Designing Curriculum for A.I.

Costas J. Spanosfor David Culler

Costas J. Spanos

Andrew S. Grove Distinguished Professor, EECS, UC Berkeley

Director, CITRIS & the Banatao Institute

CTO, Berkeley Education Alliance for Research in Singapore

5/29/2017

5/29/17 NTU 1

Deep Learning

5/29/17 NTU 2

Inferential Thinkingor, estimating from incomplete data

• Generalizations (predictions, parameter

estimates, and conclusions) that go beyond

describing the given data;

• Use of data as evidence for those generalizations;

• Conclusions that express a degree of uncertainty,

accounting for the variability or uncertainty

inherent in generalizing beyond the immediate

data to a population or a process.

5/29/17 NTU 3

Sampling and Estimation

Sampling: the act of making inferences aboutpopulations.

Random sampling: when each observation isidentically and independently distributed.

Statistic: a function of sample data containing nounknowns. (e.g. average, median, standarddeviation, etc.)

A statistic is a random variable. Its distribution is asampling distribution.

5/29/17 NTU 4

InferenceBasic Statistics

• Mean from data

• Variance from data

• Assume distribution is

normal

• Worry about estimation

precision and sample size

implications.

Real Life Generalization

5/29/17 NTU 5

Data and more Data…

6/1/17 Inclusive

Innovation

6

The AI (R)evolution: from expert to data driven

5/29/17 NTU 7

Data Science at the Heart of a

21st Century University

David Culler

University of California, Berkeley

http://data.berkeley.edu

http://data.berkeley.edu

The vision …

In the 21st century every college

student should be prepared to

understand and develop points of view

based on the analysis of data as well

as evaluate arguments made by others

12/12/16 NAS UCB

DSed

9

Vectors of ChangeNearly every field of discovery is transitioning from “data poor” to “data rich”

Astronomy:LSST

Physics:LHCOceanography:OOI

Sociology:TheWeb

Biology:SequencingEconomics:POS

terminals

Neuroscience:EEG,fMRI

3

Data Science throughout campus

Feb15,2013

AMPLabIonStoica,CSMichaelFranklin,CSMateiZaharia,CS

AdamArkin,Bioengineering

EmmanuelSaez,Economics

Reconstruc ngthemoviesinyourmind

BinYu,Sta s csJackGallant,Neuroscience

Earthqu

ake

Strong

Shaking

in

11second

s RichardAllenEarth&Plan.Science

FernandoPerez,BrainImagingCenter

CharlesMarshallRosieGillespieIntegra veBiology

11

A National Challenge

DS@UCB

In the United States, it is reported that by 2018 there will be more than 490,000 data science positions available, but only 200,000 qualified people to fill the roles. The average size of a graduate class of data science students is 23 students. With approximately only 110 universities offering data science studies, the growing market will continue to pressure the supply in the US.

11

12

Foundations of Data Science @ UCB

DS@UCB 12

14

Data Science Foundation (CS+Stat+Critical Thinking w/ data)

Today’s Majors

Data Science Core

DS MajorDS Minor

Rolling out to the studentsData Science in the Undergrad Experience

NAS UCB DSed 14

Undergrad Class

Existing course

Thesis, Research

DS in field

DS focus

ConnectorsSoc Biz CSPhys

Concentrations

Student Tracks

http://data.berkeley.edu/sites/default/files/DataScienceCurriculumSketch.pdf

http://data.berkeley.edu/sites/default/files/DataScienceCurriculumSketch.pdfhttp://data.berkeley.edu/sites/default/files/DataScienceCurriculumSketch.pdf

Enrollments in Berkeley’s Data Science Classes (1st 4 semesters)

0

200

400

600

800

1000

1200

1400

Fall 2015 Spring 2016 Fall 2016 Spring 2017

Advanced DataScience Classes(3 new in Sp 17)

Connector Courses(21 so far)

Foundations ofData Science(Data 8)

15

Foundations of Data Science

Online syllabus, labs, homeworks, videos – all at data8.orgOnline textbook – inferentialthinking.com

16

http://data8.org/http://inferentialthinking.com/

Foundations of Data Science - “Data 8”• All students should be able to

– reason sensibly based on data

– make and interpret inferences based on

data

– think critically about implications of data

– regardless of their specialization

• Designed for 1st and 2nd-year students

of any major– 4 units, MWF lecture + 2-hour W-F lab

– No CS or stats experience required

• ~700 students in Sp 17

• about 70 majors represented– Tied for first – L&S CS and Economics

– Every college, ~ 1 dozen IEOR & ORMS

• Online syllabus, labs, homeworks,

videos – all at data8.org

• Online textbook – inferentialthinking.com

17

http://data8.org/sp17/http://inferentialthinking.com/

Data and

Ethics

Making

Sense of

Cultural Data

Children in

the

Developing

World

Social

Networks

Data

Science and

the Mind

Genomics

and Data

Science

Data

Science

For Smart

Cities

Data Science

for Cognitive

Neuroscience

Comp.

