insights, predictions, and actions - test science · 2018-03-20 · insights, predictions, and...

Insights, Predict ions, and Act ions

Descript ive Definit ions of Data Science, Machine Learning, and Art ificial Intelligence

Andrew Flack

The terms “Data Science,” “Machine Learning,” and “Art ificial Intelligence” are increasingly common.

1

DoD must stay on the cut t ing edge in technology, culture, operat ions, and processes.

2

Defense Innovation Advisory Board

Recommendat ion 5: Catalyze Innovat ions in Art if icial Intel l igence and Machine Learning

“Textbook” definit ions are not terribly helpful.

4http:/ / fortune.com/ ai-artificial- intelligence-deep-machine-learning/

Textbook definit ions are not terribly helpful.

5https:/ / steemit.com/ education/ @agmakosz/ the-approaching-impact-of-a-i-on-the-education-industry

Is it just market ing?

6https:/ / twitter.com/ xaprb/ status/ 930674776317849600

Forget the formal definit ions and hierarchies.

7

What does a practitioner in each discipline actually do?

Data Science Produces InsightsMachine Learning Makes Predict ions

Art ificial Intelligence Produces Act ions

8

Data science is the process of obtaining, t ransforming, analyzing, and communicat ing data to answer a quest ion.

Data Question

Raw Data

Tidy Data

Data Analysis

• Descriptive, exploratory, predictive, inferential, causal, or mechanistic

Data Product

• Business question / analytical question

• Collecting, scraping

• Cleansing, preparation

• Report, visualization, model, tool, other decision aid

Who is a data scient ist?

10

Head of IT Analyst

The Num be rCrunche r

The Sale spe rson

Com pute r Scie nce

Profe ssor

Sta tis tics Profe ssor

Hot Air GoodConsultant

The Data Ne rd

The Hacke r

R Core Te am

The IT Guy

The Accountant

The Data

Scie ntis t

Com m unication

Statis tics Program m ing

Busine ss

Who is a data scient ist?

Isn’t this just an Operat ions Research / Systems Analyst (ORSA)? Not quite …

12

No Data Some Data “Big Data”

Op e ra tions Re se a rch

Da ta Sc ie nce

“Data Science” in Defense

13

Is this friendly activity (cordon and search, raid, road block, etc.) reducing enemy activity?

Compare the number of enemy events, before and after a friendly activity, to see if there is a significant difference.

Data Science in Defense

14

Data Science in the wild

15

Who are the most profitable customers?

Is there really a difference between profitable customers and the average customer?

Can I characterize customers in some way?

Data Science Produces InsightsMachine Learning Makes Predict ions

Art ificial Intelligence Produces Act ions

16

Discover a function 𝒚𝒚=𝒇𝒇(𝒙𝒙) that be st p re d ic ts the value of 𝒚𝒚associate d with e ach value of 𝒙𝒙.

17

If the response variable is numeric, we’re doing regression.*

18

* In this context, we’re using regression to PREDICT, not to EXPLAIN.

See Shmueli, G. 2010. “To Explain or to Predict?,” Statistical Science (25:3), pp. 289-310.

If the response is categorical, it is a classificat ion task.

19

Petal

Sepal

Iris Setosa Iris Versicolor

Find a funct ion that maps each at t ribute set to one of the predefined class labels.

20

Find “Learn” a funct ion that maps each at t ribute set to one of the predefined class labels …

21

… and generalizes to make accurate predict ions on previously unseen data.

22

Separat ion between these disciplines is blurry.

24

DataScience

MachineLearning

ArtificialIntelligence

LogisticRegression

Machine Learning in Defense

25

Machine Learning in Defense

26

“… augment or automate Processing, Exploitation, and Dissemination (PED) for tactical Unmanned Aerial System

(UAS) and Mid-Altitude Full-Motion Video (FMV) …”

“38 classes of objects the department needs to detect”

Machine Learning in the wild

27

Data Science Produces InsightsMachine Learning Makes Predict ions

Art ificial Intelligence Produces Act ions

28

Rather than teaching computers how to do everything, teach them to learn for themselves.

29

Arthur Samuel, a pioneer in Artificial Intelligence, created a program to play checkers.

He was the first to develop a machine that surpassed his ability in a task that he taught it.

Again, separat ion between these disciplines is blurry.

30

DeepLearning

DataScience

MachineLearning

ArtificialIntelligence

Art ificial Intelligence in the wild

31

Data Science Produces InsightsMachine Learning Makes Predict ions

Art ificial Intelligence Produces Act ions

32