ensure full model lineage.
The explainability piece will contain a high level overview of why this has become an important topic in machine learning, including the high profile incidents that tech companies have experienced where undesired biases have slipped into data. This will also include a high level overview of some of the tools available.
Finally, the orchestration piece will cover some of the fundamental challenges with large scale serving of models, together with some of the key tools that are available to ensure this challenge can be tackled.