man vs machine: signal detection theory and big data
DESCRIPTION
Humans are much better than computers at detecting unknown patterns in a visual data set, but, like computers we have our shortcomings. We need to optimize inputs for humans so they are fast and efficient in detecting patterns, especially with the coming data tsunami.TRANSCRIPT
![Page 1: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/1.jpg)
Can Skynet Beat Humans in Signal Detection?
Kyle Redinger, Co-Founder @VividCortex
![Page 2: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/2.jpg)
Background
Database Performance Management
Measure ‘All the Things’ in 1-Second Detail
![Page 3: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/3.jpg)
Your Mandate
Manage 300x more data
with 1.5x more people.
![Page 4: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/4.jpg)
Signal Detection Theory
Humans process up to 500GB of data per second.
![Page 5: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/5.jpg)
Find the Outliers
![Page 6: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/6.jpg)
Signal Detection
Easy Hard
![Page 7: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/7.jpg)
Easy Signal Detection
Easy signals are always easy, regardless of set size.
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
200
300
400
500
600
700
800
900
1000
% Correct Response Time
Set Size Set Size
![Page 8: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/8.jpg)
Hard Signal Detection
Hard signals get harderas set size increases.
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
200
300
400
500
600
700
800
900
1000
% Correct Response Time
Set SizeSet Size
![Page 9: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/9.jpg)
Signal Detection in Use
Statistical Process Controls are “Easy”
Upper Control Limit
Lowe Control Limit
![Page 10: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/10.jpg)
Signal Detection in Systems
1-Second CPU over 5 Minutes
What’s the cause?
![Page 11: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/11.jpg)
Looking at Work
In a database, a query is work.
But, I have 300 query classes.
Crap.
![Page 12: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/12.jpg)
Graph Everything
Query Execution Time
CP
U
![Page 13: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/13.jpg)
Reduce DataZoom & Remove the Obvious
1/10th of Data Points
Query Execution Time
CP
U
![Page 14: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/14.jpg)
Man vs MachineMan Machine
![Page 15: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/15.jpg)
Man vs MachineMan Machine
Non-linear increasing CPU Statistically significant
![Page 16: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/16.jpg)
Man vs MachineMan Machine
Non-linear increasing CPU Statistically significant
Pattern!
Not model friendlyPattern!
Pattern!
![Page 17: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/17.jpg)
Shortcomings
Elephantsin the room?
![Page 18: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/18.jpg)
Shortcomings
Humans lose with a
known pattern
![Page 19: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/19.jpg)
Key Thoughts
Our brains do much better with fewer distractors, so eliminate them!
With unknown patterns, humans find signals faster than computers.
![Page 20: Man vs Machine: Signal Detection Theory and Big Data](https://reader034.vdocuments.mx/reader034/viewer/2022052400/5597c67c1a28abc5098b4754/html5/thumbnails/20.jpg)
Thanks
@kyleredinger
http://www.flickr.com/photos/danielproulx/3524826318/http://www.sciencedirect.com/science/article/pii/S0896627301003920
http://www.himandus.net/elefunteria/eday/puzzles/hunt_24.htmlhttp://www.federaljack.com/terminator-robots-a-reality/
Remember: Less is more.