You are on page 1of 25

# The Discrete Optimization MOOC

## An Exploration in Discovery-Based Learning

Carleton Coffrin and Pascal Van Hentenryck

!1

## talk_pub - 13 February 2014

Discrete Optimization?

!
Often called the technology of better.

!2

Examples - inventory management, supply chains, energy markets Others say it is the most important field, that no one has ever heard of

## Pascal Van Hentenryck

Discrete Optimization?

!
Its also a class on Coursera, ;-)

!3

## One Size! Fits All

!
!4

- Discrete Optimization is HARD, and there are some really deep mathematical reasons for this (P vs NP question). - There is no such thing as a one size fits-all solution in this field.

## talk_pub - 13 February 2014

How can we prepare students for new optimization problems that we cant even anticipate?

!5

## Discrete Optimization in Practice

!
skills!

!6

Practitioners need to be malleable. Every problem is a new challenge. From a teaching perspective, if we want to train people to do D.O. we need to cover not only the core ideas of the field, but also general problem solving

## talk_pub - 13 February 2014

Discovery-Based Learning
Lesson: mass, volume, density

!
-

!7

A pedagogical approach, where students are given a broad problem to solve, and are left to discover a way to solve it.

## talk_pub - 13 February 2014

Discovery-Based Learning
Lesson: mass, volume, density

!
-

!7

A pedagogical approach, where students are given a broad problem to solve, and are left to discover a way to solve it.

## YOU CANNOT FAIL!

!8

- Emulate the real world your boss tells you: We need to solve this problem. I dont care how, just get it done. - We provide a digital play ground for the students to explore solving discrete optimization problems, and encourage them to try new things share their experiences - They submit the assignments as many times as they like to get as much feed back from the graders as they wish.

## Discrete Optimization Discovery-Based Learning

Lecture Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP) Assignments Knapsack (DP) Graph Coloring (GC) Travelling Salesman (TSP) Warehouse Location (WL) Capacitated Vehicle Routing (CVRP)

!9

- Students explore and discover the matching of the lecture material to the assignments.

## Discrete Optimization Course Context

50,000 17,000 4,000 795 795/17000 = 4.68%
A B C Points Qualified Certificate Distinction

10000

Student Participation

8000

6000

4000

2000

3000

4000

A 1000

2000

B C 0

5 Week

100

300

400

!10

## Discrete Optimization Course Context

50,000 17,000 4,000 1,884 795 795/17000 = 4.68% 795/1884 = 42.2%
A B C Points Qualified Certificate Distinction

## auditor active qualified

10000

Student Participation

8000

73% 61%

6000

4000

66%

2000

3000

4000

A 1000

17%

2000

14% 22%

71% 27%

77%

13%

25%

29%

27%

28%

26%

21%

5 Week

Qualied
!10

## Course Completion Survey:

Which Lecture Topics did you use on each assignment?
Lecture Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP) Assignments Knapsack (DP) Graph Coloring (GC) Travelling Salesman (TSP) Warehouse Location (WL) Capacitated Vehicle Routing (CVRP)

!11

## talk_pub - 13 February 2014

Knapsack
Best Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP)
DP BB CP LS MIP

!12

## talk_pub - 13 February 2014

Graph Coloring
Best Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP)
DP BB CP LS MIP

!13

## talk_pub - 13 February 2014

Travelling Salesman
Best Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP)
DP BB CP LS MIP

!14

## talk_pub - 13 February 2014

Warehouse Location
Best Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP)
DP BB CP LS MIP

!15

## Capacitated Vehicle Routing

Best Topics Dynamic Programming (DP) Branch & Bound (B&B) Constraint Programming (CP) Local Search (LS) Mixed Integer Programming (MIP)
DP BB CP LS MIP

!16

## talk_pub - 13 February 2014

75%
of students who nished felt condent in their ability to apply the course material to real-world problems

!17

## talk_pub - 13 February 2014

Discovery-Based Learning

Blissful Image

!18

- At this point, Discovery-based learning may seem like a pedagogical paradise, but it has a dark side too.

## Discovery-Based Learning Issues

Somewhat torn, dont feel like Im learning anything If youre looking for a new challenge: Find a way to remotivate me! I followed the recommended syllabus, ... and then attempted the assignment ... this is not sufcient to solve all of the problems.

!
Solutions:

!19

- Revise introduction material, to easy the students into the discovery based idea. - provide supplementary quick-start videos to help people get started

## talk_pub - 13 February 2014

Video of Pascal

!
assignments.

!20

- A video of Pascal van Hentenryck explaining how to keep your motivation up while approaching the challenging

## talk_pub - 13 February 2014

Overall

!21

I think the assignments were a great success. And I would encourage any other MOOC practitioner to take the extra time and effort to make discovery-based assignments in MOOCs.

## So far, my favorite part of this this course is _____________.

help computer understand
challenges

week experience

examples

homework

cant

optimization mip

learning

little

## getting graph solution

language watching

feel
level

concepts

approaches

search

coloring

professor

assignment

## nice linear lot actually none

course

tsp

lectures

time

10s

lecture

ive

information

material

knowledge

start

implement

enjoy real

makes

figure

consuming
solvers

thats forums

!22

Half way through the course the students were asked this free form text question. Visual inspection and work count analysis of around 600 responses indicate that the assignments were most students favorite part of the class! Another indication of the effectiveness of the discovery based assignments.

presentation

improve

style

coding

python fun

love

dont

spend

bit

opt

videos

hard

prof students

lack

seeing

algorithms

try

local video

## ideas able solutions

difficult
using

challenging

hours

implementation

programming

assignments

solve trying

constraint enthusiasm

example

## talk_pub - 13 February 2014

Carleton.Coffrin@nicta.com.au
!23