Hacker News new | past | comments | ask | show | jobs | submit login
Marching neural network (arogozhnikov.github.io)
113 points by signa11 on June 26, 2017 | hide | past | favorite | 11 comments



A beautiful piece of data art (as in https://lisacharlotterost.github.io/2015/12/19/Meaning-and-B...), but not much of a data visualization (I can hardly get any insight from it).


Is this meant to be useful or just look interesting? I can't really say I understand what is being shown.


Pretty. It's useless


for those wondering, what is on display are the level sets (3d contours) of a neural network.

this is a very cool use of shaders for raytracing, but it is not terribly informative.


Why is it moving?


It's a 4-dimensional object. The animation doesn't show the object moving, it just shows progressive cross-sections of the object. If you stop the animation, you can choose which layer of the object to view with the slider.

Edit: Not to derail the thread, but here's another post about 4D visualization toys. https://news.ycombinator.com/item?id=14471931


Otherwise you'd only see the level set for one specific value. Looping through the values allows to see the entire shape over time.


Already mentioned in the comments. But still. Looks nice, but it is hard to get any insight from it. For instance, I zeroed out all weights and outputs, then slightly increased one weight. First there seems to be some non-zero bias on the minimal weight, second there is some mon-linearity which goes against intuition. It would be nice to be more clear about what is going on and then it can be more useful not only nice.


This is pretty cool. It seems impossible to mouse away while maintaining the orientation. I want to be able to perturb the weights and watch from different angles. Very cool though, nonetheless.


You can change tab or window with the keyboard, move the mouse, and get back to the visualisation with the previous orientation intact.


There are some sliders for the different axis just below the visualization.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: