Summer of Math Exposition

Presented by 3Blue1Brown 3blue1brown

OptiVIz : An interactive tool to visually understand the optimization algorithms

Audience:

We have developed a website to help users visually understand different types of optimization algorithms, such as Gradient Descent, Conjugate Gradient, Newton’s Method, Golden Section, Powell’s Method, and the Nelder–Mead algorithm. Users can either choose from existing benchmark objective functions or provide a custom one. After selecting an initial guess and the appropriate algorithm-specific settings, the user can start the optimization process and interactively track the path of the optimizer.


Analytics

5.34 Overall score*
46 Rank
14 Votes
6 Comments

Comments

6.7

A nice interactive visualization of some optimization algorithms.

6.4

As this is a tool, I feel like it can’t be marked exactly like the article/video media in other submissions. That said, I will try to rank this based on the categories I do consider important for a website.

  1. Adaptability: Can work on mobile and PC, which makes this a great and versatile tool across (most, if not all) devices.
  2. UI: very intuitive, easy to use. Absolutely no complaints here; it takes a while to load, but since this might be the fault of my internet connection, I cannot take off marks for that.
  3. Applications: high. This tool seems to have optimisations for all needs, and beyond. Very useful for visualisations as well.

All of the above will add +1 to the total.

Now, ranks by official categories:

  1. Motivation: 4.5/9, low average. This is a tool, so unless you have need of it or just want to play around, there’s no real need to use it. Although I personally find it interesting to tweak the animations and watch the algorithm do its thing, this is ultimately a tool and isn’t here to “explain”, per se, the mathematics behind it.
  2. Clarity: 6.5/9, high. The points underneath the function, explaining how the algorithm works, are extremely clear and simple enough to understand. Here also I would mention how simple it is to zoom in/out of the function, with clear instructions and an intuitive layout.
  3. Novelty: 6.5/9, high. Despite the existence of optimisation tools online, this one collects different types of algorithms and explains how they work. The different kinds of optimisation tools, all in one place, is truly what makes this stand out.
  4. Memorability: 4/9, low. Once again, this being a tool makes it slightly less memorable than other forms of media here. However, its efficiency and versatility might be memorable to a satisfied user.

Total: 5.375 + 1 = 6.375, rounds to 6.38 on the scale (or 6.39, since the .38 option decided to be so elusive 😅)

4

I’m familiar with these methods, but it was still not completely clear how to use this or what it was supposed to teach me. I would probably not give this to students first learning these methods, as there is a lot happening on the page, and I’m not sure how much of it is useful for a new student. I did not like that after clicking compute I needed to drag between iterations of the method, that felt clunky to me. Overall, this seems like it was a good amount of work to put together, and could be very useful in specific situations, but not something I would plan to use in a class.

6.5

Important resource, but looks like still mostly under construction - I would add maybe a bit more emphasis in making the demo’s play automatically / show the examples in parallel so that it is easier to compare the methods against each other?

5.5

This is an important resource but it looks like it’s still in development. There are both typos in the text and bugs in the visualisation software, such as when you try to change the zoom, the rotation angle gets reset to default. Also, the ore set functions need to be chosen more judiciously even though I understand why these functions anre chosen, but some of the functions like the sphere get optimised at the first step itself for some optimisation algorithms and some don’t ever get optimised. Otherwise, the text is concise and well structured. When it is complete it’ll be a very useful resource.

5.5

A very nice visualisation of gradients. However, I’m not sure that this really fits into SOME. It doesn’t really have any vulgarisation content, as it is just a “tool” design for specialist?