Machine learning and AI in the datacenter and how it will affect you Nick Chase Head of Content, Mirantis Editor in Chief, Open Cloud Digest Author, Machine Learning for Mere Mortals
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Machine learning and AI in the datacenter and how it will affect you
Nick ChaseHead of Content, MirantisEditor in Chief, Open Cloud DigestAuthor, Machine Learning for Mere Mortals
Who am I?
Nick Chase• Long-time programmer• Head of Technical and Marketing Content for
Mirantis• Editor in Chief of Open Cloud Digest• Author of Machine Learning for Mere Mortals• Absolutely NOT a math major• nchase at mirantis dot com
2
Shhhhhh.....
Turns out, this is not as tough as you think.
Special Bonus
• Free early access to Machine Learning for Mere Mortals for live attendees
http://bit.ly/machine_learning_course
4
Agenda
• What is machine learning?• Types of machine learning• Problems facing the datacenter and how ML can
solve them • You, yes, YOU, can do this (and I'll prove it)• Intelligent Delivery datacenter and how to start• Q&A
5
Introduction Just what is machine learning,
anyway?
6
What is machine learning?
Machine learning is the study of enabling a computer to perform in situations it may not have
seen before.
7
Where can we see ML today?
• Analytics• Advertising• Bots• Pattern recognition• NLP• Search
8
Where can we see ML today?
• Analytics• Advertising• Bots• Pattern recognition• NLP• Search
9
• Image recognition • Video analysis• Threat detection• Medicine• Self driving cars
Lots of hype
• Existing companies http://indeedhi.re/2nFRKY0
10
● Amazon.com (1252)● JPMorgan Chase (733)● Goldman Sachs (674)● Microsoft (580)● Apple (322)● Google (217)● Facebook (217)● KPMG (148)● Jobspring Partners (124)● Capital One (118)● NVIDIA (116)● Strategic IT Staffing (111)● Oracle (110)● Lockheed Martin (106)● IBM (101)
Lots of hype
• New companies (http://bit.ly/2E2qFoU)• Newly accessible
• Tensorflow• SciKit-Learn• etc.
11
Machine learning vs artificial intelligence
12
What does ML have to do with "cloud native"?
• Dynamic environments
13
What does ML have to do with "cloud native"?
• Dynamic environments
• High-velocity management
14
What does ML have to do with "cloud native"?
• Dynamic environments
• High-velocity management
• Continuous automation
15
What does ML have to do with "cloud native"?
• Dynamic environments
• High-velocity management
• Continuous automation
• Observability
16
What does ML have to do with "cloud native"?
• Dynamic environments
• High-velocity management
• Continuous automation
• Observability
• Efficiency
17
What can ML do for the datacenter?
Continuous Delivery infrastructure
18
What can ML do for the datacenter?
Continuous Delivery infrastructure
Intelligent Delivery infrastructure
19
Intelligent Delivery infrastructure
• Defined architecture
20
Intelligent Delivery infrastructure
• Defined architecture
• Flexible but controllable infrastructure
21
Intelligent Delivery infrastructure
• Defined architecture
• Flexible but controllable infrastructure
• Intelligent oversight
22
Intelligent Delivery infrastructure
• Defined architecture
• Flexible but controllable infrastructure
• Intelligent oversight
• Secure footing
23
Types of Machine Learning
24
Supervised learning
Training a computer to understand data it hasn't seen by teaching it about data
that has already been labeled
25
Supervised learning: Examples
• Image recognition
26
Supervised learning: Examples
• Image recognition• Speech recognition
27
Supervised learning: Examples
• Image recognition• Speech recognition• Medical outcome prediction
28
Supervised learning: Examples
• Image recognition• Speech recognition• Medical outcome prediction• Spam filtering
29
Supervised learning: Examples
• Image recognition• Speech recognition• Medical outcome prediction• Spam filtering• Stock market/crypto pricing
30
Unsupervised learning
Letting the computer discover patterns and relationships in unlabeled data.
31
Unsupervised learning: example
• Clustering
32
Unsupervised learning: example
• Clustering • Customers into categories
33
Unsupervised learning: example
• Clustering • Customers into categories• Chemical compounds
34
Unsupervised learning: example
• Clustering • Customers into categories• Chemical compounds
• Anomaly detection
35
Unsupervised learning: example
• Clustering • Customers into categories• Chemical compounds
• Anomaly detection• Signal separation
36
Reinforcement learning
A combination of supervised and unsupervised learning, where the computer is given an objective and attempts to reach that objective, discovering
tactics as it goes along.
