index.md

title: how to make a donut style: style.css author: name: dino-dna url: https://github.com/dino-dna output: index.html controls: true

--

how to make a donut

• NOTES
  • CD
    • thx for showin' up
    • 10 minute not enough time for pres or content, we're going to blaze through it!
      • leave time for Qs!
    • PLEASE without haste or fear just interrupt us, really

--

who... are you?

• 👋🏻 chris, general guy

• 👋🏻 cory, general guy

• NOTES

  • old chums from OSU
  • old coworkers
  • ...excited to talk you about donnies

--

readiness-check!

• bloopers, queued

• donut-monster, growling

• hacks, hacking

• stories, telling

• jerks, crushed (i.e. little server worker buddies)

• comp audios, silenced

• NOTES -- why

why are you here?

• coinstac

  • Collaborative Informatics and Neuroimaging Suite Toolkit for Anonymous Computation

• NOTES

• CD

  • here show of a donut maker, but it was inspired by our prior art--a research project.
  • cory, can you explain what motivated today's presentation?

• CR

• COINSTAC - collaborative informatic suite

  • solves two major problems in neuroscience data sharing!
    • First: constraints placed on imaging data
      • researchers commit to maintaining participant privacy!
      • Study facilitors sometimes forbid file sharing, because...
      • it's been found that evil doers can reverse engineer brain images and derive personally identifiable information.
      • creates data silos, which is bad:
        • data would be useful to many other studies
    • CD
      • Second: portability
        • data too heavy and big to ship over network
        • people still shipping hard drives - WAT
      • no defacto tooling for coordinating data formats and analysis pipelines
      • ...more, but those are the biggins -- so-wat

so what did we do?

• built a distributed Machine Learning system

• NOTES
  • CD
    • SOLUTION
      • work around the constraints
        • never share the data. instead...
          • do micro analyses, and share those!
          • do a super analyses in the cloud to simulate as though we all did one big analysis together
          • double down on thwarting evil doers by adding noise to all our data as we do micro analyses
          • ?? add pictures ??

--

how?

• JS!

• ...but brain research happens in python & R

  • distributed docker pipelines!

• NOTES

  • CR
    • make an app to solve these things! but how?
      • electron app, react, redux
      • node services
        • pouch, hapi, couch etc #jslyfe
    • docker for research algorithms
      • brain research happens in python & R

--

so why are you here?

• we wanted to show it off!

• they researched "what factors contribute to Schizophrenia?"

• we researched "what factors make donuts most delicious?"

  • sneak peek of coinstac (kinda) via donuts
  • share some tinkering with DonutLearning™

• NOTES

  • CD

-- launch

so how do you make a donut?

dino-dna.github.io/donut/

• NOTES
  • CR
    • everyone, launch this app!
    • there are a few main views
      • DonutViewer
        • all donuts basic SVG elements
          • Donut SVG
          • Sprinkle SVG
    • make some donuts!
    • please, disregard the emojis
      • make the donut the way you like it, we wont judge

--

what makes a donut delicious?

• NOTES
  • CD
    • in research, you are looking for equations that explain the world
    • one technique is called regression, visualized here
    • in this example, there are a bunch of data points, and some regression algorithm is doing it's best to make a mathmatical equation that explains these data points
      • the intent here is that the reg line is the equation that the regression creates
    • so assume that there is a global, wordly phenomenon that explains what really makes a delicious donut
      • are pips where its at?
      • are blue star where its at?
      • assuming that there is a truth to donut delicousness, we can use YOUR donuts to teach a machine what characteristics make it great
        • for example, donuts with big radii are great
        • donuts with too many sprinkles are gross
    • for ease, we've cheated, and hardcoded this magical formula into the ether

--

DonutLearn™ Time

• launch the HACKS

• enable DONUTS, firehose mode

• watch 'em flow!

• Notes

  • CR
    • use upload button

--

what's happening here?

• you send us donuts

• we compute a mathmatical model/equation

  • regression learns an equation that explains deliciousness
  • f(donut) = delicious-ness-score

• we some ML goodness to maximize that equation

• we send back down our best guess to the UI

• centralized, not decentralized

• NOTES

  • CR
    • donut build process (redux, blah)
    • websockets, pure node & socketio
    • child process to docker child...
  • CD
    • where we use the defacto learning toolsuite
    • blab about scipy, numpy, etc
      • maybe blab about all MATHS in CS all ultimately use the same stuff, it just so happens the python ones get all the credit ;). there are node linear alg things... anyway...
    • we build a regression on the fly
    • we run an optimizer, looking for maximums
      • if you want to know more about the py... ask about in during questions
    • steam it back to cory
  • CR
    • and i send it back to here! pointing to screen

--

#bloopers

• open VLC, dawg.

• open img/bloopers

• bulk select and Open with...

• NOTES --

plugs

• math.

• cleaver

  • it's like a real simple stupid revealjs

• scikit learn

• docker, obviously. such great.

• create-react-app!

  • config is soooo 2016

• NOTES

--

anti-plugs, kinda

• dockerode

  • great, but tricky!

• NOTES --

questions?