1. Developing Computational Skills
in the Sciences with MATLAB
Presenters: Lisa Kempler, Alain Plattner, Benjamin Bratton, Daniel Zysman
Audience: Science educators
Lisa Alain Benjamin Daniel

2. The Session Will Give You
• Strategies for teaching data analysis, modeling,
and computation, in domain-focused courses
• Resources for teaching computation in Sciences
– Teaching activities (with code)
– Tools to address common challenges conveying
computational skills in the Sciences
– with MATLAB
• Access to a community of peer educators
http://serc.carleton.edu/NAGTWorkshops/data_models/matlab16/index.html

3. Session Flow
• Session goals
• Tour of MATLAB Resources for Sciences: SERC site
• 3 professors’ representative Teaching Activities
– Choice of geophone layout in a simple near-surface seismics setting
• Alain Plattner, Geophysics
• California State University – Fresno
– Building Modular Tools for Visualizing Computation
• Benjamin Bratton, Biology
• Princeton University
– Principal Component Analysis
• Daniel Zysman, Neuroscience
• Massachusetts Institute of Technology
• Q&A

4. Teaching Computation in the Sciences
with MATLAB Workshop (Oct 2016)
• Objectives and process
• Participants
• Outcomes

5. https://serc.carleton.edu/NAGTWorkshops/data_models/matlab15/participants.htm
he
Participants and Posted Resources

6. Resource Overview Page:
Teaching with MATLAB
http://serc.carleton.edu/NAGTWorkshops/data_models/toolsheets/MATLAB.html

7. October 2016 Workshop Outcomes:
Teaching Computation with MATLAB
http://serc.carleton.edu/teaching_computation/index.html

8. SERC Site Resources:
Teaching Computation with MATLAB
MATLAB page for educators
• http://serc.carleton.edu/NAGTWorkshops/
data_models/toolsheets/MATLAB.html
Workshop resources and outcomes
• http://serc.carleton.edu/
teaching_computation/index.html
• http://serc.carleton.edu/
matlab_computation2016/outcomes.html

9. Session Flow
• Session goals
• Tour of MATLAB Resources for Sciences: SERC site
• 3 professors’ representative Teaching Activities
– Choice of geophone layout in a simple near-surface seismics setting
• Alain Plattner, Geophysics
• California State University – Fresno
– Building Modular Tools for Visualizing Computation
• Benjamin Bratton, Biology
• Princeton University
– Principal Component Analysis
• Daniel Zysman, Neuroscience
• Massachusetts Institute of Technology
• Q&A

10. Examples using Matlab / Octave for
Experimental Design and Data Processing
in a Near-surface Applied Geophysics Class
Alain Plattner
California State University Fresno
April 27, 2017
https://github.com/NSGeophysics

11. Two Matlab / Octave m-file packages
Seism-O: Simulation of applied geophysics seismic data
Used for teaching concepts and experimental design
GPR-O: Processing and visualization of ground penetrating radar data
Used for teaching concepts and effects of data processing
On https://github.com/NSGeophysics
https://github.com/NSGeophysics

12. Seism-O: Simple Seismic Data Simulation
https://github.com/NSGeophysics

13. https://github.com/NSGeophysics
Seism-O: Simple Seismic Data Simulation

14. air wave
direct wave
refracted waveReflected wave
V1
V2
V0 = 343 m/s
https://github.com/NSGeophysics
Seism-O: Simple Seismic Data Simulation

15. air wave
direct wave
refracted waveReflected wave
V1
V2
V0 = 343 m/s
Show simulated arrival times
Simulated arrival times using Seism-O for 1 m geophone spacing
https://github.com/NSGeophysics
Seism-O: Simple Seismic Data Simulation

16. Simulated arrival times for 1 m geophone spacing Simulated waveforms for 1 m geophone spacing
https://github.com/NSGeophysics
Seism-O: Simple Seismic Data Simulation

17. Simulated waveforms for 2 m geophone spacing Simulated waveforms for 1 m geophone spacing
https://github.com/NSGeophysics
Seism-O: Simple Seismic Data Simulation

18. Simulated waveforms for 2 m geophone spacing Simulated waveforms for 1 m geophone spacing
Software and teaching activity on
https://github.com/NSGeophysics/Seism-O/wiki
https://github.com/NSGeophysics
Seism-O: Simple Seismic Data Simulation

19. GPR-O: GPR data processing & visualization
https://github.com/NSGeophysics

20. Instructions and software on
https://github.com/NSGeophysics/GPR-O/wiki
- Simple Matlab / Octave scripts
- Ground penetrating radar basic data processing
- 2-D profile and depth-slice plots
- Topography
- Educational documentation
GPR data available for example from
https://alaska.usgs.gov/portal/project.php?project_id=384
https://github.com/NSGeophysics
GPR-O: GPR data processing & visualization

21. Data: Liu et al. (in prep)
https://github.com/NSGeophysics
GPR-O: GPR data processing & visualization

22. https://github.com/NSGeophysics
GPR-O: GPR data processing & visualization
Data: Liu et al. (in prep)

23. https://github.com/NSGeophysics
GPR-O: GPR data processing & visualization
Data: Liu et al. (in prep)

