PipelineAI Continuous Machine Learning and AI - Rework Deep Learning Summit - San Francisco, CA - Jan 25, 2019 -

Traditional machine learning pipelines end with life-less models sitting on disk in the research lab. These traditional models are typically trained on stale, offline, historical batch data. Static models and stale data are not sufficient to power today's modern, AI-first Enterprises that require continuous model training, continuous model optimizations, and lightning-fast model experiments directly in production. Through a series of open source, hands-on demos and exercises, we will use PipelineAI to breathe life into these models using 4 new techniques that we’ve pioneered:
* Continuous Validation (V)
* Continuous Optimizing (O)
* Continuous Training (T)
* Continuous Explainability (E).

The Continuous "VOTE" techniques has proven to maximize pipeline efficiency, minimize pipeline costs, and increase pipeline insight at every stage from continuous model training (offline) to live model serving (online.)
Attendees will learn to create continuous machine learning pipelines in production with PipelineAI, TensorFlow, and Kafka.

1. “Halliburton chooses PipelineAI to power its Oil & Gas Vertical Cloud”
(LIFE Conference Keynote 2018)
“PipelineAI is…

Uber Michelangelo for
AI-First Enterprises.”
“PipelineAI is…

AWS SageMaker for
Industry Vertical
Clouds.”
Chris Fregly
Founder @ PipelineAI
[email protected]
Deep Learning Summit
San Francisco, CA
Jan 25, 2019

2. Problem
2
It’s Hard to Balance the 3 “Cy’s” of AI
Privacy
Accuracy Latency
Solution: Experiment in Live Production to Find the Right Balance

3. Current Solution: Cloud Lock-In
3
https://aws.amazon.com/blogs/machine-learning/automated-and-continuous-deployment-of-amazon-sagemaker-models-with-aws-step-functions/ (Dec 2018)

4. PipelineAI Solution: 1-Click & Multi-Cloud
x11 Generated Models1 Original Model x3 Clouds
4
Arbitrage cost savings
across
all public cloud providers
Find best performing model
among all generated models

5. Mission & Value Proposition
5x smaller and 3x faster models
Easy integration with Enterprise systems
Auto-tune accuracy vs. latency vs. privacy vs. cost
Safely explore new models in seconds vs. months
Unified runtime across language, framework & cloud
5
The Premium Enterprise AI Runtime

6. Perform Online Predictions using Slack
A/B and multi-armed bandit model compare

Train Online Models with Kafka Streams
Create new models quickly
Deploy to production safely
Mirror traffic to validate online performance
PipelineAI: Real-Time Machine Learning

7. Advantages of PipelineAI
Any Framework, Any Hardware, Any Cloud
Dashboard to manage the lifecycle of models
from local development to live production
Generates optimized runtimes for the models
Custom targeting rules, shadow mode, and
percentage-based rollouts to safely test features
in live production
Continuous model training, model validation, and
pipeline optimization

8. Market Validation
8
Existing AI Industry Vertical Clouds
GE Edison
Salesforce Einstein

PipelineAI-based Vertical Clouds
Halliburton Open Earth Cloud
Huawei Cloud
Large Travel Enterprise
Large Electronics Manufacturer
Consumer Product Group (CPG) Analytics

9. DEMO









https://joinslack.pipeline.ai - join the #demo channel
/predict cat vs.
dog

10. Slack - Predict with Image
Cat?
Dog?
/predict
https://images.ctfassets.net/kvimhx6nhg7h/5WclEHFxUksuS2IwsUE
CE6/a29fa96920666f9d4eb7c456403e4f9d/Tan-cat-in-a-cone.png
Model Variant
Confidence of Each Prediction
Possible Predictions
REQUEST
RESPONSE

11. COMPOSE/
ENSEMBLE
Architecture for Online Prediction
/predict <img>
Archive
Model 3
(Canary)
Model 1
Model 2
INPUT
ARCHIVE
RESPONSE
REQUEST
Select prediction with highest
confidence (via customizable
Objective Function)
Replay for future use
Compare Canary to live
Model 1 and Model 2
Mirrored Traffic
Live Traffic
Traffic
Routing
/predict: Pass an image URL to classify (cat or dog) via model prediction REST API
/predict_archive

12. Validate new model performance

13. Online Model Training with Streams
/label <img> <label>
Training Stream
Distributed
Filesystem
Deploy model
Model 3
(Canary)
Train model
Model 1
Model 2
/label: Add new training data (human feedback loop) to improve the model
/train: Create a new model with the latest training data
/deploy: Deploy the model as a Canary alongside live models
/route: Mirror the live traffic to Canary to validate model performance
/label_data

14. Slack - Train Model
/label
https://images.ctfassets.net/kvimhx6nhg7h/5WclEHFxUksuS2IwsUE
CE6/a29fa96920666f9d4eb7c456403e4f9d/Tan-cat-in-a-cone.png
cat

15. Slack API: Outbound Webhook to PipelineAI REST API

16. WORKSHOP
https://community.pipeline.ai - Notebooks => 00_Explore_Environment

17. Thank You!
17
Privacy
Accuracy Latency
Contact me:
[email protected]
https://community.pipeline.ai