Structures in

Data

Science

Probability

and Math

Stats in

Data

Science

Data

Science,

Demography

&

Immigration

Data Science Connectors

Social Data

Revolution

19

Data8 - Concepts and Computing• Fundamental co-mingling of CS & Stat concepts on real data

– Learn computing concepts by doing interesting things on data

– Learn advanced statistical concepts by observing what’s interesting

– Codify understanding of concepts symbolically

• “Explorations” in visualization, privacy, personalized medicine,…

• Entirely cloud-based computing environment built on Jupyter notebooks

plus UCB datascience Tables

NAS UCB DSed 19

http://data8.org

http://data8.org

Tables (https://github.com/dsten/datascience)

• A single, simple, powerful data structure for all

• Inspired by Excel, SQL, R, Pandas, Numpy, …NAS UCB

DSed

20

ordered collection of labeled columns of anything

label

values

Numpy arrayT[‘label’]

dict, record,tuple

select, where, take, drop, group

join

stats, binsamplepivot, pivot_bin

split

12/12/16

https://github.com/dsten/datascience

21

Explorations

BEARS DS @ UCB 21

Connectors offered in 1st 4 semsCivEng 88 Data Science for Smart Cities Stat 88 Probability and Mathematical

Statistics in DS

CivEng 88B Time Series Analysis Stat 89A Introduction to Matrices and

Graphs in DS

Cog Sci 88 Data Science and the Mind L&S 39 Race, Policing, and Data Science

CS 88 Computational Structures in

Data Science

L&S 88 Health, Human Behavior, and Data

ESPM 88A Geospatial Data Explorations L&S 88 Child Development Around the

World

ESPM 88B Data Sciences in Ecology and

the Environment

L&S 88 Literature and Data

Geog 88 Data Science Applications in

Geography

L&S 88 Genomics and Data Science

Hist 88 How Does History Count? L&S 88 Social Networks

Info 88A Data and Ethics L&S 88 Data Science for Cognitive

Neuroscience

Legal St 88 Crime and Punishment L&S 88 Data Science, Demography, and

Immigration

MCB 88 Immunotherapy of Cancer22

Data 8 Demographics, Fall 2016• 517 students from 59 majors

23

Online syllabus and more – ds100.org

NEW: The upper-division gateway, DS 100

24

http://ds100.org/

Big Concepts in DS 100

• Data preparation and representation

• Efficient and scalable data processing

• Question formulation and experimental design

• Exploratory data analysis and visualization

• Modeling fitting and inference

• Machine learning techniques and overfitting

• Validation and hypothesis testing

25

http://www.ds100.org/sp17/syllabus

Week 1

Week 2

Week 3

Week 4

Week 5

Week 6

Week 7

Week 8

Week 9

Week 10

Week 11

Week 12

Week 13

Week 14

Week 15

Week 16

Review & Midterm

SpringBreak

RRR

Big Concepts in Data Science

Working with Real Data and SQL

Basic StatisticalModeling

Inference, Prediction, &

Machine Learning

ML Techniques

ML Techniques

Analytics at Scale

HW1

HW2

HW3

HW4

HW4

HW5

HW6

HW6

HW6

“pilot” DS 100 syllabus

26

http://www.ds100.org/sp17/syllabus

Foundations of Data Science

Possible DS Major Structure

12/12/16

Mathematics College Breadth Electives

Principles and Techniques of Data Science

Statistics Depth

External Emphasis

Statistical Machine Learning Implications

Computing

CS Depth

NAS UCB

DSed

27

Biological Sciences

Math & Physical Sciences

Arts & Humanities

Social Sciences

College of Letters & Science

Computational &

Data Sciences & Eng.

Engineering Professional Schools

Institutional Refactoring

NEW: Probability for Data Science

29

NEW: Statistical Methods for Data Science

30