Designing well-behaved parameter spaces

[alice-i-cecile]

Name

Well-behaved parameter spaces are more intuitive: lessons from the mathematics of optimization for UX design

Notes

Same ideas as in mathematical optimization / ML: you need to allow for efficient exploration!

Desiderata:

1. As small as possible.
2. As orthogonal as possible.
3. All combinations of values are valid.
4. Always non-zero gradient.
5. Avoid discontinuities.
6. Local behavior should map onto global behavior in predictable ways.

Inspired by UI layout experiments and https://www.reddit.com/r/rust/comments/nofrp5/tightness_driven_development_in_rust/ + https://www.ecorax.net/tightness/.

[alice-i-cecile]

[12:48 PM] IvyLashes: A huge part of how humans learn is by creating good representations on top of complex, messy spaces: think chords in music, elements of design in art, abstraction boundaries in programming... [12:49 PM] IvyLashes: Rather than thinking in terms of "well this note is off", you're often operating at a much higher and more useful level: "this would sound better in a minor key" [12:50 PM] IvyLashes: But the things that make for good representations, where you can accomplish the tasks that you're trying to achieve in an efficient and intuitive way, are the same properties that machines need to optimize effectively

[alice-i-cecile]

[12:53 PM] IvyLashes: Anyways, you can use this as a general principle of UX design [12:53 PM] IvyLashes: For things like APIs, user interfaces, game design and so on [12:53 PM] IvyLashes: Where this "ease of exploring space" is something you can analyze in very tangible and well-motivated ways [12:54 PM] IvyLashes: For example, this is a huge part of why CSS is So Bad