24. https://github.com/NSGeophysics
Data: Liu et al. (in prep)
GPR-O: GPR data processing & visualization

25. Tutorial with test data:
https://github.com/NSGeophysics/GPR-O/blob/master/doc/GPR-O.pdf
Raw field data available for example from:
https://alaska.usgs.gov/portal/project.php?project_id=384
https://github.com/NSGeophysics
GPR-O: GPR data processing & visualization

26. Summary
- Seism-O and GPR-O: Matlab / Octave scripts for seismic data
simulation and ground penetrating radar data processing
- Both available from https://github.com/NSGeophysics
- For near-surface geophysics class or general data simulation,
experimental design, data processing class.
https://github.com/NSGeophysics

27. Building modular tools for
visualizing computation

28. Why are computational tools
daunting (for biology graduate students) to use?

29. Modular problem sets help
students quickly gain proficiency
• Small pieces with
displayed outputs
• Reusable pieces
• Springboard to
independence

30. Dynamical systems
What do x(t) and y(t) look like?

31. Dynamical systems

32. Dynamical systems

33. Dynamical systems

34. Dynamical systems

35. Specific example: Modeling
oscillations of glycolysis in yeast
For multiple versions of the code see
https://github.com/bpbratton/buildingModularVisua
lization_SERC201610

36. Benefits for modular problems
•Students learn when to take breaks
•Fail (and succeed) quickly
•Encourages code reuse

37. A principled way to principal
components analysis
Daniel Zysman
April 27, 2017

38. Teaching scientific computing in
Brain & Cog sciences
• The challenge is to deal with students
motivation, preparedness and interest:
• How to effectively break the ice.
• How to cope with inertia.
• We need to make it interesting, relevant
and transferable.

39. A two tier approach
• Tier 1:
– Learn the basics of data visualization and modelling via toy
examples.
– Build it step by step.
• Tier 2:
– Apply the fundamentals to fun and relevant problems.
Photos: Mandana Sassanfar, Quantitative Methods Workshop, CBMM, MIT.

40. Tier 1: PCA Toy Example
Toy data: Bivariate Gaussian, unknown covariance to students.
Task: Find directions of maximum variance.
-8 -6 -4 -2 2 4 6 8 10
X1
-6
-4
-2
2
4
6
X2
-8 -6 -4 -2 2 4 6 8 10
X1
-6
-4
-2
2
4
6
X2
-8 -6 -4 -2 2 4 6 8 10
X1
-6
-4
-2
2
4
6
X2
-8 -6 -4 -2 2 4 6 8 10
X1
-6
-4
-2
2
4
6
X2
Rotate
Project
Reveal
Structure

41. Take home messages:
• Rotations and Projections
• Data variability and its geometry
• Matrix duality:
– to represent data
– to transform data
• Challenge activity: make a dataset, ask fellow
student to reveal underlying structure.

42. Tier 2: PCA for image classification
28 by 28 pixels
8-bit gray scale images
These images live in
a 784 dimensional space
Task: classify digits one
from seven using PCA
Revisit:
rotations
projections
Teaching activity:
http://serc.carleton.edu/matlab_computation2016/activities/159581.html
MNIST data set
http://yann.lecun.com/exdb/mnist/

43. Pairwise pixel intensity relations
• There are more than 300000 possible pairwise pixel plots!!!
• Not clear which one to choose.
• Which choice is more informative? Try PCA.

44. The 9 images that capture most
variance

45. Projections for classification
• The first two PCs capture ~37% of the total variance.
• The data clusters nicely in this space (linear separability).

46. Learning outcomes
• PCA reveals underlying structure in the data.
• It aids in visualization and classification tasks.
• Provides a chance to address short comes,
assumptions and pitfalls.
• Emphasize the different role of matrices:
– To store data
– To transform data

47. Student’s feedback
• The two tier approach:
– Keep students engaged and motivated
to learn, while being challenged.
– Improves confidence and competency
to transfer what they have learned to
broader contexts.

48. Join the Community:
Teaching Computation in Sciences with MATLAB
Community
• http://serc.carleton.edu/matlab_computatio
n2016/community.html

49. Join the Community:
Teaching Computation in Sciences with MATLAB
Community• http://serc.carleton.edu/matlab_computatio
n2016/community.html

50. Future Community Sessions on
Teaching with MATLAB
• Earth Educators’ Rendezvous, July 17-21, 2017
– Early Bird Registration deadline is May 1st!
• Mark your calendars!
– Teaching Computation with MATLAB in the Sciences
Workshop
• October 15-17, 2017
• Carleton College, Northfield, MN
• Email Rory McFadden (rmcfadden@carleton.edu) if
you are interested in joining the community

51. SERC Resources:
Teaching Computation with MATLAB
MATLAB page for educators
• http://serc.carleton.edu/NAGTWorkshops/data_models/
toolsheets/MATLAB.html
• And researchers, too
Workshop outcomes
• Teaching Activities
– http://serc.carleton.edu/matlab_computation2016/activities.html
• Pedagogy and computational philosophy
– http://serc.carleton.edu/teaching_computation/index.html
• Community
– http://serc.carleton.edu/matlab_computation2016/community.html