Building Better Predictive Models Using Topology

[@cosmicallyderived]

"knowing shape in advance", "absence of theory to guide the data", "coordinate obscuring", "topology guides model creation", "topology to generate principled local models that don't suffer overfitting and that understand particular shapes to succeed in its partitioning for reliable local models". Fantastically great perspectives and abstractions! Beautiful stuff! Looking forward to great outcomes from you all.

[@jposadalcs]

Wonderful work! It's very exciting to see the application of TDA to real data sets.

[@ismailelabbassi7150]

Was a great presentation thank you so much.

[@posthocprior]

I didn't understand the initial example of colors being assigned and how the shape of the circle is determined, using an unsupervised method. More broadly, it seems that functors -- from algebraic topology/category theory -- could be used powerfully in this context. That is, if you can form a functor, and you can then determine its topological shape, you can then form morphisms. That is, you can find the isomorphic properties of the shape, which, I think, is important in the context of predictive learning. For instance, this would allow the shrinking or enlarging or the given shape, given a smaller but similar set of data.

[@tejask5417]

That was an amazing explanation :) Thank you!

[@ismailelabbassi7150]

I learned things that i was looking at them for 5 months

[@thomas.moerman]

What happens if one would simply replace the TDA networks with t-SNE plots? How different would the results be?

[@erickay123]

Great question, I am wondering that also. I saw a video where they were using both.

[@bluefiddleguy]

Holy Milnor batman! It's Morse theory applied to discrete data. Revelatory!

[@iramkumar78]

Nice

[@iramkumar78]

Nice