< Back to BLOG /

Machine Learning function as a Service (ML FaaS)

I am going to give a talk in upcoming Software Architecture Conference  in London. You may find the preceding part here. I hope, that this short “teaser” will be useful for you and hope to see you attending my talk in person. If you have any questions on machine learning topics, please raise your hand during the session, drop me an email, or ask me when you see me around!

Serverless functions for Machine Learning

Complexity of server environment and configurations sparked out the need and interest to create an abstraction layer, for “severless deployment”. AWS Lambda became to be one of the early adopters of such “serverless”, architecture. This can be optimal for machine learning models as well, since those are executed in the context of user requests or stream of events.

During my talk, I will present an open-source cluster for serving machine learning models within this kind of microservices architecture. This cluster can spin up machine learning lambdas by just a couple of clicks. Software engineers can then use these lambdas as microservices within of their systems.

Sidecar unit encapsulates underlying layers and details for networking, monitoring, versioning, API contracts etc. It does all the magic with pipelining under that level of abstraction. This encapsulates details for A/B testing, autoscaling, streaming as well as integration with cloud providers, DC/OS and Kubernetes data centers.

My talk will also cover and walk through some of the challenges of this kind of architecture, like a) the processes of decoupling infrastructure logic from machine learning models and b) supporting different machine learning (ML) frameworks, like Scikit-learn, Spark ML, TensorFlow etc..

Stay tuned and hope to see you in London or San Francisco.

Please contact me by email spushkarev@hydrosphere.io before or after the talk. Would be happy to chat and follow up.

Please use discount code Pushkarev20 to save 20%.