37
Reinforcement learning: example
• Just about anything humans do
38
Reinforcement learning: example
• Just about anything humans do• Self-driving cars
39
Reinforcement learning: example
• Just about anything humans do• Self-driving cars• Strategy development
40
Reinforcement learning: example
• Just about anything humans do• Self-driving cars• Strategy development
• Games
41
Reinforcement learning: example
• Just about anything humans do• Self-driving cars• Strategy development
• Games• Scheduling
42
Problems Facing the Datacenter
43
Configuration: potential issues
• Self-service
44
Configuration: potential issues
• Self-service • Microservices architectures
45
Configuration: potential issues
• Self-service• Microservices architectures• Dynamic environments and drift issues
46
Configuration: how ML/AI can help
• App configuration based on past history
47
Configuration: how ML/AI can help
• App configuration based on past history• Reinforcement learning to retroactively improve
configuration
48
Performance optimization: Potential issues
• Lots of factors in play
49
Performance optimization: Potential issues
• Lots of factors• Dynamic environments create a changing
landscape
50
Performance optimization: how ML/AI can help
• Analyze multiple factors
51
Performance optimization: how ML/AI can help
• Analyze multiple factors• Predict future load and move or scale workloads
based on patterns
52
Performance optimization: how ML/AI can help
• Analyze multiple factors• Predict future load and move or scale workloads
based on patterns• Move workloads to closer geographical resources
based on patterns
53
Cost optimization: potential issues
• Complex environments
54
Cost optimization: potential issues
• Complex environments• Changing costs
55
Cost optimization: potential issues
• Complex environments• Changing costs• Soaring storage requirements
56
Cost optimization: how AI/ML can help
• Keep track of multiple environments and costs
57
Cost optimization: how AI/ML can help
• Keep track of multiple environments and costs• Adapt to changing conditions
58
Cost optimization: how AI/ML can help
• Keep track of multiple environments and costs• Adapt to changing conditions• Optimize what data is kept
59
Fault detection: potential issues
• Hardware failure
60
Fault detection: potential issues
• Hardware failure• Software configuration drift
61
Fault detection: potential issues
• Hardware failure• Software configuration drift• Data corruption
62
Fault detection: potential issues
• Hardware failure• Software configuration drift• Data corruption
No, I am not making that up: http://bit.ly/cosmic-corruption
63
Fault detection: potential issues
• Hardware failure• Software configuration drift• Data corruption• Resource overrun
64
Fault detection: how AI/ML can help
• Autoscaling
65
Fault detection: how AI/ML can help
• Autoscaling• Pattern recognition
66
Fault detection: how AI/ML can help
• Autoscaling• Pattern recognition• Anomaly detection
67
Security: potential issues
• DDOS
68
Security: potential issues
• Distributed Denial Of Service attacks• Advanced Persistent Threats
69
Security: potential issues
• DDOS• APT• Inside jobs
70
Security: how ML/AI can help
• Anomaly detection to find APTs and inside jobs
71
Security: how ML/AI can help
• Anomaly detection to find APTs and inside jobs• Pattern recognition to predict DDOS attacks
before they're in full swing
72
Try It Yourself
73
Once upon a time...
74
Once upon a time...
• Self-configuring
75
Once upon a time...
• Self-configuring• Self-optimizing
76
Once upon a time...
• Self-configuring• Self-optimizing• Self-healing
77
Once upon a time...
• Self-configuring• Self-optimizing• Self-healing• Self-protecting
78
Once upon a time...
79
What are we trying to do?
80
What are we trying to do?
81
Defining the error
•
82
Defining the error
•
83
Defining the error
•
84
y
Defining the error
•
85
y
ypredicted
Defining the error
•
86
y
ypredicted
abs(y - ypredicted)
Defining the error
•
87
errortotal = sum(abs(y - ypredicted))
Defining the error
•
88
errortotal = sum(abs(y - (mx + b))
Hot and cold
89
Hot and cold
90
Hot and cold
91
Hot and cold
92
Hot and cold
93
Hot and cold
94
Hot and cold
95
errortotal = sum(abs(y - (mx + b))
Gradient descent
An iterative algorithm to find the minimum of the cost/error function and that parameters that
achieve that minimum.
96
Putting it together
• Data• Python3• Tensorflow• Matplotlib (optional)• Ubuntu 16.04
97
What we'll do
• Read the data• Create the model• Define the error• Define the optimizer• Repeatedly run the optimizer until we get as close
as we can
98
DEMO
99
The actual answer
y = .37x + 12m = .37b = 12
100
DEMO
101
Where do we go from here?
102
More practically
• Don't expect to go from B&W to VR in one step• Smaller steps in between
103
More practically
• Don't expect to go from B&W to VR in one step• Smaller steps in between
• Start with analytics
104
More practically
• Don't expect to go from B&W to VR in one step• Smaller steps in between
• Start with analytics• Ultimately, know what you're trying to achieve
105
Q&A
106
Thank You
107
Related Documents
