ml5js / intro-ml-arts-ima-f19 Goto Github PK

Add any ideas to this thread!

• Eye Conductor by Andreas Refsgaard

https://forensic-architecture.org/investigation/triple-chaser

Thank you @noisyneuron! https://noisyneuron.github.io/nyu-hpc/

https://vimeo.com/304131671

Similar to the talk they gave at NYU last year - really helpful for the importance of debugging data.

At the end of the New Organs video -- https://lav.io/projects/the-new-organs/ -- Brain & Lavigne make the point that the assemblages of surveillance, AI, etc are less about classifying you based on your past, but about trying to predict your future / nudge you into some prescribed future

Maybe we can lay into the problematic issues around prediction?

https://www.youtube.com/watch?v=aircAruvnKk&feature=youtu.be

I've found this video to be understandable to even people who haven't worked with ML before and helps demystify what is a "neural network". Maybe people should watch before coming to class next time to help make some of these concepts like gradient descent, hidden layers, etc etc?

While the video is nearly 20min could be helpful to watch together so that students can ask questions along the way? or maybe helpful if students have to respond to the video with an illustration of how all those components work?

@shiffman In case you're looking for an example for evaluating student work for the IMA students, here's an example of a rubric that I've found to be very helpful

https://github.com/cmda-tt/course-17-18/tree/master/assessment-1#rubric

Hi!

This documentation is in progress, but you can begin to find the info you might need for the ml5.neuralNetwork here:

https://ml5js.github.io/ml5-library/docs/#/reference/neural-network

if there are things that are unclear or confusing, please feel free to ask for clarification!

The question came up today in class: "What if I want to collect data? Is there a helpful guide / document of tips / common strategies for ethical data collection?". Please add your suggestions here:

Also, nothing these two topics I referenced:

Duke University MTMC

• https://www.dukechronicle.com/article/2019/06/duke-university-facial-recognition-data-set-study-surveillance-video-students-china-uyghur
• https://twitter.com/adamhrv/status/1135134952528470016?lang=en

Atlanta Asks Google Whether It Targeted Black Homeless People

• https://www.nytimes.com/2019/10/04/technology/google-facial-recognition-atlanta-homeless.html

http://www.tylervigen.com/spurious-correlations

This could be handy to showcase what it means to draw false conclusions from data that follow similar trends. This could lead to a larger discussion around what it means to come up with proxies for XYZ or to develop "bad models".

https://medium.com/transit-app/can-we-make-montreals-buses-more-predictable-no-but-machines-can-e42f28a1a0ba

for Ethan!

I wonder, would this repo make sense in the ml5.js org? I would like to make use of the "team" features to add collaborators. I could also create a new org. The course isn't exclusively ml5, but the goal is to use ml5 as the foundation. any thoughts @joeyklee?

Do we need to add the code snippet displayed after the model is published to our html file? I don't see it in the example and it looks buggy, the copy icon is not working and one end tag </script> is missing. Should we ignore it?

• What is the data?
• What format is the data
• dimensions and data types
• what’s missing / incorrect / problematic
• how/where/why was it collected
• “actuate” for who is this accurate — for what purpose
• does this data make sense for “ML”? (Domain Knowledge — expert)

Love how friendly and accessible this free course is from Reaktor and the University of Helsinki: Elements of AI (may also help with #28 if we develop that)

Also this! A visual introduction to machine learning

Quickly adding some links to things that came up in class that may not be documented on the notes or slides.

http://www.augustincosse.com/ml2018
https://www.coursera.org/learn/machine-learning
...

In finalizing the materials for week 7 noting these two articles for week 8.

https://medium.com/artists-and-machine-intelligence/adventures-in-narrated-reality-6516ff395ba3
https://medium.com/@ml5js/the-subtext-of-a-black-corpus-4440de02eb32

Here is the article I referenced in class. There is a lot of reading for this first week already so I'll save this for next.

• A Brief History of Neural Nets and Deep Learning

https://www.annagudnason.com/post/emotional-disconnect

HTTP Requests

base64 encoding

socket.io

Processing library

Related to #34, I'd like to make video tutorials for:

• PoseNet (demonstrating single part selection specifically)
• UNet
• BodyPix

Vibert is available to present his work in class.

http://vibertthio.com/runn/

On Slide 7 - might be cool to introduce why ML/AI has a new "resurgence". Making reference to the AI Winter could be helpful to contextualize what has changed - e.g. new compute power, keras ==> tensorflow/tfjs ==> ml5 etc etc

Making a note here to remind me to reference / discuss in class tomorrow. I've added this material in #39 to week 2 since I believe that is best for future iterations of course. Thank you @ellennickles!

The session1 README should include resources for students who want a p5.js refresher before next week.

Due to the Monday / Wednesday schedule and the labor day holiday, this course meets only once the first week (and again Thanksgiving week). I will consider tomorrow an "intro" day and wait to get more into the rhythm of M/W assignments next week? Not 100% sure about this, but thinking out lout as I start to add material.

The link in week 7 assignment instruction leads to 404 page of p5 web editor.

It would be great/helpful to define a set of critical questions for students to define when approaching ML models and data:

e.g.

• "Accurate for whom?"
• "Accurate for which purpose?"
• "Appropriate in which context?"
• "(Un)Representative of what or whom?"

As a learning outcome, it would be great if students start to always ask or address these questions in their work, reflections, and future projects.

https://stackoverflow.com/questions/41455101/what-is-the-meaning-of-the-word-logits-in-tensorflow

More of a meta discussion: maybe it is good to have a ML term dictionary with creative examples and metaphors?

e.g.

• http://18f.github.io/API-All-the-X/pages/what_are_APIs-anecdotes_and_metaphors/
• http://metaphorical.ly/
• https://sidewaysdictionary.com/#/ via @auremoser

A note to remake or find this example. See discussion in #15.

We should figure out what is meant by "location of data".

For me one of the "ah-ha" moments was seeing what is going on in a Perceptron. Seeing how simple "intelligence" can be was very helpful to demystify the idea of defining fitness.

• NOC chapter: https://natureofcode.com/book/chapter-10-neural-networks/
• Example from NOC of a vehicle "learning to steer towards targets": https://joeyklee.github.io/sandbox/nature-of-code/nn-steering-perceptron/