The neural representations code and prose are less differentiable

Decoding the Representation of Code in the Brain: An fMRI Study of Code Review and Expertise

Benjamin Floyd, Tyler Santander, Westley Weimer

University of Virginia

University of Michigan

University of Michigan

Westley Weimer	2

Source Code” (ICSE 2014)

●Found five brain regions associated with code comprehension

●Encouraged future fMRI+SE research

Source Code” (ICSE 2014)

●Today: Understanding 'Understanding Understanding Source Code' ?

Expertise

●Individual differences in programming and

●Could this apply to CS? Westley Weimer	6

Functional Magnetic

Resonance Imaging (fMRI)

●Which parts of the brain are in use?

●Your brain needs energy but does not store it

●Blood-oxygen level dependent (BOLD) signal Westley Weimer	7

A Study in Contrasts

●e.g., reading code and being nervous

●How can we tell if an observed pattern of activation corresponds to one activity?

activation that vary between the stimuli/tasks Westley Weimer	8

High-Level Question

Is reading code more like doing math or more like reading prose?

Westley Weimer	9

Code Review and Comprehension

●Developers spend more time understanding

●Mandated in Facebook, Google, etc.

●One of the most effective techniques in software

Westley Weimer		10

Experimental Design: 3 Tasks


Westley Weimer	11

Experiment Setup and Data

●Right-handed, native English speakers, corrected-

to-normal vision, IRB-HSR #18420, etc.

	12

Dead Fish and Software Bugs

	13

Results: Mind Reading

Westley Weimer	14

Results: Can we relate tasks to

brain regions?

decision-making, language, conflict monitoring, etc.)

were highly weighted

Westley Weimer		15

classification accuracy?

●How accurately our model distinguishes between Code Comprehension and Prose significantly predicted expertise (r = -0.44, p=0.016)

●The inverse relationship between accuracy and expertise suggests that, as one develops more skill in coding, the neural representations of code and prose are less differentiable. That is, programming languages are treated more like

natural languages with greater expertise. Westley Weimer	16

Costs and Reproducible Research

●Including IRB protocol application, recruitment materials, screening forms, training videos, visual stimuli, etc.

●	17

Future Studies

●Writing code (fMRI-safe keyboard)

●Transcranial magnetic stimulation (read-write)

●Does any of this sound interesting? …

Westley Weimer	18

Call To Arms

●I have funding and am looking for collaborators

●Come talk to me	19

Conclusion

●Our classifiers distinguish them based solely on brain activity

●The same brain locations distinguish these tasks

●Greater expertise accompanies a less-

differentiated neural representation Westley Weimer	20