In this session, Reza Zadeh, CEO of Matroid, presents a Kubernetes deployment on Amazon Web Services that provides customized computer vision to a large number of users. Reza offers an overview of Matroid’s pipeline and demonstrates how to customize computer vision neural network models in the browser, followed by building, training, and visualizing TensorFlow models, which are provided at scale to monitor video streams.
4. Kubernetes (K8s)
Basic unit: Pod
Pods contain one or more containers
Containers in a pod share one IP address
Pods are scalable & fault tolerant by K8s
10. Convolutional Neural Network
Slide a two-dimensional patch over pixels.
Deep Learning frameworks all support this.
Figure: Google image search for “convolutional neural network”
11. TensorFlow
Deep Learning framework from Google Brain
Happy coincidence: A Matroid is a generalization of a Tensor
Matroid incorporated Oct 2015
TensorFlow released Nov 2015
12. GPU Instances (e.g. p2 & p3)
Can ask Kubernetes to schedule a pod on a machine
with dedicated GPU as of v1.5
Multi-GPU machine support added in v1.6
Support for other co-processors coming
13. Amazon Spot Marketplace
One pod is always watching out for workers that are going
to be evicted because of Spot Instance marketplace
Brings them up as reserved instance if they are needed to
handle load
14. Matroid Product
Product is a studio for creating and using detectors
Like a metal detector, a matroid detector finds things
in media, watches streams for them