Realtime ML + Kubernetes + TensorFlow + KubeFlow + MLflow + PipelineAI + PyTorch




Realtime ML + Kubernetes + TensorFlow + KubeFlow + MLflow + PipelineAI + PyTorch

15 April 2019

New York

Added 29-Mar-2019

Talk 1: "Slack After Dark": Realtime ML + Kubernetes + TensorFlow + MLflow + Slack API + PipelineAI (Antje Barth, Developer Advocate, MapR)

Welcome to “Slack after Dark” - our slack-based dating app which showcases an end-to-end containerized & integrated ML workflow, running Online Model Predictions and Online Model Training using Keras/Tensorflow, PipelineAI, MapR Data Platform and Slack!

In this talk and live presentation you’ll also see how a streaming architecture can help you with moving from batch to real-time model training, and simplifying your overall model data logistics.

Talk 2: Real-Time, Continuous ML/AI Model Training, Optimizing, and Predicting with Kubernetes, Kafka, TensorFlow, KubeFlow, MLflow, Keras, Spark ML, PyTorch, Scikit-Learn, and GPUs (Chris Fregly, Founder @ PipelineAI)

Chris Fregly, Founder @ PipelineAI, will walk you through a real-world, complete end-to-end Pipeline-optimization example.

Through a series of live demos, Chris will install, create, and deploy a model ensemble using the PipelineAI Platform with GPUs, TensorFlow, and Scikit-Learn.

While most Hyper-parameter Optimizers stop at the training phase (ie. learning rate, tree depth, ec2 instance type, etc), we extend model validation and tuning into a new post-training optimization phase including 8-bit reduced precision weight quantization and neural network layer fusing - among many other framework and hardware-specific optimizations.

Next, we introduce hyper-parameters at the prediction phase including request-batch sizing and chipset (CPU v. GPU v. TPU). We’ll continuously learn from all phases of our pipeline - including the prediction phase. And we’ll update our model in real-time using data from a Kafka stream.

Lastly, we determine a PipelineAI Efficiency Score of our overall Pipeline including Cost, Accuracy, and Time. We show techniques to maximize this PipelineAI Efficiency Score using our massive PipelineDB along with the Pipeline-wide hyper-parameter tuning techniques mentioned in this talk.